jackkuo commited on
Commit
770e473
·
verified ·
1 Parent(s): e4ff3ba

Add files using upload-large-folder tool

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. -9AyT4oBgHgl3EQfRPbk/content/2301.00064v1.pdf +3 -0
  2. -9AyT4oBgHgl3EQfRPbk/vector_store/index.pkl +3 -0
  3. -NAzT4oBgHgl3EQfFfpP/content/tmp_files/2301.01011v1.pdf.txt +1230 -0
  4. -NAzT4oBgHgl3EQfFfpP/content/tmp_files/load_file.txt +0 -0
  5. -NE1T4oBgHgl3EQf8QVm/content/tmp_files/2301.03543v1.pdf.txt +1778 -0
  6. -NE1T4oBgHgl3EQf8QVm/content/tmp_files/load_file.txt +0 -0
  7. -dA0T4oBgHgl3EQfPP9i/content/2301.02172v1.pdf +3 -0
  8. -dA0T4oBgHgl3EQfPP9i/vector_store/index.faiss +3 -0
  9. -dA0T4oBgHgl3EQfPP9i/vector_store/index.pkl +3 -0
  10. .gitattributes +57 -0
  11. 0NAzT4oBgHgl3EQfDPod/content/2301.00971v1.pdf +3 -0
  12. 0NAzT4oBgHgl3EQfDPod/vector_store/index.pkl +3 -0
  13. 1dE2T4oBgHgl3EQfNQZD/content/tmp_files/2301.03734v1.pdf.txt +403 -0
  14. 1dE2T4oBgHgl3EQfNQZD/content/tmp_files/load_file.txt +420 -0
  15. 49AzT4oBgHgl3EQffvx7/content/2301.01457v1.pdf +3 -0
  16. 49AzT4oBgHgl3EQffvx7/vector_store/index.pkl +3 -0
  17. 4NAzT4oBgHgl3EQfffyv/content/tmp_files/2301.01454v1.pdf.txt +1097 -0
  18. 4NAzT4oBgHgl3EQfffyv/content/tmp_files/load_file.txt +0 -0
  19. 4dE0T4oBgHgl3EQfeQDL/content/2301.02389v1.pdf +3 -0
  20. 4dE0T4oBgHgl3EQfeQDL/vector_store/index.pkl +3 -0
  21. 4dFAT4oBgHgl3EQfExzK/vector_store/index.faiss +3 -0
  22. 69E1T4oBgHgl3EQfBgJT/content/2301.02852v1.pdf +3 -0
  23. 6NE0T4oBgHgl3EQfvwGn/content/tmp_files/2301.02623v1.pdf.txt +626 -0
  24. 6NE0T4oBgHgl3EQfvwGn/content/tmp_files/load_file.txt +426 -0
  25. 79E5T4oBgHgl3EQfQQ7U/vector_store/index.faiss +3 -0
  26. 79E5T4oBgHgl3EQfQQ7U/vector_store/index.pkl +3 -0
  27. 79FLT4oBgHgl3EQfsi_P/content/2301.12148v1.pdf +3 -0
  28. 79FLT4oBgHgl3EQfsi_P/vector_store/index.pkl +3 -0
  29. 7dA0T4oBgHgl3EQfOf8M/content/tmp_files/2301.02160v1.pdf.txt +1003 -0
  30. 7dA0T4oBgHgl3EQfOf8M/content/tmp_files/load_file.txt +0 -0
  31. 8tE1T4oBgHgl3EQfngT9/content/tmp_files/2301.03311v1.pdf.txt +2053 -0
  32. 8tE1T4oBgHgl3EQfngT9/content/tmp_files/load_file.txt +0 -0
  33. BNE3T4oBgHgl3EQfswvl/vector_store/index.pkl +3 -0
  34. D9AzT4oBgHgl3EQfwv5m/content/tmp_files/2301.01727v1.pdf.txt +1790 -0
  35. D9AzT4oBgHgl3EQfwv5m/content/tmp_files/load_file.txt +0 -0
  36. D9E2T4oBgHgl3EQfSgcI/content/tmp_files/2301.03792v1.pdf.txt +775 -0
  37. D9E2T4oBgHgl3EQfSgcI/content/tmp_files/load_file.txt +486 -0
  38. DNFRT4oBgHgl3EQfxDge/content/tmp_files/2301.13640v1.pdf.txt +894 -0
  39. DNFRT4oBgHgl3EQfxDge/content/tmp_files/load_file.txt +0 -0
  40. EdAzT4oBgHgl3EQfG_ss/content/2301.01036v1.pdf +3 -0
  41. G9E2T4oBgHgl3EQf-wmW/vector_store/index.faiss +3 -0
  42. GdE0T4oBgHgl3EQfzgIU/content/tmp_files/2301.02672v1.pdf.txt +1253 -0
  43. GdE0T4oBgHgl3EQfzgIU/content/tmp_files/load_file.txt +0 -0
  44. J9FOT4oBgHgl3EQfyjQh/content/2301.12928v1.pdf +3 -0
  45. J9FOT4oBgHgl3EQfyjQh/vector_store/index.faiss +3 -0
  46. K9AyT4oBgHgl3EQfTveL/content/2301.00112v1.pdf +3 -0
  47. K9AyT4oBgHgl3EQfTveL/vector_store/index.faiss +3 -0
  48. K9AyT4oBgHgl3EQfTveL/vector_store/index.pkl +3 -0
  49. K9AzT4oBgHgl3EQfVvwt/vector_store/index.pkl +3 -0
  50. KE Research Center Intro/content/知识工程研究中心汇报20230710.pdf +3 -0
-9AyT4oBgHgl3EQfRPbk/content/2301.00064v1.pdf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c5c930a349dba6b03e6c2d0b0a805b730c6e6f144502cbf3ed3958162045448e
3
+ size 833344
-9AyT4oBgHgl3EQfRPbk/vector_store/index.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9d1dfc94cddb3d0ad3c1813f84822b35bf77bc230556e749cacbe95aeeec1153
3
+ size 161558
-NAzT4oBgHgl3EQfFfpP/content/tmp_files/2301.01011v1.pdf.txt ADDED
@@ -0,0 +1,1230 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ arXiv:2301.01011v1 [math.SG] 3 Jan 2023
2
+ GEOMETRIC QUANTIZATIONS ASSOCIATED TO MIXED
3
+ POLARIZATIONS ON K¨AHLER MANIFOLDS WITH T-SYMMETRY
4
+ NAICHUNG CONAN LEUNG, AND DAN WANG
5
+ Abstract. Let M be a compact K¨ahler manifold equipped with a pre-quantum line
6
+ bundle L. In [9], using T -symmetry, we constructed a polarization Pmix on M, which
7
+ generalizes real polarizations on toric manifolds. In this paper, we obtain the following
8
+ results for the quantum space Hmix associated to Pmix. First, Hmix consists of distri-
9
+ butional sections of L with supports inside µ−1(t∗
10
+ Z). This gives Hmix = �
11
+ λ∈t∗
12
+ Z Hmix,λ.
13
+ Second, the above decomposition of Hmix coincides with the weight decomposition for
14
+ the T -symmetry. Third, an isomorphism Hmix,λ ∼= H0(M �λ T, L �λ T ), for regular λ.
15
+ Namely, geometric quantization commutes with symplectic reduction.
16
+ 1. Introduction
17
+ Let (M, ω) be a symplectic manifold equipped with a pre-quantum line bundle (L, ∇),
18
+ in particular F∇ = −iω. A polarization P on M is an integrable Lagrangian subbundle of
19
+ TM ⊗ C. Geometric quantization assigns a Hilbert space HP to these data. Namely,
20
+ (1.1)
21
+ HP = Γ(M, L) ∩ Ker(∇)|P,
22
+ where Γ(M, L) is the space of smooth sections of L 1. When (M, ω, J) is K¨ahler, we have
23
+ a K¨ahler polarization PJ = T 0,1
24
+ J M, and HPJ = H0(M, L). If, in addition, M admits T-
25
+ symmetry, we constructed a polarization Pmix using T-symmetry in [9]. In this paper, we
26
+ study the quantum space Hmix associated to Pmix on M. Concretely, we have the following
27
+ assumption throughout this paper.
28
+ (∗) (M, ω, J) is a compact K¨ahler manifold of real dimension 2m equipped with an
29
+ effective Hamiltonian n-dimensional torus action ρ : T n → Diff(M, ω, J) by isome-
30
+ tries with moment map µ : M → t∗. Let (L, ∇, h) be a T n-invariant pre-quantum
31
+ line bundle on M.
32
+ Recall from [9], a (singular) polarization Pmix is constructed in this situation, which is,
33
+ Pmix = (PJ ∩ DC) ⊕ IC,
34
+ where DC = (Ker dµ) ⊗ C and IC = (Im dρ) ⊗ C. When n = m, i.e. M is a toric variety,
35
+ Pmix coincides with the singular real polarization defined by moment map and Hmix is the
36
+ space of Bohr-Sommerfeld states.
37
+ 1In fact we need to allow distributional sections.
38
+ 1
39
+
40
+ 2
41
+ LEUNG AND WANG
42
+ Recall that there is a natural way to embed the space of smooth sections into the space
43
+ of distributional sections using the Liouville measure volM = ωm
44
+ m! . That is, for any test
45
+ section τ ∈ Γc(M, L−1),
46
+ ι : Γ(M, L) → Γc(M, L−1)′, s �→ (ιs)(τ) =
47
+
48
+ M
49
+ ⟨s, τ⟩ volM .
50
+ Then the quantum space Hmix can be described as:
51
+ Hmix = Γc(M, L−1)′ ∩ Ker(∇)|Pmix,
52
+ Our first result says that Hmix consists of distributional sections with supports inside
53
+ µ−1(t∗
54
+ Z) and Hmix,λ is the λ-weight subspace of Hmix.
55
+ Theorem 1.1. (Theorem 3.2) Under the assumption (∗),
56
+ (1) given any δ ∈ Hmix, we have supp δ ⊂ �
57
+ λ∈t∗
58
+ Z µ−1(λ).
59
+ This gives the following
60
+ decomposition
61
+ Hmix =
62
+
63
+ λ∈t∗
64
+ Z
65
+ Hmix,λ,
66
+ where Hmix,λ = {δ ∈ Hmix | supp δ ⊂ µ−1(λ)};
67
+ (2) for any λ ∈ t∗
68
+ Z, Hmix,λ is a λ-weight subspace in Hmix.
69
+ Therefore the decomposition Hmix = �
70
+ λ∈t∗
71
+ Z Hmix,λ is the weight decomposition with respect
72
+ to T n-action.
73
+ When n = m, this is a result for toric variety (see [1, 5]). Inspired by the works (see
74
+ [3]) of Guillemin and Sternberg that geometric quantizations commute with symplectic
75
+ reductions, we give a geometric description of Hmix,λ. Our main result (Theorem 1.5 or
76
+ Theorem 3.12) says that when λ is an integral regular value of µ, denoted as λ ∈ t∗
77
+ Z,reg, we
78
+ have
79
+ Hmix,λ ∼= H0(Mλ, Lλ),
80
+ where (Mλ, Lλ) = (M �λ T, L �λ T) is the symplectic reduction of (M, L). Concretely,
81
+ Mλ = µ−1/T, we also denote the level set µ−1(λ) as Mλ. The restriction (Lλ, ∇) of pre-
82
+ quantum line (L, ∇) to Mλ can be descended to the quotient space Mλ denoted by (Lλ, ∇)
83
+ (see [3]).
84
+ Our second result states that, for any s ∈ H0(Mλ, Lλ), there is an associated distribu-
85
+ tional section δs ∈ Γc(Mλ, (Lλ)−1)′ such that ı(δs) lies in Hmix,λ, where ı : Γc(Mλ, (Lλ)−1)′ ֒→
86
+ Γc(M, L−1)′ is the natural inclusion.
87
+ Definition 1.2. (Definition 3.4) For any λ ∈ t∗
88
+ Z,reg and s ∈ H0(Mλ, Lλ), we define
89
+ the distributional section δs ∈ Γc(Mλ, (Lλ)−1)′ associated to s as follows: for any τ ∈
90
+ Γc(Mλ, (Lλ)−1),
91
+ δs(τ) =
92
+
93
+ Mλ ⟨π∗s, τ⟩ volλ,
94
+
95
+ GQ ASSOCIATED TO MIXED POLARIZATIONS ON K¨AHLER MANIFOLDS WITH T-SYMMETRY
96
+ 3
97
+ where volλ is the volume form on Mλ and π : Mλ → Mλ is the quotient map.
98
+ To see ı(δs) ∈ Hmix,λ, we need to investigate the interaction between the covariant
99
+ derivative on the space of smooth sections of L and the covariant derivative on the space
100
+ of distributional sections of Lλ. Our third result (Theorem 3.6) says that the following
101
+ diagram
102
+ Γ(M, L)
103
+ Γc(M, L−1)′
104
+ Γc(M, L−1)′
105
+ Γ(Mλ, Lλ)
106
+ Γc(Mλ, (Lλ)−1)′
107
+ Γc(Mλ, (Lλ)−1)′,
108
+ volM
109
+ ∇ξ
110
+ volλ
111
+ ı
112
+ ∇ξ
113
+ ı
114
+ is a commutative diagram, for any ξ ∈ Γ(M, TM ⊗ C) satisfying ξ|Mλ ∈ Γ(Mλ, TMλ ⊗ C).
115
+ In order to show the above diagram commutes, we use the coisotropic embedding theorem
116
+ due to Weinstein [13] and further studied by Guillemin in [2] to relate the volume forms
117
+ volM and volλ. Then we obtain the following theorem:
118
+ Theorem 1.3. (Theorem 3.6) For any λ ∈ t∗
119
+ Z,reg, δ ∈ Γc(Mλ, (Lλ)−1)′ and ξ ∈ Γ(M, TM ⊗
120
+ C) satisfying ξ|Mλ ∈ Γ(Mλ, TMλ ⊗ C), we have
121
+ (1.2)
122
+ ∇ξ(ı(δ)) = ı(∇ξδ),
123
+ where ı : Γc(Mλ, (Lλ)−1)′ ֒→ Γc(M, L−1)′ is the natural inclusion.
124
+ Proposition 1.4. (Proposition 3.7) For any λ ∈ t∗
125
+ Z,reg and s ∈ H0(Mλ, Lλ), we have
126
+ ı(δs) ∈ Hmix,λ,
127
+ where ı : Γc(Mλ, (Lλ)−1)′ ֒→ Γc(M, L−1)′ is the natural inclusion.
128
+ This allows us to define a map κ : H0(Mλ, Lλ) → Hmix,λ by s �→ κ(s) = ı(δs). Finally
129
+ we show that κ is an isomorphism.
130
+ Theorem 1.5. (Theorem 3.12) For any λ ∈ t∗
131
+ Z,reg,
132
+ κ : H0(Mλ, Lλ) → Hmix,λ
133
+ is an isomorphism.
134
+ We show the surjectivity of κ via the following steps. First, we show that any element
135
+ ˜δ in Hmix,λ is locally a delta function along µ−1(λ), and does not involve any derivative of
136
+ delta functions. This implies ˜δ = ı(δ), for some δ ∈ Γc(Mλ, (Lλ)−1)′.
137
+ Second, we show that T n-invariant distributional sections of Lλ can be descended to
138
+ distributional sections of Lλ. That is, for any δ ∈ Γc(Mλ, (Lλ)−1)′ satisfying ∇ξ#δ = 0,
139
+ there exists a distributional section η ∈ Γc(Mλ, L−1
140
+ λ )′ such that δ = π∗η (Lemma 3.8).
141
+ Third, we show that if ∇ζ(π∗η) = 0 for all ζ ∈ Γ(Mλ, Pmix), then η is ¯∂-closed (Theorem
142
+ 3.11). By the regularity of elliptic operator ∆ = ¯∂∗ ¯∂, we have η is smooth (i.e. η = ι(s),
143
+ for some s ∈ H0(Mλ, Lλ)). Finally, we show that ˜δ = κ(s).
144
+
145
+ 4
146
+ LEUNG AND WANG
147
+ 1.1. Acknowledgements. We are grateful to Siye Wu for insightful comments and useful
148
+ discussions. D. Wang would like to thank Qingyuan Jiang, Yutung Yau and Ki Fung Chan
149
+ for many helpful discussions. This research was substantially supported by grants from the
150
+ Research Grants Council of the Hong Kong Special Administrative Region, China (Project
151
+ No. CUHK14301619 and CUHK14301721) and a direct grant from the Chinese University
152
+ of Hong Kong.
153
+ 2. Preliminary
154
+ 2.1. The Marsden-Weistein construction. In this subsection, we review the basic con-
155
+ cepts of Hamiltonian action and symplectic reduction in order to fix the notations in our
156
+ setting (for more details, the reader can refer to [3], [10]).
157
+ 2.1.1. Hamiltonian action. Let (M, ω) be a compact symplectic manifold. For f ∈ C∞(M, R),
158
+ the Hamiltonian vector field Xf associated to f is determined by ıXf ω = −df. This gives
159
+ a Lie algebra homeomorphism
160
+ ψ : (C∞(M; R), {·, ·}) → (Vect(M, ω), [·, ·])
161
+ defined by ψ(f) = Xf, where {, } is the Poisson bracket of two functions f, g ∈ C∞(M; R)
162
+ determined by {f, g} = ω(Xf, Xg). Let T n be a torus of real dimension n and ρ : T n →
163
+ Diff(M, ω) an action of T n on M which preserves ω.
164
+ Let t be the Lie algebra of T n.
165
+ Differentiating ρ at the identity element, we have
166
+ dρ : t → Vect(M, ω), ξ �→ ξ#,
167
+ where t is the Lie algebra of T n and ξ# is called the fundamental vector field associated to
168
+ ξ. The action of T n on M is said to be Hamiltonian if dρ factors through ψ.
169
+ Let ⟨, ⟩ : t∗ × t → R be the natural pairing between t∗ and t. For each point p ∈ M, we
170
+ can associate an element µ(p) ∈ t∗ by the formula
171
+ ⟨µ(p), ξ⟩ = −µξ(p), ∀ξ ∈ t.
172
+ This gives us a moment mapping µ : M → t∗ which is a T n-equivariant map.
173
+ 2.1.2. Symplectic reduction. We denote the set of regular values of µ by t∗
174
+ reg, that is,
175
+ t∗
176
+ reg = {λ ∈ t∗| λ is a regular value of µ}.
177
+ For any λ ∈ t∗
178
+ reg, denote the level set µ−1(λ) by Mλ.
179
+ Then Mλ is a T n-invariant
180
+ coisotropic submanifold (i : Mλ ֒→ M) and the action of T n is locally free (see [10]).
181
+ For simplicity, we assume T n acts freely on Mλ. Then the projection mapping
182
+ π : Mλ → Mλ
183
+ is a principal T n-fibration. Moreover there exists a unique symplectic form ωλ on Mλ such
184
+ that π∗ωλ = i∗ω. Denote the volume form
185
+ 1
186
+ (m−n)!ωm−n
187
+ λ
188
+ on Mλ by volλ. Take a connection
189
+
190
+ GQ ASSOCIATED TO MIXED POLARIZATIONS ON K¨AHLER MANIFOLDS WITH T-SYMMETRY
191
+ 5
192
+ α ∈ Ω1(Mλ, t) on Mλ, π∗ volλ ∧αn is a volume form on Mλ denoted by volλ (here αn is a
193
+ n-form on Mλ defined by α∧···∧α
194
+ v
195
+ , where v ∈ ∧nt is a T n-invariant top form on t).
196
+ (Mλ = µ−1(λ), volλ)
197
+ M
198
+ (Mλ = µ−1(λ)/T n, volλ).
199
+ i
200
+ π
201
+ 2.2. Pre-quantum data. In this subsection, we first review the definition of T n-invariant
202
+ pre-quantum line bundles. Then we restate the result that the pre-quantum line bundle
203
+ can always be descended to the reduction space by Guillemin and Sternberg in our setting
204
+ (K¨ahler manifold equipped with T n-symmetry).
205
+ Definition 2.1. Let (M, ��, J) be a symplectic manifold, a pre-quantum line bundle (L, ∇, h)
206
+ on M is a complex line bundle L together with a Hermitian metric h and Hermitian con-
207
+ nection ∇, such that the curvature form F∇ = −iω.
208
+ The existence of a pre-quantum line bundle L on M is equivalent to [ ω
209
+ 2π] being integral
210
+ (see [8]). When M is K¨ahler, L is an ample holomorphic line bundle. There is a canonical
211
+ representation of the Lie algebra t on space of smooth sections of L given by the operators
212
+ (2.1)
213
+ ∇ξ# + iµξ, ξ ∈ t.
214
+ The pre-quantum line bundle is said to be T n-invariant if there exists a global action of
215
+ T n on L such that the induced action of t is given by (2.1). It is always possible if the
216
+ T n-action on M is Hamiltonian (see [8]).
217
+ Let tZ be the kernel of the exponential map exp : T n → t and t∗
218
+ Z ⊂ t∗ be the dual lattice
219
+ of tZ. We denote the set of integral regular values of µ by t∗
220
+ Z,reg, that is, t∗
221
+ Z,reg = t∗
222
+ reg ∩ t∗
223
+ Z.
224
+ Guillemin and Sternberg in [3] showed that there are associated pre-quantum data on
225
+ the reduction space Mλ, for λ ∈ t∗
226
+ Z,reg.
227
+ Theorem 2.2. [3, Theorem 3.2] There is a unique line bundle with connection (Lλ, ∇λ)
228
+ on Mλ such that
229
+ (2.2)
230
+ π∗Lλ = i∗L =: Lλ, and π∗∇λ = i∗∇.
231
+ Corollary 2.3. [3, Corollary 3.4] The curvature of the connection, ∇λ, is the symplectic
232
+ form ωλ.
233
+
234
+ 6
235
+ LEUNG AND WANG
236
+ Therefore we have the following commuting diagram:
237
+ (Lλ, i∗∇)
238
+ (L, ∇)
239
+ t∗
240
+ Z,reg
241
+ (Lλ, ∇λ)
242
+ (Mλ, volλ)
243
+ (M, volM)
244
+ t∗
245
+ (Mλ, volλ)
246
+ π
247
+ i
248
+ µ
249
+ By abuse of notations, we denote both i∗∇ and ∇λ by ∇. In order to pull-back distribu-
250
+ tional sections from Mλ to Mλ later, we first recall how to push-forward sections of line
251
+ bundle Lλ.
252
+ Remark 2.4. Let π : P → B be a principal T n-bundle over B, E → B a line bundle over
253
+ B, and π∗E → P the pullback line bundle. Then we can define the dual map
254
+ π∗ : Γc(B, L−1)′ → Γc(P, (π∗L)−1)′, η �→ π∗η
255
+ by (π∗η)(τ) = η(π∗τ), for any τ ∈ Γc(P, L−1)
256
+ Throughout this paper, we fix a T n-invariant n-form dθ on T n such that
257
+
258
+ T n dθ = 1.
259
+ When we deal with the pull-back of distribution sections of Lλ, we mean in the sense of
260
+ Remark 2.4 with respect to dθ.
261
+ 2.3. Complex structures on symplectic reduction spaces. In order to study the
262
+ relationship between geometric quantization associated to Pmix and symplectic reduction,
263
+ we recall the work on the existence of complex structures on symplectic reduction spaces
264
+ Mλ (see [3]).
265
+ Recall that the anti-holomorphic Lagrangian sub-bundle TM0,1
266
+ J
267
+ ⊂ TM ⊗ C is a K¨ahler
268
+ polarization denoted by PJ. We define F ⊂ TMλ ⊗ C by
269
+ (2.3)
270
+ Fp = (PJ)p ∩ (TMλ ⊗ C)p,
271
+ for any p ∈ Mλ. F can be descended to a bundle PJ,λ over the reduction space Mλ, which
272
+ is a positive-definite Lagrangian sub-bundle of TMλ ⊗ C. Under the assumption (∗), we
273
+ have (DC ∩ PJ)p = Fp, for any p ∈ Mλ.
274
+ Theorem 2.5. [3, Theorem 3.5] There is a positive-definite polarization PJ,λ canonically
275
+ associated with PJ on the reduction space Mλ.
276
+ By Definition 4.2 and Lemma 4.3 in [3], PJ,λ determined a complex structure Jλ on Mλ
277
+ such that
278
+ (2.4)
279
+ PJ,λ = TM0,1
280
+ λ ,
281
+ where TM0,1
282
+ λ
283
+ is the anti-holomorphic sub-bundle of TMλ ⊗ C with respect to Jλ.
284
+
285
+ GQ ASSOCIATED TO MIXED POLARIZATIONS ON K¨AHLER MANIFOLDS WITH T-SYMMETRY
286
+ 7
287
+ 2.4. Polarizations on K¨ahler manifolds with T-symmetry. In [9], we constructed
288
+ polarizations Pmix on K¨ahler manifolds with T-symmetry. Throughout this sections, the
289
+ existence of pre-quantum line L in (∗) is not needed.
290
+ For any point p ∈ M, consider the map ρp : T n → M defined by ρp(g) = ρ(g)(p). Let
291
+ IR ⊂ TM be the singular distribution generated by fundamental vector fields in Im dρ,
292
+ that is (IR)p = Im dρp(e). Let DR = (Ker dµ) ⊂ TM be a distribution defined by the kernel
293
+ of dµ. Note that there is a K¨ahler polarization PJ = TM0,1
294
+ J
295
+ associated to the complex
296
+ structure J on a K¨ahler manifold (M, ω, J).
297
+ Definition 2.6. [9, Definition] We define the singular distribution Pmix ⊂ TM ⊗ C by:
298
+ (2.5)
299
+ Pmix = (PJ ∩ DC) ⊕ IC,
300
+ where DC = DR ⊗ C and IC = IR ⊗ C are the complexification of DR and IR respectively.
301
+ Let Hp be the stabilizer of T n at point p ∈ M. Denote by ˇ
302
+ M the union of n-dimensional
303
+ orbits in M, that is,
304
+ ˇ
305
+ M = {p ∈ M| dim Hp = 0},
306
+ which is an open dense subset in M.
307
+ Theorem 2.7. [9, Theorem 1.1] Under the assumption (∗), Pmix is a singular polarization
308
+ and smooth on ˇ
309
+ M. Moreover, rank(Pmix ∩ ¯Pmix ∩ TM)| ˇ
310
+ M = n.
311
+ According to Definition 4.6, Pmix is a singular real polarization on M, when n = m,
312
+ namely, M is toric manifold; Pmix is a singular mixed polarization on M, when 1 ≤ n < m.
313
+ 3. Main results
314
+ We define the quantum space associated to the polarization Pmix = (PJ ∩ DC) ⊕ IC as
315
+ follows. Let (L, ∇, h) be the pre-quantum line bundle on M. We first recall the definition
316
+ of quantum space H associated to polarization P (see [14]).
317
+ Definition 3.1. The quantum space H associated to polarization P is the following sub-
318
+ space of Γc(M, L−1)′:
319
+ H = {δ ∈ Γc(M, L−1)′ | ∇ξδ = 0, ∀ ξ ∈ Γ(M, P)},
320
+ where ∇ξ is the covariant derivative operator acting on the space of distributional sections
321
+ defined by equation (3.2).
322
+ In our setting, even through the polarization Pmix is singular, we continue to use the
323
+ above definition for the quantum space. We denote it by Hmix. When n = m, M is toric
324
+ variety and Pmix is a singular real polarization. The definition of Hmix coincides with the
325
+ definition of the quantum spaces associated to singular real polarizations studied in [1].
326
+
327
+ 8
328
+ LEUNG AND WANG
329
+ Moreover, for any λ ∈ t∗, we define the subspace of those sections with supports on µ−1(λ)
330
+ as:
331
+ Hmix,λ = {δ ∈ Hmix | supp δ ⊂ µ−1(λ)}.
332
+ 3.1. Distributional sections in Hmix,λ associated to sections in H0(Mλ, Lλ). In this
333
+ subsection, we first confirm that for any distributional section δ ∈ Hmix we have (see
334
+ Theorem 3.2):
335
+ supp δ ⊂
336
+
337
+ λ∈t∗
338
+ Z
339
+ µ−1(λ).
340
+ After extending the T n-action from the space of smooth sections to the space of distri-
341
+ butional sections of L, we show that Hmix,λ is a λ-weight subspace of Hmix, for any λ ∈ t∗
342
+ Z.
343
+ This gives the weight decomposition of Hmix, i.e. Hmix = �
344
+ λ∈t∗
345
+ Z Hmix,λ.
346
+ Inspired by the work on geometric quantizations commute with symplectic reductions by
347
+ Guillemin and Sternberg in [3], we expect to establish the isomorphism between H0(Mλ, Lλ)
348
+ and Hmix,λ, where (Mλ, Lλ) is the symplectic reduction of M at a regular integral level λ.
349
+ At the end of this subsection, given any holomorphic section s ∈ H0(Mλ, Lλ), we de-
350
+ fine an associated distributional section δs ∈ Γc(Mλ, (Lλ)−1)′ (see Definition 3.4) with
351
+ respect to the volume form volλ. Then we show that distributional sections in Hmix,λ as-
352
+ sociated to sections in H0(Mλ, Lλ) (see Proposition 3.7). That is ı(δs) ∈ Hmix,λ, where
353
+ ı : Γc(Mλ, (Lλ)−1)′ ֒→ Γc(M, L−1)′ is the natural inclusion.
354
+ In order to study the quantum space Hmix, we recall how to extend covariant differenti-
355
+ ation to distributional sections of L (see [1]). First of all, there is a natural way to embed
356
+ the space of smooth sections into the space of distributional sections using the Liouville
357
+ measure volM = ωm
358
+ m! :
359
+ ι : Γ(M, L) → Γc(M, L−1)′
360
+ s �→ (ιs)(τ) =
361
+
362
+ M
363
+ ⟨s, τ⟩ volM .
364
+ Here ⟨, ⟩ : L×L−1 → C is the natural paring between L and L−1. Let ∇ be the connection
365
+ on L−1 such that d⟨s, τ⟩ = ⟨∇s, τ⟩+⟨s, ∇τ⟩. It is necessary to require that the operator ∇
366
+ acting on the distributional sections ι(s) which come from any smooth section s coincides
367
+ with the operator ∇ acting on s, i.e. the following diagram
368
+ Γ(M, L)
369
+ Γc(M, L−1)′
370
+ Γ(M, L)
371
+ Γc(M, L−1)′
372
+ ι
373
+ ∇ξ
374
+ ∇ξ
375
+ ι
376
+
377
+ GQ ASSOCIATED TO MIXED POLARIZATIONS ON K¨AHLER MANIFOLDS WITH T-SYMMETRY
378
+ 9
379
+ commutes, for any ξ ∈ Γ(M, TM ⊗ C). Let div ξ be the divergence of ξ with respect to
380
+ volM = ωm
381
+ m! , equivalently, Lξ(volM) = (div ξ) volM. It can be seen that:
382
+ 0 =
383
+
384
+ M
385
+ Lξ(⟨s, τ⟩ volM) =
386
+
387
+ M
388
+ (Lξ⟨s, τ⟩) volM +
389
+
390
+ M
391
+ ⟨s, τ⟩Lξ(volM)
392
+ =
393
+
394
+ M
395
+ ⟨∇ξs, τ⟩ volM +
396
+
397
+ M
398
+ ⟨s, ∇ξτ⟩ volM +
399
+
400
+ M
401
+ ⟨s, τ⟩(div ξ) volM .
402
+ This gives, for any smooth section s ∈ Γ(M, L) and smooth test section τ ∈ Γc(M, L−1),
403
+ (3.1)
404
+ (∇ξι(s))(τ) =
405
+
406
+ M
407
+ ⟨∇ξs, τ⟩ volM =
408
+
409
+ M
410
+ ⟨s, −((div ξ)τ + ∇τ)⟩ volM .
411
+ To determine ∇ξσ for a general distributional section σ not of the form ι(s), we built its
412
+ transpose by integrating the operator ∇ξ by parts. Namely, ∇ξ is characterized by its
413
+ transpose t∇ξ as follows: for any τ ∈ Γc(M, L−1),
414
+ (3.2)
415
+ (∇ξσ)(τ) = σ(t∇ξτ), with t∇ξτ = −(div ξτ + ∇ξτ).
416
+ Similarly we can extend the T n-action on space of smooth sections to space of distri-
417
+ butional sections such that the inclusion ι : Γ(M, L) ֒→ Γc(M, L−1)′ (with respect to the
418
+ Liouville volume form volM) is T n-equivariant. That is, for any ξ ∈ t, the following diagram
419
+ commute.
420
+ Γ(M, L)
421
+ Γc(M, L−1)′
422
+ Γ(M, L)
423
+ Γc(M, L−1)′
424
+ ξ·
425
+ volM
426
+ ξ·
427
+ volM
428
+ , i.e. ξ · (ι(s)) = ι(ξ · s).
429
+ Namely, for any δ ∈ Γc(M, L−1)′, τ ∈ Γc(M, L−1), and ξ ∈ t, ξ · δ is characterized by:
430
+ (3.3)
431
+ (ξ · δ)(τ) = δ(ξ · τ), with ξ · τ = ∇ξ#τ + iµξτ.
432
+ The T n-action on L preserve connection ∇, which implies that T n acts on Hmix. We obtain
433
+ the following results.
434
+ Theorem 3.2. Under the assumption (∗),
435
+ (1) given any δ ∈ Hmix, we have supp δ ⊂ �
436
+ λ∈t∗
437
+ Z µ−1(λ).
438
+ This gives the following
439
+ decomposition
440
+ Hmix =
441
+
442
+ λ∈t∗
443
+ Z
444
+ Hmix,λ,
445
+ where Hmix,λ = {δ ∈ Hmix | supp δ ⊂ µ−1(λ)};
446
+ (2) for any λ ∈ t∗
447
+ Z, Hmix,λ is a λ-weight subspace in Hmix.
448
+ Therefore the decomposition Hmix = �
449
+ λ∈t∗
450
+ Z Hmix,λ is the weight decomposition with respect
451
+ to T n-action.
452
+
453
+ 10
454
+ LEUNG AND WANG
455
+ Proof. (1) For a loop γb ⊂ T n specified by a vector b ∈ tZ, for any test function τ ∈
456
+ Γc(M, L−1), parallel transporting τ(p) with respect to the connection ∇ around a loop
457
+ γb · p ⊂ M results in multiplication of τ(p) by e−2iπ⟨µ(p),b⟩, where ⟨, ⟩ : t∗ × t → R is the
458
+ natural pairing between t∗ and t. The reason is as follows. Recall given T 1-equivariant
459
+ line bundle L → M with equivariant curvature FA + µ, the holonomy around any T 1-orbit
460
+ at p ∈ M is given by e2πiµ(p). Applying this to our case, for a loop γb ⊂ T n specified by
461
+ b ∈ tZ, holonomies of (L−1, ∇) around the loops in M specified by b ∈ tZ define a smooth
462
+ function:
463
+ fb : M → C, p �→ fb(p) := e−2iπ⟨µ(p),b⟩.
464
+ Therefore, ∇b#τ = 0 implies fb · τ = τ.
465
+ By transporting this to the dual space, we
466
+ have fb · δ = δ for any δ ∈ Γc(M, L−1)′ satisfying ∇b#δ = 0. For δ ∈ Hmix, we have
467
+ ∇ξ#δ = 0, ∀ξ ∈ t, in particular ∇b#δ = 0, ∀b ∈ tZ. This implies that fb is constant 1 on
468
+ supp δ, for any b ∈ tZ. Therefore we conclude that δ should be supported in the set where
469
+ µ takes integral value. That is,
470
+ supp δ ⊂
471
+
472
+ λ∈t∗
473
+ Z
474
+ µ−1(λ).
475
+ To prove (2), given any λ ∈ t∗
476
+ Z and δ ∈ Hmix,λ, we need to show for any τ ∈ Γc(M, L−1)
477
+ and ξ ∈ t,
478
+ (ξ · δ)(τ) = i⟨λ, ξ⟩δ(τ),
479
+ where ⟨, ⟩ : t∗ × t → R is the natural pairing. Note that the Liouville volume form volM is
480
+ T n-invariant, div ξ# = 0. This implies,
481
+ (3.4)
482
+ (∇ξ#δ)(τ) = −δ(∇ξ#τ) = 0, and (ξ · δ)(τ) = −δ(ξ · τ).
483
+ Recall that ξ · τ = ∇ξ#τ + iµξτ. By equation (3.4), we have
484
+ (3.5)
485
+ (ξ · δ)(τk) = −δ(ξ · τk) = −δ(∇ξ#τk + iµξτk) = −δ(iµξτk).
486
+ Suppose τ = τk ∈ Γc(M, L−1) has weight k, i.e. ξ · τk = i⟨k, ξ⟩τk, by equation (3.4), one
487
+ has
488
+ (3.6)
489
+ (ξ · δ)(τk) = −δ(ξ · τk) = −δ(i⟨k, ξ⟩τk).
490
+ Combine equations (3.4) and (3.6), we obtain
491
+ (3.7)
492
+ δ(i(µξ − ⟨k, ξ⟩)τk) = 0, ∀ξ ∈ t.
493
+ For any k ̸= λ, there exists ξ ∈ t, such that ⟨λ, ξ⟩ ̸= ⟨k, ξ⟩. For such ξ, as µξ|Mλ = ⟨λ, ξ⟩,
494
+ µξ − ⟨k, ξ⟩ is no-where vanishing on a T n-invariant open neighbourhood of Mλ. One has:
495
+ (3.8)
496
+ δ(τk) = δ(i(µξ − ⟨k, ξ⟩)
497
+ 1
498
+ i(µξ − ⟨k, ξ⟩)τk).
499
+
500
+ GQ ASSOCIATED TO MIXED POLARIZATIONS ON K¨AHLER MANIFOLDS WITH T-SYMMETRY 11
501
+ Since the moment map is T n-invariant,
502
+ 1
503
+ i(µξ−⟨k,ξ⟩)τk still has weight k. Hence, by the above
504
+ discussion (i.e. in equation 3.7, replacing τk by
505
+ 1
506
+ i(µξ−⟨k,ξ⟩)τk), one has:
507
+ (3.9)
508
+ δ(τk) = 0.
509
+ Given the weight decomposition of τ = �
510
+ k τk (i.e. ξ · τk = i⟨k, ξ⟩τk), by linearity of δ
511
+ and equation (3.9), we obtain:
512
+ (ξ · δ)(τ) = i⟨λ, ξ⟩δ(τ⟨λ,ξ⟩) = i⟨λ, ξ⟩δ(τ).
513
+ Therefore we have: ξ · δ = i⟨λ, ξ⟩δ.
514
+
515
+ The next corollary says that any element in Hmix,λ is locally a delta function along
516
+ µ−1(λ), and does not involve any derivative of delta functions.
517
+ Corollary 3.3. For any λ ∈ t∗
518
+ Z, δ ∈ Hmix,λ, and any test section τ ∈ Γc(M, L−1) satisfying
519
+ τ|Mλ = 0, we have
520
+ (3.10)
521
+ δ(τ) = 0.
522
+ Proof. For any τ ∈ Γc(M, L−1) satisfying τ|Mλ = 0, let τ = �
523
+ k τk be its weight decompo-
524
+ sition, where τk =
525
+
526
+ eit∈T n(eit · τ)e−iktdt. By Theorem 3.2, δ has weight λ with respect to
527
+ T n-action. This implies, for k ̸= λ,
528
+ δ(τk) = 0.
529
+ Note that
530
+ τk(p) =
531
+
532
+ eit∈T n(e−it · τ)(p)eiktdt =
533
+
534
+ eit∈T n τ(e−it · p)eiktdt.
535
+ For any p ∈ Mλ and t ∈ t, eit · p ∈ Mλ since Mλ is T n-invariant. Therefore τk|Mλ = 0,
536
+ as τ|Mλ = 0. In particular, τλ|Mλ = 0. So there exists weight λ test section τ ′ such that
537
+ τλ = i(µξ − ⟨λ, ξ⟩)τ ′
538
+ λ for some ξ ∈ t. Hence, for k = λ, by equation (3.7), we obtain
539
+ (3.11)
540
+ δ(τλ) = δ(i(µξ − ⟨λ, ξ⟩)τ ′
541
+ λ) = 0.
542
+ Therefore, by the linearity of δ, we have δ(τ) = 0.
543
+
544
+ In order to establish the isomorphism between H0(Mλ, Lλ) and Hmix,λ, where (Mλ, Lλ) =
545
+ (M, L)//λT is the symplectic reduction of M at a regular integral level λ. We first define
546
+ the distributional section δs ∈ Γc(Mλ, (Lλ)−1)′ on Mλ ⊂ M associated to s ∈ H0(Mλ, Lλ)
547
+ as follows.
548
+ Definition 3.4. For any λ ∈ t∗
549
+ Z,reg and s ∈ H0(Mλ, Lλ), we define the distributional section
550
+ δs ∈ Γc(Mλ, (Lλ)−1)′ associated to s as follows: for any τ ∈ Γc(Mλ, (Lλ)−1),
551
+ (3.12)
552
+ δs(τ) =
553
+
554
+ Mλ ⟨π∗s, τ⟩ volλ .
555
+
556
+ 12
557
+ LEUNG AND WANG
558
+ In fact, δs = ι(π∗s), under the embedding ι : Γ(Mλ, Lλ)
559
+ Γc(Mλ, (Lλ)−1)′
560
+ volλ
561
+ defined
562
+ by σ �→ (ισ)(τ) =
563
+
564
+ M⟨σ, τ⟩ volλ with respect to volλ. Note that ı(δs) ∈ Γc(M, L−1)′, where
565
+ ı : Γc(Mλ, (Lλ)−1)′ ֒→ Γc(M, L−1)′ is a natural inclusion defined by:
566
+ (3.13)
567
+ (ı(δ))(τ) = δ(τ|Mλ), ∀δ ∈ Γc(Mλ, (Lλ)−1)′.
568
+ In order to show ı(δs) ∈ Hmix,λ, we first need to extend covariant derivative ∇ξ on the
569
+ space of smooth sections to the space of distributional sections of Lλ with respect to volλ
570
+ as before. That is, for any σ ∈ Γc(Mλ, (Lλ)−1)′, and ξ ∈ Γ(Mλ, TMλ ⊗ C),
571
+ (3.14)
572
+ (∇ξσ)(τ) = σ(t∇ξτ), with t∇ξτ = −(div ξτ + ∇ξτ),
573
+ where div2 ξ = Lξ volλ
574
+ volλ .
575
+ In particular, ı(δs) ∈ Γc(M, L−1)′.
576
+ we also need to study the relationship between
577
+ the covariant derivative on the space of distributional sections of Lλ and the space of
578
+ distributional sections of L. We expect the following diagram
579
+ Γ(M, L)
580
+ Γc(M, L−1)′
581
+ Γc(M, L−1)′
582
+ Γ(Mλ, Lλ)
583
+ Γc(Mλ, (Lλ)−1)′
584
+ Γc(Mλ, (Lλ)−1)′,
585
+ volM
586
+ ∇ξ
587
+ volλ
588
+ ı
589
+ ∇ξ
590
+ ı
591
+ commute, for any ξ ∈ Γ(M, TM ⊗ C) satisfying ξ|Mλ ∈ Γ(Mλ, TMλ ⊗ C). In order to
592
+ show that the above diagram commute, we use the coisotropic embedding theorem due to
593
+ Weinstein [13] and further studied by Guillemin in [2] to (relate volM and volλ) show the
594
+ following lemma.
595
+ 3.1.1. Restriction commutes with taking divergence. For any ξ ∈ Γ(M, TM ⊗C), we denote
596
+ the restriction of the divergence of ξ (with respect to volM) to Mλ by div1
597
+ ξ and denote the
598
+ divergence of ξ|Mλ (with respect to volλ) by div2
599
+ ξ i.e.
600
+ div1
601
+ ξ = d(iξ volM)
602
+ volM
603
+ |Mλ, and div2 ξ =
604
+ d(iξ|Mλ volλ)
605
+ .
606
+ volλ.
607
+ Lemma 3.5. Under the assumption (∗), for any ξ ∈ Γ(M, Pmix) and λ ∈ t∗
608
+ reg, we have
609
+ div1 ξ = div2 ξ,
610
+ as functions on Mλ.
611
+ Proof. Without loss of generality, we assume λ = 0 and n = 1. In order to show that
612
+ div1 ξ = div2 ξ, we shall first relate the volume form volM of M and the volume form volMλ
613
+ of Mλ. Taking a principal T 1-connection α ∈ Ω1(M0, t) on M0, choose a basis ξ1 of t and
614
+ denote the corresponding dual basis of t∗ by ξ∗
615
+ 1 with coordinate function t. In terms of ξ1,
616
+
617
+ GQ ASSOCIATED TO MIXED POLARIZATIONS ON K¨AHLER MANIFOLDS WITH T-SYMMETRY 13
618
+ we write α = ξ1 ⊗ α1, where α1 is a scalar valued form. By abuse of notations, we denote
619
+ α1 by α. Consider M0 as a submanifold of M0 × t∗ via the embedding
620
+ i0 : M0 → M0 × t∗, i0(p) = (p, 0).
621
+ The two-form
622
+ ˜ω = π∗ω0 + d(tα)
623
+ is symplectic on a neighbourhood U of M0 in M0 × t∗ and satisfies i∗
624
+ 0˜ω = π∗ω0.
625
+ Note that (dt ∧ α)2 = 0 and (i∗ω)m = 0.
626
+ Then we restrict our attention to show
627
+ Lξt = 0, ∀ ξ ∈ Γ(M, Pmix). We extend the T 1-action on M0 to M0 ×t∗ in a trivial manner.
628
+ Then ˜ω is T 1-invariant and the action of T 1 on M0 × t∗ is Hamiltonian with moment map
629
+ µ0 : M0 × t∗ → t∗, (p, t) �→ t.
630
+ According to Theorem 2.2 of [2], in a neighborhood U of M0, the Hamiltonian T 1-spaces
631
+ (M, ω) and (M0 × t∗, ˜ω) are isomorphic (see Appendix).
632
+ This gives Lζt = 0, for any
633
+ ζ ∈ Γ(U, DC) and
634
+ volM = 1
635
+ m!(i∗ω + tdα)m−1 ∧ α ∧ dt,
636
+ in a neighbourhood U of M0. As Pmix ⊂ DC, it is obvious that Γ(U, Pmix) ⊂ Γ(U, DC).
637
+ This gives us that, for any ξ ∈ Γ(U, Pmix),
638
+ (3.15)
639
+ Lξt = 0.
640
+ It follows that:
641
+ Lξ(i∗ω + tdα)m−1 = (m − 1)(Lξ(i∗ω) + tLξdα) ∧ (i∗ω + tdα)m−2, ∀ξ ∈ Γ(U, Pmix).
642
+ Therefore, we obtain:
643
+ 1
644
+ volM
645
+ d(iξ volM) =
646
+ 1
647
+ volM
648
+ Lξ volM =
649
+ 1
650
+ volM
651
+ 1
652
+ m!Lξ((i∗ω + tdα)m−1 ∧ α ∧ dt)
653
+ =
654
+ 1
655
+ volM
656
+ 1
657
+ m!(Lξ(i∗ω + tdα)m−1 ∧ α ∧ dt + (i∗ω + tdα)m−1 ∧ Lξα ∧ dt)
658
+ = (m − 1)(Lξi∗ω + tLξdα) ∧ (i∗ω + tdα)m−2 ∧ α + (i∗ω + tdα)m−1 ∧ Lξα)
659
+ (i∗ω + tdα)m−1 ∧ α
660
+ .
661
+ Recall that vol0 =
662
+ 1
663
+ (m−1)!(i∗ω)m−1 ∧ α. By abuse of notation, iξ vol0 means iξ|M0 vol0. Then
664
+ by a straight computation,
665
+ div2 ξ =
666
+ 1
667
+ vol0d(iξ vol0) =
668
+ 1
669
+ vol0Lξ vol0
670
+ =
671
+ 1
672
+ vol0
673
+ 1
674
+ (m − 1)!Lξ((i∗ω)m−1 ∧ α)
675
+ = (m − 1)(Lξi∗ω) ∧ (i∗ω)m−2 ∧ α + (i∗ω)m−1 ∧ Lξα)
676
+ (i∗ω)m−1 ∧ α
677
+ .
678
+
679
+ 14
680
+ LEUNG AND WANG
681
+ Therefore, for ξ ∈ Γ(M, Pmix), we have:
682
+ div1 ξ =
683
+
684
+ 1
685
+ volM
686
+ d (iξ volM)
687
+
688
+ |M0 =
689
+
690
+ 1
691
+ volM
692
+ d (iξ volM)
693
+
694
+ |t=0 = div2 ξ.
695
+
696
+ Then we obtain the following theorem:
697
+ Theorem 3.6. For any λ ∈ t∗
698
+ Z,reg, δ ∈ Γc(Mλ, (Lλ)−1)′ and ξ ∈ Γ(M, TM ⊗ C) satisfying
699
+ ξ|Mλ ∈ Γ(Mλ, TMλ ⊗ C), we have
700
+ (3.16)
701
+ ∇ξ(ı(δ)) = ı(∇ξδ),
702
+ where ı : Γc(Mλ, (Lλ)−1)′ ֒→ Γc(M, L−1)′ is the natural inclusion.
703
+ Proof. For any test section τ ∈ Γc(M, L−1), according to equation (3.2), one has
704
+ (3.17)
705
+ (∇ξ(ı(δ)))(τ) = ı(δ)(t∇ξτ) = δ(i∗(t∇ξτ)),
706
+ and
707
+ (3.18)
708
+ (ı(∇ξδ))(τ) = (∇ξδ)(i∗τ) = δ(t∇ξ(i∗τ)),
709
+ where i : Mλ ֒→ M is the inclusion.
710
+ To show equation (3.16), by (3.17) and (3.18), it is enough to prove that
711
+ (3.19)
712
+ i∗(t∇ξτ) =t ∇ξ(i∗τ).
713
+ According to equation (3.2), we have:
714
+ (3.20)
715
+ i∗(t∇ξτ) = −i∗ (div ξτ + ∇ξτ) ,
716
+ where div ξ = iξ volM
717
+ vol M . Similarly, applying the equation (3.1) to L|Mλ, we have:
718
+ (3.21)
719
+ t∇ξ(i∗τ) = −((div2 ξ)(i∗τ) + ∇ξ(i∗τ))
720
+ where div2 ξ = iξ volλ
721
+ volλ , i∗τ = τ|Mλ = τ by abuse of notation. Denote i∗(div ξ) by div1 ξ i.e.
722
+ div1 ξ = iξ volM
723
+ volM |Mλ. As i∗(∇ξτ) = ∇ξ(i∗τ) by abuse of notation ξ|Mλ = ξ, we have
724
+ (3.22)
725
+ − i∗ (div ξτ + ∇ξτ) = −(div1 ξ(i∗τ) + ∇ξ(i∗τ)).
726
+ By Lemma 3.5,
727
+ (3.23)
728
+ div1 ξ = div2 ξ.
729
+ Combining equations (3.20), (3.21), (3.22) with (3.23), one has
730
+ δ(i∗(t∇ξτ)) = δ(t∇ξ(i∗τ)).
731
+ Therefore we have: ∇ξ(ı(δ)) = ı(∇ξδ).
732
+
733
+
734
+ GQ ASSOCIATED TO MIXED POLARIZATIONS ON K¨AHLER MANIFOLDS WITH T-SYMMETRY 15
735
+ Proposition 3.7. For any regular λ ∈ t∗
736
+ Z,reg and s ∈ H0(Mλ, Lλ), we have:
737
+ ı(δs) ∈ Hmix,λ,
738
+ where ı : Γc(Mλ, (Lλ)−1)′ ֒→ Γc(M, L−1)′ is the natural inclusion.
739
+ Proof. By the definition of δs, we have ı(δs) ∈ Γc(M, L−1)′ and supp ı(δs) ⊂ µ−1(λ). It
740
+ remains to show that, for any ξ ∈ Γ(M, Pmix),
741
+ (3.24)
742
+ ∇ξ(ı(δs)) = 0.
743
+ Note that for any ξ ∈ (M, Pmix), ξ|Mλ ∈ Γ(Mλ, TMλ ⊗ C). To check equation (3.24), by
744
+ Theorem 3.6, it is equivalent to prove, for any ξ ∈ Γ(Mλ, Pmix)
745
+ (3.25)
746
+ ∇ξδs = 0.
747
+ Take any test section τ ∈ Γc(Mλ, (Lλ)−1), according to equation (3.2), we have:
748
+ (3.26)
749
+ (∇ξδs)(τ) = δs �t∇ξτ
750
+
751
+ = −δs �
752
+ (div2 ξ)τ + ∇ξτ
753
+
754
+ ,
755
+ where div2 ξ = iξ volλ
756
+ volλ . By definition of δs, it can be seen that:
757
+ (3.27)
758
+ − δs �
759
+ (div2 ξ)τ + ∇ξτ
760
+
761
+ = −
762
+
763
+
764
+
765
+ π∗s, (div2 ξ)τ + ∇ξτ
766
+
767
+ volλ .
768
+ Similarly, applying the equation (3.1) to L|Mλ, we have:
769
+ (3.28)
770
+
771
+ Mλ ⟨∇ξ(π∗s), τ⟩ volλ = −
772
+
773
+
774
+
775
+ π∗s, (div2 ξ)τ + ∇ξτ
776
+
777
+ volλ .
778
+ Combining equations (3.26), (3.27), with (3.28), we have
779
+ (3.29)
780
+ (∇ξδs)(τ) =
781
+
782
+ Mλ ⟨∇ξ(π∗s), τ⟩ volλ .
783
+ Since s ∈ H0(Mλ, Lλ) is a holomorphic section, we have ∇s ∈ Γ(Mλ, T ∗M1,0
784
+ λ
785
+ ⊗ Lλ). For
786
+ any ξ ∈ Γ(M, Pmix) and q ∈ Mλ, as (Pmix)q ⊂ (DC)q = TqMλ ⊗ C, we have π∗(ξq) ∈
787
+ Tπ(q)M0,1
788
+ λ .
789
+ This implies ∇ξ(π∗s) = 0 on Mλ, for any ξ ∈ Γ(M, Pmix).
790
+ Then, for all
791
+ τ ∈ Γc(M, L−1), by equation (3.29),
792
+ (∇ξδs)(τ) = 0.
793
+ Therefore we have: ı(δs) ∈ Hmix,λ.
794
+
795
+
796
+ 16
797
+ LEUNG AND WANG
798
+ 3.2. λ-weight quantum subspace Hmix,λ. In this subsection, we are going to show that
799
+ (see Theorem 3.12) for any regular λ ∈ t∗
800
+ Z,reg,
801
+ κ : H0(Mλ, Lλ) → Hmix,λ
802
+ given by s �→ κ(s) = ı(δs) is an isomorphism.
803
+ Firstly, we show that T n-invariant distributional sections of Lλ can be descended to
804
+ distributional sections of Lλ. That is, for any δ ∈ Γc(Mλ, (Lλ)−1)′ satisfying ∇ξ#δ = 0,
805
+ there exists a distributional section η ∈ Γc(Mλ, L−1
806
+ λ )′ such that δ = π∗η (Lemma 3.8).
807
+ Secondly, we show that if ∇ζ(π∗η) = 0 for all ζ ∈ Γ(Mλ, Pmix), then η is ¯∂-closed (Theorem
808
+ 3.11). Finally, we show that H0(Mλ, Lλ) ∼= Hmix,λ (Theorem 3.12).
809
+ 3.2.1. Descending distributional sections from Mλ to Mλ. For any λ ∈ t∗
810
+ reg, let π : Mλ →
811
+ Mλ be the principal T n-bundle. Recall that (Lλ, ∇) can be descended to Mλ which we
812
+ denote as (Lλ, ∇). According to Remark 2.4, we have π∗ : Γ(Mλ, (Lλ)−1) → Γ(Mλ, L−1
813
+ λ )
814
+ and dually we have π∗ : Γc(Mλ, L−1
815
+ λ )′ → Γc(Mλ, (Lλ)−1)′.
816
+ In fact, our above claim δ = π∗η holds true for any T n-principal bundle P → B. Let
817
+ π : P → B be a principal T n-bundle with a fiberwise T n-invariant volume form dθ such
818
+ that
819
+
820
+ P dθ = 1 ∈ C∞(B). Let (E, ∇) be a line bundle over B. We can push-forward
821
+ sections of π∗E to sections of E with respect to dθ. Furthermore we have:
822
+ Lemma 3.8. Taking δ ∈ Γc(P, (π∗E)−1)′, if ∇ξ#δ = 0 for any ξ ∈ t, then there exists a
823
+ distributional section η ∈ Γc(B, E−1)′ such that
824
+ δ = π∗η.
825
+ Proof. By partition of unity, it is enough to show that on any open subset U of B, for
826
+ δ ∈ Γc(π−1(U), (π∗E)−1)′, if ∇ξ#δ = 0 for any ξ ∈ t, there exists a distributional section
827
+ η ∈ Γc(U, E−1)′ such that δ = π∗η. That is, for any τ ∈ Γc(π−1(U), (π∗E)−1),
828
+ δ(τ) = η(π∗τ).
829
+ Fixing a local frame σ0 ∈ Γ(U, E) of E on an open subset U ⊂ B, let σ := π∗σ0 and
830
+ σ−1 be the corresponding local frames of π∗E and (π∗E)−1 respectively on π−1(U). With
831
+ respect to local frames σ and σ−1, the distributional section δ ∈ Γc(π−1(U), (π∗E)−1)′
832
+ corresponds to the distributional function fδ ∈ Γc(π−1(U), C)′, where fδ is determined by:
833
+ (3.30)
834
+ fδ(gτ) = δ(gτσ−1),
835
+ for any text function gτ ∈ Γc(π−1(U), C). We restrict our attention to show that ∇ξ#δ = 0
836
+ if and only if ξ#fδ = 0. Applying the equation (3.2) to line bundle π∗L and trivial bundle
837
+ over π−1(U) respectively, it can be seen that:
838
+ (3.31)
839
+
840
+ ∇ξ#δ
841
+
842
+ (τ) = −δ
843
+
844
+ (div ξ#)τ + ∇ξ#τ
845
+
846
+ ,
847
+
848
+ GQ ASSOCIATED TO MIXED POLARIZATIONS ON K¨AHLER MANIFOLDS WITH T-SYMMETRY 17
849
+ and
850
+ (3.32)
851
+ (ξ#fδ) (gτ) = fδ
852
+
853
+
854
+
855
+ div ξ#gτ + ξ#gτ
856
+ ��
857
+ .
858
+ Since ∇ξ#σ = ∇ξ#(π∗σ0) = 0, one has ∇ξ#σ−1 = 0 and
859
+ (3.33)
860
+ ∇ξ#τ = ∇ξ#
861
+
862
+ gτσ−1�
863
+ =
864
+
865
+ ξ#gτ
866
+
867
+ σ−1.
868
+ Combining equations ( 3.30), ( 3.31), (3.32), with (3.33), we obtain that
869
+ (3.34)
870
+
871
+ ∇ξ#δ
872
+
873
+ (τ) = (ξ#fδ) (gτ) ,
874
+ for any τ ∈ Γc (π−1(U), (π∗E)−1).
875
+ It turns out that ∇ξ#δ = 0 iff ξ#fδ = 0 for any
876
+ ξ ∈ t. Then by Lemma 3.9, there exists a distributional function fη ∈ Γc(U, C)′ such that
877
+ fδ = π∗(fη). Define η ∈ Γc(U, (π∗E)−1)′ to be distributional section associated to fη with
878
+ respect to the nowhere vanishing section σ−1
879
+ 0 , that is η(hτσ−1
880
+ 0 ) = fη(hτ). For any test
881
+ section τ ∈ Γc(π−1(U), π∗E), it can be check that:
882
+ δ(τ) = (π∗η)τ.
883
+ Therefore we have δ = π∗η.
884
+
885
+ Lemma 3.9. Let π : P → B be the principal T n-bundle and let U be any open subset of
886
+ B. Let δ ∈ Γc(π−1(U), C)′ be a distributional function. If ξ#δ = 0 for any ξ ∈ t, there
887
+ exists a distributional function η ∈ Γc(U, C)′, such that δ = π∗η. Namely,
888
+ δ(g) = η(π∗g), ∀ g ∈ Γc(π−1(U), C).
889
+ Proof. For any δ ∈ Γc(π−1(U), C)′, there exist δǫ ∈ Γ(π−1(U), C) (see [6, 11]) such that
890
+ limǫ→0 δǫ = δ and
891
+ (3.35)
892
+ (ξ#δǫ)(g) = (ξ#δ)(gǫ),
893
+ for any g ∈ Γc(π−1(U), C). As ξ#δ = 0, we obtain ξ#δǫ = 0. Since δǫ is smooth, there
894
+ exists a smooth function ηǫ ∈ Γ(U, C), such that δǫ = π∗ηǫ ∈ Γ(π−1(U), C). It can be check
895
+ that
896
+ lim
897
+ ǫ→0 ηǫ(h) = lim
898
+ ǫ→0 δǫ(π∗h),
899
+ for any h ∈ Γc(U, C). Hence we have limǫ→0 ηǫ exists and denoted by η. It follows
900
+ δ = π∗η.
901
+
902
+
903
+ 18
904
+ LEUNG AND WANG
905
+ 3.2.2. Pulling back commutes with taking divergence. Fix λ ∈ t∗
906
+ Z,reg, let α ∈ Ω1(Mλ, t) be
907
+ a connection on the principal T n-bundle π : Mλ → Mλ. For any ζ ∈ Γ(Mλ, TMλ), the
908
+ horizontal lifting of ζ with respect to α is denoted by ˜ζ. Denote the divergence of ζ on Mλ
909
+ with respect to volλ by div ζ (i.e. div ζ = Lζ volλ
910
+ volλ ) and denote the divergence of ˜ζ on Mλ
911
+ with respect to volλ by div ˜ζ (i.e. div ˜ζ =
912
+ L˜ζ volλ
913
+ volλ ).
914
+ Lemma 3.10. Let div ζ and div ˜ζ be defined as above. Then we have
915
+ π∗(div ζ) = div ˜ζ,
916
+ as smooth functions on Mλ.
917
+ Proof. As T n is abelian, the horizontal lifting ˜ζ of ζ with respect to the connection one
918
+ form α is T n-invariant. That is
919
+ (3.36)
920
+ Lξ# ˜ζ = 0,
921
+ for all ξ ∈ t, where ξ# is the fundamental vector field associate to ξ. According to the
922
+ property of principal T n-connection and equation (3.37), we have
923
+ (3.37)
924
+ (L˜ζα)(ξ#) = L˜ζ(α(ξ#)) − α(L˜ζξ#) = 0.
925
+ Recall that volλ = π∗ volλ ∧αn. By equation (3.37), one has
926
+ (3.38)
927
+ L˜ζ volλ = (L˜ζ(π∗ volλ)) ∧ αn.
928
+ On the other hand, by Cartan formula and volλ being the volume form on B, we have:
929
+ (3.39)
930
+ L˜ζ(π∗ volλ) = d(i˜ζ(π∗ volλ)) = π∗(Lζ volλ),
931
+ Recall that
932
+ (3.40)
933
+ Lζ volλ = (div ζ) volλ, L˜ζ volλ = (div ˜ζ) volλ .
934
+ Combining equation (3.38), (3.39), with (3.40), one has
935
+ (div ˜ζ) volλ = L˜ζ volλ = π∗(Lζ volλ) ∧ αn
936
+ = π∗(div ζ)π∗ volλ ∧αn
937
+ = π∗(div ζ) volλ .
938
+ Therefore we obtain: π∗(div ζ) = div ˜ζ.
939
+
940
+ Theorem 3.11. For any λ ∈ t∗
941
+ Z,reg and distributional function η ∈ Γc(Mλ, C)′, if ∇ξ(π∗η) =
942
+ 0, for any ξ ∈ Γ(Mλ, Pmix), then we have ∇ζη = 0, for all ζ ∈ Γ(Mλ, TM0,1
943
+ λ ).
944
+
945
+ GQ ASSOCIATED TO MIXED POLARIZATIONS ON K¨AHLER MANIFOLDS WITH T-SYMMETRY 19
946
+ Proof. To prove this statement, fixing the connection one form α ∈ Ω1(Mλ, t) on principal
947
+ T n-bundle π : Mλ → Mλ, we denote the horizontal lifting of ζ with respect to the connec-
948
+ tion α by ˜ζ, for any ζ ∈ Γ(Mλ, TM0,1
949
+ λ ). In order to show ∇ζη = 0, it is enough to show ,
950
+ for any test function φ ∈ Dc(Mλ),
951
+ (∇ζη) (φ) =
952
+
953
+ ∇˜ζ (π∗η)
954
+
955
+ (π∗φ) .
956
+ Let volλ and volλ be volume forms of Mλ and Mλ respectively as defined before.
957
+ In
958
+ particular, volλ = π∗ volλ ∧αn with respect to the principal T n-connection α ∈ Ω1(Mλ, t).
959
+ Applying equation (3.2) to the trivial bundle of Mλ and Mλ respectively, we obtain:
960
+ (3.41)
961
+ (∇ζη) (φ) = η (− (div ζ) φ − ∇ζφ) ,
962
+ and
963
+ (3.42)
964
+
965
+ ∇˜ζ (π∗η)
966
+
967
+ (π∗φ) = (π∗η)
968
+
969
+
970
+
971
+ div ˜ζ
972
+
973
+ π∗φ − ∇˜ζ (π∗φ)
974
+
975
+ ,
976
+ where div ζ (div ˜ζ resp.) is the divergence of ζ (˜ζ resp.) with respect to volλ (volλ resp.).
977
+ According to the Remark 2.4, we have:
978
+ (3.43)
979
+ (π∗η) (π∗ (− (div ζ) φ − ∇ζφ)) = η (− (div ζ) φ − ∇ζφ) .
980
+ By Lemma 3.10,
981
+ (3.44)
982
+ π∗(div ζ) = div ˜ζ.
983
+ Note that π∗(ζ(φ)) = π∗ζ(π∗φ). By equation (3.44), one has
984
+ (3.45)
985
+ π∗ (− (div ζ) φ − ∇ζφ) = −
986
+
987
+ div ˜ζ
988
+
989
+ π∗φ − ∇˜ζ (π∗φ) .
990
+ Furthermore:
991
+ (3.46)
992
+ (π∗η) (π∗ (− (div ζ) φ − ∇ζφ)) = (π∗η)
993
+
994
+
995
+
996
+ div ˜ζ
997
+
998
+ π∗φ − ∇˜ζ (π∗φ)
999
+
1000
+ .
1001
+ Combining (3.41), (3.42), (3.43), with (3.46), we are able to conclude:
1002
+ (3.47)
1003
+ (∇ζη) (φ) =
1004
+
1005
+ ∇˜ζ (π∗η)
1006
+
1007
+ (π∗φ) .
1008
+ Then we restrict our attention to show ˜ζ ∈ Γ(Mλ, Pmix). As T n acts freely on Mλ, Mλ×t ∼=
1009
+ IR|Mλ. Note that π∗(˜ζ) = ζ ∈ Γ(Mλ, TM0,1
1010
+ λ ) and α(˜ζ) = 0. Since TpMλ ⊗ C ⊂ (DC)p and
1011
+ (Pmix)p = (DC ∩ TM0,1)p ⊕ (IC)p, for any p ∈ M0, we have ˜ζ ∈ Γ(Mλ, Pmix). According to
1012
+ what we assume, we have ∇˜ζ (π∗η) = 0. Therefore, by equation (3.47), we have
1013
+ (∇ζη) (φ) =
1014
+
1015
+ ∇˜ζ (π∗η)
1016
+
1017
+ (π∗φ) = 0, ∀φ ∈ Dc(Mλ), ζ ∈ Γ(Mλ, TM0,1
1018
+ λ ).
1019
+
1020
+
1021
+ 20
1022
+ LEUNG AND WANG
1023
+ 3.2.3. Building the isomorphism H0(Mλ, Lλ) ∼= Hmix,λ. Recall given any s ∈ H0(Mλ, Lλ),
1024
+ by Proposition 3.7, the associated distributional section ı(δs) belongs to Hmix,λ. Therefore
1025
+ we can define a homomorphism
1026
+ κ : H0(Mλ, Lλ) → Hmix,λ
1027
+ given by s �→ κ(s) = ı(δs), where ı : Γc(Mλ, (Lλ)−1)′ ֒→ Γc(M, L−1)′ is the natural
1028
+ inclusion. It can be checked that κ is injective.
1029
+ Theorem 3.12. For any λ ∈ t∗
1030
+ Z,reg, κ : H0(Mλ, Lλ) → Hmix,λ is an isomorphism.
1031
+ Proof. Given any ˜δ ∈ Hmix,λ, we need to construct s ∈ H0(Mλ, Lλ) such that ˜δ = κ(s).
1032
+ Firstly we show that, there exists δ ∈ Γc(Mλ, (Lλ)−1)′ such that ˜δ = ı(δ) as follows: we
1033
+ define the distributional section δ ∈ Γc(Mλ, (Lλ)−1)′ by:
1034
+ δ(τ) = ˜δ(˜τ),
1035
+ for any τ ∈ Γc(Mλ, (Lλ)−1), where ˜τ ∈ Γc(M, L−1) is any test section satisfying ˜τ|Mλ = τ
1036
+ By Corollary 3.3, δ is well defined. Moreover, one has
1037
+ (3.48)
1038
+ ˜δ = ı(δ).
1039
+ That is, for any test section ˜τ ′ ∈ Γc(M, L−1), (ı(δ))(˜τ ′) = δ(˜τ ′|Mλ) = ˜δ(˜τ ′). Secondly we
1040
+ show that there exists η ∈ Γc(Mλ, L−1
1041
+ λ )′ such that δ = π∗η, where π : Mλ → Mλ is the
1042
+ projection. For any ˜δ ∈ Hmix,λ, since ξ# ∈ Γ(M, Pmix), we have ∇ξ#˜δ = 0, for any ξ ∈ t.
1043
+ By Theorem 3.6, one has
1044
+ (3.49)
1045
+ 0 = ∇ξ#˜δ = ∇ξ#(ı(δ)) = ı(∇ξ#δ), ∀ξ ∈ t.
1046
+ By the injectivity of ı, we obtain, for any ξ ∈ t,
1047
+ (3.50)
1048
+ ∇ξ#δ = 0.
1049
+ According to Lemma 3.8, there exists a distributional section η ∈ Γc(Mλ, L−1
1050
+ λ )′, such that
1051
+ (3.51)
1052
+ δ = π∗η.
1053
+ Next we show that there exists a holomorphic section s ∈ H0(Mλ, Lλ) such that η = ι(s)
1054
+ under the inclusion map ι : Γ(Mλ, Lλ) → Γc(Mλ, L−1
1055
+ λ )′ with respect to volλ.
1056
+ By the
1057
+ definition of Pmix, for any ξ ∈ Γ(M, Pmix), we have ξ|Mλ ∈ Γ(Mλ, Pmix) ⊂ Γ(Mλ, TMλ⊗C).
1058
+ By abuse of notation, we denote ξ|Mλ by ξ. According to Theorem 3.11 and equation (3.51),
1059
+ we have
1060
+ (3.52)
1061
+ ∇ξ˜δ = ∇ξ(ı(δ)) = ı(∇ξδ) = ı(∇ξ(π∗η)).
1062
+ Since ˜δ ∈ Hmix,λ, ∇ξ˜δ = 0, for ξ ∈ Γ(M, Pmix). By the injectivity of ı and equation (3.52),
1063
+ we obtain:
1064
+ (3.53)
1065
+ ∇ξ(π∗η) = 0, ∀ξ ∈ Γ(Mλ, Pmix).
1066
+
1067
+ GQ ASSOCIATED TO MIXED POLARIZATIONS ON K¨AHLER MANIFOLDS WITH T-SYMMETRY 21
1068
+ Then by Theorem 3.11, we have ∇ζη = 0, for any ζ ∈ Γ(Mλ, TM0,1
1069
+ λ ).
1070
+ This implies
1071
+ ∇0,1η = 0. By the regularity of elliptic operator ∆ = ¯∂∗ ¯∂, η is smooth. Therefore there
1072
+ exists a holomorphic section s ∈ H0(Mλ, Lλ) such that η = ι(s) under the inclusion map
1073
+ ι : Γ(Mλ, Lλ) → Γc(Mλ, L−1
1074
+ λ )′ with respect to volλ. It is remain to show ˜δ = κ(s).
1075
+ According to the above discussion, we have ˜δ = ı(π∗(ι(s))). Recall that κ(s) = ı(δs),
1076
+ where δs with respect to volume form volλ is defined by
1077
+ (3.54)
1078
+ δs(τ) =
1079
+
1080
+ Mλ⟨π∗s, τ⟩ volλ,
1081
+ for any test section τ ∈ Γc(Mλ, Lλ)′. By the injectivity of ı, to show ˜δ = κ(s), it is enough
1082
+ to show:
1083
+ (3.55)
1084
+ π∗(ι(s)) = δs.
1085
+ By remark 2.4, we have
1086
+ (3.56)
1087
+
1088
+ Mλ⟨π∗s, τ⟩ volλ = (π∗s)(τ) = s(π∗τ) =
1089
+
1090
+
1091
+ ⟨s, π∗τ⟩ volλ
1092
+ And
1093
+ (3.57)
1094
+ π∗(ι(s))(τ) = (ι(s))(π∗τ) =
1095
+
1096
+
1097
+ ⟨s, π∗τ⟩ volλ .
1098
+ According to equations (3.54 ), (3.56), and (3.57), we have π∗(ι(s)) = δs.
1099
+
1100
+ 4. Appendix
1101
+ 4.1. Polarizations on symplectic manifolds. A step in the process of geometric quanti-
1102
+ zation is to choose a polarization. We first recall the definitions polarizations on symplectic
1103
+ manifolds (M, ω) (See [12, 14]). All polarizations discussed in this subsection are smooth.
1104
+ Definition 4.1. A complex polarization on M is a complex sub-bundle of the complexified
1105
+ tangent bundle TM ⊗ C satisfying the following conditions:
1106
+ (1) P is involutive, i.e. if u, v ∈ Γ(M, P), then [u, v] ∈ Γ(M, P);
1107
+ (2) for every x ∈ M, Px ⊆ TxM ⊗ C is Lagrangian; and
1108
+ (3) rkR (P) := rank(P ∩ P ∩ TM) is constant.
1109
+ Furthermore, P is called
1110
+ · real polarization, if P = P, i.e. rkR (P) = m;
1111
+ · K¨ahler polarization, if P ∩ P = 0, i.e. rkR (P) = 0;
1112
+ · mixed polarization, if 0 < rank(P ∩ P ∩ TM) < m, i.e. 0 < rkR (P) < m.
1113
+
1114
+ 22
1115
+ LEUNG AND WANG
1116
+ 4.2. Singular polarizations on symplectic manifolds. In subsection, we review the
1117
+ definitions of singular polarizations, smooth sections of singular polarizations which were
1118
+ used in the proof of the main results (see [9]).
1119
+ Definition 4.2. P ⊂ TM ⊗ C is a singular complex distribution on M if it satisfies: Pp is
1120
+ a vector subspace of TpM ⊗ C, for all point p ∈ M. Such a P is called smooth on an open
1121
+ subset ˇ
1122
+ M ⊂ M if P| ˇ
1123
+ M is a smooth sub-bundle of the tangent bundle T ˇ
1124
+ M ⊗ C.
1125
+ Remark 4.3. In this paper, we only consider such distributions with mild singularities in
1126
+ the sense that they are only singular outside an open dense subset ˇ
1127
+ M ⊂ M. Under our
1128
+ setting, we define smooth sections of singular distributions and involutive distributions as
1129
+ follows.
1130
+ Definition 4.4. Let P be a singular complex distribution of TM ⊗C. For any open subset
1131
+ U of M, the space of smooth sections of P on U is defined by the smooth section of TM ⊗C
1132
+ with value in P, that is,
1133
+ Γ(U, P) = {v ∈ Γ(U, TM ⊗ C) | vp ∈ (P)p, ∀p ∈ U}.
1134
+ Definition 4.5. Let P be a singular complex distribution on M. P is involutive if it
1135
+ satisfies:
1136
+ [u, v] ∈ Γ(M, P), for any u, v ∈ Γ(M, P).
1137
+ Definition 4.6. Let P be a singular complex distribution P on M and smooth on ˇ
1138
+ M.
1139
+ Such a P is called a singular polarization on M, if it satisfies the following conditions:
1140
+ (a) P is involutive, i.e. if u, v ∈ Γ(M, P), then [u, v] ∈ Γ(M, P);
1141
+ (b) for every x ∈ ˇ
1142
+ M, Pp ⊆ TpM ⊗ C is Lagrangian; and
1143
+ (c) the real rank rkR(P) := rank(P ∩ P ∩ TM)| ˇ
1144
+ M is a constant.
1145
+ Furthermore, such a singular P is called
1146
+ · real polarization, if P| ˇ
1147
+ M = P| ˇ
1148
+ M, i.e. rkR(P| ˇ
1149
+ M) = m;
1150
+ · K¨ahler polarization, if P ˇ
1151
+ M ∩ P| ˇ
1152
+ M = 0 on ˇ
1153
+ M, i.e. r (P| ˇ
1154
+ M) = 0;
1155
+ · mixed polarization, if 0 < rank(P ∩ P ∩ TM)| ˇ
1156
+ M < m, i.e. 0 < rkR(P| ˇ
1157
+ M) < m.
1158
+ 4.3. Coisotropic embedding theorem. We review the coisotropic embedding theorem
1159
+ studied by Guillemin in [2], which was used in the proof of taking divergence. Let (M, ω)
1160
+ be a symplectic manifold of dimensional 2m equipped with Hamiltonian T n-action with
1161
+ moment map µ. Without loss of generality, we assume n = 1. Choose a principal T 1-
1162
+ connection α ∈ Ω1(M0, t) on M0, where M0 = µ−1(0). Consider M0 as a submanifold of
1163
+ M0 × R via the embedding
1164
+ i : M0 → M0 × R, i(p) = (p, 0).
1165
+
1166
+ GQ ASSOCIATED TO MIXED POLARIZATIONS ON K¨AHLER MANIFOLDS WITH T-SYMMETRY 23
1167
+ On the product space ˜
1168
+ M = M0 × (−ǫ, ǫ), the two-form
1169
+ ˜ω = π∗ω0 + d(tα), −ǫ < t < ǫ
1170
+ is symplectic on ˜
1171
+ M and satisfies i∗˜ω = π∗ω0. Extending the T 1-action on M0 to M0×t∗ in a
1172
+ trivial manner. Then ˜ω is T 1-invariant and that the action of T 1 on M0 ×t∗ is Hamiltonian
1173
+ with moment map
1174
+ µ0 : M0 × t∗ → t∗, (p, t) �→ t.
1175
+ Theorem 4.7. [3, Theorem 2.2] In a neighborhood of M0, the Hamiltonian T n-spaces
1176
+ (M, ω) and ( ˜
1177
+ M, ˜ω) are isomorphic.
1178
+ 4.4. Geometric quantization commute with symplectic reduction. In this subsec-
1179
+ tion, we review the work on geometric quantization commute with symplectic reduction by
1180
+ Guillemin and Sternberg in [3]. Let (L, ∇) and (Lλ, ∇λ) be the pre-quantum line bundle
1181
+ on M and Mλ respectively as discussed before, for λ ∈ t∗
1182
+ Z,reg. Then the quantum space
1183
+ HPJ associated to PJ is the space of J-holomorphic sections of L:
1184
+ HPJ = {s ∈ Γ(M, L) | ¯∂Js = 0} = H0(M, L).
1185
+ One can perform two processes on the pre-quantum line bundle (L, ∇); one is geometric
1186
+ quantization, and the other is symplectic reduction. Guillemin and Sternberg in [3] showed
1187
+ that these two processes commute with each other, that is,
1188
+ (4.1)
1189
+ (HPJ)λ ∼= HPJ,λ,
1190
+ where (HPJ)λ (Jλ-holomorphic sections of Lλ) is the λ-weight subspace of HPJ and HPJ,λ
1191
+ is the quantum space associated to reduced K¨ahler polarization PJ,λ, i.e.
1192
+ (4.2)
1193
+ HPJ,λ = {s ∈ Γ(Mλ, Lλ) | ¯∂Jλs = 0} = H0(Mλ, Lλ).
1194
+ References
1195
+ 1. T. Baier, C. Florentino, J. M. Mour˜ao and J. P. Nunes, Toric K¨ahler metrics seen from infinity,
1196
+ quantization and compact tropical amoebas, J. Diff. Geom., 89 (3), 411-454, 2011.
1197
+ 2. V. Guillemin, Moment Maps and Combinatorial Invariants of Hamiltonian T n- spaces, Progress in
1198
+ Math., 122, Birkh¨auser, 1994.
1199
+ 3. V. Guillemin and S. Sternberg, Geometric Quantization and Multiplicities of Group Representations,
1200
+ Inventiones mathematicae, 67.3 (1982): 515-538.
1201
+ 4. V. Guillemin and S. Sternberg, Symplectic Techniques in Physics, Cambridge University Press, Cam-
1202
+ bridge University Press, Cambridge, 1984
1203
+ 5. M. D. Hamilton, Locally toric manifolds and singular Bohr-Sommerfeld leaves, Mem. Amer. Math.
1204
+ Soc. 207 (2010), no. 971, vi+60pp.
1205
+ 6. J. Horv´ath, Topological vector spaces and distributions, Courier Corporation, 2012.
1206
+ 7. A. A. Kirillov, Geometric quantization, in: Encyclopaedia of Mathematical Sciences, vol. 4 Dynamical
1207
+ systems, Springer-Verlag, 1990, 137-172.
1208
+
1209
+ 24
1210
+ LEUNG AND WANG
1211
+ 8. B. Kostant, Quantization and unitary representations, In: Modern analysis and applications. Lecture
1212
+ Notes in Math., Vol. 170, pp. 87-207. Berlin-Heidelberg-Mew York: Springer 1970.
1213
+ 9. N.C. Leung and D. Wang Geodesic rays in space of K¨ahler metrics with T-symmetry, arXiv preprint
1214
+ arXiv: 2211.05324 (2022).
1215
+ 10. J. Marsden and A. Weinstein, Reduction of symplectic manifolds with symmetry. Report on Math.
1216
+ Phys. 5,121-130 (1974).
1217
+ 11. W. Rudin, Functional analysis, Second edition. International Series in Pure and Applied Mathematics.
1218
+ McGraw-Hill, Inc., New York, 1991.
1219
+ 12. D. Simms and N. Woodhouse, Lectures on geometric quantization, Lectures Notes in Physics, Vol. 53.
1220
+ Berlin-Heidelberg-New York: Springer 1976.
1221
+ 13. A. Weinstein, Symplectic manifolds and their Lagrangian submanifolds, Advances in Math. 6 (1971),
1222
+ 329-346.
1223
+ 14. N. M. J. Woodhouse, Geometric quantization, Second Edition, Clarendon Press, Oxford, 1991.
1224
+ The Institute of Mathematical Sciences and Department of Mathematics, The Chinese
1225
+ University of Hong Kong, Shatin, Hong Kong
1226
+ Email address: [email protected]
1227
+ The Institute of Mathematical Sciences and Department of Mathematics, The Chinese
1228
+ University of Hong Kong, Shatin, Hong Kong
1229
+ Email address: [email protected]
1230
+
-NAzT4oBgHgl3EQfFfpP/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
-NE1T4oBgHgl3EQf8QVm/content/tmp_files/2301.03543v1.pdf.txt ADDED
@@ -0,0 +1,1778 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 1
2
+ Efficient Mutation Testing via Pre-Trained
3
+ Language Models
4
+ Ahmed Khanfir , Renzo Degiovanni , Mike Papadakis and Yves Le Traon
5
+ SnT, University of Luxembourg, Luxembourg
6
+ Abstract—Mutation testing is an established fault-based testing technique. It operates by seeding faults into the programs under test
7
+ and asking developers to write tests that reveal these faults. These tests have the potential to reveal a large number of faults – those
8
+ that couple with the seeded ones – and thus are deemed important. To this end, mutation testing should seed faults that are both
9
+ “natural” in a sense easily understood by developers and strong (have high chances to reveal faults). To achieve this we propose using
10
+ pre-trained generative language models (i.e. CodeBERT) that have the ability to produce developer-like code that operates similarly,
11
+ but not exactly, as the target code. This means that the models have the ability to seed natural faults, thereby offering opportunities to
12
+ perform mutation testing. We realise this idea by implementing µBERT, a mutation testing technique that performs mutation testing
13
+ using CodeBert and empirically evaluated it using 689 faulty program versions. Our results show that the fault revelation ability of
14
+ µBERT is higher than that of a state-of-the-art mutation testing (PiTest), yielding tests that have up to 17% higher fault detection
15
+ potential than that of PiTest. Moreover, we observe that µBERT can complement PiTest, being able to detect 47 bugs missed by PiTest,
16
+ while at the same time, PiTest can find 13 bugs missed by µBERT.
17
+ Index Terms—Fault Injection, Mutation Testing, Pre-Trained Language Models
18
+ !
19
+ 1
20
+ INTRODUCTION
21
+ Mutation testing aims at seeding faults using simple syntac-
22
+ tic transformations [19]. These transformations, also known
23
+ as mutation operators are typically constructed based on
24
+ syntactic rules crafted based on the grammar of the target
25
+ programming language [8], i.e. replacing an arithmetic op-
26
+ erator with another such as a + by a -. Unfortunately, such
27
+ techniques generate mutants (seeded faults), many of which
28
+ are “unatural”, i.e., non-conforming to the way developers
29
+ code, thereby perceived as unrealistic by developers [11].
30
+ At the same time, the syntactic-based fault seeding fails to
31
+ capture the semantics of the code snippets that they apply,
32
+ leading to numerous trivial or low utility faults [46].
33
+ To deal with the above issue we propose forming natural
34
+ mutations by using big code. Thus, we aim at introducing
35
+ modifications that follow the implicit rules, norms and cod-
36
+ ing conventions followed by programmers, by leveraging
37
+ the capabilities of pre-trained language models to capture
38
+ the underlying distribution of code and its writing, as
39
+ learned by the pre-training process on big code.
40
+ To this end, we rely on CodeBERT [22], an NL-PL
41
+ bimodal language model that has been trained on over
42
+ 6.4 million programs. More precisely, we use its Masking
43
+ Modelling Language (MLM) functionality, which given a
44
+ code sequence with a masked token, predicts alternative
45
+ replacements to that token, that is best matching the se-
46
+ quence context. This is important, since the predictions
47
+ do not follow fixed predefined patterns as is the case of
48
+ conventional mutation testing, but are instead adapted to
49
+ fit best the target code. For instance, given a sequence
50
+
51
+ A. Khanfir, R. Degiovanni, M. Papadakis, Y. Le Traon are with the
52
+ University of Luxembourg, Luxembourg.
53
+ int a = 1;, we pass a masked version of it as int a =
54
+ <mask>;, then CodeBERT by default proposes 5 predictions
55
+ sorted by likelihood score: 0, 1, b, 2, and 10. Being the most
56
+ likely fitting tokens to the code context, our intuition is that
57
+ replacing the masked token with these predictions would
58
+ induce “natural” mutants.
59
+ Precisely, we introduce µBERT, a mutation testing ap-
60
+ proach that uses a pre-trained language model (CodeBERT)
61
+ to generate mutants by masking and replacing tokens with
62
+ the aim of forming natural mutants. µBERT iterates through
63
+ the program statements and modifies their token. In par-
64
+ ticular, µBERT proceeds as follows: (1) it selects and masks
65
+ one token at a time; (2) feeds CodeBERT with the masked
66
+ sequence and obtains the predictions; (3) creates mutants by
67
+ replacing the masked token with the predicted ones; and (4)
68
+ discards non-compilable, duplicate and equivalent mutants
69
+ (mutants syntactically equal to original code).
70
+ Recent research [32] has shown that some real faults
71
+ are only captured by using complex patterns, i.e. patterns
72
+ that require more than one token mutation. To account for
73
+ such cases, µBERT is equipped with additive mutations, i.e.,
74
+ mutations that add code (instead of deleting or altering). For
75
+ example, consider a boolean expression e1 (typically present
76
+ in if, do, while and return statements), which is mutated
77
+ by µBERT by adding a new condition e2, thereby generating
78
+ a new condition e1||e2 (or e1&&e2), which is then masked
79
+ and completed by CodeBERT. For instance, given a condi-
80
+ tion if(a == b), ��BERT produces a new condition if(a
81
+ == b || a > 0) that is masked and produces if(a ==
82
+ b || b > 0).
83
+ We implement µBERT, and evaluate its ability to serve
84
+ the main purposes of mutation testing, i.e. guiding the
85
+ testing towards finding faults. We thus, evaluate it using
86
+ 689 faults from Defects4J and asses µBERT effectiveness and
87
+ arXiv:2301.03543v1 [cs.SE] 9 Jan 2023
88
+
89
+ 2
90
+ cost-efficiency to reveal1 them. Our results show that µBERT
91
+ is very effective in terms of fault revelation, finding on
92
+ average 84% of the faults. This implies that µBERT mutants
93
+ cover efficiently faulty behaviours caused by real bugs.
94
+ More importantly, the approach is noticeably more effec-
95
+ tive and cost-efficient than a traditional mutation testing
96
+ technique, namely PiTest [17], that we use as a baseline
97
+ in our evaluation. Precisely, we consider three different
98
+ configurations for PiTest that uses different sets of mutation
99
+ operators (DEFAULT, ALL and RV). In fact, test suites that
100
+ kill all mutants of µBERT find on average between 5.5% to
101
+ 33% more faults than those generated to kill all mutants
102
+ introduced by PiTest. Moreover, even when analysing the
103
+ same number of mutants, µBERT induces test suites that find
104
+ on average 6% to 16% more faults than PiTest. These results
105
+ are promising and endorse the usage of µBERT over the
106
+ considered mutation testing technique, as a test generation
107
+ and assessment criterion.
108
+ We also study the impact of the condition-seeding-based
109
+ mutations in the fault detection capability of µBERT. We
110
+ observe that test-suites designed to kill both kinds of µBERT
111
+ mutants – induced by 1) direct CodeBERT predictions and
112
+ 2) a combination of conditions-seeding with CodeBERT
113
+ predictions – find on average over 9% more bugs than the
114
+ ones designed to kill direct CodeBERT prediction mutants
115
+ only (1).
116
+ Overall, our main contributions are:
117
+ • We introduce µBERT, the first mutation testing ap-
118
+ proach that uses pre-trained language models. It lever-
119
+ ages the model’s code knowledge captured during its
120
+ pretraining on large code corpora and its ability to
121
+ capture the program context, to produce “natural” mu-
122
+ tants.
123
+ • We propose new additive mutations which operate
124
+ by seeding new conditions in the existing conditional
125
+ expressions of the target code, then masking and re-
126
+ placing their tokens with the model predictions.
127
+ • We provide empirical evidence that µBERT mutants can
128
+ guide testing towards higher fault detection capabili-
129
+ ties, outperforming those achieved by SOA techniques
130
+ (i.e. PiTest), in terms of effectiveness and cost-efficiency.
131
+ In our empirical study, we validate also the advantage
132
+ of employing the new additive mutation patterns, w.r.t
133
+ improving the effectiveness and cost-efficiency in writ-
134
+ ing test suites with higher fault revelation capability.
135
+ 2
136
+ BACKGROUND
137
+ 2.1
138
+ Mutation Testing
139
+ Mutation analysis [47] is a test adequacy criterion repre-
140
+ senting test requirements by the mean of mutants, which
141
+ are obtained by performing slight syntactic modifications
142
+ to the original program. For instance, an expression like
143
+ x > 0 can be mutated to x < 0 by replacing the relational
144
+ operator > with <. These mutants are then used to assess the
145
+ effectiveness and thoroughness of a test suite in detecting
146
+ their corresponding code modification.
147
+ 1. Tests are written/generated to kill (reveal) the mutants. A bug is
148
+ revealed by a mutation testing approach, if the written tests to kill its
149
+ mutants also reveal the bug.
150
+ A test case detects a mutant if it is capable of producing
151
+ distinguishable observable outputs between the mutant and
152
+ the original program. A mutant is said to be killed if it is
153
+ detected by a test case or a test suite; otherwise, it is called
154
+ live or survived. Some mutants cannot be killed as they are
155
+ functionally equivalent to the original program. The mutation
156
+ score measures the test suite adequacy and is computed
157
+ as the ratio of killed mutants over the total number of
158
+ generated mutants.
159
+ 2.2
160
+ Generative Language Models
161
+ Advances in deep learning approaches gave birth to new
162
+ language models for code generation
163
+ [1], [4], [15], [22].
164
+ These models are trained on large corpora counting multiple
165
+ projects, thereby acquiring a decent knowledge of code,
166
+ enabling them to predict accurately source code to devel-
167
+ opers. Among these pre-trained models, CodeBERT [22], a
168
+ language model that has been recently introduced and made
169
+ openly accessible for researchers by Microsoft.
170
+ CodeBERT is an NL-PL bimodal pre-trained language
171
+ model (Natural Language Programming Language) that
172
+ supports multiple applications such as code search, code
173
+ documentation generation, etc. Same as most large pre-
174
+ trained models, i.e. BERT [20], CodeBERT’s developing
175
+ adopts a Multilayer Transformer [55] architecture. It has
176
+ been trained on a large corpus collected from over 6.4
177
+ million projects available on GitHub, counting 6 different
178
+ programming languages, including Java. The model was
179
+ trained in a cross-modal fashion, through bimodal NL-PL
180
+ data, where the input data is formed by pairs of source code
181
+ and its related documentation, as well-as unimodal data,
182
+ including either natural language or programming language
183
+ sequences per input. This way, it enables the model to offer
184
+ both – PL and NL-PL – functionalities. The training targets
185
+ a hybrid objective function, that is based on replaced token
186
+ detection.
187
+ µBERT incorporates the Masked Language Modeling
188
+ (MLM) functionality [2] of CodeBERT in its workflow, to
189
+ generate “natural” mutants. The CodeBERT MLM pipeline
190
+ takes as input a code sequence of maximum 512 tokens,
191
+ including among them one masked as <mask>, whose
192
+ value will be predicted by the model based on the context
193
+ captured from the remaining tokens. CodeBERT provides
194
+ by default 5 predictions per token, among which we use the
195
+ inaccurate and compilable predicted codes as mutants.
196
+ 3
197
+ APPROACH
198
+ We propose µBERT, a generative language-model-based mu-
199
+ tation testing approach, which is described step by step
200
+ in Figure 1. Given an input source code, µBERT leverages
201
+ CodeBERT’s knowledge of code and its capability in captur-
202
+ ing the program’s context to produce “natural” mutations,
203
+ i.e. that are similar to eventual developer mistakes.To do so,
204
+ µBERT proceeds as follows in six steps:
205
+ 1) First, it extracts relevant locations (AST 2 nodes) where
206
+ to mutate
207
+ 2) Second, it masks the identified node-tokens, creating
208
+ one masked version per selected token.
209
+ 2. AST: Abstract Syntax Tree.
210
+
211
+ 3
212
+ if (a != b)
213
+ return a != d || b>0;
214
+ Source-code
215
+ AST nodes
216
+ locations
217
+ 1. AST Nodes
218
+ selection
219
+ Y
220
+ Y
221
+ 3. Masked code 

222
+ prediction
223
+ Model
224
+ Predictions
225
+ Predict°
226
+ 4. Conditions seeding
227
+ 5. Injection &
228
+ Compilation
229
+ check
230
+ Injected faults
231
+ if (a != b)
232
+ return a != d;
233
+ if (a != b && b>i)
234
+ return a != d;
235
+ a = b + c;
236
+ return a <mask> d;
237
+ 2. AST Nodes
238
+ Masking
239
+ a = b + c;
240
+ return a == d;
241
+ a = <mask> + c;
242
+ return a == d;
243
+ Fig. 1: µBERT Workflow: (1) it parses the Java code given as input, and extracts the expressions to mutate; (2) it creates
244
+ simple-replacement mutants by masking the tokens of interest and invoking CodeBERT; (3) it generates the mutants by
245
+ replacing the masked token with CodeBERT predictions; (4) it generates complex mutants via a) conditions-seeding, b)
246
+ tokens masking then c) replacing by CodeBERT predictions; and finally, (5) it discards not compiling and syntactically
247
+ identical mutants.
248
+ 3) Then, it invokes CodeBERT to predict replacements for
249
+ these masked tokens.
250
+ 4) In addition to the mutants produced in Step (3), µBERT
251
+ also implements some condition-seeding additive mu-
252
+ tations that modify more than one token. Precisely, it
253
+ modifies the conditional expressions in the control flow
254
+ (typically present in if, do, while and return state-
255
+ ments) by extending the original condition with a new
256
+ one, combined with the logical operator && or ||. Then,
257
+ the new conditional expression is mutated by following
258
+ the same steps (2) and (3) – masking and replacing the
259
+ masked tokens by the CodeBERT predictions.
260
+ 5) Finally, the approach discards duplicate predictions
261
+ or those inducing similar code to the original one,
262
+ or not compiling, and outputs the remaining ones as
263
+ mutants, from diverse locations of the target code. More
264
+ precisely, it iterates through the statements in random
265
+ order and outputs in every iteration one mutant per
266
+ line, until achieving the desired number of mutants or
267
+ all mutants are outputted.
268
+ 3.1
269
+ AST Nodes Selection
270
+ µBERT parses the AST of the input source code and selects
271
+ the lines that are more likely to carry the program’s specifi-
272
+ cation implementation, excluding the import statements and
273
+ the declaration ones, e.g. the statements declaring a class, a
274
+ method, an attribute, etc. This way, the approach focuses the
275
+ mutation on the business-logic portion of the program and
276
+ excludes the lines that are probably of lower impact on the
277
+ program behaviour. It proceeds then, by selecting from each
278
+ of these statements, the relevant nodes to mutate, i.e. the
279
+ operators, the operands, the method calls and variables, etc.,
280
+ and excluding the language-specific ones, like the separators
281
+ and the flow controls, i.e. semicolons, brackets, if, else,
282
+ etc. Table 1 summarises the type of targeted AST nodes
283
+ by µBERT, with corresponding example expressions and
284
+ induced mutants. We refer to these as the conventional
285
+ mutations provided by µBERT, denoted by µBERTconv in
286
+ our evaluation, previously introduced in the preliminary
287
+ version of the approach [18].
288
+ 3.2
289
+ Token Masking
290
+ In this step, we mask the selected nodes one by one, pro-
291
+ ducing a masked version from the original source code for
292
+ each node of interest. This means that every masked version
293
+ contains the original code with one missing node, replaced
294
+ by the placeholder <mask>.
295
+ This way, µBERT can generate several mutants in the
296
+ same program location. For instance, for an assignment ex-
297
+ pression like res = a + b, µBERT will create (potentially
298
+ 25) mutants from the following masked sequences:
299
+ • <mask> = a + b
300
+ • res <mask>= a + b
301
+ • res = <mask> + b
302
+ • res = a <mask> b
303
+ • res = a + <mask>
304
+ 3.3
305
+ CodeBERT-MLM prediction
306
+ µBERT invokes CodeBERT to predict replacements for the
307
+ masked nodes. To do so, it tokenizes every masked version
308
+ into a tokens vector then crops it to a subset one that fits the
309
+ maximum size allowed by the model (512) and counts the
310
+ masked token with the surrounding code-tokens. Next, our
311
+ approach feeds these vectors to CodeBERT MLM to predict
312
+ the most probable replacements of the masked token. Our
313
+ intuition is that the larger the code portion accompanying
314
+ the mask placeholder, the better CodeBERT would be able
315
+ to capture the code context, and consequently, the more
316
+ meaningful its predictions would be. This step ends with
317
+ the generation of five predictions per masked token.
318
+ 3.4
319
+ Condition seeding
320
+ µBERT generates second-order mutants by combining con-
321
+ dition seeding with CodeBERT prediction capabilities. To do
322
+ so, our approach modifies the conditions in control flow and
323
+
324
+ ava4
325
+ TABLE 1: Example of µBERT conventional mutations, available in the preliminary version of the approach [18], denoted by
326
+ µBERTconv.
327
+ Ast node
328
+ Expression
329
+ Masked Expression
330
+ Mutant Example
331
+ literals
332
+ res + 10
333
+ res + <mask>
334
+ res + 0
335
+ identifiers
336
+ res + 10
337
+ <mask> + 10
338
+ a + 10
339
+ binary expressions
340
+ a && b
341
+ a <mask> b
342
+ a || b
343
+ unary expressions
344
+ --a
345
+ <mask>a
346
+ ++a
347
+ assignments
348
+ sum += current
349
+ sum <mask>= current
350
+ sum -= current
351
+ object fields
352
+ node.next
353
+ node.<mask>
354
+ node.prev
355
+ method calls
356
+ list.add(node)
357
+ list.<mask>(node)
358
+ list.push(node)
359
+ array access
360
+ arr[index + 1]
361
+ arr[<mask>]
362
+ arr[index]
363
+ static type references
364
+ Math.random() * 10
365
+ <mask>.random() * 10
366
+ Random.random() * 10
367
+ return statements, including if, do, while and return
368
+ conditional expressions. For every one of these statements,
369
+ it starts by extending the original condition by a new one,
370
+ separated with the logical operator && or ||, in both orders
371
+ (original condition first or the other way around) and with
372
+ or without negation (!).
373
+ Next, all substitute conditions are put one by one in place
374
+ in the original code, forming multiple condition-seeded
375
+ code versions, that we pass as input to Step (2), in which
376
+ their tokens are masked and then (3) passed each to Code-
377
+ BERT to predict the best substitute of their corresponding
378
+ masked tokens.
379
+ The seeded conditions are created in two ways:
380
+ 3.4.1
381
+ Using existing conditions in the same class
382
+ To mutate a given condition – if, do, while and return
383
+ conditional expressions –, we collect all other conditions
384
+ existing in the same class, then combine each one of them
385
+ with the target condition, using logical operators.
386
+ Precisely, let Expt a conditional expression to mutate
387
+ and SE = {Exp0, ..., Expn} the set of other conditional
388
+ expressions appearing in the same class, excluding the null-
389
+ check ones (i.e. var == null). The alternative replacement
390
+ conditions generated for Expt are the combinations of:
391
+ • Expt op neg Expi and
392
+ • Expi op neg Expt,
393
+ where op is a binary logical operator taking the values in
394
+ {&&,||}, neg is either the negation operator ! or nothing
395
+ and Expi is a condition from SE.
396
+ 3.4.2
397
+ Using existing variables in the same class
398
+ When the target if conditional expression to mutate con-
399
+ tains variables (including fields), we create new additional
400
+ conditions by combining these variables with others of the
401
+ same type from the same class. Then we combine each one
402
+ of the newly created conditions with the original one, using
403
+ logical operators.
404
+ Precisely, let Expt be a conditional expression to mutate
405
+ containing a set of variables Svt. For every variable vart in
406
+ Svt, we load Sv = {var0, ..., varn} the set of other variables
407
+ appearing in the same class and of the same type T as vart,
408
+ then we generate the following new conditions:
409
+ • Expt op (vart relop vari) and
410
+ • (vart relop vari) op Expt,
411
+ where op is a binary logical operator taking the values in
412
+ {&&,||}, relop is a relational operator applicable on the
413
+ type T and vari is a variable from Sv.
414
+ 3.5
415
+ Mutant filtering
416
+ In this step, our approach starts by discarding accurate and
417
+ duplicate predictions; the redundant predictions and the
418
+ ones that are exactly the same as the original code. Then, it
419
+ iterates through the statements and selects in every iteration
420
+ one compilable prediction by line, while discarding not
421
+ compilable ones. Once all first-order mutants are selected
422
+ (issued by one single token replacement), our approach
423
+ proceeds by selecting second-order ones (issued by the com-
424
+ bination of condition seeding and one token replacement) in
425
+ the same iterative manner. µBERT continues iterating until
426
+ achieving the desired number of mutants or all mutants are
427
+ outputted.
428
+ 4
429
+ RESEARCH QUESTIONS
430
+ We start our analysis by investigating the advantage
431
+ brought by the additive mutations (a.k.a. conditions seeding
432
+ ones) w.r.t. the fault detection capabilities of test suites
433
+ designed to kill µBERT’s mutants. Thus, we ask:
434
+ RQ1 (µBERT Additive mutations) What is the added value of
435
+ the additive mutations on the fault detection capabili-
436
+ ties of test suites designed to kill µBERT’s mutants?
437
+ To answer this question, we generate two sets of mutants
438
+ using µBERT: 1) the first set using all possible mutations that
439
+ we denote as µBERT and 2) a second one using only the
440
+ conventional µBERT’ mutations – part of our preliminary
441
+ implementation [18], excluding the additive ones – that we
442
+ denote as µBERTconv. Then we evaluate the fault detection
443
+ ability of test suites selected to kill the mutants from each
444
+ set.
445
+ The answer of this question provides evidence that the
446
+ additive mutations increase the fault detection capability of
447
+ µBERT. Yet, to assess its general performance we compare
448
+ it to state-of-the-art (SOA) mutation testing, particularly
449
+ PiTest [17], and thus, we ask:
450
+ RQ2 (Fault detection) How does µBERT compare with state-
451
+ of-the-art mutation testing, in terms of fault detection?
452
+ To answer this question we generate mutants using the
453
+ latest version of PiTest [17], on the same target projects as
454
+ for RQ1. As we are interested in comparing the approaches
455
+ and not the implementations of the tools, we exclude the
456
+ subjects on which PiTest did not run correctly or did not
457
+ generate any mutant. This way we ensure having a fair base
458
+ of comparison by counting exactly the same study subjects
459
+ for both approaches (further details are given in Section 5).
460
+ Then, we compare the fault detection capability of test suites
461
+
462
+ 5
463
+ selected to kill the same number of mutants produced by
464
+ each approach.
465
+ Finally, we qualitatively analyse some of the mutants
466
+ generated with µBERT and ask:
467
+ RQ3 (Qualitative analysis) Does µBERT generate different
468
+ mutants than traditional mutation testing operators?
469
+ To answer this question, we showcase the mutants gen-
470
+ erated by µBERT that help in detecting faults not found
471
+ by PiTest. Additionally, we discuss the program-context-
472
+ capturing importance in µBERT’s functioning, by rerunning
473
+ it with a reduced size of the masked codes passed to the
474
+ model, and comparing examples of yielded mutants with
475
+ those obtained in our original setup.
476
+ 5
477
+ EXPERIMENTAL SETUP
478
+ 5.1
479
+ Dataset & Benchmark
480
+ To evaluate µBERT’s fault detection, we use real bugs from a
481
+ popular dataset in the software engineering research area
482
+ – Defects4J [29] v2.0.0. In this benchmark, every subject
483
+ bug is provided with a buggy version of the source code,
484
+ its corresponding fixed version, and equipped with a test
485
+ suite that passes on the fixed version and fails with at least
486
+ one test on the buggy one. The dataset includes over 800
487
+ bugs from which, we exclude the ones presenting issues, i.e.
488
+ with wrong revision ids, not compiling or with execution
489
+ issues, or having failing tests on the fixed version, at the
490
+ reporting time. Next, we run µBERT and PiTest on the
491
+ corresponding classes impacted by the bug from the fixed
492
+ versions of the remaining bugs and exclude the ones where
493
+ no tool generated any mutant, ending up with 689 bugs
494
+ covered by µBERT and 457 covered by PitTest. As we’re
495
+ interested in comparing the approaches and not the tools’
496
+ implementations, and to exclude eventual threats related to
497
+ the environment (i.e. supported java and juint versions by
498
+ each technique, etc.) or the limitations and shortages of the
499
+ dataset, we establish every comparison study on a dataset
500
+ counting only bugs covered by all considered approaches:
501
+ 689 bugs to answer RQ1 and 457 to answer RQ2 and RQ3.
502
+ 5.2
503
+ Experimental Procedure
504
+ To assess the complementary and added value in terms of
505
+ fault revelation of the condition-seeding-based mutations
506
+ (answer to RQ1), we run our approach with and without
507
+ those additional mutations – that we name respectively
508
+ µBERT and µBERTconv–, and thus, generating all possible
509
+ mutants on our dataset programs’ fixed versions. Next,
510
+ we compare the average effectiveness of the test suites
511
+ generated to kill the mutants of each set; induced by µBERT
512
+ and µBERTconv.
513
+ Once the added value of the proposed condition-
514
+ seeding-based mutations is validated, we compare its per-
515
+ formance to S.O.A. mutation testing (answer to RQ2 and
516
+ RQ3). We use PiTest [17], a stable and mature Java mutation
517
+ testing tool, because it has been more effective at finding
518
+ faults than other tools [33] and it is among the most com-
519
+ monly used by researchers and practitioners [47], [52], as of
520
+ today. The tool proposes different configurations to adapt
521
+ the produced mutations and their general cost to the target
522
+ users, by excluding or including mutators. Among these
523
+ configurations we used the three following:
524
+ • Pit-all (ALL) which counts all available mutation oper-
525
+ ators available in the current version3.
526
+ • Pit-default (DEFAULTS) whose mutators are selected to
527
+ form a stable and cost-efficient subset of operators by
528
+ producing less but more relevant mutants.
529
+ • Pit-rv-all (ALL) which is a version4 that includes the
530
+ mutators of Pit-all and extra experimental [7] ones that
531
+ are made available for research studies.
532
+ To compare the different approaches, we evaluate their
533
+ effectiveness and cost-efficiency in achieving one of the
534
+ main purposes of mutation testing, i.e., to guide the testing
535
+ towards higher fault detection capabilities. For this reason,
536
+ we simulate a mutation testing use-case scenario, where
537
+ a developer/tester selects mutants and writes tests to kill
538
+ them [13], [34].
539
+ We run every approach on the fixed versions and test
540
+ suites provided by Defects4J, then collect the mutants and
541
+ their test execution results; whether the mutant is killed
542
+ (breaks at least one test of the test suite) and if yes by which
543
+ tests. Next, we suppose that the not killed mutants are
544
+ equivalent or irrelevant, explaining why no tests have been
545
+ written to kill them by the developers. Then, we simulate the
546
+ scenario of a developer testing the fixed version, in a state
547
+ where 1) it did not have any test 2) thus all mutants did not
548
+ have killing tests and 3) the developer had no knowledge
549
+ of which mutants are equivalent or not. This way, we can
550
+ reproduce the developer flow of
551
+ 1) selecting and analysing one mutant,
552
+ 2) to either (a) discard it from the mutant set if it is
553
+ equivalent (not killed in the actual test suite) or (b) write
554
+ a test to kill it (by selecting one of the actual killing tests
555
+ of the mutant),
556
+ 3) then discarding all killed mutants by that test and
557
+ 4) iterating similarly over the remaining mutants until all
558
+ of them are analysed.
559
+ We say that a bug is found by a mutation testing technique
560
+ if the resulting test suite – formed by the written (selected)
561
+ tests by the developer – contains at least one test that reveals
562
+ it; a test that breaks when executed on the buggy version.
563
+ We express the testing cost in terms of mutants analysed,
564
+ and hence, we consider the effort required to find a bug as
565
+ the number of mutants analysed until the first bug-revealing
566
+ test is written. To set a common basis of comparison be-
567
+ tween the approaches, accounting for the different number
568
+ of generated mutants, we run the simulations until the same
569
+ maximum effort is reached (maximum number of mutants
570
+ to analyse), which we set to the least cost required to kill
571
+ all the mutants by one of the compared approaches. During
572
+ our evaluation study, we use the same mutation selection
573
+ strategy for all compared approaches, iterating through the
574
+ lines in random order and selecting 1 arbitrary mutant per
575
+ line per iteration. To reduce the process randomness impact
576
+ 3. Version
577
+ 1.9.4
578
+ available
579
+ in
580
+ PitTest’s
581
+ [6]
582
+ GitHub
583
+ repository
584
+ (branch=master,
585
+ repo=https://github.com/hcoles/pitest.git,
586
+ rev-
587
+ id=17e1eecf)
588
+ 4. Version
589
+ 1.7.4
590
+ available
591
+ in
592
+ PitTest’s
593
+ [6]
594
+ GitHub
595
+ repository
596
+ (branch=master,
597
+ repo=https://github.com/hcoles/pitest.git,
598
+ rev-
599
+ id=2ec1178a)
600
+
601
+ 6
602
+ on our results (in the selection of mutants and tests), we
603
+ run every simulation 100 times, then average their results
604
+ for every target-bug and considered approach. Finally, we
605
+ aggregate these averages computed on all target bugs and
606
+ normalise them as global percentages of achieved fault
607
+ detection by spent effort, in terms of mutants analysed.
608
+ Finally, to answer RQ3, we select example mutants that
609
+ enabled µBERT to find bugs exclusively (not found by any
610
+ of PiTest versions), from the results of RQ2. Then we discuss
611
+ the added value of µBERT mutations through the analysis of
612
+ the mutants’ behavioural difference from the fixed version
613
+ and similarity with the buggy one.
614
+ 5.3
615
+ Implementation
616
+ We implemented µBERT’s approach as described in Sec-
617
+ tion 3: we have used Spoon [51] and Jdt [21] libraries to
618
+ parse and extract the business logic related AST nodes and
619
+ apply condition-seeding mutators. To predict the masked
620
+ tokens we have used the implementation proposed by
621
+ CodeBERT-nt [3], [31], using CodeBERT Masked Language
622
+ Modeling (MLM) functionality [2], [22].
623
+ We provide the implementation of our approach and the
624
+ reproduction package of its evaluation at https://github.
625
+ com/Ahmedfir/mBERTa.
626
+ 6
627
+ EXPERIMENTAL RESULTS
628
+ 6.1
629
+ RQ1: µBERT Additive mutations
630
+ To answer this question we compare the fault detection
631
+ effectiveness of test suites written to kill mutants generated
632
+ by µBERT with and without additive mutations, noted re-
633
+ spectively µBERT and µBERTconv. Figure 2 depicts the fault
634
+ detection improvement when extending µBERT mutations
635
+ by the additive ones. In fact, µBERT fault detection increased
636
+ on average by over 9% compared to the one achieved by
637
+ µBERTconv, achieving 84.64% on average. We can also see
638
+ that besides outliers, the majority of bugs are found in 100%
639
+ of the times. Moreover, when examining the bugs separately,
640
+ we find that µBERT finds 20 more bugs than µBERTconv
641
+ (with fault detection > 0%), and 70 more when considering
642
+ bugs found with fault detection percentages above 90%.
643
+ This confirms that the additive patterns induce relevant
644
+ mutants ensuring the detection of some bugs always or in
645
+ most of the cases, as well as representing better new types
646
+ of faults, which were not detectable otherwise.
647
+ To check the significance of the fault detection advantage
648
+ brought by the additive patterns, we performed a statistical
649
+ test (Wilcoxon paired test) on the data of Figure 2a to vali-
650
+ date the hypothesis ”the fault detection yielded by µBERT
651
+ is greater than the one by µBERTconv ”. The very small
652
+ obtained p-values of 5.92e-21 (≪ 0.05) showed that the dif-
653
+ ferences are significant, indicating the low probability of this
654
+ fault detection amelioration to be happening by chance. The
655
+ difference size confirms also the same advantage, with ˆA12
656
+ values of 0.5827 (> 0.5), indicating that µBERT induces test-
657
+ suites with higher fault detection capability in the majority
658
+ of the cases.
659
+ Next, we compare the fault detection performance of
660
+ µBERT and µBERTconv when analysing the same number
661
+ of mutants, and illustrate in Figure 3 their average fault
662
+ BERT
663
+ BERTconv
664
+ tool
665
+ 0
666
+ 20
667
+ 40
668
+ 60
669
+ 80
670
+ 100
671
+ Fault detection %
672
+ 84.64%
673
+ 75.30%
674
+ (a) Effectiveness: mean fault-detection per subject.
675
+ 0
676
+ 20
677
+ 40
678
+ 60
679
+ 80
680
+ 100
681
+ Effort % (number of analysed mutants)
682
+ 0
683
+ 20
684
+ 40
685
+ 60
686
+ 80
687
+ Fault detection %
688
+ tool
689
+ BERT
690
+ BERTconv
691
+ (b) Cost-efficiency: fault detection by the number of mutants
692
+ analysed.
693
+ Fig. 2: Fault-detection performance improvement when us-
694
+ ing additive patterns. Comparison between µBERT and
695
+ µBERTconv, w.r.t. the fault-detection of test suites written to
696
+ kill all generated mutants.
697
+ detection effectiveness and cost-efficiency in terms of anal-
698
+ ysed mutants. The box-plots of the Subfigure 3a show that
699
+ even when spending the same effort as µBERTconv, µBERT
700
+ keeps a similar advantage of on average 6.05% higher fault
701
+ detection, achieving a maximum of 81.35%. From the line-
702
+ plots of the Subfigure 3b, we can see that both approaches
703
+ achieve a comparable fault detection (≈ 70%) at (≤≈ 40%)
704
+ of the maximum costs. At higher costs, µBERTconv’s curve
705
+ increases slowly until achieving a plateau at ≈ 60% of
706
+ the effort, whereas µBERT’s curve keeps increasing to-
707
+ wards higher fault detection ratios even when achieving the
708
+ ≈ 100% of the fixed maximum effort.
709
+ To validate these findings we re-conducted the same
710
+ statistical tests on the data of Subfigure 3a and found that
711
+ µBERT outperforms significantly µBERTconv with negligible
712
+ p-values of 1.15e-19 and ˆA12 values of 0.5711.
713
+
714
+ 7
715
+ BERT
716
+ BERTconv
717
+ tool
718
+ 0
719
+ 20
720
+ 40
721
+ 60
722
+ 80
723
+ 100
724
+ Fault detection %
725
+ 81.35%
726
+ 75.30%
727
+ (a) Effectiveness: mean fault-detection per subject.
728
+ 0
729
+ 20
730
+ 40
731
+ 60
732
+ 80
733
+ 100
734
+ Effort % (number of analysed mutants)
735
+ 0
736
+ 10
737
+ 20
738
+ 30
739
+ 40
740
+ 50
741
+ 60
742
+ 70
743
+ 80
744
+ Fault detection %
745
+ tool
746
+ BERT
747
+ BERTconv
748
+ (b) Cost-efficiency: fault detection by the number of mutants
749
+ analysed.
750
+ Fig. 3: Fault-detection comparison between µBERT and
751
+ µBERTconv, with the same effort: where the maximum effort
752
+ is limited to the minimum effort required to analyse all mutants
753
+ of any of them, which is µBERTconv in most of the cases.
754
+ 6.2
755
+ RQ2: Fault Detection comparison with PiTest
756
+ To answer this research question we reduce our dataset to
757
+ the bugs covered by µBERT and the 3 considered versions of
758
+ PitTest approaches: ”Pit-default” which contains the default
759
+ mutation operators of PiTest, ”Pit-all” containing all PiTest
760
+ operators including the default ones and ”Pit-rv-all” which
761
+ contains experimental operators [7] in addition to the ”Pit-
762
+ all” ones. Then, we perform the same study as in RQ1,
763
+ where we compare the considered approaches’ effectiveness
764
+ and cost-efficiency based on the fault detection capability of
765
+ test suites written to kill their generated mutants. To have a
766
+ fair base of comparison, we compare the approaches under
767
+ the same effort in analysing mutants, which is equal to
768
+ the least average effort required to kill all mutants of one
769
+ of the approaches (which is the one of Pit-default in the
770
+ majority of the cases). As we are interested in comparing
771
+ the mutation testing approaches and not mutant selection
772
+ strategies, we run the simulation with the same one-mutant-
773
+ BERT
774
+ Pit-all
775
+ Pit-default
776
+ Pit-rv-all
777
+ tool
778
+ 0
779
+ 20
780
+ 40
781
+ 60
782
+ 80
783
+ 100
784
+ Fault detection %
785
+ 66.43%
786
+ 60.87%
787
+ 49.90%
788
+ 56.33%
789
+ (a) Effectiveness: mean fault-detection per subject.
790
+ 0
791
+ 20
792
+ 40
793
+ 60
794
+ 80
795
+ 100
796
+ Effort % (number of analysed mutants)
797
+ 0
798
+ 10
799
+ 20
800
+ 30
801
+ 40
802
+ 50
803
+ 60
804
+ Fault detection %
805
+ tool
806
+ BERT
807
+ Pit-all
808
+ Pit-default
809
+ Pit-rv-all
810
+ (b) Cost-efficiency: fault detection by the number of mutants
811
+ analysed.
812
+ Fig. 4: Fault-detection comparison between µBERT and
813
+ PiTest, with the same effort: where the maximum effort is
814
+ limited to the minimum effort required to analyse all mutants
815
+ of any of them, which is Pit-default in most of the cases.
816
+ per-line random sampling of mutants for all techniques (see
817
+ Subsection 5.2).
818
+ Figure 4b shows that with small effort (≤≈ 5%) all
819
+ approaches yield comparable fault detection scores (≈ 10%).
820
+ However, the difference becomes more noticeable when
821
+ spending more effort, with µBERT outperforming all ver-
822
+ sions of PiTest; achieving on average 16.53% higher fault
823
+ detection scores than Pit-default, 10.10% higher than Pit-rv-
824
+ all and 5.56% higher than Pit-all (see Figure 4a).
825
+ To validate these results, we performed the same statis-
826
+ tical tests as in RQ1, checking the hypothesis that ”µBERT
827
+ yields better fault detection capabilities than the other ap-
828
+ proaches”. We illustrate in the first row of Tables 2a and 2b
829
+ the corresponding computed Wilcoxon paired test p-values
830
+ and Vargha and Delaney ˆA12 values. Our results show that
831
+ µBERT has a significant advantage over the considered SOA
832
+ approaches with p-values under 0.05. Additionally, µBERT
833
+ scores ˆA12 values above 0.5 which confirms that guiding
834
+
835
+ 8
836
+ TABLE 2: Paired (per subject bug) statistical tests of the
837
+ average fault detection of test suites written to kill the same
838
+ number of mutants generated by each approach (data of
839
+ Figure 4a).
840
+ (a) Wilcoxon paired test p-values computed on every dataset
841
+ subject, comparing each approach (A1) from the first column
842
+ to the other approaches (A2). p-values smaller than 0.05 in-
843
+ dicate that (A1) yields an average fault detection significantly
844
+ higher than that of (A2).
845
+ p-values
846
+ Pit-rv-all
847
+ Pit-default
848
+ Pit-all
849
+ µBERT
850
+ 7.78e-11
851
+ 1.18e-12
852
+ 3.32e-02
853
+ Pit-all
854
+ 1.54e-22
855
+ 8.87e-06
856
+
857
+ Pit-default
858
+ 9.55e-01
859
+
860
+
861
+ (b) Vargha and Delaney ˆA12 values computed on every dataset
862
+ subject, comparing each approach (A1) from the first column
863
+ to the other approaches (A2). ˆA12 values higher than 0.5
864
+ indicate that (A1) yields an average fault detection higher than
865
+ that of (A2) in the majority of the cases.
866
+ ˆA12
867
+ Pit-rv-all
868
+ Pit-default
869
+ Pit-all
870
+ µBERT
871
+ 0.6488
872
+ 0.5514
873
+ 0.5066
874
+ Pit-all
875
+ 0.7210
876
+ 0.4956
877
+
878
+ Pit-default
879
+ 0.5449
880
+
881
+
882
+ the testing by µBERT mutants instead of those generated by
883
+ SOA techniques yields comparable or higher fault detection
884
+ ratios, in the majority of the cases. Indeed, the ˆA12 differ-
885
+ ence between Pit-all and µBERT is small (0.5066), indicating
886
+ that both approaches perform similarly or better on some
887
+ studied subjects and worst on others.
888
+ We notice also from the sub-figure 4b that Pit-default
889
+ achieves a plateau at around 60% of the effort while the
890
+ other tools keep increasing, showing that they are able to
891
+ achieve higher fault detection capabilities, at a higher cost.
892
+ This is very noticeable when we compare the sub-figures
893
+ (a) and (b) of Figure 4 with the figure 2, where the average
894
+ fault detection of µBERT is way lower than what it achieves
895
+ in RQ1 – around 66% instead of 84%. This is a direct
896
+ consequence of the fact that Pit default produces fewer
897
+ mutants than the other approaches, limiting considerably
898
+ the maximum effort of the mutation campaigns and thus
899
+ the fault detection ratios, in the majority of the cases. Indeed,
900
+ as illustrated in Figure 5, all approaches score higher fault
901
+ detection percentages when spending more effort, achieving
902
+ on average ≈65% for Pit-all, ≈66% for Pit-rv-all and ≈83%
903
+ for µBERT. We explain the small decrease of 1.72% in the
904
+ mean fault detection achieved by µBERT in comparison
905
+ with RQ1 (82,92% in RQ2 instead of 84.64% in RQ1) by the
906
+ difference in the considered dataset for each RQ.
907
+ In Table 3, we illustrate the ˆA12 and p-values computed
908
+ on data of the boxplots in Sub-figure 5a. The results confirm
909
+ that µBERT outperforms significantly SOA mutation testing
910
+ w.r.t the fault detection capability of test suites written to all
911
+ kill mutants generated by each approach.
912
+ Next, we turned our interest to the set of particular bugs
913
+ that every approach can and cannot reveal when spending
914
+ the same effort. Hence, we map each bug with its revealing
915
+ tool, from the simulation results of Figure 4a and illustrate
916
+ their corresponding Venn diagrams in Figure 6.
917
+ BERT
918
+ Pit-all
919
+ Pit-default
920
+ Pit-rv-all
921
+ tool
922
+ 0
923
+ 20
924
+ 40
925
+ 60
926
+ 80
927
+ 100
928
+ Fault detection %
929
+ 82.92%
930
+ 65.49%
931
+ 49.90%
932
+ 66.35%
933
+ (a) Effectiveness: mean fault-detection per subject.
934
+ 0
935
+ 20
936
+ 40
937
+ 60
938
+ 80
939
+ 100
940
+ Effort % (number of analysed mutants)
941
+ 0
942
+ 20
943
+ 40
944
+ 60
945
+ 80
946
+ Fault detection %
947
+ tool
948
+ BERT
949
+ Pit-all
950
+ Pit-default
951
+ Pit-rv-all
952
+ (b) Cost-efficiency: fault detection by the number of mutants
953
+ analysed.
954
+ Fig. 5: Comparison between µBERT and PiTest, relative to
955
+ the fault-detection of test suites written to kill all generated
956
+ mutants.
957
+ From the disjoint sets in Sub-figure 6a, we notice a
958
+ clear advantage in using µBERT over the considered SOA
959
+ baselines, as it finds most of the bugs they find in addition to
960
+ finding exclusively 47 bugs when spending the same effort.
961
+ More precisely, µBERT finds 52, 77 and 52 more bugs than
962
+ Pit-all, Pit-default and Pit-rv-all, respectively, whereas they
963
+ find each 13, 10 and 13 bugs that µBERT missed.
964
+ This endorses the fact that µBERT introduces mutants
965
+ that represent more real bugs than SOA mutation tech-
966
+ niques, and encourages the investigation of the eventual
967
+ complementary between the approaches. This observation
968
+ is more noticeable when considering the overlapping be-
969
+ tween bugs found by each approach in at least 90% of the
970
+ simulations (Sub-figure 6b). We notice that the approaches
971
+ perform comparably, with a particular distinction of Pit-all
972
+ and Pit-default results which find exclusively 19 and 21 bugs
973
+ with these high fault detection percentages instead of 0, as
974
+ observed in Sub-figure 6a. Nevertheless, µBERT conserves
975
+ the same advantage over the considered baselines in this
976
+
977
+ 9
978
+ TABLE 3: Paired (per subject bug) statistical tests of the
979
+ average fault detection of test suites written to kill all the
980
+ mutants generated by each approach (data of Figure 5a).
981
+ (a) Wilcoxon paired test p-values computed on every dataset
982
+ subject, comparing each approach (A1) from the first column
983
+ to the other approaches (A2). p-values smaller than 0.05 in-
984
+ dicate that (A1) yields an average fault detection significantly
985
+ higher than that of (A2).
986
+ p-values
987
+ Pit-rv-all
988
+ Pit-default
989
+ Pit-all
990
+ µBERT
991
+ 2.49e-13
992
+ 2.14e-33
993
+ 1.47e-14
994
+ Pit-all
995
+ 4.71e-01
996
+ 2.76e-23
997
+
998
+ Pit-default
999
+ 1.00e+00
1000
+
1001
+
1002
+ (b) Vargha and Delaney ˆA12 values computed on every dataset
1003
+ subject, comparing each approach (A1) from the first column
1004
+ to the other approaches (A2). ˆA12 values higher than 0.5
1005
+ indicate that (A1) yields an average fault detection higher than
1006
+ that of (A2) in the majority of the cases.
1007
+ ˆA12
1008
+ Pit-rv-all
1009
+ Pit-default
1010
+ Pit-all
1011
+ µBERT
1012
+ 0.6028
1013
+ 0.7123
1014
+ 0.6061
1015
+ Pit-all
1016
+ 0.5077
1017
+ 0.6400
1018
+
1019
+ Pit-default
1020
+ 0.3676
1021
+
1022
+
1023
+ regard, finding exclusively 42 bugs more. It finds also 50, 63
1024
+ and 69 more bugs than respectively Pit-all, Pit-default and
1025
+ Pit-rv-all, whereas they find each 59, 58 and 27 bugs that
1026
+ µBERT missed.
1027
+ 6.3
1028
+ RQ3: Qualitative Analysis of µBERT Mutants
1029
+ To answer this research question we investigate the mutants
1030
+ generated by µBERT, which induced test suites able to find
1031
+ bugs that were not detected otherwise, i.e. by the considered
1032
+ SOA approaches (see Figure 6). Meaning that the mutants
1033
+ break similar tests as the target real buggy version.
1034
+ As a simple bug example (requiring only one change
1035
+ to
1036
+ fix
1037
+ it),
1038
+ we
1039
+ consider
1040
+ Lang-49
1041
+ from
1042
+ Defects4J
1043
+ and
1044
+ we investigate mutants that have been generated by
1045
+ µBERT and helped in generating tests that reveal it. This
1046
+ bug impacts the results of the method reduce() from
1047
+ the class org.apache.commons.lang.math.Fraction,
1048
+ which returns a new reduced fraction instance, if possible,
1049
+ or the same instance, otherwise. The bug is caused by a
1050
+ miss-implementation of a specific corner case, which con-
1051
+ sists of calling the method on a fraction instance that has
1052
+ 0 as numerator. In Table 4, we illustrate example mutants
1053
+ generated by µBERT that helped in revealing this bug. Every
1054
+ mutant is represented by a diff between the fixed and the
1055
+ mutated version by µBERT.
1056
+ As can be seen, µBERT can generate mutants that can be
1057
+ induced by applying conventional pattern-based mutations,
1058
+ i.e., Mutant 1 replaces a relational operator (==) with an-
1059
+ other (>) and Mutant 2 replaces an integer operand (0) with
1060
+ another one (1).
1061
+ In addition, it proposes more complex mutations that
1062
+ are unlikely achievable without any knowledge of either the
1063
+ AST or the context of the considered program. For instance,
1064
+ it can generate Mutant 4 by changing a conditional return
1065
+ statement with (this) the current instance of Fraction,
1066
+ which matches the return type of the method. Similarly, to
1067
+ 47
1068
+ 0
1069
+ 1
1070
+ 0
1071
+ 1
1072
+ 0
1073
+ 3
1074
+ 0
1075
+ 3
1076
+ 3
1077
+ 23
1078
+ 0
1079
+ 1
1080
+ 10
1081
+ 354
1082
+ Pit-all
1083
+ Pit-default
1084
+ Pit-rv-all
1085
+ BERT
1086
+ (a) Faults discovered at least once per 100 runs
1087
+ (Fault detection > 0%).
1088
+ 42
1089
+ 2
1090
+ 3
1091
+ 21
1092
+ 3
1093
+ 0
1094
+ 2
1095
+ 19
1096
+ 10
1097
+ 3
1098
+ 8
1099
+ 15
1100
+ 14
1101
+ 22
1102
+ 114
1103
+ Pit-all
1104
+ Pit-default
1105
+ Pit-rv-all
1106
+ BERT
1107
+ (b) Faults discovered in over 90% of the runs
1108
+ (Fault detection≥ 90%).
1109
+ Fig. 6: Number of faults discovered by test-suites written to
1110
+ kill mutants generated by µBERT and PiTest versions when
1111
+ analysing the same number of mutants (same effort).
1112
+ generate Mutant 5, it replaces (this) the current instance of
1113
+ the class Fraction by an existent instance of the same type
1114
+ (ONE), making the statement returning either the object ONE
1115
+ or the object ZERO.
1116
+ To produce more complex mutants, µBERT applies a
1117
+ condition seeding followed by token-masking and Code-
1118
+ BERT prediction, such as adding || (numerator ==
1119
+ other.numerator) to the original condition of a return
1120
+ statement, inducing Mutant 8, or adding || !(numerator
1121
+ == Integer.MIN_VALUE) to the original condition of an
1122
+ if statement, inducing Mutant 3.
1123
+ To investigate further the impact of the code context
1124
+ captured by the model on the generated mutants, we have
1125
+ rerun µBERT on 5 subjects from our dataset, with a max-
1126
+ imum number of surrounding tokens equal to 10 (instead
1127
+ of 512). Then, we compared manually the induced mutants
1128
+ with those generated by our default setup, in the same
1129
+ locations. From our results, we observed a noticeable de-
1130
+ crease in the number of compilable predictions, indicating
1131
+ the general performance decrease of the model when it lacks
1132
+ information about the code context. Particularly, we notice
1133
+
1134
+ 10
1135
+ TABLE 4: Example of mutants generated by µBERT that
1136
+ helped find the bug Lang-49 from Defects4J.
1137
+ Mutant 1: replacing binary operator
1138
+ @@ org . apache . commons . lang . math . Fraction
1139
+ :
1140
+ 466 @@
1141
+ − i f
1142
+ ( numerator == 0) {
1143
+ + i f
1144
+ ( numerator > 0) {
1145
+ Mutant 2: replacing literal implementation
1146
+ @@ org . apache . commons . lang . math . Fraction
1147
+ :
1148
+ 466 @@
1149
+ − i f
1150
+ ( numerator == 0) {
1151
+ + i f
1152
+ ( numerator == 1) {
1153
+ Mutant 3: adding a condition to an if statement
1154
+ @@ org . apache . commons . lang . math . Fraction
1155
+ :
1156
+ 466 @@
1157
+ − i f
1158
+ ( numerator == 0) {
1159
+ + i f
1160
+ ( ( numerator == 0)
1161
+ +
1162
+ | |
1163
+ ! ( numerator==Integer .MIN VALUE) )
1164
+ {
1165
+ Mutant 4: replacing a condition
1166
+ @@ org . apache . commons . lang . math . Fraction
1167
+ :
1168
+ 467 @@
1169
+ − return
1170
+ equals (ZERO)
1171
+ ?
1172
+ t h i s : ZERO;
1173
+ + return
1174
+ t h i s ;
1175
+ Mutant 5: replacing this access by another object
1176
+ @@ org . apache . commons . lang . math . Fraction
1177
+ :
1178
+ 467 @@
1179
+ − return
1180
+ equals (ZERO)
1181
+ ?
1182
+ t h i s : ZERO;
1183
+ + return
1184
+ equals (ZERO)
1185
+ ? ONE: ZERO;
1186
+ Mutant 6: replacing method argument
1187
+ @@ org . apache . commons . lang . math . Fraction
1188
+ :
1189
+ 469 @@
1190
+ int gcd = greatestCommonDivisor (
1191
+ − Math . abs ( numerator ) ,
1192
+ denominator ) ;
1193
+ + Math . abs ( numerator ) ,
1194
+ 1 ) ;
1195
+ Mutant 7: replacing a variable
1196
+ @@ org . apache . commons . lang . math . Fraction
1197
+ :
1198
+ 473 @@
1199
+ − return
1200
+ Fraction . getFraction ( numerator / gcd ,
1201
+ + return
1202
+ Fraction . getFraction ( numerator / 3 ,
1203
+ denominator / gcd ) ;
1204
+ Mutant 8: adding a condition to a return statement
1205
+ @@ org . apache . commons . lang . math . Fraction
1206
+ :
1207
+ 840 @@
1208
+ return
1209
+ ( getNumerator ( ) == other . getNumerator ( )
1210
+
1211
+ && getDenominator ( ) == other . getDenominator ( ) ) ;
1212
+ +
1213
+ && getDenominator ( ) == other . getDenominator ( ) ) )
1214
+ +
1215
+ | |
1216
+ ( numerator == other . numerator ) ;
1217
+ that it is not able to produce program-specific mutants, i.e.
1218
+ by changing an object by another or a method call with
1219
+ another. In Table 5, we illustrate some example mutants that
1220
+ helped find each of the studied subjects (breaking same tests
1221
+ as the original bug), which µBERT failed to generate when
1222
+ the maximum number of surrounding tokens is limited to
1223
+ 10.
1224
+ 7
1225
+ THREATS TO VALIDITY
1226
+ One external threat to validity concerns the generalisation
1227
+ of our findings and results in the empirical evaluation. To
1228
+ reduce this threat, we used a large number of real bugs
1229
+ from popular open-source projects with their associated
1230
+ developer test-suites, provided by an established and in-
1231
+ dependently built benchmark (i.e. Defects4J [29]). Never-
1232
+ theless, we acknowledge that the results may be different
1233
+ considering projects in different domains.
1234
+ Other threats may arise from our way of assessing the
1235
+ fault detection capability of mutation testing approaches,
1236
+ based on their capability of guiding the testing via a devel-
1237
+ oper/tester simulation in which we assume that the current
1238
+ test suites are complete and the not killed mutants are
1239
+ equivalent. Although we acknowledge that this may not
1240
+ be the case in real-world scenarios, we believe that this
1241
+ process is sufficient to evaluate our approach, particularly
1242
+ considering the fact the test suites provided by Defects4J
1243
+ are relatively strong. Additionally, to mitigate any com-
1244
+ parison threat between the considered approaches, we use
1245
+ consistently and similarly the same test-suites, setups and
1246
+ simulation assumptions in all our study.
1247
+ The choice of our comparison baseline may form other
1248
+ threats to the validity of our findings. While different fault-
1249
+ seeding approaches have been proposed recently, PiTest
1250
+ remains among the most mature and stable mutation test-
1251
+ ing tools for Java programs, thus, forming an appropriate
1252
+ comparison baseline to evaluate our work. Furthermore, we
1253
+ compared our results with those obtained by mutants from
1254
+ different configurations proposed by PiTest, enlarging our
1255
+ study to the different audiences targeted by this latter. We
1256
+ acknowledge however that the results may change when
1257
+ considering other techniques and consider the evaluation
1258
+ of the effectiveness and cost-efficiency of different mutation
1259
+ testing techniques as out of the scope of this paper.
1260
+ Other construct threats may arise from considering the
1261
+ number of mutants analysed as metric to measure the effort
1262
+ required to find a fault. In addition to the fact that this metric
1263
+ has been widely used by the literature [9], [34], [47], we
1264
+ believe that it is intuitive and representative of the global
1265
+ manual effort of the tester in analysing the mutants, dis-
1266
+ carding them or writing tests to kill them. While being the
1267
+ standard in the literature, we acknowledge that this measure
1268
+ does not account for the cost difference between mutants,
1269
+ attributing the same cost to all mutants. This is simply
1270
+ because we do not know the specific effort required to
1271
+ analyse every specific mutant or to write every specific test.
1272
+ Additionally, our cost-efficiency results may be impacted
1273
+ by costs that are not captured with this metric, such as
1274
+ the execution or the developing effort of either CodeBERT,
1275
+ the key component of µBERT, or the set of patterns and
1276
+ execution enhancements over the different releases of PiTest.
1277
+ Nevertheless, we tried to mitigate any major threats that
1278
+ can be induced by the implementation of the different tools,
1279
+ i.e. we reduce the dataset subjects to those on which every
1280
+ approach generated at least one mutant and consider any
1281
+ implementation difference between the approaches as out
1282
+ of the current scope.
1283
+ 8
1284
+ RELATED WORK
1285
+ Since the 1970s, mutation testing has been the main focus
1286
+ of multiple research works [57]. Their findings have proven
1287
+ that artificial faults can be useful in multiple software en-
1288
+ gineering applications, such as testing [47], debugging [37],
1289
+ [48], assessing fault tolerance [42], risk analysis [16], [56] and
1290
+ dependability evaluation [10].
1291
+ Despite this long history of research, the generation
1292
+ of relevant mutants remains an open question. Most of
1293
+ the related research has focused on the design of fault
1294
+
1295
+ 11
1296
+ TABLE 5: Example of mutants generated by µBERT that helped in finding bugs from Defects4J and could not be generated
1297
+ when limiting the maximum number of surrounding tokens to 10.
1298
+ Mutant 1 (JacksonCore-4) : replacing a method call
1299
+ @@ com . fasterxml . jackson . core . u t i l . TextBuffer
1300
+ :
1301
+ 515 @@
1302
+ − unshare ( 1 ) ;
1303
+ + expand ( 1 ) ;
1304
+ Mutant 2 (Closure-26) : replacing an object
1305
+ @@ com . google . j a v a s c r i p t . jscomp . ProcessCommonJSModules
1306
+ :
1307
+ 89 @@
1308
+ − . replaceAll ( Pattern . quote ( F i l e . separator ) , MODULE NAME SEPARATOR)
1309
+ + . replaceAll ( Pattern . quote ( filename ) , MODULE NAME SEPARATOR)
1310
+ Mutant 3 (Closure-35) : replacing a method call
1311
+ @@ com . google . j a v a s c r i p t . jscomp . TypeInference
1312
+ :
1313
+ 1092 @@
1314
+ − scope = traverseChildren (n ,
1315
+ scope ) ;
1316
+ + scope = traverse (n ,
1317
+ scope ) ;
1318
+ Mutant 4 (Lang-27) : replacing a method call
1319
+ @@ org . apache . commons . lang3 . math . NumberUtils
1320
+ :
1321
+ 526 @@
1322
+ − i f
1323
+ ( ! ( f . i s I n f i n i t e ( )
1324
+ | |
1325
+ ( f . floatValue ( ) == 0.0 F && ! allZeros ) ) )
1326
+ {
1327
+ + i f
1328
+ ( ! ( f . i s I n f i n i t e ( )
1329
+ | |
1330
+ ( f . round ( ) == 0.0 F && ! allZeros ) ) )
1331
+ {
1332
+ / /
1333
+ a l s o
1334
+ ” f . f l o a t V a l u e ( ) ”
1335
+ to ” f . s c a l e ( ) ”
1336
+ Mutant 5 (Math-64) : replacing an object
1337
+ @@ org . apache . commons . lang . math . Fraction
1338
+ :
1339
+ 852 @@
1340
+ − for
1341
+ ( i nt
1342
+ j = k ;
1343
+ j < jacobian . length ; ++ j ) {
1344
+ + for
1345
+ ( i nt
1346
+ j = k ;
1347
+ j < beta . length ; ++ j ) {
1348
+ Mutant 6 (Lang-27) : replacing an object
1349
+ @@ org . apache . commons . lang3 . math . NumberUtils
1350
+ :
1351
+ 526 @@
1352
+ − i f
1353
+ ( ! ( f . i s I n f i n i t e ( )
1354
+ | |
1355
+ ( f . floatValue ( ) == 0.0 F && ! allZeros ) ) )
1356
+ {
1357
+ + i f
1358
+ ( ! ( f . i s I n f i n i t e ( )
1359
+ | |
1360
+ ( f . round ( ) == 0.0 F && ! zero ) ) )
1361
+ {
1362
+ patterns (mutation operators) which are usually defined
1363
+ based on the target language grammar [8], [47] then refined
1364
+ through empirical studies [33], [40], [44] aiming at reducing
1365
+ the redundancy and noise among their generated mutants.
1366
+ The continuous advances in this sense were followed by
1367
+ a constant emergence of pattern-based mutation testing
1368
+ tools and releases [17], [35], [39], among which some are
1369
+ becoming popular and widely adopted by researchers and
1370
+ practitioners, such as PiTest [17], from which we consider
1371
+ three configurations as our comparison baseline.
1372
+ Recent research has focused their interest on improving
1373
+ the representativeness of artificial faults aiming at reducing
1374
+ the mutation space to real-like faults. For instance, instead of
1375
+ basing the mutation operators’ design on the programming
1376
+ language grammar, Brown et al. [12] proposed inferring
1377
+ them from real bug fixes. Similarly, Tufano et al. [54] pro-
1378
+ posed a neural machine translation technique that learns
1379
+ how to inject faults from real bug fixes. Along the same
1380
+ line, Patra et al. [50] proposed a semantic-aware learning
1381
+ approach, that learns and then adapts fault patterns to the
1382
+ project of interest. Their results are promising, however,
1383
+ the fact that these techniques depend on the availability
1384
+ of numerous, diverse, comprehensive and untangled fix
1385
+ commits [27] of not coupled faults [43], which is often hard
1386
+ to fulfil in practice, may hinder their performance. Acknowl-
1387
+ edging for the injection location [13], [42], Khanfir et al. [32]
1388
+ combined the usage of information retrieved from bug
1389
+ reports with inverted automated-program-repair patterns to
1390
+ replicate real faults fixable by the original fix-patterns. Their
1391
+ results showed that they can generate faults that mimic real
1392
+ ones, however, their approach remains dependent and lim-
1393
+ ited to the presence of good bug reports. Overall, designing
1394
+ the mutation operators based on the known faults space
1395
+ yields more diverse mutants that represent more fault types.
1396
+ However, these extended operator sets tend to increase the
1397
+ number of generated mutants and consequently the general
1398
+ cost of the mutation campaign i.e. the fault patterns pro-
1399
+ posed by Brown et al. and Khanfir et al. counted also most of
1400
+ the conventional mutators in addition to new ones. Unlike
1401
+ these techniques, µBERT leverages pre-trained models to
1402
+ introduce mutants based on code knowledge instead of the
1403
+ faults one. As code is more available than faults, it offers a
1404
+ more flexible and complete knowledge base than faults, i.e.
1405
+ it perms to overcome the limitations and efforts required 1)
1406
+ to collect clean bug-fixing commits, 2) to capture the faulty
1407
+ behaviour and 3) design fault patterns, be it manually or via
1408
+ machine learning techniques.
1409
+ Aiming at reducing the number of generated mutants,
1410
+ researchers have proposed different strategies to generate
1411
+ relevant mutants. For instance, studies that show that mu-
1412
+ tant strength resides in not only its inducing pattern but also
1413
+ where it is injected [13], [42], motivated the importance of
1414
+ selecting relevant locations to mutate. In this regard, Sun
1415
+ et al. [53] suggest mutating multiple places within diverse
1416
+ program execution paths. Gong et al. [26] also propose the
1417
+ mutation in diverse locations of the program extracted from
1418
+ graph analysis. Similarly, Mirshokraie et al. [41] compute
1419
+ complexity metrics from program executions to extract loca-
1420
+
1421
+ 12
1422
+ tions with good observability to mutate. Other approaches
1423
+ restrict the fault injection on specific locations of the pro-
1424
+ gram, such as the code impacted by the last commits [38],
1425
+ [58] for better usability in continuous integration, or target-
1426
+ ing locations related to a given bug-report [32] to target a
1427
+ specific feature or behaviour, etc. More recent advances have
1428
+ resulted in powerful techniques for cost-effectively selecting
1429
+ mutants, i.e., by avoiding the analysis of redundant mutants
1430
+ (basically, equivalent and subsumed ones) [24], [25], [28]. In
1431
+ particular, the work of Garg et al. [24] utilises the knowledge
1432
+ of mutants’ surrounding context, embedded into the vector
1433
+ space, to predict whether a mutant is likely subsuming
1434
+ or not. In this work, we do not target any specific code
1435
+ part or any narrow use case, but instead, perform fault
1436
+ injection in a brute-force way similarly to mutation testing,
1437
+ by iterating every program statement and masking every
1438
+ involved token.
1439
+ Multiple studies have been focused on the relationship
1440
+ between artificial and real faults [47]. The results of the stud-
1441
+ ies conducted by Ojdanic et al. [45], Papadakis et al. [49],
1442
+ Just et al. [30] and Andrews et al. [9] showed that there
1443
+ is a correlation between tests broken by a bug and tests
1444
+ killing mutants. Meaning that artificial faults can be used
1445
+ as alternatives to real faults in controlled studies. Moreover,
1446
+ the findings of Chekam et al. [14], Frankl et al. [23] and
1447
+ Li et al. [36] show that guiding testing by mutants leads
1448
+ to significantly higher fault revelation capability than the
1449
+ ones of other test adequacy criteria. Based on these findings,
1450
+ we assess our approach based on the relation between the
1451
+ injected and real faults, in terms of breaking tests. More
1452
+ precisely, we conduct a fault detection effectiveness and
1453
+ cost-efficiency study to evaluate our approach’s mutants in
1454
+ guiding testing and compare it to state-of-the-art techniques.
1455
+ Furthermore, we discuss the diversity and readability of
1456
+ µBERT mutants through real examples.
1457
+ The closest related work is a preliminary implementation
1458
+ of µBERT that was recently presented in the 2022 mutation
1459
+ workshop [18]. This implementation, denoted as µBERTconv
1460
+ in our evaluation, includes the conventional mutations (to
1461
+ mask and replace tokens by the model predictiosn), but it
1462
+ does not include the condition-seeding additive mutations
1463
+ that provide major benefits for fault detection. Moreover,
1464
+ µBERTconv was evaluated only on 40 bugs from Defects4J,
1465
+ and compared only to an early version of PiTest (similar
1466
+ to Pit-rv-all). In this work, we perform an extensive exper-
1467
+ imental evaluation including 689 bugs from Defects4J and
1468
+ compare µBERT effectiveness with three different configura-
1469
+ tions from PiTest. Moreover, we show that µBERT finds on
1470
+ average more bugs than µBERTconv without requiring more
1471
+ effort.
1472
+ 9
1473
+ CONCLUSION
1474
+ We presented µBERT; a pre-trained language model based
1475
+ fault injection approach. µBERT provides researchers and
1476
+ practitioners with easy-to-understand “natural” mutantsto
1477
+ help them in writing tests of higher fault revelation capabil-
1478
+ ities.
1479
+ Unlike state-of-the-art approaches, it does neither re-
1480
+ quire nor depend on any kind of faults knowledge or
1481
+ language grammar but instead on the actual code definition
1482
+ and distribution, as written by developers in numerous
1483
+ projects. This facilitates its developing, maintainability, inte-
1484
+ gration and extension to different programming languages.
1485
+ In fact, it reduces the overhead of learning how to mutate,
1486
+ be it via creating and selecting patterns or collecting good
1487
+ bug-fixes and learning from their patches.
1488
+ In a nutshell, µBERT takes as input a given program and
1489
+ replaces different pieces of its code base with predictions
1490
+ made by a pretrained generative language model, produc-
1491
+ ing multiple likely-to-occur mutations. The approach targets
1492
+ diverse business code locations and injects either simple
1493
+ one-token replacement mutants or more complex ones by
1494
+ extending the control-flow conditions. This provides proba-
1495
+ ble developer-like faults impacting different functionalities
1496
+ of the program with higher relevance and lower cost to
1497
+ developers. This is further endorsed by our results where
1498
+ µBERT induces high fault detection test suites at low effort,
1499
+ outperforming state-of-the-art techniques (PiTest), in this
1500
+ regard.
1501
+ We have made our implementation and results avail-
1502
+ able [5] to enable reproducibility and support future re-
1503
+ search.
1504
+ ACKNOWLEDGMENT
1505
+ This work was supported by the Luxembourg National
1506
+ Research Fund (FNR) projects C20/IS/14761415/TestFlakes
1507
+ and TestFast, ref. 12630949.
1508
+ REFERENCES
1509
+ [1]
1510
+ Amazon
1511
+ codewhisperer.
1512
+ https://aws.amazon.com/
1513
+ codewhisperer/.
1514
+ [2]
1515
+ Codebert. https://github.com/microsoft/CodeBERT.
1516
+ [3]
1517
+ Codebert-nt. https://github.com/Ahmedfir/CodeBERT-nt.
1518
+ [4]
1519
+ Github copilot. https://github.com/features/copilot.
1520
+ [5]
1521
+ mberta. https://github.com/Ahmedfir/mBERTa.
1522
+ [6]
1523
+ Pitest. http://pitest.org/.
1524
+ [7]
1525
+ Pitest-rv-plugin. https://github.com/pitest/pitest-rv-plugin.
1526
+ [8]
1527
+ Paul Ammann and Jeff Offutt.
1528
+ Introduction to Software Testing.
1529
+ Cambridge University Press, 2008.
1530
+ [9]
1531
+ James H. Andrews, Lionel C. Briand, Yvan Labiche, and Ak-
1532
+ bar Siami Namin.
1533
+ Using mutation analysis for assessing and
1534
+ comparing testing coverage criteria.
1535
+ IEEE Trans. Software Eng.,
1536
+ 32(8):608–624, 2006.
1537
+ [10] Jean Arlat, Alain Costes, Yves Crouzet, Jean-Claude Laprie, and
1538
+ David Powell.
1539
+ Fault injection and dependability evaluation of
1540
+ fault-tolerant systems. IEEE Trans. Computers, 42(8):913–923, 1993.
1541
+ [11] Moritz Beller, Chu-Pan Wong, Johannes Bader, Andrew Scott,
1542
+ Mateusz Machalica, Satish Chandra, and Erik Meijer.
1543
+ What it
1544
+ would take to use mutation testing in industry - A study at
1545
+ facebook. In 43rd IEEE/ACM International Conference on Software
1546
+ Engineering: Software Engineering in Practice, ICSE (SEIP), pages
1547
+ 268–277. IEEE, 2021.
1548
+ [12] David
1549
+ Bingham
1550
+ Brown,
1551
+ Michael
1552
+ Vaughn,
1553
+ Ben
1554
+ Liblit,
1555
+ and
1556
+ Thomas W. Reps. The care and feeding of wild-caught mutants. In
1557
+ Proceedings of the 2017 11th Joint Meeting on Foundations of Software
1558
+ Engineering, ESEC/FSE, pages 511–522. ACM, 2017.
1559
+ [13] Thierry Titcheu Chekam, Mike Papadakis, Tegawend´e F. Bis-
1560
+ syand´e, Yves Le Traon, and Koushik Sen. Selecting fault revealing
1561
+ mutants. Empirical Software Engineering, 25(1):434–487, 2020.
1562
+ [14] Thierry Titcheu Chekam, Mike Papadakis, Yves Le Traon, and
1563
+ Mark Harman. An empirical study on mutation, statement and
1564
+ branch coverage fault revelation that avoids the unreliable clean
1565
+ program assumption. In Proceedings of the 39th International Con-
1566
+ ference on Software Engineering, ICSE 2017, Buenos Aires, Argentina,
1567
+ May 20-28, 2017, pages 597–608, 2017.
1568
+
1569
+ 13
1570
+ [15] Mark Chen, Jerry Tworek, Heewoo Jun, Qiming Yuan, Hen-
1571
+ rique Ponde de Oliveira Pinto, J Kaplan, H Edwards, Y Burda,
1572
+ N Joseph, G Brockman, et al. Evaluating large language models
1573
+ trained on code.(2021). arXiv preprint arXiv:2107.03374, 2021.
1574
+ [16] J¨orgen Christmansson and Ram Chillarege. Generation of error
1575
+ set that emulates software faults based on field data. In Digest of
1576
+ Papers: FTCS-26, The Twenty-Sixth Annual International Symposium
1577
+ on Fault-Tolerant Computing, 1996, pages 304–313. IEEE Computer
1578
+ Society, 1996.
1579
+ [17] Henry Coles, Thomas Laurent, Christopher Henard, Mike Pa-
1580
+ padakis, and Anthony Ventresque.
1581
+ Pit: A practical mutation
1582
+ testing tool for java (demo). In Proceedings of the 25th International
1583
+ Symposium on Software Testing and Analysis, ISSTA 2016, page
1584
+ 449–452, New York, NY, USA, 2016. Association for Computing
1585
+ Machinery.
1586
+ [18] Renzo Degiovanni and Mike Papadakis. µbert: Mutation testing
1587
+ using pre-trained language models.
1588
+ In 15th IEEE International
1589
+ Conference on Software Testing, Verification and Validation Workshops
1590
+ ICST Workshops 2022, Valencia, Spain, April 4-13, 2022, pages 160–
1591
+ 169. IEEE, 2022.
1592
+ [19] Richard A. DeMillo, Richard J. Lipton, and Frederick G. Sayward.
1593
+ Hints on test data selection: Help for the practicing programmer.
1594
+ IEEE Computer, 11(4):34–41, 1978.
1595
+ [20] Jacob
1596
+ Devlin,
1597
+ Ming-Wei
1598
+ Chang,
1599
+ Kenton
1600
+ Lee,
1601
+ and
1602
+ Kristina
1603
+ Toutanova. Bert: Pre-training of deep bidirectional transformers
1604
+ for language understanding. arXiv preprint arXiv:1810.04805, 2018.
1605
+ [21] JDT Eclipse. Eclipse java development tools (jdt), 2013.
1606
+ [22] Zhangyin Feng, Daya Guo, Duyu Tang, Nan Duan, Xiaocheng
1607
+ Feng, Ming Gong, Linjun Shou, Bing Qin, Ting Liu, Daxin Jiang,
1608
+ and Ming Zhou. Codebert: A pre-trained model for programming
1609
+ and natural languages.
1610
+ In Proceedings of the 2020 Conference on
1611
+ Empirical Methods in Natural Language Processing: Findings, EMNLP,
1612
+ volume EMNLP 2020 of Findings of ACL, pages 1536–1547. Asso-
1613
+ ciation for Computational Linguistics, 2020.
1614
+ [23] Phyllis G. Frankl, Stewart N. Weiss, and Cang Hu. All-uses vs
1615
+ mutation testing: An experimental comparison of effectiveness. J.
1616
+ Syst. Softw., 38(3):235–253, 1997.
1617
+ [24] Aayush Garg, Milos Ojdanic, Renzo Degiovanni, Thierry Titcheu
1618
+ Chekam, Mike Papadakis, and Yves Le Traon.
1619
+ Cerebro: Static
1620
+ subsuming mutant selection. IEEE Trans. Software Eng.
1621
+ [25] Rohit Gheyi, M´arcio Ribeiro, Beatriz Souza, Marcio Augusto
1622
+ Guimar˜aes, Leo Fernandes, Marcelo d’Amorim, Vander Alves,
1623
+ Leopoldo Teixeira, and Baldoino Fonseca.
1624
+ Identifying method-
1625
+ level mutation subsumption relations using Z3. Inf. Softw. Technol.,
1626
+ 132:106496, 2021.
1627
+ [26] Dunwei Gong, Gongjie Zhang, Xiangjuan Yao, and Fanlin Meng.
1628
+ Mutant reduction based on dominance relation for weak mutation
1629
+ testing. Information & Software Technology, 81:82–96, 2017.
1630
+ [27] Kim Herzig and Andreas Zeller. Untangling changes. Unpublished
1631
+ manuscript, September, 37:38–40, 2011.
1632
+ [28] Claudinei Brito Junior, Vinicius H. S. Durelli, Rafael Serapilha
1633
+ Durelli, Simone R. S. Souza, Auri M. R. Vincenzi, and M´arcio Ed-
1634
+ uardo Delamaro. A preliminary investigation into using machine
1635
+ learning algorithms to identify minimal and equivalent mutants.
1636
+ In 13th IEEE International Conference on Software Testing, Verification
1637
+ and Validation Workshops, ICSTW, pages 304–313. IEEE, 2020.
1638
+ [29] Ren´e Just, Darioush Jalali, and Michael D. Ernst.
1639
+ Defects4j: A
1640
+ database of existing faults to enable controlled testing studies for
1641
+ java programs. In Proceedings of the 2014 International Symposium
1642
+ on Software Testing and Analysis, ISSTA 2014, page 437–440, New
1643
+ York, NY, USA, 2014. Association for Computing Machinery.
1644
+ [30] Ren´e Just, Darioush Jalali, Laura Inozemtseva, Michael D. Ernst,
1645
+ Reid Holmes, and Gordon Fraser. Are mutants a valid substitute
1646
+ for real faults in software testing?
1647
+ In Proceedings of the 22nd
1648
+ ACM SIGSOFT International Symposium on Foundations of Software
1649
+ Engineering, (FSE-22), Hong Kong, China, November 16 - 22, 2014,
1650
+ pages 654–665. ACM, 2014.
1651
+ [31] Ahmed Khanfir, Matthieu Jimenez, Mike Papadakis, and Yves Le
1652
+ Traon. Codebert-nt: code naturalness via codebert. arXiv preprint
1653
+ arXiv:2208.06042, 2022.
1654
+ [32] Ahmed Khanfir, Anil Koyuncu, Mike Papadakis, Maxime Cordy,
1655
+ Tegawende F. Bissyand´e, Jacques Klein, and Yves Le Traon. Ibir:
1656
+ Bug report driven fault injection. ACM Trans. Softw. Eng. Methodol.,
1657
+ may 2022.
1658
+ [33] Marinos Kintis, Mike Papadakis, Andreas Papadopoulos, Evan-
1659
+ gelos Valvis, Nicos Malevris, and Yves Le Traon. How effective
1660
+ are mutation testing tools? an empirical analysis of java mutation
1661
+ testing tools with manual analysis and real faults. Empir. Softw.
1662
+ Eng., 23(4):2426–2463, 2018.
1663
+ [34] Bob Kurtz, Paul Ammann, Jeff Offutt, M´arcio Eduardo Delamaro,
1664
+ Mariet Kurtz, and Nida G¨okc¸e. Analyzing the validity of selective
1665
+ mutation with dominator mutants.
1666
+ In Proceedings of the 24th
1667
+ ACM SIGSOFT International Symposium on Foundations of Software
1668
+ Engineering, FSE 2016, Seattle, WA, USA, November 13-18, 2016,
1669
+ pages 571–582, 2016.
1670
+ [35] T. Laurent, M. Papadakis, M. Kintis, C. Henard, Y. L. Traon, and
1671
+ A. Ventresque.
1672
+ Assessing and improving the mutation testing
1673
+ practice of pit. In 2017 IEEE International Conference on Software
1674
+ Testing, Verification and Validation (ICST), pages 430–435, March
1675
+ 2017.
1676
+ [36] Nan Li, Upsorn Praphamontripong, and Jeff Offutt. An experi-
1677
+ mental comparison of four unit test criteria: Mutation, edge-pair,
1678
+ all-uses and prime path coverage. In Second International Conference
1679
+ on Software Testing Verification and Validation, ICST, 2009, Workshops
1680
+ Proceedings, pages 220–229. IEEE Computer Society, 2009.
1681
+ [37] Yiling Lou, Ali Ghanbari, Xia Li, Lingming Zhang, Haotian Zhang,
1682
+ Dan Hao, and Lu Zhang. Can automated program repair refine
1683
+ fault localization? a unified debugging approach. In ISSTA ’20:
1684
+ 29th ACM SIGSOFT International Symposium on Software Testing and
1685
+ Analysis, Virtual Event, USA, July 18-22, 2020, pages 75–87. ACM,
1686
+ 2020.
1687
+ [38] Wei Ma, Thomas Laurent, Milos Ojdanic, Thierry Titcheu Chekam,
1688
+ Anthony Ventresque, and Mike Papadakis. Commit-aware muta-
1689
+ tion testing. In Proceedings of the 36th IEEE International Conference
1690
+ on Software Maintenance and Evolution, ICSME, 2020.
1691
+ [39] Yu-Seung Ma, Jeff Offutt, and Yong Rae Kwon.
1692
+ Mujava: an
1693
+ automated class mutation system. Softw. Test. Verification Reliab.,
1694
+ 15(2):97–133, 2005.
1695
+ [40] Micha¨el Marcozzi, S´ebastien Bardin, Nikolai Kosmatov, Mike Pa-
1696
+ padakis, Virgile Prevosto, and Lo¨ıc Correnson.
1697
+ Time to clean
1698
+ your test objectives. In Michel Chaudron, Ivica Crnkovic, Marsha
1699
+ Chechik, and Mark Harman, editors, Proceedings of the 40th Inter-
1700
+ national Conference on Software Engineering, ICSE 2018, Gothenburg,
1701
+ Sweden, May 27 - June 03, 2018, pages 456–467. ACM, 2018.
1702
+ [41] Shabnam Mirshokraie, Ali Mesbah, and Karthik Pattabiraman.
1703
+ Guided mutation testing for javascript web applications.
1704
+ IEEE
1705
+ Trans. Software Eng., 41(5):429–444, 2015.
1706
+ [42] Roberto Natella, Domenico Cotroneo, Jo˜ao Dur˜aes, and Henrique
1707
+ Madeira. On fault representativeness of software fault injection.
1708
+ IEEE Trans. Software Eng., 39(1):80–96, 2013.
1709
+ [43] A. Jefferson Offutt. Investigations of the software testing coupling
1710
+ effect. ACM Trans. Softw. Eng. Methodol., 1(1):5–20, 1992.
1711
+ [44] A. Jefferson Offutt, Ammei Lee, Gregg Rothermel, Roland H.
1712
+ Untch, and Christian Zapf.
1713
+ An experimental determination of
1714
+ sufficient mutant operators.
1715
+ ACM Trans. Softw. Eng. Methodol.,
1716
+ 5(2):99–118, 1996.
1717
+ [45] Milos Ojdanic, Aayush Garg, Ahmed Khanfir, Renzo Degiovanni,
1718
+ Mike Papadakis, and Yves Le Traon.
1719
+ Syntactic vs. semantic
1720
+ similarity of artificial and real faults in mutation testing studies.
1721
+ arXiv preprint arXiv:2112.14508, 2021.
1722
+ [46] Mike Papadakis, Thierry Titcheu Chekam, and Yves Le Traon.
1723
+ Mutant quality indicators. In 2018 IEEE International Conference on
1724
+ Software Testing, Verification and Validation Workshops, pages 32–39.
1725
+ IEEE Computer Society, 2018.
1726
+ [47] Mike Papadakis, Marinos Kintis, Jie Zhang, Yue Jia, Yves Le Traon,
1727
+ and Mark Harman. Chapter six - mutation testing advances: An
1728
+ analysis and survey. Advances in Computers, 112:275–378, 2019.
1729
+ [48] Mike Papadakis and Yves Le Traon. Metallaxis-fl: mutation-based
1730
+ fault localization. Software Testing, Verification and Reliability, 25(5-
1731
+ 7):605–628, 2015.
1732
+ [49] Mike Papadakis, Donghwan Shin, Shin Yoo, and Doo-Hwan Bae.
1733
+ Are mutation scores correlated with real fault detection?: a large
1734
+ scale empirical study on the relationship between mutants and
1735
+ real faults.
1736
+ In Proceedings of the 40th International Conference on
1737
+ Software Engineering, ICSE 2018, Gothenburg, Sweden, May 27 - June
1738
+ 03, 2018, pages 537–548, 2018.
1739
+ [50] Jibesh Patra and Michael Pradel.
1740
+ Semantic bug seeding: A
1741
+ learning-based approach for creating realistic bugs.
1742
+ ESEC/FSE
1743
+ 2021, page 906–918, New York, NY, USA, 2021. Association for
1744
+ Computing Machinery.
1745
+ [51] Renaud Pawlak, Martin Monperrus, Nicolas Petitprez, Carlos
1746
+ Noguera, and Lionel Seinturier. Spoon: A Library for Implement-
1747
+ ing Analyses and Transformations of Java Source Code. Software:
1748
+ Practice and Experience, 46:1155–1179, 2015.
1749
+
1750
+ 14
1751
+ [52] Ana Bel´en S´anchez, Pedro Delgado-P´erez, Inmaculada Medina-
1752
+ Bulo, and Sergio Segura. Mutation testing in the wild: findings
1753
+ from github. Empir. Softw. Eng., 27(6):132, 2022.
1754
+ [53] Chang-ai Sun, Feifei Xue, Huai Liu, and Xiangyu Zhang.
1755
+ A
1756
+ path-aware approach to mutant reduction in mutation testing.
1757
+ Information & Software Technology, 81:65–81, 2017.
1758
+ [54] Michele Tufano, Cody Watson, Gabriele Bavota, Massimiliano Di
1759
+ Penta, Martin White, and Denys Poshyvanyk. Learning how to
1760
+ mutate source code from bug-fixes.
1761
+ In 2019 IEEE International
1762
+ Conference on Software Maintenance and Evolution, ICSME 2019,
1763
+ pages 301–312. IEEE, 2019.
1764
+ [55] Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit,
1765
+ Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin.
1766
+ Attention is all you need. Advances in neural information processing
1767
+ systems, 30, 2017.
1768
+ [56] Jeffrey M. Voas, Frank Charron, Gary McGraw, Keith W. Miller,
1769
+ and Michael Friedman. Predicting how badly ”good” software
1770
+ can behave. IEEE Softw., 14(4):73–83, 1997.
1771
+ [57] Jeffrey M. Voas and Gary McGraw. Software Fault Injection: Inocu-
1772
+ lating Programs against Errors. John Wiley & Sons, Inc., USA, 1997.
1773
+ [58] Lingming Zhang, Darko Marinov, Lu Zhang, and Sarfraz Khur-
1774
+ shid.
1775
+ Regression mutation testing.
1776
+ ISSTA 2012, page 331–341,
1777
+ New York, NY, USA, 2012. Association for Computing Machinery.
1778
+
-NE1T4oBgHgl3EQf8QVm/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
-dA0T4oBgHgl3EQfPP9i/content/2301.02172v1.pdf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:92157812b00c20600fd85ab54fd0d5af8ab73b11fce8445fe698949727a0f515
3
+ size 1478315
-dA0T4oBgHgl3EQfPP9i/vector_store/index.faiss ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a70158c6ddb8955f8e391c1af50d669500a09b945d650514d074da2b9f75e751
3
+ size 3735597
-dA0T4oBgHgl3EQfPP9i/vector_store/index.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d9781357efdb40e7b915e92e0188bca2b36b784b5c4a112f60fe55a031467187
3
+ size 126440
.gitattributes CHANGED
@@ -4670,3 +4670,60 @@ w9FPT4oBgHgl3EQfPzQl/content/2301.13039v1.pdf filter=lfs diff=lfs merge=lfs -tex
4670
  YtFJT4oBgHgl3EQf6y0f/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
4671
  79E5T4oBgHgl3EQfQQ7U/content/2301.05512v1.pdf filter=lfs diff=lfs merge=lfs -text
4672
  etFST4oBgHgl3EQfGDhd/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4670
  YtFJT4oBgHgl3EQf6y0f/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
4671
  79E5T4oBgHgl3EQfQQ7U/content/2301.05512v1.pdf filter=lfs diff=lfs merge=lfs -text
4672
  etFST4oBgHgl3EQfGDhd/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
4673
+ J9FOT4oBgHgl3EQfyjQh/content/2301.12928v1.pdf filter=lfs diff=lfs merge=lfs -text
4674
+ w9FRT4oBgHgl3EQfhDe7/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
4675
+ G9E2T4oBgHgl3EQf-wmW/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
4676
+ VNAzT4oBgHgl3EQfl_0n/content/2301.01555v1.pdf filter=lfs diff=lfs merge=lfs -text
4677
+ _NE5T4oBgHgl3EQfSA5Z/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
4678
+ w9FPT4oBgHgl3EQfPzQl/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
4679
+ 79FLT4oBgHgl3EQfsi_P/content/2301.12148v1.pdf filter=lfs diff=lfs merge=lfs -text
4680
+ SdE2T4oBgHgl3EQfswgq/content/2301.04062v1.pdf filter=lfs diff=lfs merge=lfs -text
4681
+ 69E1T4oBgHgl3EQfBgJT/content/2301.02852v1.pdf filter=lfs diff=lfs merge=lfs -text
4682
+ VNAzT4oBgHgl3EQfl_0n/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
4683
+ 79E5T4oBgHgl3EQfQQ7U/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
4684
+ 4dFAT4oBgHgl3EQfExzK/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
4685
+ etFST4oBgHgl3EQfGDhd/content/2301.13720v1.pdf filter=lfs diff=lfs merge=lfs -text
4686
+ K9AyT4oBgHgl3EQfTveL/content/2301.00112v1.pdf filter=lfs diff=lfs merge=lfs -text
4687
+ kNAyT4oBgHgl3EQfyPlA/content/2301.00680v1.pdf filter=lfs diff=lfs merge=lfs -text
4688
+ U9E_T4oBgHgl3EQfxhwl/content/2301.08312v1.pdf filter=lfs diff=lfs merge=lfs -text
4689
+ xtE0T4oBgHgl3EQf-gJ8/content/2301.02815v1.pdf filter=lfs diff=lfs merge=lfs -text
4690
+ hNE3T4oBgHgl3EQfgArH/content/2301.04558v1.pdf filter=lfs diff=lfs merge=lfs -text
4691
+ M9AzT4oBgHgl3EQfkv2l/content/2301.01537v1.pdf filter=lfs diff=lfs merge=lfs -text
4692
+ pNE4T4oBgHgl3EQfvQ0x/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
4693
+ p9E2T4oBgHgl3EQf0ggt/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
4694
+ K9AyT4oBgHgl3EQfTveL/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
4695
+ xtE0T4oBgHgl3EQf-gJ8/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
4696
+ J9FOT4oBgHgl3EQfyjQh/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
4697
+ _9AyT4oBgHgl3EQfqvjk/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
4698
+ ddFAT4oBgHgl3EQf6h5C/content/2301.08739v1.pdf filter=lfs diff=lfs merge=lfs -text
4699
+ MdFIT4oBgHgl3EQfcCt7/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
4700
+ p9E2T4oBgHgl3EQf0ggt/content/2301.04141v1.pdf filter=lfs diff=lfs merge=lfs -text
4701
+ KdFIT4oBgHgl3EQfaSsg/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
4702
+ VdE5T4oBgHgl3EQfbw_i/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
4703
+ M9AzT4oBgHgl3EQfkv2l/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
4704
+ etE0T4oBgHgl3EQfXAB2/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
4705
+ ktAzT4oBgHgl3EQfNfuY/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
4706
+ ktAzT4oBgHgl3EQfNfuY/content/2301.01150v1.pdf filter=lfs diff=lfs merge=lfs -text
4707
+ 49AzT4oBgHgl3EQffvx7/content/2301.01457v1.pdf filter=lfs diff=lfs merge=lfs -text
4708
+ -9AyT4oBgHgl3EQfRPbk/content/2301.00064v1.pdf filter=lfs diff=lfs merge=lfs -text
4709
+ qtE4T4oBgHgl3EQfvw2m/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
4710
+ 0NAzT4oBgHgl3EQfDPod/content/2301.00971v1.pdf filter=lfs diff=lfs merge=lfs -text
4711
+ KE[[:space:]]Research[[:space:]]Center[[:space:]]Intro/content/知识工程研究中心汇报20230710.pdf filter=lfs diff=lfs merge=lfs -text
4712
+ adE4T4oBgHgl3EQfOwzB/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
4713
+ u9FAT4oBgHgl3EQfiB2Y/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
4714
+ UtE1T4oBgHgl3EQfuwWT/content/2301.03392v1.pdf filter=lfs diff=lfs merge=lfs -text
4715
+ 4dE0T4oBgHgl3EQfeQDL/content/2301.02389v1.pdf filter=lfs diff=lfs merge=lfs -text
4716
+ EdAzT4oBgHgl3EQfG_ss/content/2301.01036v1.pdf filter=lfs diff=lfs merge=lfs -text
4717
+ -dA0T4oBgHgl3EQfPP9i/content/2301.02172v1.pdf filter=lfs diff=lfs merge=lfs -text
4718
+ U9AyT4oBgHgl3EQf8vpk/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
4719
+ X9A0T4oBgHgl3EQfFf93/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
4720
+ S9E3T4oBgHgl3EQfZwr5/content/2301.04501v1.pdf filter=lfs diff=lfs merge=lfs -text
4721
+ qdFQT4oBgHgl3EQftDZs/content/2301.13390v1.pdf filter=lfs diff=lfs merge=lfs -text
4722
+ eNE0T4oBgHgl3EQf5gKT/content/2301.02752v1.pdf filter=lfs diff=lfs merge=lfs -text
4723
+ qdFQT4oBgHgl3EQftDZs/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
4724
+ UtE1T4oBgHgl3EQfuwWT/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
4725
+ X9A0T4oBgHgl3EQfFf93/content/2301.02033v1.pdf filter=lfs diff=lfs merge=lfs -text
4726
+ KE[[:space:]]Research[[:space:]]Center[[:space:]]Intro/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
4727
+ VdE1T4oBgHgl3EQfbATf/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
4728
+ -dA0T4oBgHgl3EQfPP9i/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
4729
+ VdE1T4oBgHgl3EQfbATf/content/2301.03169v1.pdf filter=lfs diff=lfs merge=lfs -text
0NAzT4oBgHgl3EQfDPod/content/2301.00971v1.pdf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e81e74f1f6d0644b86937b12a30433baa60d91c8fae4a60a9f2532680728d695
3
+ size 2309881
0NAzT4oBgHgl3EQfDPod/vector_store/index.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:940488208b17c62fc473f6cb5d06bc8faceeb91ce0c15c1c5737ab8e01b2f61f
3
+ size 173067
1dE2T4oBgHgl3EQfNQZD/content/tmp_files/2301.03734v1.pdf.txt ADDED
@@ -0,0 +1,403 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Exoshuffle-CloudSort
2
+ FRANK SIFEI LUAN∗, UC Berkeley
3
+ STEPHANIE WANG, UC Berkeley and Anyscale
4
+ SAMYUKTA YAGATI, UC Berkeley
5
+ SEAN KIM, UC Berkeley
6
+ KENNETH LIEN, UC Berkeley
7
+ ISAAC ONG, UC Berkeley
8
+ TONY HONG, UC Berkeley
9
+ SANGBIN CHO, Anyscale
10
+ ERIC LIANG, Anyscale
11
+ ION STOICA, UC Berkeley and Anyscale
12
+ 1
13
+ INTRODUCTION
14
+ We present Exoshuffle-CloudSort, a sorting application running on top of Ray using the Exoshuffle archi-
15
+ tecture [4]. Exoshuffle-CloudSort runs on Amazon EC2, with input and output data stored on Amazon S3.
16
+ Using 40× i4i.4xlarge workers, Exoshuffle-CloudSort completes the 100 TB CloudSort Benchmark (Indy
17
+ category [6]) in 5378 seconds, with an average total cost of $97.
18
+ 2
19
+ IMPLEMENTATION
20
+ 2.1
21
+ Overview
22
+ Exoshuffle-CloudSort is a distributed futures program running on top of Ray, a task-based distributed
23
+ execution system. The program acts as the control plane to coordinate map and reduce tasks; the Ray system
24
+ acts as the data plane, responsible for executing tasks, transferring blocks, and recovering from failures.
25
+ Exoshuffle-CloudSort implements a two-stage external sort algorithm. The first stage is map and shuffle.
26
+ Each map task reads an input partition, sorts it, and partitions the result into 𝑊 output partitions, each sent
27
+ to a merger on a worker node. A merger receives 𝑊 map output partitions, merges and sorts them, and
28
+ further partition the result into 𝑅/𝑊 output partitions, all of which are spilled to local disk.
29
+ The second stage is reduce. Once the map and shuffle stage finishes, each reduce task reads 𝑊 shuffled
30
+ partitions, merges and sorts them, and writes the final output partition.
31
+ For the 100 TB CloudSort Benchmark, we set the following parameters:
32
+ • Total data size is 100 TB.
33
+ • Number of input partitions 𝑀 = 50 000. Each input partition is 2 GB.
34
+ ∗Author’s address: [email protected], 465 Soda Hall, Berkeley, CA, USA.
35
+ 1
36
+ arXiv:2301.03734v1 [cs.DC] 10 Jan 2023
37
+
38
+ 2
39
+ Luan et al.
40
+ • Number of workers 𝑊 = 40.
41
+ • Number of output partitions 𝑅 = 25 000.
42
+ 2.2
43
+ Preparation
44
+ The first step in Exoshuffle-CloudSort is to compute the partition boundary values. For a sort record with
45
+ 10-byte key, we view the first 8 bytes as a 64-bit unsigned integer partition key. We partition the key space
46
+ [0, 264 − 1) into 𝑅 = 25 000 equal ranges, such that all the records within a key range should be sent to one
47
+ reducer.
48
+ Every 𝑅1 = 𝑅/𝑊 = 625 reducer ranges are combined into a worker range, and records in each worker
49
+ range will be sent to one worker node. This yields 𝑊 = 40 equally-partitioned worker ranges.
50
+ 2.3
51
+ Map and Shuffle Stage
52
+ In the map and shuffle stage, Exoshuffle-CloudSort schedules the 𝑀 = 50 000 map tasks onto all worker
53
+ nodes. In our experiments we set the map parallelism, i.e. the number of map tasks running on a single
54
+ worker node, to be 3/4 of the total number of vCPU cores. Extra tasks are queued on the driver node.
55
+ Whenever a worker node finishes a map task, the driver assigns a new task from the queue to this node.
56
+ In a map task, we first download the input partition from S3. We then sort the input data in memory,
57
+ then partition it into 𝑊 = 40 slices. Each slice is eagerly sent to a merge controller on each worker. The
58
+ map task returns when all slices are sent.
59
+ On the receiving end, the merge controller accumulates the map blocks in memory until a threshold
60
+ is reached. We set the threshold to 40 blocks, or about 2 GB of data. Once the threshold is reached, the
61
+ controller launches a merge task to merge the already-sorted map blocks, and further partitions it into
62
+ 𝑅1 = 625 merged blocks, each corresponding to a reduce task on this node. These blocks are spilled to the
63
+ local SSD for use by the reducers.
64
+ The merge parallelism is set to be the same as the map parallelism. When the number of merge tasks
65
+ reaches the maximum parallelism, and the merge controller’s in-memory buffer is filled up, it will hold off
66
+ acknowledging the receipt of a map block until a merge task finishes and a new merge task can launch. This
67
+ effectively creates back pressure to the map task scheduler to ensure the map, shuffle, and merge progresses
68
+ are in sync.
69
+ In our experiments, the average map task duration is 24 seconds; 15 seconds are used for downloading
70
+ input data. The average shuffle time (i.e. time to send and receive blocks) is 7 seconds. The merge task takes
71
+ 17 seconds on average.
72
+ 2.4
73
+ Reduce Stage
74
+ Once all map and merge tasks finish, Exoshuffle-CloudSort enters the reduce stage. Each reduce task loads
75
+ 𝑅1 = 625 from the local SSD, merges them, and uploads the sorted output partition to S3. In our experiments,
76
+ each reduce task takes 22 seconds on average.
77
+
78
+ Exoshuffle-CloudSort
79
+ 3
80
+ 2.5
81
+ The Execution System
82
+ A highlight of the Exoshuffle architecture is that the application program only implements the control plane
83
+ logic, and the distributed futures system, Ray, handles execution. This is reflected in Exoshuffle-CloudSort.
84
+ Here is an incomplete list of features provided by Ray that we take “for free”:
85
+ • Task scheduling: The program specifies when and where to schedule tasks; the system handles the
86
+ RPC, serialization, and other bookkeeping.
87
+ • Network transfer: The program instructs data to be transferred by passing distributed futures as task
88
+ arguments; the system implements high-performance network transfer.
89
+ • Memory management and disk spilling: The program manipulates data references in a virtual, infinite
90
+ address space; the system uses reference counting to manage distributed memory, spills objects to
91
+ local disks when memory is low, and restores objects from local disks when they are needed.
92
+ • Pipelining of network and disk I/O: The network transfer, spilling and recovery of objects are trans-
93
+ parent to the application and are performed asynchronously. For example, the system shuffles map
94
+ output blocks while other map and merge tasks are running; it spills merge task output to disk while
95
+ other merge tasks are executing, and it restores merged blocks while reduce tasks are executing.
96
+ • Fault tolerance: this is transparent to the application: the system automatically retries the operation
97
+ when it encounters network failures and worker process failures.
98
+ For more details, we refer the reader to the Ray Architecture Whitepaper [7], the ownership design for
99
+ distributed futures systems [8], and the Exoshuffle paper [4].
100
+ 2.6
101
+ Source Code
102
+ Exoshuffle-CloudSort is implemented in about 1000 lines of Python, and about 300 lines of C++. The
103
+ C++ component implements two functionalities: sorting and partitioning records, and merging sorted
104
+ record arrays. Exoshuffle-CloudSort runs on top of Ray, which is implemented in Python and C++. All of
105
+ Exoshuffle-CloudSort’s source code is available at https://github.com/exoshuffle/cloudsort.
106
+ 3
107
+ EVALUATION
108
+ 3.1
109
+ Environment Setup
110
+ We run Exoshuffle-CloudSort on AWS on a compute cluster configured as follows:
111
+ • 1× r6i.2xlarge master node. This node runs on 8 cores of an Intel Xeon 8375C CPU at 2.9 GHz, and
112
+ 64 GiB memory.
113
+ • 40× i4i.4xlarge worker nodes. Each node runs on 16 cores of an Intel Xeon 8375C CPU at 2.9 GHz,
114
+ and 128 GiB memory. Each node has a directly-attached 3.75 TB AWS Nitro NVMe SSD.
115
+ • Each node is attached with a 40 GiB Amazon EBS General Purpose SSD (gp3) volume.
116
+ The software stack is configured as follows:
117
+
118
+ 4
119
+ Luan et al.
120
+ • Ubuntu 22.04.1 LTS, Linux kernel version 5.15.0-1022-aws.
121
+ • XFS 5.13.0 filesystem.
122
+ • Intel oneAPI DPC++/C++ Compiler 2022.2.0.20220730.
123
+ • Python 3.9.13.
124
+ • Ray 2.1.0.
125
+ We measure the raw system I/O performance on the worker nodes using standard benchmarking tools:
126
+ • Network bandwidth: 25 Gbps between nodes, benchmarked with iperf.
127
+ • SSD: 2.9 GB/s read, 2.2 GB/s write, benchmarked with fio.
128
+ For storage, we use 40 buckets on Amazon S3 and randomly distribute the input and output partitions
129
+ across the buckets.
130
+ 3.2
131
+ Benchmark Setup
132
+ Generating Input. We use gensort version 1.5 as provided by the Sort Benchmark committee [5]. We
133
+ run the command gensort -c -b{offset} {size} {path} to generate each partition. {size} is fixed at
134
+ 𝑃 = 20 000 000 such that each partition is exactly 2 GB. {offset} takes the values {𝑖 · 𝑃 : 0 ≤ 𝑖 < 𝑀} where
135
+ the number of input partitions 𝑀 = 50 000. {path} is a unique path in tmpfs. -c provides data checksum for
136
+ validation. After generating an input file, we randomly choose a bucket and upload the partition to S3. We
137
+ use Ray to schedule the 50 000 input generation tasks to all 40 worker nodes. The result is aggregated as an
138
+ input manifest file, saved for use by Exoshuffle-CloudSort to locate the sort input.
139
+ Validating Output. Exoshuffle-CloudSort produces an output manifest file containing the bucket and keys
140
+ of each output partition on S3. In each validation task, we first download the output partition to tmpfs, then
141
+ run the command valsort -o {sumpath} {path} to validate the ordering of records in each partition. We
142
+ use Ray to schedule the 25 000 output validation tasks to all 40 worker nodes. We concatenate the contents
143
+ of the summary files from each validation task, then run valsort -s to validate the total ordering, and
144
+ generate the total output checksum. Finally, we compare the output checksum with the input checksum to
145
+ verify data integrity.
146
+ 3.3
147
+ Experimental Results
148
+ 3.3.1
149
+ Job Completion Time. On November 10, 2022, we ran the 100 TB CloudSort Benchmark in the AWS
150
+ US West (Oregon, us-west-2) region with the setup described above. We first generated the input data on
151
+ Amazon S3, then ran Exoshuffle-CloudSort 3 times, each followed by a validation step. All 3 runs succeeded
152
+ with the same output checksum as the input, indicating all bytes are preserved in the sort. Table 1 reports
153
+ the job completion times of each run. The average job completion time is 5378 seconds, or 1.4939 hours.
154
+ Figure 1 shows the system utilizations of all worker nodes in the compute cluster during run #1 of the
155
+ 100 TB CloudSort Benchmark.
156
+
157
+ Exoshuffle-CloudSort
158
+ 5
159
+ Run
160
+ Map & Shuffle Time
161
+ Reduce Time
162
+ Total Job Completion Time
163
+ #1
164
+ 3509 s
165
+ 1852 s
166
+ 5361 s
167
+ #2
168
+ 3496 s
169
+ 1852 s
170
+ 5348 s
171
+ #3
172
+ 3520 s
173
+ 1906 s
174
+ 5426 s
175
+ Average
176
+ 3508 s
177
+ 1870 s
178
+ 5378 s
179
+ Table 1. Job completion times of Exoshuffle-CloudSort on the 100 TB CloudSort Benchmark.
180
+ Fig. 1. Cluster utilization during run #1 of the 100 TB CloudSort Benchmark. Each thick line represents the median
181
+ system utilization of all worker nodes; the highest and lowest lines represent the maximum and minimum utilization
182
+ among all worker nodes, respectively.
183
+ 3.3.2
184
+ Total Cost of Ownership. The total job cost comprises of two parts: compute cost (Amazon EC2), and
185
+ the storage cost (Amazon S3). The storage cost is further divided into data storage cost and data access cost.
186
+ Compute Cost. The compute cost is calculated as the compute cluster’s hourly cost times the job completion
187
+ time. The total hourly cost is calculated as follows:
188
+ Total Hourly Compute Cost = Master Node Hourly Cost
189
+ + Worker Node Hourly Cost × Number of Workers
190
+ + EBS Volume Hourly Cost × (Number of Workers + 1)
191
+ (1)
192
+ We obtain the compute instance hourly costs from the Amazon EC2 on-demand pricing information [2].
193
+ For EBS, we use the Amazon EBS monthly price [1] divided by the average number of hours in a month
194
+ ( 365×24
195
+ 12
196
+ = 730) as the hourly price. The hourly cost of a 40 GiB gp3 volume is $0.08/730 × 40 = $0.0044. Now
197
+ we plug the cost variables into Equation (1):
198
+
199
+ CPU
200
+ Memory
201
+ Application Progress
202
+ 50000
203
+ 100%
204
+ 70 GB
205
+ 40000
206
+ 60 GB
207
+ 80%
208
+ 30000
209
+ 50 GB
210
+ 20000
211
+ 60%
212
+ 40 GB
213
+ 10000
214
+ 30 GB
215
+ 40%
216
+ 20 GB
217
+ 02:40
218
+ 02:50
219
+ 03:00
220
+ 03:10
221
+ 03:20
222
+ 03:30
223
+ 03:40
224
+ 03:5004:00
225
+ 10 GB
226
+ - map_in_progress
227
+ 20%
228
+ reduce_in_progress
229
+ reduce_in_progress
230
+ reduce_in_progress
231
+ 0 B
232
+ 02:40
233
+ 02:50
234
+ 03:0003:1003:20
235
+ 03:3003:40
236
+ 03:5004:00
237
+ map_completed
238
+ map_completec
239
+ map_completed
240
+ map_completed
241
+ 0%
242
+ 02:4002:50
243
+ 03:10
244
+ 03:20
245
+ 03:30
246
+ 03:40
247
+ 03:50
248
+ 04:00
249
+ median objmem
250
+ - reducer_completed
251
+ reducer_completed reducer_completed
252
+ - reducer_completed
253
+ 03:00
254
+ min objmem
255
+ max objmem
256
+ min workmem
257
+ median cpu - min cpu - max cpu
258
+ max workmem
259
+ merge_in_progress
260
+ merge_in_progress
261
+ merge_in_progress
262
+ nerge_in_progress
263
+ NVMe Disk I/0
264
+ Network I/0
265
+ Disk Usage
266
+ 7 GB/s
267
+ 3 GB/s
268
+ 100%
269
+ 6 GB/s
270
+ 2.50 GB/s
271
+ 80%
272
+ 5 GB/s
273
+ 2 GB/s
274
+ 4 GB/s
275
+ 60%
276
+ 1.50 GB/s
277
+ 3 GB/s
278
+ 2 GB/s
279
+ 40%
280
+ 1 GB/s
281
+ 1 GB/s
282
+ 500 MB/s
283
+ 20%
284
+ 0 B/s
285
+ 02:40
286
+ 02:50
287
+ 03:00
288
+ 03:10
289
+ 03:20
290
+ 03:30
291
+ 03:40
292
+ 03:50
293
+ 04:00
294
+ 0 B/s
295
+ 02:4002:50
296
+ ¥03:0003:1003:2003:30
297
+ 03:4003:50
298
+ 04:00
299
+ median network in
300
+ min network in
301
+ max network in
302
+ median network out
303
+ median disk write - min disk write - max disk write - median disk read
304
+ min network out
305
+ min network total
306
+ max network out
307
+ 02:40
308
+ 02:50
309
+ 03:00
310
+ 03:10
311
+ 03:20
312
+ 03:30
313
+ 03:40
314
+ 03:50
315
+ 04:00
316
+ - min disk read
317
+ max disk read -
318
+ median disk total - max disk total
319
+ - max network total6
320
+ Luan et al.
321
+ • Master node (r6i.2xlarge) hourly cost is $0.504.
322
+ • Worker node (i4i.4xlarge) hourly cost is $1.373.
323
+ • Number of workers is 40.
324
+ • EBS volume hourly cost is $0.0044.
325
+ Hence, the total hourly compute cost is $55.6044. We multiply this hourly cost by the job completion
326
+ time of 1.4939 hours to obtain the total compute cost of $83.0674.
327
+ Data Storage Cost. The storage cost comprises of data storage cost and data access cost. We first consider
328
+ the data storage cost. Amazon S3 employs a pay-as-you-go pricing model, i.e. the user does not need to
329
+ provision storage capacity ahead of time, and only pays for the storage cost of objects based on their sizes
330
+ and storage duration. Amazon S3 charges $0.023 per GB-month for the first 50 TB, then $0.022 per GB-month
331
+ for the next 450 TB [3]. Since the total data size is 100 TB, we take the average price between the first two
332
+ tiers, i.e. $0.0225 per GB-month, or $3.0822 per hour per 100 TB.
333
+ • Input: The storage cost of the 100 TB input data is simply the cost to store 100 TB for the duration of
334
+ the sort: $3.0822 × 1.4939 = $4.6045.
335
+ • Output: The 100 TB output data is uploaded to and stored on Amazon S3 during the reduce stage
336
+ of the sort. We use the duration of the reduce stage as the storage time of the 100 TB output data.
337
+ This is an over-estimation because the output partitions are uploaded as the reduce stage progresses,
338
+ and therefore most of the 100 TB is stored on S3 for less time than the entire reduce stage duration.
339
+ Table 1 shows the average reduce stage time is 1870 seconds, or 0.5194 hours. Hence we get the output
340
+ storage cost: $3.0822 × 0.5194 = $1.6009.
341
+ Adding up the input and output data storage cost, we get the total data storage cost: $6.2054.
342
+ Data Access Cost. We consider GET and PUT requests to Amazon S3. Exoshuffle-CloudSort downloads
343
+ the 100 TB input data in 50 000 map tasks. Each map task downloads a 2 GB input partition in 16 MiB chunks,
344
+ resulting in 120 GET requests per task, or 6 000 000 GET requests in total. Amazon S3 charges $0.0004 per
345
+ 1000 GET requests [3]. Hence the total GET cost is $2.4000.
346
+ Exoshuffle-CloudSort uploads the output data in 25 000 reduce tasks. Each reduce task uploads approxi-
347
+ mately 4 GB data in 100 MB chunks, resulting in 40 PUT requests, or 1 000 000 PUT requests in total. Amazon
348
+ S3 charges $0.005 per 1000 PUT requests [3]. Hence the total PUT cost is $5.0000.
349
+ The actual number of requests could be marginally higher due to request failures and retries, but the
350
+ amount should be negligible. Hence, the total data access cost is $7.4000.
351
+ Total Cost of Ownership. Adding up the compute cost and storage cost, we get the total cost of ownership
352
+ for the 100 TB CloudSort Benchmark: $96.6728. Table 2 presents a summary of the cost analysis.
353
+
354
+ Exoshuffle-CloudSort
355
+ 7
356
+ Service
357
+ Unit Price
358
+ Amount
359
+ Total Price
360
+ Compute VM Cluster
361
+ $55.6044 / hr
362
+ 1.4939 hours
363
+ $83.0674
364
+ Data Storage (Input)
365
+ $3.0822 / hr
366
+ 1.4939 hours
367
+ $4.6045
368
+ Data Storage (Output)
369
+ $3.0822 / hr
370
+ 0.5194 hours
371
+ $1.6009
372
+ Data Access (Input)
373
+ $0.0004 / 1000 requests
374
+ 6 000 000 requests
375
+ $2.4000
376
+ Data Access (Output)
377
+ $0.005 / 1000 requests
378
+ 1 000 000 requests
379
+ $5.0000
380
+ Total
381
+
382
+
383
+ $96.6728
384
+ Table 2. Cost breakdown of Exoshuffle-CloudSort on the 100 TB CloudSort Benchmark.
385
+ ACKNOWLEDGMENTS
386
+ This work is done in the Sky Computing Lab at UC Berkeley, sponsored by Astronomer, Google, IBM, Intel,
387
+ Lacework, Nexla, Samsung SDS, and VMware. This work is done in collaboration with Anyscale.
388
+ REFERENCES
389
+ [1] Amazon. 2022. Amazon EBS High-Performance Block Storage Pricing. Amazon Web Services. https://aws.amazon.com/ebs/pricing/
390
+ [2] Amazon. 2022. Amazon EC2 On-Demand Instance Pricing. Amazon Web Services. https://aws.amazon.com/ec2/pricing/on-demand/
391
+ [3] Amazon. 2022. Amazon S3 Simple Storage Service Pricing. Amazon Web Services. https://aws.amazon.com/s3/pricing/
392
+ [4] Frank Sifei Luan, Stephanie Wang, Samyukta Yagati, Sean Kim, Kenneth Lien, Isaac Ong, SangBin Cho, Eric Liang, and Ion Stoica.
393
+ 2022. Exoshuffle: Large-Scale Shuffle at the Application Level. https://doi.org/10.48550/ARXIV.2203.05072
394
+ [5] Chris Nyberg. 2022. Sort Benchmark Data Generator and Output Validator. Ordinal Technology Corp. http://www.ordinal.com/
395
+ gensort.html
396
+ [6] Mehul A. Shah, Amiato, and Chris Nyberg. 2014. CloudSort: A TCO Sort Benchmark. http://sortbenchmark.org/2014_06_
397
+ CloudSort_v_0_4.pdf. (Accessed on 11/10/2022).
398
+ [7] Ray Team. 2022. Ray v2 Architecture. Anyscale. https://docs.google.com/document/d/1tBw9A4j62ruI5omIJbMxly-la5w4q_TjyJgJL_
399
+ jN2fI/preview
400
+ [8] Stephanie Wang, Eric Liang, Edward Oakes, Ben Hindman, Frank Sifei Luan, Audrey Cheng, and Ion Stoica. 2021. Ownership: A
401
+ Distributed Futures System for Fine-Grained Tasks. In 18th USENIX Symposium on Networked Systems Design and Implementation
402
+ (NSDI 21). USENIX Association, Virtual, 671–686. https://www.usenix.org/conference/nsdi21/presentation/cheng
403
+
1dE2T4oBgHgl3EQfNQZD/content/tmp_files/load_file.txt ADDED
@@ -0,0 +1,420 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf,len=419
2
+ page_content='Exoshuffle-CloudSort FRANK SIFEI LUAN∗, UC Berkeley STEPHANIE WANG, UC Berkeley and Anyscale SAMYUKTA YAGATI, UC Berkeley SEAN KIM, UC Berkeley KENNETH LIEN, UC Berkeley ISAAC ONG, UC Berkeley TONY HONG, UC Berkeley SANGBIN CHO, Anyscale ERIC LIANG, Anyscale ION STOICA, UC Berkeley and Anyscale 1 INTRODUCTION We present Exoshuffle-CloudSort, a sorting application running on top of Ray using the Exoshuffle archi- tecture [4].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
3
+ page_content=' Exoshuffle-CloudSort runs on Amazon EC2, with input and output data stored on Amazon S3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
4
+ page_content=' Using 40× i4i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
5
+ page_content='4xlarge workers, Exoshuffle-CloudSort completes the 100 TB CloudSort Benchmark (Indy category [6]) in 5378 seconds, with an average total cost of $97.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
6
+ page_content=' 2 IMPLEMENTATION 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
7
+ page_content='1 Overview Exoshuffle-CloudSort is a distributed futures program running on top of Ray, a task-based distributed execution system.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
8
+ page_content=' The program acts as the control plane to coordinate map and reduce tasks;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
9
+ page_content=' the Ray system acts as the data plane, responsible for executing tasks, transferring blocks, and recovering from failures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
10
+ page_content=' Exoshuffle-CloudSort implements a two-stage external sort algorithm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
11
+ page_content=' The first stage is map and shuffle.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
12
+ page_content=' Each map task reads an input partition, sorts it, and partitions the result into 𝑊 output partitions, each sent to a merger on a worker node.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
13
+ page_content=' A merger receives 𝑊 map output partitions, merges and sorts them, and further partition the result into 𝑅/𝑊 output partitions, all of which are spilled to local disk.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
14
+ page_content=' The second stage is reduce.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
15
+ page_content=' Once the map and shuffle stage finishes, each reduce task reads 𝑊 shuffled partitions, merges and sorts them, and writes the final output partition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
16
+ page_content=' For the 100 TB CloudSort Benchmark, we set the following parameters: Total data size is 100 TB.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
17
+ page_content=' Number of input partitions 𝑀 = 50 000.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
18
+ page_content=' Each input partition is 2 GB.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
19
+ page_content=' ∗Author’s address: lsf@berkeley.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
20
+ page_content='edu, 465 Soda Hall, Berkeley, CA, USA.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
21
+ page_content=' 1 arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
22
+ page_content='03734v1 [cs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
23
+ page_content='DC] 10 Jan 2023 2 Luan et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
24
+ page_content=' Number of workers 𝑊 = 40.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
25
+ page_content=' Number of output partitions 𝑅 = 25 000.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
26
+ page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
27
+ page_content='2 Preparation The first step in Exoshuffle-CloudSort is to compute the partition boundary values.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
28
+ page_content=' For a sort record with 10-byte key, we view the first 8 bytes as a 64-bit unsigned integer partition key.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
29
+ page_content=' We partition the key space [0, 264 − 1) into 𝑅 = 25 000 equal ranges, such that all the records within a key range should be sent to one reducer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
30
+ page_content=' Every 𝑅1 = 𝑅/𝑊 = 625 reducer ranges are combined into a worker range, and records in each worker range will be sent to one worker node.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
31
+ page_content=' This yields 𝑊 = 40 equally-partitioned worker ranges.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
32
+ page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
33
+ page_content='3 Map and Shuffle Stage In the map and shuffle stage, Exoshuffle-CloudSort schedules the 𝑀 = 50 000 map tasks onto all worker nodes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
34
+ page_content=' In our experiments we set the map parallelism, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
35
+ page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
36
+ page_content=' the number of map tasks running on a single worker node, to be 3/4 of the total number of vCPU cores.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
37
+ page_content=' Extra tasks are queued on the driver node.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
38
+ page_content=' Whenever a worker node finishes a map task, the driver assigns a new task from the queue to this node.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
39
+ page_content=' In a map task, we first download the input partition from S3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
40
+ page_content=' We then sort the input data in memory, then partition it into 𝑊 = 40 slices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
41
+ page_content=' Each slice is eagerly sent to a merge controller on each worker.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
42
+ page_content=' The map task returns when all slices are sent.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
43
+ page_content=' On the receiving end, the merge controller accumulates the map blocks in memory until a threshold is reached.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
44
+ page_content=' We set the threshold to 40 blocks, or about 2 GB of data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
45
+ page_content=' Once the threshold is reached, the controller launches a merge task to merge the already-sorted map blocks, and further partitions it into 𝑅1 = 625 merged blocks, each corresponding to a reduce task on this node.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
46
+ page_content=' These blocks are spilled to the local SSD for use by the reducers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
47
+ page_content=' The merge parallelism is set to be the same as the map parallelism.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
48
+ page_content=' When the number of merge tasks reaches the maximum parallelism, and the merge controller’s in-memory buffer is filled up, it will hold off acknowledging the receipt of a map block until a merge task finishes and a new merge task can launch.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
49
+ page_content=' This effectively creates back pressure to the map task scheduler to ensure the map, shuffle, and merge progresses are in sync.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
50
+ page_content=' In our experiments, the average map task duration is 24 seconds;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
51
+ page_content=' 15 seconds are used for downloading input data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
52
+ page_content=' The average shuffle time (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
53
+ page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
54
+ page_content=' time to send and receive blocks) is 7 seconds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
55
+ page_content=' The merge task takes 17 seconds on average.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
56
+ page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
57
+ page_content='4 Reduce Stage Once all map and merge tasks finish, Exoshuffle-CloudSort enters the reduce stage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
58
+ page_content=' Each reduce task loads 𝑅1 = 625 from the local SSD, merges them, and uploads the sorted output partition to S3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
59
+ page_content=' In our experiments, each reduce task takes 22 seconds on average.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
60
+ page_content=' Exoshuffle-CloudSort 3 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
61
+ page_content='5 The Execution System A highlight of the Exoshuffle architecture is that the application program only implements the control plane logic, and the distributed futures system, Ray, handles execution.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
62
+ page_content=' This is reflected in Exoshuffle-CloudSort.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
63
+ page_content=' Here is an incomplete list of features provided by Ray that we take “for free”: Task scheduling: The program specifies when and where to schedule tasks;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
64
+ page_content=' the system handles the RPC, serialization, and other bookkeeping.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
65
+ page_content=' Network transfer: The program instructs data to be transferred by passing distributed futures as task arguments;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
66
+ page_content=' the system implements high-performance network transfer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
67
+ page_content=' Memory management and disk spilling: The program manipulates data references in a virtual, infinite address space;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
68
+ page_content=' the system uses reference counting to manage distributed memory, spills objects to local disks when memory is low, and restores objects from local disks when they are needed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
69
+ page_content=' Pipelining of network and disk I/O: The network transfer, spilling and recovery of objects are trans- parent to the application and are performed asynchronously.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
70
+ page_content=' For example, the system shuffles map output blocks while other map and merge tasks are running;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
71
+ page_content=' it spills merge task output to disk while other merge tasks are executing, and it restores merged blocks while reduce tasks are executing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
72
+ page_content=' Fault tolerance: this is transparent to the application: the system automatically retries the operation when it encounters network failures and worker process failures.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
73
+ page_content=' For more details, we refer the reader to the Ray Architecture Whitepaper [7], the ownership design for distributed futures systems [8], and the Exoshuffle paper [4].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
74
+ page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
75
+ page_content='6 Source Code Exoshuffle-CloudSort is implemented in about 1000 lines of Python, and about 300 lines of C++.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
76
+ page_content=' The C++ component implements two functionalities: sorting and partitioning records, and merging sorted record arrays.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
77
+ page_content=' Exoshuffle-CloudSort runs on top of Ray, which is implemented in Python and C++.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
78
+ page_content=' All of Exoshuffle-CloudSort’s source code is available at https://github.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
79
+ page_content='com/exoshuffle/cloudsort.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
80
+ page_content=' 3 EVALUATION 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
81
+ page_content='1 Environment Setup We run Exoshuffle-CloudSort on AWS on a compute cluster configured as follows: 1× r6i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
82
+ page_content='2xlarge master node.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
83
+ page_content=' This node runs on 8 cores of an Intel Xeon 8375C CPU at 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
84
+ page_content='9 GHz, and 64 GiB memory.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
85
+ page_content=' 40× i4i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
86
+ page_content='4xlarge worker nodes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
87
+ page_content=' Each node runs on 16 cores of an Intel Xeon 8375C CPU at 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
88
+ page_content='9 GHz, and 128 GiB memory.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
89
+ page_content=' Each node has a directly-attached 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
90
+ page_content='75 TB AWS Nitro NVMe SSD.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
91
+ page_content=' Each node is attached with a 40 GiB Amazon EBS General Purpose SSD (gp3) volume.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
92
+ page_content=' The software stack is configured as follows: 4 Luan et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
93
+ page_content=' Ubuntu 22.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
94
+ page_content='04.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
95
+ page_content='1 LTS, Linux kernel version 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
96
+ page_content='15.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
97
+ page_content='0-1022-aws.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
98
+ page_content=' XFS 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
99
+ page_content='13.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
100
+ page_content='0 filesystem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
101
+ page_content=' Intel oneAPI DPC++/C++ Compiler 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
102
+ page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
103
+ page_content='0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
104
+ page_content='20220730.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
105
+ page_content=' Python 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
106
+ page_content='9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
107
+ page_content='13.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
108
+ page_content=' Ray 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
109
+ page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
110
+ page_content='0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
111
+ page_content=' We measure the raw system I/O performance on the worker nodes using standard benchmarking tools: Network bandwidth: 25 Gbps between nodes, benchmarked with iperf.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
112
+ page_content=' SSD: 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
113
+ page_content='9 GB/s read, 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
114
+ page_content='2 GB/s write, benchmarked with fio.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
115
+ page_content=' For storage, we use 40 buckets on Amazon S3 and randomly distribute the input and output partitions across the buckets.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
116
+ page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
117
+ page_content='2 Benchmark Setup Generating Input.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
118
+ page_content=' We use gensort version 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
119
+ page_content='5 as provided by the Sort Benchmark committee [5].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
120
+ page_content=' We run the command gensort -c -b{offset} {size} {path} to generate each partition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
121
+ page_content=' {size} is fixed at 𝑃 = 20 000 000 such that each partition is exactly 2 GB.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
122
+ page_content=' {offset} takes the values {𝑖 · 𝑃 : 0 ≤ 𝑖 < 𝑀} where the number of input partitions 𝑀 = 50 000.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
123
+ page_content=' {path} is a unique path in tmpfs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
124
+ page_content=' -c provides data checksum for validation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
125
+ page_content=' After generating an input file, we randomly choose a bucket and upload the partition to S3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
126
+ page_content=' We use Ray to schedule the 50 000 input generation tasks to all 40 worker nodes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
127
+ page_content=' The result is aggregated as an input manifest file, saved for use by Exoshuffle-CloudSort to locate the sort input.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
128
+ page_content=' Validating Output.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
129
+ page_content=' Exoshuffle-CloudSort produces an output manifest file containing the bucket and keys of each output partition on S3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
130
+ page_content=' In each validation task, we first download the output partition to tmpfs, then run the command valsort -o {sumpath} {path} to validate the ordering of records in each partition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
131
+ page_content=' We use Ray to schedule the 25 000 output validation tasks to all 40 worker nodes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
132
+ page_content=' We concatenate the contents of the summary files from each validation task, then run valsort -s to validate the total ordering, and generate the total output checksum.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
133
+ page_content=' Finally, we compare the output checksum with the input checksum to verify data integrity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
134
+ page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
135
+ page_content='3 Experimental Results 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
136
+ page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
137
+ page_content='1 Job Completion Time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
138
+ page_content=' On November 10, 2022, we ran the 100 TB CloudSort Benchmark in the AWS US West (Oregon, us-west-2) region with the setup described above.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
139
+ page_content=' We first generated the input data on Amazon S3, then ran Exoshuffle-CloudSort 3 times, each followed by a validation step.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
140
+ page_content=' All 3 runs succeeded with the same output checksum as the input, indicating all bytes are preserved in the sort.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
141
+ page_content=' Table 1 reports the job completion times of each run.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
142
+ page_content=' The average job completion time is 5378 seconds, or 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
143
+ page_content='4939 hours.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
144
+ page_content=' Figure 1 shows the system utilizations of all worker nodes in the compute cluster during run #1 of the 100 TB CloudSort Benchmark.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
145
+ page_content=' Exoshuffle-CloudSort 5 Run Map & Shuffle Time Reduce Time Total Job Completion Time #1 3509 s 1852 s 5361 s #2 3496 s 1852 s 5348 s #3 3520 s 1906 s 5426 s Average 3508 s 1870 s 5378 s Table 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
146
+ page_content=' Job completion times of Exoshuffle-CloudSort on the 100 TB CloudSort Benchmark.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
147
+ page_content=' Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
148
+ page_content=' 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
149
+ page_content=' Cluster utilization during run #1 of the 100 TB CloudSort Benchmark.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
150
+ page_content=' Each thick line represents the median system utilization of all worker nodes;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
151
+ page_content=' the highest and lowest lines represent the maximum and minimum utilization among all worker nodes, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
152
+ page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
153
+ page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
154
+ page_content='2 Total Cost of Ownership.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
155
+ page_content=' The total job cost comprises of two parts: compute cost (Amazon EC2), and the storage cost (Amazon S3).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
156
+ page_content=' The storage cost is further divided into data storage cost and data access cost.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
157
+ page_content=' Compute Cost.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
158
+ page_content=' The compute cost is calculated as the compute cluster’s hourly cost times the job completion time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
159
+ page_content=' The total hourly cost is calculated as follows: Total Hourly Compute Cost = Master Node Hourly Cost + Worker Node Hourly Cost × Number of Workers + EBS Volume Hourly Cost × (Number of Workers + 1) (1) We obtain the compute instance hourly costs from the Amazon EC2 on-demand pricing information [2].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
160
+ page_content=' For EBS, we use the Amazon EBS monthly price [1] divided by the average number of hours in a month ( 365×24 12 = 730) as the hourly price.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
161
+ page_content=' The hourly cost of a 40 GiB gp3 volume is $0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
162
+ page_content='08/730 × 40 = $0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
163
+ page_content='0044.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
164
+ page_content=' Now ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
165
+ page_content='we plug the cost variables into Equation (1): ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
166
+ page_content='CPU ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
167
+ page_content='Memory ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
168
+ page_content='Application Progress ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
169
+ page_content='50000 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
170
+ page_content='100% ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
171
+ page_content='70 GB ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
172
+ page_content='40000 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
173
+ page_content='60 GB ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
174
+ page_content='80% ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
175
+ page_content='30000 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
176
+ page_content='50 GB ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
177
+ page_content='20000 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
178
+ page_content='60% ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
179
+ page_content='40 GB ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
180
+ page_content='10000 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
181
+ page_content='30 GB ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
182
+ page_content='40% ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
183
+ page_content='20 GB ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
184
+ page_content='02:40 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
185
+ page_content='02:50 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
186
+ page_content='03:00 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
187
+ page_content='03:10 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
188
+ page_content='03:20 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
189
+ page_content='03:30 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
190
+ page_content='03:40 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
191
+ page_content='03:5004:00 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
192
+ page_content='10 GB ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
193
+ page_content='map_in_progress ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
194
+ page_content='20% ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
195
+ page_content='reduce_in_progress ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
196
+ page_content='reduce_in_progress ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
197
+ page_content='reduce_in_progress ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
198
+ page_content='0 B ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
199
+ page_content='02:40 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
200
+ page_content='02:50 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
201
+ page_content='03:0003:1003:20 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
202
+ page_content='03:3003:40 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
203
+ page_content='03:5004:00 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
204
+ page_content='map_completed ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
205
+ page_content='map_completec ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
206
+ page_content='map_completed ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
207
+ page_content='map_completed ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
208
+ page_content='0% ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
209
+ page_content='02:4002:50 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
210
+ page_content='03:10 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
211
+ page_content='03:20 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
212
+ page_content='03:30 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
213
+ page_content='03:40 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
214
+ page_content='03:50 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
215
+ page_content='04:00 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
216
+ page_content='median objmem ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
217
+ page_content='reducer_completed ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
218
+ page_content='reducer_completed ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
219
+ page_content='reducer_completed ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
220
+ page_content='reducer_completed ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
221
+ page_content='03:00 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
222
+ page_content='min objmem ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
223
+ page_content='max objmem ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
224
+ page_content='min workmem ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
225
+ page_content='median cpu - min cpu - max cpu ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
226
+ page_content='max workmem ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
227
+ page_content='merge_in_progress ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
228
+ page_content='merge_in_progress ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
229
+ page_content='merge_in_progress ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
230
+ page_content='nerge_in_progress ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
231
+ page_content='NVMe Disk I/0 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
232
+ page_content='Network I/0 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
233
+ page_content='Disk Usage ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
234
+ page_content='7 GB/s ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
235
+ page_content='3 GB/s ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
236
+ page_content='100% ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
237
+ page_content='6 GB/s ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
238
+ page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
239
+ page_content='50 GB/s 80% 5 GB/s 2 GB/s 4 GB/s 60% 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
240
+ page_content='50 GB/s ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
241
+ page_content='3 GB/s ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
242
+ page_content='2 GB/s ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
243
+ page_content='40% ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
244
+ page_content='1 GB/s ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
245
+ page_content='1 GB/s ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
246
+ page_content='500 MB/s ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
247
+ page_content='20% ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
248
+ page_content='0 B/s ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
249
+ page_content='02:40 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
250
+ page_content='02:50 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
251
+ page_content='03:00 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
252
+ page_content='03:10 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
253
+ page_content='03:20 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
254
+ page_content='03:30 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
255
+ page_content='03:40 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
256
+ page_content='03:50 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
257
+ page_content='04:00 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
258
+ page_content='0 B/s ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
259
+ page_content='02:4002:50 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
260
+ page_content='¥03:0003:1003:2003:30 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
261
+ page_content='03:4003:50 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
262
+ page_content='04:00 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
263
+ page_content='median network in ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
264
+ page_content='min network in ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
265
+ page_content='max network in ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
266
+ page_content='median network out ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
267
+ page_content='median disk write - min disk write - max disk write - median disk read ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
268
+ page_content='min network out ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
269
+ page_content='min network total ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
270
+ page_content='max network out ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
271
+ page_content='02:40 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
272
+ page_content='02:50 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
273
+ page_content='03:00 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
274
+ page_content='03:10 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
275
+ page_content='03:20 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
276
+ page_content='03:30 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
277
+ page_content='03:40 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
278
+ page_content='03:50 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
279
+ page_content='04:00 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
280
+ page_content='min disk read ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
281
+ page_content='max disk read - ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
282
+ page_content='median disk total - max disk total ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
283
+ page_content='max network total6 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
284
+ page_content='Luan et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
285
+ page_content=' Master node (r6i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
286
+ page_content='2xlarge) hourly cost is $0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
287
+ page_content='504.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
288
+ page_content=' Worker node (i4i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
289
+ page_content='4xlarge) hourly cost is $1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
290
+ page_content='373.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
291
+ page_content=' Number of workers is 40.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
292
+ page_content=' EBS volume hourly cost is $0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
293
+ page_content='0044.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
294
+ page_content=' Hence, the total hourly compute cost is $55.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
295
+ page_content='6044.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
296
+ page_content=' We multiply this hourly cost by the job completion time of 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
297
+ page_content='4939 hours to obtain the total compute cost of $83.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
298
+ page_content='0674.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
299
+ page_content=' Data Storage Cost.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
300
+ page_content=' The storage cost comprises of data storage cost and data access cost.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
301
+ page_content=' We first consider the data storage cost.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
302
+ page_content=' Amazon S3 employs a pay-as-you-go pricing model, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
303
+ page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
304
+ page_content=' the user does not need to provision storage capacity ahead of time, and only pays for the storage cost of objects based on their sizes and storage duration.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
305
+ page_content=' Amazon S3 charges $0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
306
+ page_content='023 per GB-month for the first 50 TB, then $0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
307
+ page_content='022 per GB-month for the next 450 TB [3].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
308
+ page_content=' Since the total data size is 100 TB, we take the average price between the first two tiers, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
309
+ page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
310
+ page_content=' $0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
311
+ page_content='0225 per GB-month, or $3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
312
+ page_content='0822 per hour per 100 TB.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
313
+ page_content=' Input: The storage cost of the 100 TB input data is simply the cost to store 100 TB for the duration of the sort: $3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
314
+ page_content='0822 × 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
315
+ page_content='4939 = $4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
316
+ page_content='6045.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
317
+ page_content=' Output: The 100 TB output data is uploaded to and stored on Amazon S3 during the reduce stage of the sort.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
318
+ page_content=' We use the duration of the reduce stage as the storage time of the 100 TB output data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
319
+ page_content=' This is an over-estimation because the output partitions are uploaded as the reduce stage progresses, and therefore most of the 100 TB is stored on S3 for less time than the entire reduce stage duration.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
320
+ page_content=' Table 1 shows the average reduce stage time is 1870 seconds, or 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
321
+ page_content='5194 hours.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
322
+ page_content=' Hence we get the output storage cost: $3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
323
+ page_content='0822 × 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
324
+ page_content='5194 = $1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
325
+ page_content='6009.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
326
+ page_content=' Adding up the input and output data storage cost, we get the total data storage cost: $6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
327
+ page_content='2054.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
328
+ page_content=' Data Access Cost.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
329
+ page_content=' We consider GET and PUT requests to Amazon S3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
330
+ page_content=' Exoshuffle-CloudSort downloads the 100 TB input data in 50 000 map tasks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
331
+ page_content=' Each map task downloads a 2 GB input partition in 16 MiB chunks, resulting in 120 GET requests per task, or 6 000 000 GET requests in total.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
332
+ page_content=' Amazon S3 charges $0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
333
+ page_content='0004 per 1000 GET requests [3].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
334
+ page_content=' Hence the total GET cost is $2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
335
+ page_content='4000.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
336
+ page_content=' Exoshuffle-CloudSort uploads the output data in 25 000 reduce tasks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
337
+ page_content=' Each reduce task uploads approxi- mately 4 GB data in 100 MB chunks, resulting in 40 PUT requests, or 1 000 000 PUT requests in total.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
338
+ page_content=' Amazon S3 charges $0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
339
+ page_content='005 per 1000 PUT requests [3].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
340
+ page_content=' Hence the total PUT cost is $5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
341
+ page_content='0000.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
342
+ page_content=' The actual number of requests could be marginally higher due to request failures and retries, but the amount should be negligible.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
343
+ page_content=' Hence, the total data access cost is $7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
344
+ page_content='4000.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
345
+ page_content=' Total Cost of Ownership.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
346
+ page_content=' Adding up the compute cost and storage cost, we get the total cost of ownership for the 100 TB CloudSort Benchmark: $96.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
347
+ page_content='6728.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
348
+ page_content=' Table 2 presents a summary of the cost analysis.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
349
+ page_content=' Exoshuffle-CloudSort 7 Service Unit Price Amount Total Price Compute VM Cluster $55.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
350
+ page_content='6044 / hr 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
351
+ page_content='4939 hours $83.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
352
+ page_content='0674 Data Storage (Input) $3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
353
+ page_content='0822 / hr 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
354
+ page_content='4939 hours $4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
355
+ page_content='6045 Data Storage (Output) $3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
356
+ page_content='0822 / hr 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
357
+ page_content='5194 hours $1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
358
+ page_content='6009 Data Access (Input) $0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
359
+ page_content='0004 / 1000 requests 6 000 000 requests $2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
360
+ page_content='4000 Data Access (Output) $0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
361
+ page_content='005 / 1000 requests 1 000 000 requests $5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
362
+ page_content='0000 Total – – $96.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
363
+ page_content='6728 Table 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
364
+ page_content=' Cost breakdown of Exoshuffle-CloudSort on the 100 TB CloudSort Benchmark.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
365
+ page_content=' ACKNOWLEDGMENTS This work is done in the Sky Computing Lab at UC Berkeley, sponsored by Astronomer, Google, IBM, Intel, Lacework, Nexla, Samsung SDS, and VMware.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
366
+ page_content=' This work is done in collaboration with Anyscale.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
367
+ page_content=' REFERENCES [1] Amazon.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
368
+ page_content=' 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
369
+ page_content=' Amazon EBS High-Performance Block Storage Pricing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
370
+ page_content=' Amazon Web Services.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
371
+ page_content=' https://aws.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
372
+ page_content='amazon.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
373
+ page_content='com/ebs/pricing/ [2] Amazon.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
374
+ page_content=' 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
375
+ page_content=' Amazon EC2 On-Demand Instance Pricing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
376
+ page_content=' Amazon Web Services.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
377
+ page_content=' https://aws.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
378
+ page_content='amazon.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
379
+ page_content='com/ec2/pricing/on-demand/ [3] Amazon.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
380
+ page_content=' 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
381
+ page_content=' Amazon S3 Simple Storage Service Pricing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
382
+ page_content=' Amazon Web Services.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
383
+ page_content=' https://aws.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
384
+ page_content='amazon.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
385
+ page_content='com/s3/pricing/ [4] Frank Sifei Luan, Stephanie Wang, Samyukta Yagati, Sean Kim, Kenneth Lien, Isaac Ong, SangBin Cho, Eric Liang, and Ion Stoica.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
386
+ page_content=' 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
387
+ page_content=' Exoshuffle: Large-Scale Shuffle at the Application Level.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
388
+ page_content=' https://doi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
389
+ page_content='org/10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
390
+ page_content='48550/ARXIV.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
391
+ page_content='2203.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
392
+ page_content='05072 [5] Chris Nyberg.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
393
+ page_content=' 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
394
+ page_content=' Sort Benchmark Data Generator and Output Validator.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
395
+ page_content=' Ordinal Technology Corp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
396
+ page_content=' http://www.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
397
+ page_content='ordinal.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
398
+ page_content='com/ gensort.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
399
+ page_content='html [6] Mehul A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
400
+ page_content=' Shah, Amiato, and Chris Nyberg.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
401
+ page_content=' 2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
402
+ page_content=' CloudSort: A TCO Sort Benchmark.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
403
+ page_content=' http://sortbenchmark.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
404
+ page_content='org/2014_06_ CloudSort_v_0_4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
405
+ page_content='pdf.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
406
+ page_content=' (Accessed on 11/10/2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
407
+ page_content=' [7] Ray Team.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
408
+ page_content=' 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
409
+ page_content=' Ray v2 Architecture.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
410
+ page_content=' Anyscale.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
411
+ page_content=' https://docs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
412
+ page_content='google.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
413
+ page_content='com/document/d/1tBw9A4j62ruI5omIJbMxly-la5w4q_TjyJgJL_ jN2fI/preview [8] Stephanie Wang, Eric Liang, Edward Oakes, Ben Hindman, Frank Sifei Luan, Audrey Cheng, and Ion Stoica.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
414
+ page_content=' 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
415
+ page_content=' Ownership: A Distributed Futures System for Fine-Grained Tasks.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
416
+ page_content=' In 18th USENIX Symposium on Networked Systems Design and Implementation (NSDI 21).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
417
+ page_content=' USENIX Association, Virtual, 671–686.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
418
+ page_content=' https://www.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
419
+ page_content='usenix.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
420
+ page_content='org/conference/nsdi21/presentation/cheng' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/1dE2T4oBgHgl3EQfNQZD/content/2301.03734v1.pdf'}
49AzT4oBgHgl3EQffvx7/content/2301.01457v1.pdf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:69d533687a7c2207d8fcd9d8e9be6bac0c8cced12e9f0878cb452e2f9bc612e8
3
+ size 1978081
49AzT4oBgHgl3EQffvx7/vector_store/index.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c2b160dcfb7fcd866e3452ddb93eba54b176455dee995ae5ea0fe222a23f057b
3
+ size 252938
4NAzT4oBgHgl3EQfffyv/content/tmp_files/2301.01454v1.pdf.txt ADDED
@@ -0,0 +1,1097 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Accurate, Low-latency, Efficient SAR Automatic
2
+ Target Recognition on FPGA
3
+ Bingyi Zhang∗, Rajgopal Kannan†, Viktor Prasanna∗, Carl Busart†
4
+ ∗University of Southern California †DEVCOM US Army Research Lab
5
+ ∗{bingyizh, prasanna}@usc.edu †{rajgopal.kannan.civ, carl.e.busart.civ}@army.mil
6
+ Abstract—Synthetic aperture radar (SAR) automatic target
7
+ recognition (ATR) is the key technique for remote-sensing image
8
+ recognition. The state-of-the-art convolutional neural networks
9
+ (CNNs) for SAR ATR suffer from high computation cost and
10
+ large memory footprint, making them unsuitable to be deployed
11
+ on resource-limited platforms, such as small/micro satellites.
12
+ In this paper, we propose a comprehensive GNN-based model-
13
+ architecture co-design on FPGA to address the above issues.
14
+ Model design: we design a novel graph neural network (GNN) for
15
+ SAR ATR. The proposed GNN model incorporates GraphSAGE
16
+ layer operators and attention mechanism, achieving comparable
17
+ accuracy as the state-of-the-art work with near 1/100 computa-
18
+ tion cost. Then, we propose a pruning approach including weight
19
+ pruning and input pruning. While weight pruning through lasso
20
+ regression reduces most parameters without accuracy drop, input
21
+ pruning eliminates most input pixels with negligible accuracy
22
+ drop. Architecture design: to fully unleash the computation
23
+ parallelism within the proposed model, we develop a novel unified
24
+ hardware architecture that can execute various computation
25
+ kernels (feature aggregation, feature transformation, graph pool-
26
+ ing). The proposed hardware design adopts the Scatter-Gather
27
+ paradigm to efficiently handle the irregular computation patterns
28
+ of various computation kernels. We deploy the proposed design
29
+ on an embedded FPGA (AMD Xilinx ZCU104) and evaluate
30
+ the performance using MSTAR dataset. Compared with the
31
+ state-of-the-art CNNs, the proposed GNN achieves comparable
32
+ accuracy with 1/3258 computation cost and 1/83 model size.
33
+ Compared with the state-of-the-art CPU/GPU, our FPGA accel-
34
+ erator achieves 14.8×/2.5× speedup (latency) and is 62×/39×
35
+ more energy efficient.
36
+ Index Terms—SAR ATR, graph neural network (GNN), hard-
37
+ ware architecture
38
+ I. INTRODUCTION
39
+ Synthetic aperture radar (SAR) can acquire remote-sensing
40
+ data in all-weather conditions to observe target on the earth
41
+ ground. SAR has been widely used in real-world applications,
42
+ such as agriculture [1], [2], civilization [3], [4], etc. SAR
43
+ automatic target recognition (ATR) is the key technique to
44
+ classify the target in a SAR image. Convolutional neural
45
+ networks (CNNs) [5]–[9] have been extensively studied for
46
+ ATR SAR since CNNs can extract discriminative features
47
+ from an image. However, the CNN-based approaches [5]–
48
+ [9] suffer from two issues: (1) high computation cost: to
49
+ achieve high accuracy, the authors [5]–[9] develop large CNN
50
+ models with high computation complexity, (2) large memory
51
+ requirement: these large CNN models have large number of
52
+ parameters, which require large memory footprint. Therefore,
53
+ it is unsuitable to deploy large CNNs on resource-limited
54
+ platforms, such as small/micro satellites [10]–[14].
55
+ The causes of the above issues are (1) heavy convolutional
56
+ operations in CNNs, and (2) CNNs are hard to exploit data
57
+ sparsity in SAR images because CNNs need to use the whole
58
+ image as the input. In a SAR image (Figure 1), only a
59
+ small set of pixels belongs to the target (defined as pixels of
60
+ interest, POI), which can be easily extracted through applying
61
+ a constant threshold [15]. However, the extracted POI has
62
+ irregular structure that is hard to be processed by CNNs,
63
+ where Graph Neural Network (GNN) provides an opportunity.
64
+ Intuitively, we can use the POI to construct a graph and
65
+ use GNN to perform classification for the graph. Fortunately,
66
+ GNNs have been proven to be powerful models [16] to
67
+ classify graphs based on graph structural information and
68
+ vertex features. Therefore GNNs [17]–[19] have been applied
69
+ to many graph classification tasks [20]–[24]. Recently, GNNs
70
+ have been successfully applied to many image classification
71
+ tasks [25]–[27]. Motivated by that, we design a novel GNN
72
+ model for SAR ATR (Section III-A). We propose a graph
73
+ representation G(V, E) for a SAR image. The proposed GNN
74
+ model can extract the structural information of the target from
75
+ the constructed graph. To improve classification accuracy, we
76
+ leverage the attention mechanism including spatial attention
77
+ and channel attention to identify the important vertices and
78
+ features. To further reduce the computation complexity, we
79
+ perform weight pruning by training the GNN model through
80
+ lasso regression and pruning the GNN model weights that have
81
+ small absolute values. Taking advantage of the GNN model,
82
+ we perform input pruning (POI extraction). By eliminating the
83
+ vertices that have small value, the computation complexity is
84
+ reduced by 92.8% with small accuracy loss (< 0.17%).
85
+ The proposed GNN has the following advantages: (1) even
86
+ without weight/input pruning, the proposed GNN has near
87
+ 1/100 computation cost as the state-of-the-art CNNs with
88
+ similar accuracy, (2) while weight pruning can potentially be
89
+ exploited by CNNs, input pruning is hard to be exploited by
90
+ CNNs because CNNs need to use the whole image as the
91
+ input. GNN is flexible to use a small set of input pixels as
92
+ the input. Therefore, despite that we can accelerate the CNNs
93
+ [5]–[9] on advanced CNN accelerators [28], their latency is
94
+ still significant (Section VI-D).
95
+ While the proposed GNN is lightweight that can be de-
96
+ ployed on the resource limited platforms, accelerating GNNs
97
+ is challenging. GNNs have irregular computation pattern and
98
+ heterogeneous computation kernels [29], making them ineffi-
99
+ cient to be deployed on the general purpose processors. The
100
+ arXiv:2301.01454v1 [cs.AR] 4 Jan 2023
101
+
102
+ pruned GNN model introduces additional irregularity through
103
+ weight pruning. Moreover, the proposed model has various
104
+ heterogeneous computation kernels (feature aggregation, fea-
105
+ ture transformation, graph pooling) that need to be mapped
106
+ on an accelerator. While there are many GNN accelerators
107
+ [29]–[35] proposed, none of them exploits the sparsity of the
108
+ weight matrices or deals with graph pooling, which are still
109
+ inefficient for the proposed model. While the proposed GNN
110
+ achieves high accuracy with small computation complexity,
111
+ we believe that low-latency execution of SAR ATR must be
112
+ achieved through careful model-architecture co-design.
113
+ Therefore, we develop a novel unified hardware architecture
114
+ for the proposed GNN model. We demonstrate the methods
115
+ of mapping various computation kernels onto the proposed
116
+ accelerator. In the accelerator design, we adopt Scatter-Gather
117
+ paradigm to efficient deal with the irregular computation
118
+ patterns of various kernels. To the best of our knowledge, this
119
+ is the first GNN-based model-architecture co-design for SAR
120
+ ATR. Our main contributions are:
121
+ • We propose a lightweight GNN for SAR ATR that
122
+ achieves comparable accuracy as state-of-the-art GNNs
123
+ with significant less computation complexity.
124
+ • We perform weight pruning and input pruning to dramat-
125
+ ically reduce the computation complexity and the number
126
+ of model weights.
127
+ • We design a unified hardware architecture that can exe-
128
+ cute various computation kernels in the proposed model.
129
+ We adopt Scatter-Gather paradigm to deal with the irreg-
130
+ ular computation patterns.
131
+ • Taking advantage of the proposed hardware mapping
132
+ strategy, we further optimize the load balance of various
133
+ computation kernels (Section V-A).
134
+ • We deploy our co-design on Xilinx ZCU104. We evaluate
135
+ our co-design using MSTAR dataset. Compared with
136
+ the state-of-the-art CNNs, the proposed GNN achieves
137
+ comparable accuracy with 1/3258 computation cost and
138
+ 1/83 model size. Compared with the state-of-the-art
139
+ CPU/GPU, our FPGA accelerator achieves 14.8×/2.5×
140
+ speedup (latency) and is 62×/39× more energy efficient.
141
+ II. BACKGROUND AND RELATED WORK
142
+ A. Related Work
143
+ Fig. 1: The SAR images of various targets (vehicles)
144
+ SAR ATR is to automatically classify the target in a given
145
+ SAR images (Figure 1). To achieve high accuracy, deep
146
+ learning based methods have been extensively studied. David
147
+ [6] demonstrates that CNNs outperform traditional methods,
148
+ such as Support Vector Machine, etc. TAI-SARNET [9] is a
149
+ TABLE I: Notations
150
+ Notation
151
+ Description
152
+ Notation
153
+ Description
154
+ G(V, E, X0)
155
+ input graph
156
+ vi
157
+ ith vertex
158
+ V
159
+ set of vertices
160
+ eij
161
+ edge from vi to vj
162
+ E
163
+ set of edges
164
+ L
165
+ number of GNN layers
166
+ hl
167
+ i
168
+ feature vector of vi at layer l
169
+ N(i)
170
+ neighbors of vi
171
+ CNN model that incorporates atrous convolution and inception
172
+ module to achieve high accuracy for SAR ATR. The authors
173
+ [8] combine multi-view features to classify the target in SAR
174
+ images. The authors [5] propose the Convolutional Block At-
175
+ tention Module by exploiting the spatial attention and channel
176
+ attention. However, the state-of-the-art CNNs [5], [8], [9] suf-
177
+ fer from high computation cost, making them unsuitable to be
178
+ deployed on resource-limited platforms. Recently, the authors
179
+ [15] exploit GNN for SAR ATR. They construct graphs from
180
+ SAR images by connecting the pixels by the declined order of
181
+ pixel grayscale value. However, the constructed graphs lose the
182
+ structural information of targets, making it extremely sensitive
183
+ to the variations of input pixel values.
184
+ B. Graph Neural Network
185
+ The notations are defined in Table I. Graph Neural Networks
186
+ (GNN) [17]–[19] are proposed for representation learning on
187
+ graph G(V, E, X0). GNNs can learn from the structural infor-
188
+ mation and vertex/edge features of the graph, and embed these
189
+ information into low-dimension vector representation/graph
190
+ embedding (For example, hL
191
+ i is the embedding of vertex vi).
192
+ The vector representation can be used for many downstream
193
+ tasks, such as node classification [17], [18], link prediction
194
+ [36], graph classification [37], etc. GNNs follow the message-
195
+ passing paradigm that vertices recursively aggregate informa-
196
+ tion from the neighbors, for example:
197
+ GraphSAGE: GraphSAGE is proposed in [18] for inductive
198
+ representation learning on graphs. The GraphSAGE layer
199
+ follows the aggregate-update paradigm:
200
+ aggregate:zl
201
+ i = Mean
202
+
203
+ hl−1
204
+ j
205
+ : j ∈ N(i) ∪ {i}
206
+
207
+ update:hl
208
+ i = ReLU
209
+
210
+ zl
211
+ iW l
212
+ neighbor + bl
213
+ neighbor||hl−1
214
+ i
215
+ W l
216
+ self + bl
217
+ self
218
+ � (1)
219
+ III. MODEL-ARCHITECTURE CO-DESIGN
220
+ To achieve accurate and efficient SAR ATR on FPGA
221
+ platform, we perform comprehensive model-architecture co-
222
+ design. The proposed co-design consists of a novel GNN
223
+ model for SAR ATR (Section III-A), a pruning strategy to
224
+ reduce the computation complexity (Section III-B), a novel
225
+ hardware design to efficiently execute the proposed GNN
226
+ (Section III-C), and the strategy to keep load balance within
227
+ various computation kernels (Section V-A). The key novelty of
228
+ our hardware design is that it can execute various computation
229
+ kernels in the proposed model, and it can efficiently handle
230
+ the irregular computation patterns caused by the sparsity of
231
+ weight matrices. We use the widely used MSTAR dataset [38]
232
+ for performance evaluation. We target various performance
233
+
234
+ BTR70
235
+ BRDM2
236
+ D7
237
+ T62
238
+ U
239
+ 20
240
+ 20
241
+ 20
242
+ 40
243
+ 40
244
+ D
245
+ 40
246
+ 60
247
+ 60
248
+ 60
249
+ 08
250
+ 80
251
+ 80
252
+ 100
253
+ 100
254
+ 0
255
+ 100
256
+ 120
257
+ 120
258
+ 0
259
+ 120
260
+ 0
261
+ 2550
262
+ 100
263
+ 125
264
+ 25
265
+ 50
266
+ 75
267
+ 100
268
+ 0
269
+ 255075100
270
+ 125
271
+ 0
272
+ 2550
273
+ 75100
274
+ 125metrics: (1) Accuracy: the accuracy on MSTAR dataset, (2)
275
+ Computation complexity: the total computation complexity for
276
+ inferring a SAR image, (3) Number of parameters: the total
277
+ number of parameters in the model, (4) Latency: the latency for
278
+ inferring a SAR image, (5) Energy Consumption: the energy
279
+ consumption for inferring a SAR image.
280
+ A. GNN Model Design
281
+ Graph representation
282
+ GNNL
283
+ Pooling
284
+ Attention
285
+ GNNL
286
+ Pooling
287
+ Attention
288
+
289
+
290
+
291
+ GNNL
292
+ MLP
293
+ Classification result
294
+ SAR image
295
+ Spatial
296
+ Attention
297
+ Channel
298
+ Attention
299
+ x
300
+ x
301
+ +
302
+ Attention module
303
+ GNNL-1
304
+ Pooling-1
305
+ Attention-1
306
+ GNNL-2
307
+ Pooling-2
308
+ Attention-2
309
+ GNNL-L
310
+ Pooling
311
+ within each
312
+ 2 × 2 range
313
+ Fig. 2: The proposed GNN model
314
+ Graph representation: We represent a SAR image as a graph
315
+ G(V, E), with each pixel viewed as a vertex. Each pixel/vertex
316
+ is connected to its four neighbors (up, down, left, right)
317
+ with edges. The feature of a vertex is the grayscale value
318
+ of the pixel. Such graph representation maintains structural
319
+ information of the target that can be learned by GNN for
320
+ classification. It also provides the opportunity for input pruning
321
+ (Section III-B).
322
+ GNN model: As shown in Figure 2, the proposed GNN model
323
+ has a sequence of layers, including GNN layer (GNNL), graph
324
+ pooling layer (Pooling), Attention module (Attention). For
325
+ GNN layer, we use the GraphSAGE layer operators [18],
326
+ which have been proven to achieve superior accuracy in
327
+ various application domains. For graph pooling layer, since
328
+ the input graph has 2-D grid structure, we adopt the similar
329
+ pooling strategy as the CNN for 2-D image. Within each local
330
+ s × s range having s2 vertices, the pooling operator (e.g.,
331
+ Max(), Min()) is performed on the s2 vertices to obtain an
332
+ output vertex. Figure 2 demonstrates the pooling operation of
333
+ size 2×2 with stride 2. Motivated by the attention mechanism
334
+ in CNN [39], the proposed Attention module consists of a
335
+ Channel Attention module and a Spatial Attention module.
336
+ Suppose the input to Attention Module is {hi : vi ∈ G},
337
+ where hi ∈ Rc is the feature vector of vi and c is the
338
+ length of the feature vector. The Channel Attention calculates
339
+ the attention score Fch of each feature through a Multi-
340
+ layer perceptron. Then, each vertex is multiplied by Fch to
341
+ obtain {(hi)′ : (hi)′ = hi ⊗ Fch, vi ∈ G} where ⊗ is
342
+ the element-wise multiplication. The Spatial Attention module
343
+ calculates the attention score of each vertex using a GNN layer
344
+ (GraphSAGE layer operators):
345
+ {αi : vi ∈ G} = sigmoid(GNNL({hi : vi ∈ G})),
346
+ Then, each vertex feature vector is multiplied by its attention
347
+ score: {(hi)′′ : (hi)′′ = αihi, vi ∈ G}. The output of the
348
+ Attention module is calculated by:
349
+ {houtput
350
+ i
351
+ : houtput
352
+ i
353
+ = hi + (hi)′ + (hi)′′, vi ∈ G}
354
+ (2)
355
+ After GNNL-L, all the feature vectors are flattened to a
356
+ vector which becomes the input to the last MLP (Multi-layer
357
+ Perceptron) for classification.
358
+ B. Network Pruning
359
+ Weight pruning: To reduce the total computation complexity,
360
+ we perform weight pruning by training the model using lasso
361
+ regression [40]. We add a L1 penalty term to the loss function:
362
+ loss =
363
+ N
364
+
365
+ i=1
366
+ (yi − Model(Gi))2 + λ
367
+ W
368
+
369
+ w
370
+ |w|
371
+ The penalty term results in weight shrinkage. Some model
372
+ weights become zeros and are eliminated from the model.
373
+ After training, we set a threshold Iweight and the weights with
374
+ absolute values smaller than Iweight are pruned.
375
+ Input pruning: In a SAR image, most pixels outside of the
376
+ target have negligible grayscale values. Therefore, in the graph
377
+ representation G(V, E) of a SAR image, we set a threshold
378
+ Ivertex and prune the vertices that have grayscale values smaller
379
+ than Ivertex. The edges connected to the pruned vertices are also
380
+ pruned. After input pruning, the eliminated vertices maintain
381
+ the same positions in the graph pooling layer and do not
382
+ participate in the pooling operation. For example, in a local
383
+ 2 × 2 range, if a vertex is pruned, the pooling operator will
384
+ operate on the remaining three vertices. For the input to last
385
+ MLP, the feature vectors of the pruned vertices are padded
386
+ using zeros.
387
+ C. Architecture design
388
+ The objective of the architecture design is to (1) support
389
+ various computation kernels in the proposed model, (2) han-
390
+ dle the irregular computation patterns caused by the feature
391
+ aggregation in the GNN layer and the sparsity of the weight
392
+ matrices. Figure 3 shows the proposed architecture design
393
+ on the embedded FPGA platform. The system consists of an
394
+ Application Processing Unit (APU) and an FPGA accelerator
395
+ in Programmable Logic Region. The FPGA accelerator exe-
396
+ cutes the inference process of the GNN model. In the FPGA
397
+ accelerator, there is a Weight/Edge Buffer (WEB) to store
398
+ the model weights and edges of input graph, an Input Buffer
399
+ (IB) to store the input vertex feature vectors, a Results Buffer
400
+ (RB) to store the output vertex feature vectors. The Matrix
401
+ Transformation Unit (MTU) performs matrix transformation
402
+ to prepare the require data layout for the next layer. Thanks
403
+ to the proposed lightweight model, the trained model is fully
404
+
405
+ APU
406
+ DDR controller
407
+ Scatter 1
408
+ Scatter 2
409
+ ……
410
+ Scatter ������������
411
+ Gather 1
412
+ Gather 2
413
+ ……
414
+ Gather p
415
+ Routing
416
+ Network
417
+ MTU
418
+ Bank 1
419
+ Bank 2
420
+ ……
421
+ Bank ������������
422
+ Result Buffer
423
+ Input
424
+ Buffer
425
+ Weight
426
+ /Edge
427
+ Buffer
428
+ DMA
429
+ Programmable Logic
430
+ Scatter
431
+ Gather 1
432
+ demux
433
+ mux
434
+ x
435
+ demux
436
+ mux
437
+ x
438
+ …..
439
+ ……
440
+ MTU
441
+ Matrix Transformation Unit
442
+ FPGA
443
+ APU
444
+ Application Processing Unit (e.g., ARM Cortex-A53)
445
+ mux
446
+ demux
447
+ mux
448
+ ACC
449
+ Max
450
+ mux
451
+ ReLU
452
+ Sigmoid
453
+ demux
454
+ mux
455
+ ACC
456
+ Max
457
+ mux
458
+ ReLU
459
+ Sigmoid
460
+ Fig. 3: The diagram of the system architecture
461
+ stored in the Weight Buffer, eliminating the memory traffic of
462
+ loading the model weights at runtime.
463
+ Run Time: At runtime, the APU receives an input SAR image
464
+ and transform it into the graph presentation. During the trans-
465
+ formation, the pixels that have grayscale value smaller than
466
+ Ivertex are pruned. Then, the APU sends the input graph to the
467
+ Input Buffer of the accelerator. The accelerator executes each
468
+ layer using Scatter-Gather paradigm (SGP). The accelerator
469
+ exploits the computation parallelism within each layer. After
470
+ finishing the execution of all layers, the accelerator sends the
471
+ classification result back to the APU.
472
+ IV. HARDWARE MAPPING
473
+ A. Computation kernels
474
+ We categorize the computation kernels into two classes:
475
+ Vertex aggregation kernel (VAK): VAKs include (1) feature
476
+ aggregation (in GNN layer, and in Spatial Attention module)
477
+ (2) graph pooling. In VAKs, each vertex propagates its feature
478
+ vector to the neighbors or within a local range (graph pooling).
479
+ Vertex updating kernel (VUK): VUKs include (1) feature
480
+ update (in GNN layer, and in Spatial Attention module) (2)
481
+ Channel attention of Attention module, (3) the last MLP. In
482
+ the VUKs, the feature vector of each vertex is multiplied by a
483
+ weight matrix to obtain the updated feature vector. Due to our
484
+ weight pruning, the weight matrices have high data sparsity
485
+ (1%-33% data density).
486
+ B. Kernel Mapping using Scatter-Gather Paradigm
487
+ Algorithm 1 Scatter-Gather paradigm
488
+ while not done do
489
+ Scatter Unit:
490
+ for each edge e⟨src, dst, weight⟩ do
491
+ Produce update u ←Scatter(src.vector, e.weight)
492
+ end for
493
+ Gather Unit:
494
+ for each update u⟨dst, vector⟩ do
495
+ Update vertex vdst ← Gather(u.vector)
496
+ end for
497
+ end while
498
+ The accelerator design is based on the Scatter-Gather
499
+ paradigm (Algorithm 1). There are p parallel pipelines. Each
500
+ pipeline consists of a Scatter Unit and a Gather Unit. The
501
+ Routing Network routes the intermediate results to the des-
502
+ tination based on index dst. To map the VAKs and VUKs
503
+ ������������1
504
+ ������������2
505
+ ������������3
506
+ ������������4
507
+ ������������1
508
+ ������������2
509
+ ������������3
510
+ ������������4
511
+ ������������1 ������������2 ������������3������������4
512
+ ������������������������������������
513
+ ������������������������������������
514
+ Input Feature vectors
515
+ ������������1
516
+ ������������2
517
+ ������������3
518
+ ������������4
519
+ Output Feature vectors
520
+ Vertex aggregation kernel
521
+ ������������1
522
+ ������������2
523
+ ������������3
524
+ ������������4
525
+ 1
526
+ 2
527
+ 3
528
+ 4
529
+ 5
530
+ 1 2 3 4
531
+ Adjacency matrix
532
+ �������������������������������������
533
+ ������������������������������������������������ = ������������������������������������
534
+ ������������������������������������ = 5
535
+ ������������������������������������
536
+ ������������������������������������
537
+ ������������1
538
+ ������������2
539
+ ������������3
540
+ ������������4
541
+ ������������������������������������������������ = 4
542
+ Weight matrix
543
+ 1 2 3 4 5
544
+ 2
545
+ 1
546
+ 3 4
547
+ Vertex updating
548
+ kernel
549
+ ������������1
550
+ ������������������������������������������������������������
551
+ ������������1
552
+ ������������������������������������������������������������������������
553
+ Input Feature
554
+ vectors
555
+ Output Feature vectors
556
+ Fig. 4: The diagram of mapping the two types of kernels using
557
+ Scatter-Gather paradigm
558
+ to the accelerator, we propose the following mapping strategy
559
+ (An example is shown in Figure 4):
560
+ Mapping VAK: VAK can be directly mapped to the accelera-
561
+ tor. For each edge e⟨src, dst, weight⟩, the Scatter Unit loads
562
+ the feature vector of vsrc from input buffer and produces an
563
+ update u⟨dst, vector⟩. The update u⟨dst, vector⟩ is routed to
564
+ the corresponding Gather Unit and the Gather Unit applies the
565
+ update to the destination vertex vdst.
566
+ Mapping VUK: For VUK, we group a batch of vertices batch
567
+ and the feature vector of each vertex {hinput
568
+ i
569
+ : vi ∈ batch}
570
+ is multiplied by the weight matrix W simultaneously. The
571
+ output feature vectors are {houtput
572
+ i
573
+ : hinput
574
+ i
575
+ W , vi ∈ batch}.
576
+ To apply the Scatter-Gather paradigm, we perform feature
577
+ concatenation. For example, we concatenate the first feature
578
+ of each vertex {hi(1) : vi ∈ batch} as a vector rinput
579
+ 1
580
+ .
581
+ The vector rinput
582
+ 1
583
+ has src index 1 since its contains the 1st
584
+ feature of each input feature vector. For the weight matrix
585
+ W , we represent each non-zero element in the weight matrix
586
+ as an edge e⟨src, dst, weight⟩. During execution, for each
587
+ non-zero weight e⟨src, dst, weight⟩, the Scatter Unit loads
588
+ the rinput
589
+ src
590
+ from the input buffer and produces an update
591
+ u⟨dst, vector = e.weight × rinput
592
+ src
593
+ ⟩. Then, the Gather Unit
594
+ applies the update u⟨dst, vector⟩ to the destination routput
595
+ dst
596
+ .
597
+ routput
598
+ dst
599
+ contains the dstth features of each output feature
600
+ vector in the batch.
601
+ Note that VAK and VUK have different data layouts. In
602
+ VAK, the input/output feature vectors are stored in vertex-
603
+ major order. In VUK, the input/output feature vectors are
604
+ stored in feature-major order. To switch between the two data
605
+ layouts, we implement a Matrix Transformation Unit (MTU)
606
+
607
+ to perform data layout transformation.
608
+ C. hardware modules
609
+ Scatter/Gather Unit: A Scatter Unit has an array of q
610
+ processing elements. Each processing element has a multiplier
611
+ to perform the multiplication between an edge/weight and a
612
+ vertex feature. Similar to the Scatter Unit, a Gather Unit has an
613
+ array of q processing elements. Each processing element has
614
+ an Accumulator (ACC), a Max Unit, a ReLU Unit, a sigmoid
615
+ Unit. The multiplexer (MUX) and demultiplexer (DEMUX)
616
+ select the datapath for the current layer.
617
+ Routing Network: The routing network is implemented using
618
+ a hardware-efficient butterfly network [41].
619
+ Sigmoid Unit: We exploit the piecewise linear approximation
620
+ (PLA) [42] for Sigmoid Function.
621
+ V. LOAD BALANCE AND PERFORMANCE MODEL
622
+ A. Load Balance
623
+ Load balance in VAK: The workload balance of VAK
624
+ depends on how to partition the vertices into p memory
625
+ banks of the Result Buffer. Load imbalance is a significant
626
+ issue in GNN [43] if the graph has highly imbalanced degree
627
+ distribution. Thanks to our graph representation, the vertices
628
+ in the graph have degrees ranging from 0 to 4. We use a
629
+ greedy approach to keep the load balance of the p parallel
630
+ pipelines. For VAK, the destination vertices that have same
631
+ degree i (0 ⩽ i ⩽ 4) are evenly partitioned into p banks of
632
+ the Result Buffer. Through the proposed partitioning strategy,
633
+ each pipeline has the same amount of workload. The graph
634
+ partitioning has a small overhead O(|V|Lp) and is performed
635
+ by the APU, where Lp is the number of graph pooling layers in
636
+ the model. The proposed partitioning algorithm can be easily
637
+ parallelized using multiple threads on APU.
638
+ Load balance in VUK: To execute VUK, we need to partition
639
+ the weight matrix along the dst dimension (Figure 4). Each
640
+ Gather Unit is responsible for accumulating the partial results
641
+ of a partition. To achieve perfect load balance, each partition
642
+ should have the same number of non-zero elements. Since
643
+ the partitioning of weight matrix is an offline process, we are
644
+ able to adopt complexity algorithm to find the near optimal
645
+ data partitioning. In this work, we exploit Longest-processing-
646
+ time (LPT) first algorithm that is proved to achieve 4/3
647
+ approximation factor [44] to the optimal partition solution.
648
+ B. Performance Model
649
+ Modeling VAK: For a VAK kernel, the length of input feature
650
+ vector cin is same as the length of output feature vector cout:
651
+ cin = cout. A Scatter Unit or a Gather Unit can process q
652
+ features in each clock cycle. The p parallel pipelines can
653
+ process p edges simultaneously. Therefore, the execution time
654
+ of a VAK kernel is:
655
+ tVAK =
656
+ �|E|
657
+ p
658
+
659
+ ·
660
+ �cin
661
+ q
662
+
663
+ (3)
664
+ Modeling VUK: To execute a VUK, the accelerator groups
665
+ a batch of q vertices at a time to fully utilize the Scatter
666
+ Unit/Gather Unit. The p parallel pipelines can process p non-
667
+ zero elements in the weight matrix. Therefore, the execution
668
+ time of a VUK kernel is:
669
+ tVUK =
670
+ �|V|
671
+ q
672
+
673
+ ·
674
+ �nnz(W )
675
+ p
676
+
677
+ (4)
678
+ where nnz(W ) is the number of non-zero elements in the
679
+ weight matrix W . Since our accelerator exploits the computa-
680
+ tion parallelism within each kernel, the total execution time is
681
+ the sum of the execution time of all kernels and preprocessing
682
+ overhead.
683
+ VI. IMPLEMENTATION AND EXPERIMENTAL RESULTS
684
+ A. Implementation Details and Resource Utilizations
685
+ We implement our accelerator on an embedded FPGA plat-
686
+ form – Xilinx ZCU104. We implement 8 pipelines (8 Scatter
687
+ Units and 8 Gather Units). Each Scatter/Gather Unit has 16
688
+ processing elements (PEs). In a Scatter Unit, a PE consumes
689
+ 3 DSPs and in a Gather Unit, a PE consumes 7 DSPs. The
690
+ routing network has 8 input ports and 8 output ports. Each
691
+ port is 512-bit that can
692
+ receive/send 16 32-bit data. The
693
+ APU is a quad-core ARM-A53 processor running at 1.3 GHz.
694
+ The accelerator is developed using High-Level Synthesis. The
695
+ accelerator consumes 1280 DSPs, 96 URAMs, 221 BRAMs,
696
+ 178K LUTs. The accelerator runs at 125 MHz. The resource
697
+ utilization and frequency are reported after Place&Route.
698
+ B. Benchmark and Baseline Platform
699
+ Benchmark: We conduct experiments using the widely used
700
+ MSTAR dataset. The setting of MSTAR dataset follows the
701
+ state-of-the-art work [5], [6], [8], [9]. The dataset contains the
702
+ SAR images of 10 classes of ground vehicles. The training set
703
+ has 2747 images and the testing set has 2427 images. Each
704
+ SAR image has size 128×128 and each pixel has a grayscale
705
+ value indicating the magnitude of the SAR signal.
706
+ TABLE II: Specifications of various platforms
707
+ Platforms
708
+ CPU
709
+ AMD Ryzen 3990x
710
+ GPU
711
+ Nvidia RTX3090
712
+ FPGA
713
+ ZCU 104
714
+ Release Year
715
+ 2020
716
+ 2020
717
+ 2018
718
+ Technology
719
+ TSMC 7 nm
720
+ TSMC 7 nm
721
+ TSMC 16 nm
722
+ Frequency
723
+ 2.9 GHz
724
+ 1.7 GHz
725
+ 125 MHz
726
+ On-chip Memory
727
+ 256 MB L3 cache
728
+ 6 MB L2 cache
729
+ 4.8 MB
730
+ Baseline Platform: We compare our performance with the
731
+ state-of-the-art CPU and GPU platforms as shown in Table II.
732
+ On the CPU platform and GPU platform, we run the proposed
733
+ model using Pytorch Geometry (PyG) [45] of 1.8.0 version.
734
+ For CPU platform, PyG uses the Intel MKL as the backend
735
+ and for the GPU platform, PyG uses the CUDA 11.1 as the
736
+ backend. To exploit the sparsity of the weight matrices on the
737
+ CPU and GPU platforms, we modify the GraphSAGE layer1
738
+ of PyG by using the torch.sspaddmm() for efficient
739
+ multiplication of feature vectors and sparse weight matrices.
740
+ 1https://pytorch-geometric.readthedocs.io/en/latest/
741
+ modules/torch geometric/nn/conv/sage conv.html#SAGEConv
742
+
743
+ C. Accuracy, Computation Complexity, Model Size
744
+ Weight/Input pruning: The magnitude of the SAR signal
745
+ ranges from 0 to 8. we set the Ivertex as 0.1 because it can filter
746
+ out most irrelevant pixels. We compare Accuracy, computation
747
+ Type
748
+ Accuracy
749
+ # of FLOPs
750
+ # of Para.
751
+ Model Size
752
+ [5]
753
+ CNN
754
+ 92.3%
755
+ 1
756
+ 12 ×
757
+ 0.5 × 106
758
+ 16 Mb
759
+ [8]
760
+ CNN
761
+ 97.97%
762
+ 1
763
+ 10 ×
764
+ 0.65 × 106
765
+ 20.8 Mb
766
+ [9]
767
+ CNN
768
+ 98.52%
769
+ 1
770
+ 3 ×
771
+ 2.1 × 106
772
+ 67.2 Mb
773
+ [6]
774
+ CNN
775
+ 99.3%
776
+ 1× (6.94 GFLOPs)
777
+ 2.5 × 106
778
+ 80 Mb
779
+ This work
780
+ GNN
781
+ 99.09%
782
+ 1
783
+ 3258 ×
784
+ 0.03 × 106
785
+ 0.96 Mb
786
+ complexity, number of parameters with state-of-the-art work
787
+ [5], [6], [8], [9]. Compared with the state-of-the-art CNN [6],
788
+ the proposed model achieves comparable accuracy with only
789
+ 1
790
+ 3258 computation complexity and
791
+ 1
792
+ 83 number of parameters
793
+ on average.
794
+ D. Evaluation of Latency
795
+ Fig. 5: X-axis is the index of the SAR image (training set +
796
+ testing set). Y-axis is the inference latency of a SAR image.
797
+ To compare the latency of various platforms, we set the
798
+ batch size as 1. The measured latency on FPGA accelerator is
799
+ end-to-end from the time when APU receives the SAR image
800
+ to the time when APU gets the classification results from
801
+ the accelerator, which means the preprocessing overhead is
802
+ included in the measured latency. We measure the inference
803
+ latency on all images in training and testing sets. The com-
804
+ parison results are shown in Figure 5. On average, our FPGA
805
+ accelerator is 14.8×, 2.5× faster than the CPU and GPU
806
+ platforms in terms of latency. Since we use the input pruning,
807
+ the graph representations of the images after input pruning
808
+ have various number of vertices. Therefore, the inference
809
+ latency fluctuates with images. Compared with CPU/GPU, our
810
+ accelerator has lower latency. Because CPU/GPU has complex
811
+ cache hierarchy and large cache latency (e.g., CPU has high
812
+ cache latency: L3 cache 32ns, L2 cache 12ns). Therefore,
813
+ loading feature vectors and weight matrices leads to large
814
+ latency. In contrast, our FPGA accelerator can access data in
815
+ one-clock cycle due to our customized on-chip memory orga-
816
+ nization. Moreover, our FPGA accelerator adopts the Scatter-
817
+ Gather paradigm to efficiently deal with irregular computation
818
+ in various computation kernels.
819
+ Impact of model design: To compare the inference latency
820
+ with the state-of-the-art CNNs, we deploy AMD Xilinx DPU
821
+ [28] (2 * B4096 @ 300 MHz configuration) on the same
822
+ TABLE III: Latency comparison on ZCU 104 and GPU
823
+ Model
824
+ [5]
825
+ [8]
826
+ [9]
827
+ [6]
828
+ Proposed model
829
+ [Xilinx DPU]
830
+ [Proposed design]
831
+ ZCU104
832
+ 0.88 ms
833
+ 1.23 ms
834
+ 3.09 ms
835
+ 12.1 ms
836
+ 0.105 ms
837
+ GPU (RTX3090)
838
+ 1.53 ms
839
+ 2.5 ms
840
+ 9.5 ms
841
+ 31.2 ms
842
+ 0.269 ms
843
+ FPGA platform (ZCU 104) to execute the CNN models in [5],
844
+ [6], [8], [9]. AMD Xilinx DPU is the state-of-the-art FPGA
845
+ overlay accelerator for CNNs. The average inference latency is
846
+ shown in Table III. The proposed GNN on the proposed design
847
+ (The column 6 of Table III) is 115× faster than [6] on DPU.
848
+ Note that DPU uses 8-bit data quantization for the weights and
849
+ activations. Our work uses 32-bit floating point data format.
850
+ DPU has more computation parallelism by operating on 8-bit
851
+ data.
852
+ Preprocessing Overhead: We measure the preprocessing
853
+ overhead on APU. For a SAR image, APU transforms it
854
+ into graph representation (Section III-A) with input pruning
855
+ (Section III-B), and graph partitioning (V-A). The average
856
+ preprocessing time is 11.8 us for a SAR image, which is
857
+ negligible compared with the total latency.
858
+ TABLE IV: Comparison of Energy Consumption
859
+ Platform
860
+ Inference Speed
861
+ Power
862
+ Energy (mJ/image)
863
+ Ryzen 3990X
864
+ 644 (image/s)
865
+ 26.5W
866
+ 41.1 (mJ/image)
867
+ Nvidia RTX3090
868
+ 3717 (image/s)
869
+ 97W
870
+ 26.0 (mJ/image)
871
+ ZCU104
872
+ 9500 (image/s)
873
+ 6.3W
874
+ 0.66 (mJ/image)
875
+ E. Evaluation of Energy Consumption
876
+ Table IV shows the comparison of energy consumption
877
+ on various platforms. On the CPU platform, we measure
878
+ the power consumption of the inference program using
879
+ PowerTOP [46]. On the GPU platform, we measure power
880
+ consumption using nvidia-smi [47] command tool. For the
881
+ FPGA board (ZCU 104), we use an external power meter
882
+ to measure its power consumption. The reported numbers in
883
+ Table IV are the average power consumption during inference.
884
+ The results show that our FPGA accelerator is 62×, 39× more
885
+ energy efficient than CPU and GPU platform, respectively.
886
+ VII. CONCLUSION
887
+ In this paper, we propose a novel model-architecture co-
888
+ design for SAR ATR on FPGA. The proposed lightweight
889
+ GNN model achieves similar accuracy with state-of-the-art
890
+ models with only 1/3258 computation complexity and 1/83
891
+ model size. The proposed accelerator on an embedded FPGA
892
+ platform has lower latency than the state-of-the-art CPU/GPU
893
+ with significant less energy consumption.
894
+ ACKNOWLEDGMENT
895
+ This work is supported by the National Science Foundation
896
+ (NSF) under grants OAC-1911229, CNS-2009057, and in
897
+ part by DEVCOM Army Research Lab (ARL) under ARL-
898
+ USC collaborative grant DIRA-ECI:DEC21-CI-037. The au-
899
+ thor Bingyi Zhang is supported by the Summer Research
900
+ Program from the Army Research Lab West (ARL West).
901
+
902
+ Comparison of Latency
903
+ FPGA (ZCU104)
904
+ CPU (Ryzen 3990X)
905
+ GPU (RTX3090)
906
+ (second
907
+ atency
908
+ a
909
+ 10
910
+ 500
911
+ 1000
912
+ 1500
913
+ 2000
914
+ 2500
915
+ 3000
916
+ 3500
917
+ 4000
918
+ 4500
919
+ 5000
920
+ SAR image indexREFERENCES
921
+ [1] L. Landuyt, A. Van Wesemael, G. J.-P. Schumann, R. Hostache, N. E.
922
+ Verhoest, and F. M. Van Coillie, “Flood mapping based on synthetic
923
+ aperture radar: An assessment of established approaches,” IEEE Trans-
924
+ actions on Geoscience and Remote Sensing, vol. 57, no. 2, pp. 722–739,
925
+ 2018.
926
+ [2] P. Zhan, W. Zhu, and N. Li, “An automated rice mapping method based
927
+ on flooding signals in synthetic aperture radar time series,” Remote
928
+ Sensing of Environment, vol. 252, p. 112112, 2021.
929
+ [3] N. Li, Z. Guo, J. Zhao, L. Wu, and Z. Guo, “Characterizing ancient
930
+ channel of the yellow river from spaceborne sar: Case study of chi-
931
+ nese gaofen-3 satellite,” IEEE Geoscience and Remote Sensing Letters,
932
+ vol. 19, pp. 1–5, 2021.
933
+ [4] T. Zhang, X. Zhang, J. Shi, and S. Wei, “Hyperli-net: A hyper-light deep
934
+ learning network for high-accurate and high-speed ship detection from
935
+ synthetic aperture radar imagery,” ISPRS Journal of Photogrammetry
936
+ and Remote Sensing, vol. 167, pp. 123–153, 2020.
937
+ [5] M. Zhang, J. An, L. D. Yang, L. Wu, X. Q. Lu et al., “Convolutional
938
+ neural network with attention mechanism for sar automatic target
939
+ recognition,” IEEE Geoscience and Remote Sensing Letters, vol. 19,
940
+ pp. 1–5, 2020.
941
+ [6] D. A. Morgan, “Deep convolutional neural networks for atr from sar
942
+ imagery,” in Algorithms for Synthetic Aperture Radar Imagery XXII,
943
+ vol. 9475.
944
+ SPIE, 2015, pp. 116–128.
945
+ [7] J. Hu, L. Shen, and G. Sun, “Squeeze-and-excitation networks,” in
946
+ Proceedings of the IEEE conference on computer vision and pattern
947
+ recognition, 2018, pp. 7132–7141.
948
+ [8] J. Pei, Y. Huang, W. Huo, Y. Zhang, J. Yang, and T.-S. Yeo, “Sar auto-
949
+ matic target recognition based on multiview deep learning framework,”
950
+ IEEE Transactions on Geoscience and Remote Sensing, vol. 56, no. 4,
951
+ pp. 2196–2210, 2017.
952
+ [9] Z. Ying, C. Xuan, Y. Zhai, B. Sun, J. Li, W. Deng, C. Mai, F. Wang, R. D.
953
+ Labati, V. Piuri et al., “Tai-sarnet: Deep transferred atrous-inception cnn
954
+ for small samples sar atr,” Sensors, vol. 20, no. 6, p. 1724, 2020.
955
+ [10] F. Bardi, W. Frodella, A. Ciampalini, S. Bianchini, C. Del Ventisette,
956
+ G. Gigli, R. Fanti, S. Moretti, G. Basile, and N. Casagli, “Integration
957
+ between ground based and satellite sar data in landslide mapping: The
958
+ san fratello case study,” Geomorphology, vol. 223, pp. 45–60, 2014.
959
+ [11] H. Septanto and O. Sudjana, “Simulation-based energy balance analysis
960
+ of sar micro-satellite,” in Sixth International Symposium on LAPAN-IPB
961
+ Satellite, vol. 11372.
962
+ International Society for Optics and Photonics,
963
+ 2019, p. 113721W.
964
+ [12] Y. Yokota, Y. Okada, K. Iribe, M. Tsuji, A. Ando, and Y. Kunii, “Newly
965
+ developed x-band sar system onboard japanese small satellite “asnaro-
966
+ 2”,” in Conference Proceedings of 2013 Asia-Pacific Conference on
967
+ Synthetic Aperture Radar (APSAR).
968
+ IEEE, 2013, pp. 81–83.
969
+ [13] P. R. Akbar, H. Saito, M. Zhang, J. Hirokawa, and M. Ando, “Parallel-
970
+ plate slot array antenna for deployable sar antenna onboard small
971
+ satellite,” IEEE Transactions on Antennas and Propagation, vol. 64,
972
+ no. 5, pp. 1661–1671, 2016.
973
+ [14] K. Tanaka, H. Saito, H. Watanabe, K. Ijichi, K. Hirako, and S. Shi-
974
+ rasaka, “Development of 1kw high power x-band sar installed on small
975
+ satellite for on-demand observation,” in Proceedings of the International
976
+ Astronautical Congress, IAC, vol. 2018.
977
+ International Astronautical
978
+ Federation, IAF, 2018.
979
+ [15] H. Zhu, N. Lin, H. Leung, R. Leung, and S. Theodoidis, “Target
980
+ classification from sar imagery based on the pixel grayscale decline
981
+ by graph convolutional neural network,” IEEE Sensors Letters, vol. 4,
982
+ no. 6, pp. 1–4, 2020.
983
+ [16] K. Xu, W. Hu, J. Leskovec, and S. Jegelka, “How powerful are graph
984
+ neural networks?” arXiv preprint arXiv:1810.00826, 2018.
985
+ [17] T. N. Kipf and M. Welling, “Semi-supervised classification with graph
986
+ convolutional networks,” arXiv preprint arXiv:1609.02907, 2016.
987
+ [18] W. L. Hamilton, R. Ying, and J. Leskovec, “Inductive representation
988
+ learning on large graphs,” in Proceedings of the 31st International
989
+ Conference on Neural Information Processing Systems, 2017, pp. 1025–
990
+ 1035.
991
+ [19] P. Veliˇckovi´c, G. Cucurull, A. Casanova, A. Romero, P. Lio, and Y. Ben-
992
+ gio, “Graph attention networks,” arXiv preprint arXiv:1710.10903, 2017.
993
+ [20] V. Gligorijevi´c, P. D. Renfrew, T. Kosciolek, J. K. Leman, D. Berenberg,
994
+ T. Vatanen, C. Chandler, B. C. Taylor, I. M. Fisk, H. Vlamakis et al.,
995
+ “Structure-based protein function prediction using graph convolutional
996
+ networks,” Nature communications, vol. 12, no. 1, pp. 1–14, 2021.
997
+ [21] H. Zhu, X. Du, and Y. Yao, “Convsppis: identifying protein-protein
998
+ interaction sites by an ensemble convolutional neural network with
999
+ feature graph,” Current Bioinformatics, vol. 15, no. 4, pp. 368–378,
1000
+ 2020.
1001
+ [22] T. Zhao, Y. Hu, L. R. Valsdottir, T. Zang, and J. Peng, “Identifying
1002
+ drug–target interactions based on graph convolutional network and deep
1003
+ neural network,” Briefings in bioinformatics, vol. 22, no. 2, pp. 2141–
1004
+ 2150, 2021.
1005
+ [23] C. R. Qi, H. Su, K. Mo, and L. J. Guibas, “Pointnet: Deep learning on
1006
+ point sets for 3d classification and segmentation,” in Proceedings of the
1007
+ IEEE conference on computer vision and pattern recognition, 2017, pp.
1008
+ 652–660.
1009
+ [24] C. R. Qi, L. Yi, H. Su, and L. J. Guibas, “Pointnet++: Deep hierarchical
1010
+ feature learning on point sets in a metric space,” Advances in neural
1011
+ information processing systems, vol. 30, 2017.
1012
+ [25] Y. Ding, X. Zhao, Z. Zhang, W. Cai, N. Yang, and Y. Zhan, “Semi-
1013
+ supervised locality preserving dense graph neural network with arma
1014
+ filters and context-aware learning for hyperspectral image classification,”
1015
+ IEEE Transactions on Geoscience and Remote Sensing, 2021.
1016
+ [26] Y. Ding, X. Zhao, Z. Zhang, W. Cai, and N. Yang, “Graph sample
1017
+ and aggregate-attention network for hyperspectral image classification,”
1018
+ IEEE Geoscience and Remote Sensing Letters, vol. 19, pp. 1–5, 2021.
1019
+ [27] H. D. Nguyen, X.-S. Vu, and D.-T. Le, “Modular graph transformer
1020
+ networks for multi-label image classification,�� in Proceedings of the
1021
+ AAAI Conference on Artificial Intelligence, vol. 35, no. 10, 2021, pp.
1022
+ 9092–9100.
1023
+ [28] “Xilinx dpu.” [Online]. Available: https://github.com/Xilinx/Vitis-AI
1024
+ [29] M. Yan, L. Deng, X. Hu, L. Liang, Y. Feng, X. Ye, Z. Zhang, D. Fan,
1025
+ and Y. Xie, “Hygcn: A gcn accelerator with hybrid architecture,” in
1026
+ 2020 IEEE International Symposium on High Performance Computer
1027
+ Architecture (HPCA).
1028
+ IEEE, 2020, pp. 15–29.
1029
+ [30] B. Zhang, R. Kannan, and V. Prasanna, “Boostgcn: A framework
1030
+ for optimizing gcn inference on fpga,” in 2021 IEEE 29th Annual
1031
+ International Symposium on Field-Programmable Custom Computing
1032
+ Machines (FCCM).
1033
+ IEEE, 2021, pp. 29–39.
1034
+ [31] Y.-C. Lin, B. Zhang, and V. Prasanna, “Hp-gnn: Generating high
1035
+ throughput gnn training implementation on cpu-fpga heterogeneous plat-
1036
+ form,” in Proceedings of the 2022 ACM/SIGDA International Symposium
1037
+ on Field-Programmable Gate Arrays, 2022, pp. 123–133.
1038
+ [32] H. Zeng and V. Prasanna, “Graphact: Accelerating gcn training on cpu-
1039
+ fpga heterogeneous platforms,” in Proceedings of the 2020 ACM/SIGDA
1040
+ International Symposium on Field-Programmable Gate Arrays, 2020,
1041
+ pp. 255–265.
1042
+ [33] B. Zhang, H. Zeng, and V. Prasanna, “Hardware acceleration of large
1043
+ scale gcn inference,” in 2020 IEEE 31st International Conference
1044
+ on Application-specific Systems, Architectures and Processors (ASAP).
1045
+ IEEE, 2020, pp. 61–68.
1046
+ [34] B. Zhang, S. R. Kuppannagari, R. Kannan, and V. Prasanna, “Efficient
1047
+ neighbor-sampling-based gnn training on cpu-fpga heterogeneous plat-
1048
+ form,” in 2021 IEEE High Performance Extreme Computing Conference
1049
+ (HPEC).
1050
+ IEEE, 2021, pp. 1–7.
1051
+ [35] H. Zhou, B. Zhang, R. Kannan, V. Prasanna, and C. Busart, “Model-
1052
+ architecture co-design for high performance temporal gnn inference on
1053
+ fpga,” arXiv preprint arXiv:2203.05095, 2022.
1054
+ [36] M. Zhang and Y. Chen, “Link prediction based on graph neural net-
1055
+ works,” Advances in Neural Information Processing Systems, vol. 31,
1056
+ pp. 5165–5175, 2018.
1057
+ [37] R. Ying, J. You, C. Morris, X. Ren, W. L. Hamilton, and J. Leskovec,
1058
+ “Hierarchical graph representation learning with differentiable pooling,”
1059
+ arXiv preprint arXiv:1806.08804, 2018.
1060
+ [38] “Mstar dataset.” [Online]. Available: https://www.sdms.afrl.af.mil/index.
1061
+ php?collection=public-data&page=public-data-list
1062
+ [39] S. Woo, J. Park, J.-Y. Lee, and I. S. Kweon, “Cbam: Convolutional
1063
+ block attention module,” in Proceedings of the European conference on
1064
+ computer vision (ECCV), 2018, pp. 3–19.
1065
+ [40] R. Tibshirani, “Regression shrinkage and selection via the lasso,” Jour-
1066
+ nal of the Royal Statistical Society: Series B (Methodological), vol. 58,
1067
+ no. 1, pp. 267–288, 1996.
1068
+ [41] Y.-k. Choi, Y. Chi, W. Qiao, N. Samardzic, and J. Cong, “Hbm
1069
+ connect: High-performance hls interconnect for fpga hbm,” in The 2021
1070
+ ACM/SIGDA International Symposium on Field-Programmable Gate
1071
+ Arrays, 2021, pp. 116–126.
1072
+ [42] Y. Zhang, C. Wang, L. Gong, Y. Lu, F. Sun, C. Xu, X. Li, and
1073
+ X. Zhou, “Implementation and optimization of the accelerator based
1074
+
1075
+ on fpga hardware for lstm network,” in 2017 IEEE International
1076
+ Symposium on Parallel and Distributed Processing with Applications
1077
+ and 2017 IEEE International Conference on Ubiquitous Computing and
1078
+ Communications (ISPA/IUCC).
1079
+ IEEE, 2017, pp. 614–621.
1080
+ [43] T. Geng, A. Li, R. Shi, C. Wu, T. Wang, Y. Li, P. Haghi, A. Tumeo,
1081
+ S. Che, S. Reinhardt et al., “Awb-gcn: A graph convolutional network
1082
+ accelerator with runtime workload rebalancing,” in 2020 53rd Annual
1083
+ IEEE/ACM International Symposium on Microarchitecture (MICRO).
1084
+ IEEE, 2020, pp. 922–936.
1085
+ [44] B. T. Eck and M. Pinedo, “On the minimization of the makespan subject
1086
+ to flowtime optimality,” Operations Research, vol. 41, no. 4, pp. 797–
1087
+ 801, 1993.
1088
+ [45] M. Fey and J. E. Lenssen, “Fast graph representation learning with
1089
+ PyTorch Geometric,” in ICLR Workshop on Representation Learning
1090
+ on Graphs and Manifolds, 2019.
1091
+ [46] “Powertop.” [Online]. Available: https://github.com/fenrus75/powertop
1092
+ [47] “nvidia-smi.”
1093
+ [Online].
1094
+ Available:
1095
+ https://developer.download.nvidia.
1096
+ com/compute/DCGM/docs/nvidia-smi-367.38.pdf
1097
+
4NAzT4oBgHgl3EQfffyv/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
4dE0T4oBgHgl3EQfeQDL/content/2301.02389v1.pdf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7355ae0f3bfc87fc62c8ece5aad833d24f00d0d80145c0594f24bf61f781e99b
3
+ size 279094
4dE0T4oBgHgl3EQfeQDL/vector_store/index.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1523d651dd0919d77267a60a18bf164e6d00933da584501314b68eb1b2fa275b
3
+ size 134074
4dFAT4oBgHgl3EQfExzK/vector_store/index.faiss ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6af6c14cee702e9d3e2fc3d7dfa0f522c2dc4f681e0d4e280ce5d9eaf6fd8b1b
3
+ size 3342381
69E1T4oBgHgl3EQfBgJT/content/2301.02852v1.pdf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:045bf5acdcab79701e900112360da3e508fe2e638f1675aaabc41681c01271b6
3
+ size 2488568
6NE0T4oBgHgl3EQfvwGn/content/tmp_files/2301.02623v1.pdf.txt ADDED
@@ -0,0 +1,626 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Does Gaia Play Dice? : Simple Models of non-Darwinian
2
+ Selection
3
+ Rudy Arthur,1∗Arwen Nicholson,2†
4
+ 1University of Exeter, Department of Computer Science
5
+ 2University of Exeter, Department of Physics and Astronomy
6
+ January 9, 2023
7
+ Abstract
8
+ In this paper we introduce some simple models, based on rolling dice, to explore mechanisms
9
+ proposed to explain planetary habitability. The idea is to study these selection mechanisms
10
+ in an analytically tractable setting, isolating their consequences from other details which can
11
+ confound or obscure their effect in more realistic models. We find that while the observable
12
+ of interest, the face value shown on the die, ‘improves’ over time in all models, for two of the
13
+ more popular ideas: Selection by Survival and Sequential Selection, this is down to sampling
14
+ effects. A modified version of Sequential Selection, Sequential Selection with Memory, implies
15
+ a statistical tendency for systems to improve over time. We discuss the implications of this and
16
+ its relationship to the ideas of the ‘inhabitance paradox’ and the ‘Gaian bottleneck’.
17
+ 1
18
+ Introduction
19
+ Relatively recent discussion about the persistence of life over long periods has brought to the fore
20
+ various selection principles [1, 2, 3, 4]. With the recent launch of the James Webb Space Telescope
21
+ [5, 6] these questions not only have important implications for our understanding of Earth history,
22
+ but also for the search for other inhabited planets. In the future, with a large enough catalogue of
23
+ inhabited planets, it may be possible to experimentally investigate alternative trajectories for life
24
+ [7]. Until then, understanding general principles behind planetary habitability and inhabitance is a
25
+ way to provide working hypotheses that can explain both how ‘lucky’ the Earth is to be inhabited
26
+ and what we might expect on planets orbiting other stars.
27
+ The most discussed of these selection principles is called ‘Selection by Survival’ (SBS) in [3]
28
+ though the idea has many names, see [8]. The essential idea is that a population where the entities
29
+ have different rates of survival will be ‘purified’ so that, in the long run, surviving entities have
30
+ must have properties conducive to survival. Several works e.g. [1, 2, 9], attempt to disentangle
31
+ Darwinian selection from this ‘differential persistence’ (essentially a synonym for SBS). [2] and [9]
32
+ emphasise the importance of this selection principle acting on higher order phenomena like whole
33
+ ecosystems, planetary scale biogeochemical cycles and the entire life-Earth coupled system i.e. Gaia
34
+ [10]. They argue that in systems of hereditary replicators Darwinian selection is more powerful,
35
+ however for entities like populations, ecosystems or bio-geochemical cycles [11], which do not have
36
+ ∗E-mail: [email protected]
37
+ †E-mail: [email protected]
38
+ 1
39
+ arXiv:2301.02623v1 [q-bio.PE] 6 Jan 2023
40
+
41
+ strict heredity and reproduction, SBS will operate to favour certain macroscopic features. Specific
42
+ examples suggested by e.g. [8] are sexual reproduction and macroevolutionary freezing.
43
+ [3] defines another, related, selection principle called Sequential Selection (SS) [12]. This is a
44
+ similar idea to SBS, but motivated by the frequent upheavals in the history of life on Earth and
45
+ meant to account for life’s apparent stabilising effect on Earth’s habitability. [3] propose a simple
46
+ algorithm - evolutionary innovations have a stabilizing or destabilizing effect on the environment.
47
+ If they have a destabilizing effect, habitability is reduced, eventually eliminating the destabilizing
48
+ innovation. In this way destabilizing effects are eliminated by ‘near fatal resets’ while stabilizing
49
+ innovations persist and accumulate.
50
+ In [13, 4] we argue for a refinement of this algorithm, emphasising that the resets are‘near fatal
51
+ i.e. the evolutionary innovations developed during the previous stable period are not completely lost.
52
+ The algorithm of [3] applies: destabilizing innovations lead to resets which greatly reduce species
53
+ abundance but have a lesser effect on species diversity. The life-earth system which arises after
54
+ the reset is selected from a larger ‘pool’, which has the potential to generate better, more stable
55
+ ecosystems. Higher species and functional diversity give Gaia more tools to generate stability. The
56
+ process is completely blind, so unstable states can also be selected, however, by definition, these
57
+ are short lived and eventually a long-lived stable state will arise. During this stable period species
58
+ diversity can increase again leading to a kind of ratcheting effect. To emphasise this cumulative
59
+ process, in contrast to the sequential selection algorithm of [3], we call this ‘Sequential Selection
60
+ with Memory’ (SSM).
61
+ A variety of abstract models of varying complexity have been proposed to explore these selection
62
+ principles e.g.
63
+ [1, 14, 15, 4, 16].
64
+ Though these models have great value, it can sometimes be
65
+ unclear which of their features are programmed in (as alleged by [11] of the famous Daisyworld
66
+ [17] model) and which are emergent. It must also be said that the mathematical or computational
67
+ complexity of these models can give them an air of mystery - especially to biologists not well versed
68
+ in these methods. Indeed, the very fact that the key model features are often emergent means that
69
+ understanding how they emerge requires a detailed understanding of each model’s dynamics.
70
+ Here we propose an extremely simple probability model as a setting to study selection princi-
71
+ ples. The aim is to strip out as much complexity as possible to understand the core meaning of
72
+ these principles and their consequences. A very loose analogy would be trying to understand the
73
+ approximately Gaussian distribution of, say, human height. This has some genetic and environ-
74
+ mental causes which, with great difficultly, could be experimentally isolated and formulated into a
75
+ mechanistic model of height, simulated and shown to result in a Gaussian distribution. However, a
76
+ much simpler, and in many ways more satisfactory, explanation is that a Gaussian distribution is
77
+ the expected outcome for an observable which is a sum of independent effects.
78
+ Continuing the height analogy, by simulating sums of random variables and showing this results
79
+ in a Gaussian distribution we might start to suspect that a more general principle is operating,
80
+ one which isn’t affected by the particular details of our model, in this case, the Central Limit
81
+ Theorem. This paper doesn’t propose anything as general as a statistical convergence theorem,
82
+ what we do propose are models simple enough to be analytically solved but complex enough to
83
+ see selection principles operating. Theses models will be shown to exhibit interesting behaviour
84
+ which is also observed in more complex models. The aim is to provide some clarity on exactly what
85
+ non-Darwinian selection principles can do in a clear and tractable setting.
86
+ 2
87
+
88
+ 2
89
+ Introducing the Model
90
+ Consider an M sided die with the rule that, once rolled, whatever number is showing on the top face
91
+ gives the number of steps to wait before rolling again or finishing the game. For r ≥ 1 dice we roll
92
+ each one independently to get x1, x2, . . . , xr, and take the highest face value: max(x1, x2, . . . , xr).
93
+ Based on this consider the following dice games:
94
+ 1. Selection By Survival(SBS): Roll N (where N is a very large number) independent dice
95
+ once each.
96
+ 2. Sequential Selection (SS): Roll one die repeatedly for T time steps.
97
+ 3. Sequential Selection with Memory (SSM):
98
+ (a) Starting with r = 1, roll r dice repeatedly for T time steps. Add a new die every time
99
+ the top face shows the maximum value, M.
100
+ (b) Starting with M = 1, roll an M sided die repeatedly for T times steps. Every time the
101
+ top face shows the maximum value M, increase M by 1.
102
+ The SSM games are reminiscent of the Polya Urn model, though have not been studied before to our
103
+ knowledge. The quantity of interest will be the expected face value at time t. The names chosen are
104
+ based on the discussion in the Introduction and follow the conventions of [3] and [4]. Our version of
105
+ Selection By Survival is much simpler than the (mostly verbal) models proposed by others e.g. [9]
106
+ and most closely follows the graphical model from [2].
107
+ As a rough mapping to reality - a die represents an inhabited ‘planet’. Each roll is a period
108
+ of stability for the planet’s biosphere. The face value represents something akin to the ‘fitness’ of
109
+ the biosphere on that planet, i.e. how long it will persist. If we observe an inhabited planet at
110
+ some random point in its history we may see a biosphere with properties conducive to long term
111
+ stability (high face value) or only short term stability (low face value). The question of interest for
112
+ astrobiology is, if we were to survey a large catalogue of inhabited planets, what would be the average
113
+ ‘fitness’? For Earth history (or for the history of any inhabited planet) the equivalent question is,
114
+ if we were to observe a planet at a random point in its history, what should we expect about the
115
+ habitability properties of that planet?
116
+ 3
117
+
118
+ 3
119
+ Selection by Survival
120
+ t = 1
121
+ t = 2
122
+ t = 3
123
+ t = 4
124
+ t = 5
125
+ t = 6
126
+ Figure 1: One possible unfolding of the SBS game with N = 25 and M = 6. At t = 1 we have our
127
+ initial ensemble, at t = 2 we have removed all the 1s, at t = 3 we remove all the 2s etc.
128
+ Figure 1 shows one realisation of the SBS game. At t = 1 all of the dice are in play and the average
129
+ face value (over very large N or many different realisations of the same game) is
130
+ (1 + 2 + . . . + M)/M
131
+ at t = 2 all of the dice showing 1 on the top face are removed. Restricting our survey to inhabited
132
+ planets, the average face value is now
133
+ (2 + 3 + . . . + M)/(M − 1)
134
+ At time t ≤ M the average face value is
135
+ (t + (t + 1) + . . . + M)
136
+ (M − t)
137
+ = M + t
138
+ 2
139
+ (1)
140
+ So that average face value increases linearly with time.
141
+ 4
142
+
143
+ :
144
+ :围
145
+ 880
146
+ 围Figure 2: Average face value in the SBS game as a function of t for an M = 10 sided dice over
147
+ N = 1000 dice rolls.
148
+ Figure 2 shows the result of simulations of the game compared to equation 1. In terms of ‘planets’
149
+ this model is simply stating the (obvious) fact that planets which survive have properties (high face
150
+ value) which allow them to survive! Looking at the catalogue of inhabited planets will necessarily
151
+ yield planets with properties conducive to maintaining life, without the need for any additional
152
+ mechanism.
153
+ This realisation has all planets are seeded with life at the same time.
154
+ More complex games
155
+ could be devised (say a constant rate of habitable planet generation) to study how the generateion
156
+ rate interacts with this simple selection mechanism. For this paper, SBS represents a basic null
157
+ model - older inhabited planets must have features which have enabled them to remain inhabited.
158
+ The growth in fitness of the ‘surviving’ planets is simply a sampling artefact, the average fitness
159
+ of an inhabited planet increases because we throw away more and more of the unfit planets from
160
+ our average. Considering our solar system according to SBS, the single inhabited planet we see is
161
+ habitable because if it wasn’t, we wouldn’t be looking at it, or living on it. Thus in this context,
162
+ SBS is nothing more than an observer effect or anthropic principle.
163
+ 5
164
+
165
+ 10
166
+ Exact
167
+ Average of 1000 runs
168
+ 8
169
+ Face Value
170
+ 6
171
+ 4
172
+ 2
173
+ 0
174
+ 2
175
+ 6
176
+ 8
177
+ 10
178
+ 4
179
+ t4
180
+ Sequential Selection
181
+ The Earth has experienced numerous mass extinction events, had very different planetary regulation
182
+ mechanisms, atmospheric composition, levels of volcanic activity and life has persisted the entire
183
+ time [18]. We seek to model these sequential resets with another simple game: repeatedly rolling a
184
+ single die.
185
+ Figure 3: One possible unfolding of the SS game with T = 20 and M = 6. At t = 1 we roll 2 which
186
+ shows for 2 steps, we roll 1 which shows for 1 step, then 3 for 3 steps etc. The game is played a
187
+ large number, N, of times as in figure 1 so we are interested in average behaviour.
188
+ When observing the die at a random time t, what should we expect the face of the die to show,
189
+ on average? The chance of the die showing k is proportional to the probability of rolling a k, p(k),
190
+ times the number of ‘slots’ where the observation could occur e.g. if the die is showing 3 this could
191
+ be an observation of the die on the first, second or third step where it is face up. We normalise this
192
+ probability and compute the expected value for the top face as
193
+ M
194
+
195
+ k=1
196
+ k
197
+ kp(k)
198
+ �m
199
+ k=1 kp(k)
200
+ (2)
201
+ For one die p(k) = 1/M and this simplifies to
202
+ 2M + 1
203
+ 3
204
+ (3)
205
+ Note this is larger than the expected value of a single dice roll, M+1
206
+ 2
207
+ for M > 1.
208
+ 6
209
+
210
+ ·Figure 4: Average face value in the SS game as a function of t for an M = 10 sided dice, averaged
211
+ over N = 1000 independent instances. The expected value of a single 10 sided dice roll is 5.5 which
212
+ is less than the expected face value in the sequential sampling game, 7. Note the logged x-axis.
213
+ Figure 4 shows the results of simulations of the game compared to equation 3. The results here
214
+ imply that when observing a ‘planet’ at a random time, it is more likely to be in a state with stability
215
+ enhancing properties (high face value). Again this reflects the obvious fact that if we depict Earth’s
216
+ history as a time line and pick a random point on the line we are more likely to pick a point in a
217
+ long stable period. In particular, our present time is most likely to be a stable period, without the
218
+ need for any additional mechanism. Like with SBS, Earth’s current stability is simply an observer
219
+ effect.
220
+ One thing missing from this game (and from the algorithm of [3]) is the possibility of total extinc-
221
+ tion, that is, finishing the game early. We could implement an additional rule, say when we roll a 1,
222
+ stop the game. This would give a model where SBS and SS are both operating simultaneously. Here,
223
+ for simplicity and clarity, we don’t account for total extinction, so as not to mix the mechanisms.
224
+ All of our SS games persist for the same amount of time. More complex models e.g. [4, 19] do have
225
+ the possibility of early stopping and come to very similar conclusions.
226
+ 7
227
+
228
+ 7.4
229
+ Average of 1000 runs
230
+ Exact
231
+ 7.2
232
+ 7.0
233
+ 6.8
234
+ Face Value
235
+ 6.6
236
+ 6.4
237
+ 6.2
238
+ 6.0
239
+ 100
240
+ 101
241
+ 102
242
+ 103
243
+ t5
244
+ Sequential Selection with Memory
245
+ The continued inhabitance of Earth and the fact that biodiversity has increased over time motivates
246
+ the final games. Each reset does not start from scratch, but builds on evolutionary and ecological
247
+ innovations that came before.
248
+ We propose 2 models with an extremely simple ‘memory’.
249
+ This
250
+ memory is implemented in two ways, first by adding extra dice at fixed M, second by increasing M.
251
+ 5.1
252
+ Game A: Adding dice
253
+ The face value in this game is determined by rolling multiple dice and choosing the one with the
254
+ maximum face value. The idea is that stable biospheres outlast unstable ones. One could imagine
255
+ independent ecosystems co-existing with the final ‘reset’ only occurring when the most stable sub-
256
+ system collapses. If this seems contrived, in more complex models e.g. [4], a similar feature emerges
257
+ as a consequence of model dynamics rather than being enforced.
258
+ Figure 5: One possible unfolding of SSM game A with T = 20 and M = 6. At t = 3 we roll 6
259
+ which adds an extra die. At t = 13 we roll six again which adds a third die. The bottom row is the
260
+ observable, the other rows show dice rolls which are not observed.
261
+ The analysis is a little more complex that the previous two games. The first thing we need is the
262
+ probability to get the face value f when rolling r dice and applying the rule f = max(x1, x2, . . . , xr).
263
+ This is
264
+ pr(f) =
265
+ r
266
+
267
+ i=1
268
+ �r
269
+ i
270
+
271
+ p(f)ip(x < f)r−i
272
+ (4)
273
+ where p(f) = 1/M and p(x < f) = f−1
274
+ M . This is just the probability to get at least one f and
275
+ nothing higher, multiplied by a combinatoric factor. To simplify this, consider arranging all the
276
+ possible outcomes of r rolls in an r-dimensional hypercube. The number of ways to obtain f is given
277
+ by the difference in volumes between an f and f − 1 sided hypercube so
278
+ pr(f) = f r − (f − 1)r
279
+ M r
280
+ (5)
281
+ It is shown in Appendix A that the expected face value for large M is
282
+ E[f|r, M ≫ 1] = M
283
+ r
284
+ r + 1 + 1
285
+ 2.
286
+ (6)
287
+ In a game with r dice, the expected number of dice rolls before hitting the value M, where we
288
+ add an extra die, is 1/pr(M). Therefore, the expected time spent playing with exactly r dice is
289
+ T(r) =
290
+ �M
291
+ i=1 ipr(i)
292
+ pr(M)
293
+ .
294
+ (7)
295
+ 8
296
+
297
+ JFor large M (using the summation result from Appendix A) this is
298
+ T(r) ≃ M 2
299
+ r + 1
300
+ (8)
301
+ To calculate the expected face value at t, we first compute the expected number of dice at time t by
302
+ solving
303
+ r
304
+
305
+ k
306
+ T(k) = t
307
+ (9)
308
+ for r. Using the large M approximation
309
+ r
310
+
311
+ k
312
+ T(k) ≃ M 2
313
+ r
314
+
315
+ k
316
+ 1
317
+ i + 1 = M 2(Hr − 1)
318
+ (10)
319
+ where Hr is the rth harmonic number. This has a standard approximation, valid for large r but quite
320
+ accurate even at r = 1: Hr ≃ γ +ln(r), where γ is the Euler-Mascheroni constant. Substituting and
321
+ solving for r gives
322
+ r(t) = exp
323
+ � t
324
+ M 2 + 1 − γ
325
+
326
+ = A exp
327
+ � t
328
+ M 2
329
+
330
+ (11)
331
+ where A = exp(1 − γ). The number of dice grows exponentially, with growth rate 1/M 2. The more
332
+ faces a die has, the longer we have to wait to land on a specific one, for example the time to go from
333
+ r dice to 2r dice is ≃ M 2 ln 2.
334
+ The expected face value at time t is the expected face value with r(t) dice. Still working in the
335
+ large M limit this is
336
+ E[f : M ≫ 1](t) ≃ M
337
+ r
338
+ r + 1 = M
339
+ A exp
340
+
341
+ t
342
+ M 2
343
+
344
+ 1 + A exp
345
+
346
+ t
347
+ M 2
348
+
349
+ (12)
350
+ This is a sigmoid function in the variable t/M 2. At large values of t the value is M, as expected,
351
+ we have so many dice we are virtually guaranteed to roll at least one M. The small t behaviour
352
+ is interesting, the function is roughly linear which means, despite the exponential growth in the
353
+ number of dice suggested by equation 11, the expected face value grows much more slowly.
354
+ 9
355
+
356
+ Figure 6: Average face value in the SSM game as a function of t for an M = 10 sided dice, averaged
357
+ over N = 1000 independent instances.
358
+ Figure 6 shows the results of 1000 simulations of the game with M = 10 compared to the ‘exact’
359
+ answer, equation 12. We observe convergence to the upper bound M at a rate that is roughly linear
360
+ in log time. Such slow convergence is seen in more complex evolutionary models, especially the
361
+ Tangled Nature Model [20, 21] and its variants [13, 4]. There, it arises from the simulated ecosystem
362
+ successively crossing ‘entropic barriers’ [22]. Each time a barrier is crossed the system is likely to be
363
+ in a more stable configuration with a higher barrier. This behaviour has been discussed before in
364
+ the language of record statistics and is also observed in physical systems like spin glasses, colloids
365
+ and high temperature superconductors [23].
366
+ This model is simple enough for an approximate analytic solution. This shows that there is
367
+ competition between the growth in the number of dice over time against the growth in the time taken
368
+ between trials. There is also a trade-off between large values of M, leading to higher average rolls,
369
+ versus time taken to add a new die. What this model suggests is that selection plus accumulation
370
+ leads to slow growth in stability. This model implies that older inhabited planets should be more
371
+ habitable, so our presence on Earth is not just an observer effect but a statistically more likely
372
+ outcome.
373
+ 10
374
+
375
+ 10.0
376
+ Average of 1000 runs
377
+ Exact
378
+ 9.5
379
+ 9.0
380
+ 8.5
381
+ Value
382
+ 8.0
383
+ Face
384
+ 7.5
385
+ 7.0
386
+ 6.5
387
+ 6.0
388
+ 100
389
+ 101
390
+ 102
391
+ 103
392
+ t5.2
393
+ Game B: Increasing M
394
+ This game similar to the previous one, except instead of adding extra dice we have just one die
395
+ and add a extra faces to it, which makes this harder to play with real dice! A rough analogy to
396
+ a real ecosystem is to assume that species diversity is not lost after each collapse (dice roll) and
397
+ that species persist at low abundance, in dormant states or isolated refugia. Reaching a fitness peak
398
+ (hitting the max value of M) generates more latent diversity and allows ecosystems to explore more
399
+ of the so-called fitness landscape [24]. Thus each reset has the potential to find a more stable state
400
+ because the space of possibilities is wider.
401
+ Figure 7: One possible unfolding of SSM game B with T = 20. The bottom row shows the actual
402
+ face values and the top row shows the number of sides of the die. For example at t = 11 we roll a 4
403
+ on a 4-sided die, increasing the number of sides to 5 for the next roll.
404
+ If the die has M sides, the expected number of rolls required to hit the M face is just 1/M.
405
+ Each roll is expected to last M+1
406
+ 2
407
+ steps so the expected waiting time before increasing the number
408
+ of faces is
409
+ T(M) = M M + 1
410
+ 2
411
+ (13)
412
+ Summing up the wait times from each M gives the total duration of the experiment
413
+ t =
414
+ M
415
+
416
+ i=1
417
+ i(i + 1)
418
+ 2
419
+ = 1
420
+ 2
421
+ �M(M + 1)(2M + 1)
422
+ 6
423
+ + M(M + 1)
424
+ 2
425
+
426
+ (14)
427
+ Keeping only the terms of highest order in M and solving for t gives
428
+ M =
429
+ 3√
430
+ 6t
431
+ (15)
432
+ Substituting into equation 3 gives
433
+ E[f](t) = 2
434
+ 3√
435
+ 6t + 1
436
+ 3
437
+ (16)
438
+ 11
439
+
440
+ M=1
441
+ 2 2
442
+ 33
443
+ 4
444
+ 4
445
+ 5
446
+ 5
447
+ ....r.Figure 8: Average face value in SSM game B as a function of t, averaged over N = 1000 independent
448
+ instances.
449
+ Figure 8 shows the results of simulations of the game compared to the exact answer, equation 16.
450
+ Unlike game A there is no convergence and the expected face value grows without bound, though
451
+ fairly slowly. Again there is a trade off between increasing M by performing a large number of trials
452
+ and the time it takes to complete those trials. This is again reminiscent of Tangled Nature Model
453
+ dynamics [22, 13, 4] and other physical systems which cross energetic or entropic barriers [23].
454
+ 6
455
+ Discussion
456
+ These three mechanisms give three reasonable ideas about what to expect when surveying large
457
+ catalogues of inhabited planets, or looking at an inhabited one at a random point in its history.
458
+ The first two (SBS, SS) have no role for life.
459
+ The stability properties of inhabited planets are
460
+ down to observer effects - unless they had these properties we wouldn’t be looking at them. The
461
+ third mechanism is more interesting. Once a planet is inhabited life can have a positive effect on
462
+ habitability. In particular - inhabited planets have properties conducive to stability because of their
463
+ history of inhabitance.
464
+ This idea has appeared previously as ‘The inhabitance paradox’ in [25] and is closely related
465
+ 12
466
+
467
+ 16
468
+ Average of 1000 runs
469
+ Exact
470
+ 14
471
+ 12
472
+ Face Value
473
+ 10
474
+ 8
475
+ 6
476
+ 4
477
+ 2
478
+ 100
479
+ 101
480
+ 102
481
+ 103
482
+ tto the idea of the Gaian bottleneck [26].
483
+ This paradox says that for a planet to be habitable,
484
+ it must be inhabited. This means life must seize the reins and exert a stabilising effect early in a
485
+ planet’s history or go extinct due to deteriorating geophysical conditions - an effect dubbed the Gaian
486
+ bottleneck. The SSM game shows a very simple mechanism by which this could occur, combining
487
+ the sequential selection algorithm of [3] with some method of making cumulative improvements will
488
+ tend to generate more stable systems. We have argued previously [4] that such cumulative processes
489
+ are widespread on Earth, for example: microbial seed banks, dormancy [27] and lateral gene transfer
490
+ [28] all contribute to the maintenance of microbial diversity and therefore the stabilising effect of
491
+ functional redundancy.
492
+ We hope that this model and its analysis provides some clarity on selection principles as well as
493
+ providing a sandbox for studying selection effects. In particular, we believe that Sequential Selection
494
+ with Memory provides a plausible way for a complex system, like an inhabited planet, to become
495
+ more stable over time. We propose that Gaia - the stabilising and symbiotic feedback of life and the
496
+ environment - can be born through this kind of natural, but non-Darwinian, selection.
497
+ A
498
+ Expected value for the max of r, M-sided dice.
499
+ Rolling r, M sided dice gives the face value f with probability
500
+ p(f) = kr − (k − 1)r
501
+ M r
502
+ as discussed in the text. The expectation for the face value is therefore
503
+ E[f] =
504
+ M
505
+
506
+ k=1
507
+ k kr − (k − 1)r
508
+ M r
509
+ Writing out the sum explicitly
510
+ 1.1r + 2.2r + 3.3r + . . . + M.M r
511
+ −(1.0r + 2.1r + 3.2r + . . . + M.(M − 1)r)
512
+ Shows that we can regroup and rewrite as
513
+ E[f] =
514
+ 1
515
+ M r
516
+
517
+ M r+1 −
518
+ M−1
519
+
520
+ k=1
521
+ kr
522
+
523
+ The sum can be simplified using Faulhaber’s formula [29]
524
+ M−1
525
+
526
+ k=1
527
+ kr = (M − 1)r+1
528
+ r + 1
529
+ + (M − 1)r
530
+ 2
531
+ + O(M r−1)
532
+ where the lower order terms are fairly complex coefficients involving the Bernoulli numbers. Substi-
533
+ tuting and taking the limit of large M we get
534
+ E[f] = M
535
+ r
536
+ r + 1 + 1
537
+ 2
538
+ as stated in the text.
539
+ 13
540
+
541
+ References
542
+ [1] Pierrick Bourrat. From survivors to replicators: evolution by natural selection revisited. Biology
543
+ & Philosophy, 29(4):517–538, 2014.
544
+ [2] W Ford Doolittle. Natural selection through survival alone, and the possibility of gaia. Biology
545
+ & Philosophy, 29(3):415–423, 2014.
546
+ [3] Timothy M Lenton, Stuart J Daines, James G Dyke, Arwen E Nicholson, David M Wilkinson,
547
+ and Hywel TP Williams. Selection for gaia across multiple scales. Trends in Ecology & Evolution,
548
+ 33(8):633–645, 2018.
549
+ [4] Rudy Arthur and Arwen Nicholson. Selection principles for gaia. Journal of Theoretical Biology,
550
+ 533:110940, 2022.
551
+ [5] Ignas A. G. Snellen, F. Snik, M. Kenworthy, S. Albrecht, G. Anglada-Escud´e, I. Baraffe, P. Bau-
552
+ doz, W. Benz, J. L. Beuzit, B. Biller, J. L. Birkby, A. Boccaletti, R. van Boekel, J. de Boer,
553
+ Matteo Brogi, L. Buchhave, L. Carone, M. Claire, R. Claudi, J. M. Demory, B. O. D´esert,
554
+ S. Desidera, B. S. Gaudi, R. Gratton, M. Gillon, J. L. Grenfell, O. Guyon, T. Henning, S. Hink-
555
+ ley, E. Huby, M. Janson, C. Helling, K. Heng, M. Kasper, C. U. Keller, O. Krause, L. Kreidberg,
556
+ N. Madhusudhan, A. M. Lagrange, R. Launhardt, T. M. Lenton, M. Lopez-Puertas, A. L. Maire,
557
+ N. Mayne, V. Meadows, B. Mennesson, G. Micela, Y. Miguel, J. Milli, M. Min, E. de Mooij,
558
+ D. Mouillet, M. N’Diaye, V. D’Orazi, E. Palle, I. Pagano, G. Piotto, D. Queloz, H. Rauer,
559
+ I. Ribas, G. Ruane, F. Selsis, A. Sozzetti, D. Stam, C. C. Stark, A. Vigan, and Pieter de Visser.
560
+ Detecting life outside our solar system with a large high-contrast-imaging mission. Experimental
561
+ Astronomy, October 2021.
562
+ [6] Sascha P. Quanz, Olivier Absil, Willy Benz, Xavier Bonfils, Jean-Philippe Berger, Denis Defr`ere,
563
+ Ewine van Dishoeck, David Ehrenreich, Jonathan Fortney, Adrian Glauser, John Lee Grenfell,
564
+ Markus Janson, Stefan Kraus, Oliver Krause, Lucas Labadie, Sylvestre Lacour, Michael Line,
565
+ Hendrik Linz, J´erˆome Loicq, Yamila Miguel, Enric Pall´e, Didier Queloz, Heike Rauer, Ignasi
566
+ Ribas, Sarah Rugheimer, Franck Selsis, Ignas Snellen, Alessandro Sozzetti, Karl R. Stapelfeldt,
567
+ Stephane Udry, and Mark Wyatt. Atmospheric characterization of terrestrial exoplanets in the
568
+ mid-infrared: biosignatures, habitability, and diversity. Experimental Astronomy, pages 1–25,
569
+ September 2021.
570
+ [7] A Lenardic, JW Crowley, AM Jellinek, and M Weller. The solar system of forking paths: bifur-
571
+ cations in planetary evolution and the search for life-bearing planets in our galaxy. Astrobiology,
572
+ 16(7):551–559, 2016.
573
+ [8] Jan Toman and Jaroslav Flegr. Stability-based sorting: The forgotten process behind (not only)
574
+ biological evolution. Journal of theoretical biology, 435:29–41, 2017.
575
+ [9] Fr´ed´eric Bouchard. Ecosystem evolution is about variation and persistence, not populations
576
+ and reproduction. Biological Theory, 9(4):382–391, 2014.
577
+ [10] James E Lovelock and Lynn Margulis. Atmospheric homeostasis by and for the biosphere: the
578
+ gaia hypothesis. Tellus, 26(1-2):2–10, 1974.
579
+ [11] W Ford Doolittle. Darwinizing gaia. Journal of Theoretical Biology, 434:11–19, 2017.
580
+ [12] Richard A Betts and Timothy M Lenton.
581
+ Second chances for lucky gaia: a hypothesis of
582
+ sequential selection. Met Office, 2008.
583
+ 14
584
+
585
+ [13] Rudy Arthur and Arwen Nicholson. An entropic model of gaia. Journal of theoretical biology,
586
+ 430:177–184, 2017.
587
+ [14] Arwen E Nicholson, David M Wilkinson, Hywel TP Williams, and Timothy M Lenton. Alter-
588
+ native mechanisms for gaia. Journal of theoretical biology, 457:249–257, 2018.
589
+ [15] Timothy M Lenton, Timothy A Kohler, Pablo A Marquet, Richard A Boyle, Michel Crucifix,
590
+ David M Wilkinson, and Marten Scheffer.
591
+ Survival of the systems.
592
+ Trends in Ecology &
593
+ Evolution, 36(4):333–344, 2021.
594
+ [16] Richard A Boyle and Timothy M Lenton.
595
+ The evolution of biogeochemical recycling by
596
+ persistence-based selection. Communications Earth & Environment, 3(1):1–14, 2022.
597
+ [17] Andrew J Watson and James E Lovelock. Biological homeostasis of the global environment:
598
+ the parable of daisyworld. Tellus B: Chemical and Physical Meteorology, 35(4):284–289, 1983.
599
+ [18] Tim Lenton and Andrew Watson. Revolutions that made the Earth. OUP Oxford, 2013.
600
+ [19] Rudy Arthur and Arwen Nicholson. A gaian habitable zone, 2023.
601
+ [20] Kim Christensen, Simone A Di Collobiano, Matt Hall, and Henrik J Jensen. Tangled nature:
602
+ a model of evolutionary ecology. Journal of theoretical Biology, 216(1):73–84, 2002.
603
+ [21] Rudy Arthur, Arwen Nicholson, Paolo Sibani, and Michael Christensen.
604
+ The tangled na-
605
+ ture model for organizational ecology. Computational and Mathematical Organization Theory,
606
+ 23(1):1–31, 2017.
607
+ [22] Nikolaj Becker and Paolo Sibani. Evolution and non-equilibrium physics: A study of the tangled
608
+ nature model. EPL (Europhysics Letters), 105(1):18005, 2014.
609
+ [23] Paolo Sibani, Stefan Boettcher, and Henrik Jeldtoft Jensen.
610
+ Record dynamics of evolving
611
+ metastable systems: theory and applications. The European Physical Journal B, 94(1):1–23,
612
+ 2021.
613
+ [24] Rudy Arthur and Paolo Sibani. Decision making on fitness landscapes. Physica A: Statistical
614
+ Mechanics and its Applications, 471:696–704, 2017.
615
+ [25] Colin Goldblatt. The inhabitance paradox: How habitability and inhabitancy are inseparable.
616
+ arXiv preprint arXiv:1603.00950, 2016.
617
+ [26] Aditya Chopra and Charles H Lineweaver. The case for a gaian bottleneck: the biology of
618
+ habitability. Astrobiology, 16(1):7–22, 2016.
619
+ [27] Jay T Lennon and Stuart E Jones.
620
+ Microbial seed banks: the ecological and evolutionary
621
+ implications of dormancy. Nature reviews microbiology, 9(2):119–130, 2011.
622
+ [28] Nigel Goldenfeld and Carl Woese. Biology’s next revolution. Nature, 445(7126):369–369, 2007.
623
+ [29] Eric W. Weisstein. Faulhaber’s formula. From MathWorld—A Wolfram Web Resource. Last
624
+ visited on 1/1/2023.
625
+ 15
626
+
6NE0T4oBgHgl3EQfvwGn/content/tmp_files/load_file.txt ADDED
@@ -0,0 +1,426 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf,len=425
2
+ page_content='Does Gaia Play Dice?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
3
+ page_content=' : Simple Models of non-Darwinian Selection Rudy Arthur,1∗Arwen Nicholson,2† 1University of Exeter, Department of Computer Science 2University of Exeter, Department of Physics and Astronomy January 9, 2023 Abstract In this paper we introduce some simple models, based on rolling dice, to explore mechanisms proposed to explain planetary habitability.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
4
+ page_content=' The idea is to study these selection mechanisms in an analytically tractable setting, isolating their consequences from other details which can confound or obscure their effect in more realistic models.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
5
+ page_content=' We find that while the observable of interest, the face value shown on the die, ‘improves’ over time in all models, for two of the more popular ideas: Selection by Survival and Sequential Selection, this is down to sampling effects.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
6
+ page_content=' A modified version of Sequential Selection, Sequential Selection with Memory, implies a statistical tendency for systems to improve over time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
7
+ page_content=' We discuss the implications of this and its relationship to the ideas of the ‘inhabitance paradox’ and the ‘Gaian bottleneck’.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
8
+ page_content=' 1 Introduction Relatively recent discussion about the persistence of life over long periods has brought to the fore various selection principles [1, 2, 3, 4].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
9
+ page_content=' With the recent launch of the James Webb Space Telescope [5, 6] these questions not only have important implications for our understanding of Earth history, but also for the search for other inhabited planets.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
10
+ page_content=' In the future, with a large enough catalogue of inhabited planets, it may be possible to experimentally investigate alternative trajectories for life [7].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
11
+ page_content=' Until then, understanding general principles behind planetary habitability and inhabitance is a way to provide working hypotheses that can explain both how ‘lucky’ the Earth is to be inhabited and what we might expect on planets orbiting other stars.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
12
+ page_content=' The most discussed of these selection principles is called ‘Selection by Survival’ (SBS) in [3] though the idea has many names, see [8].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
13
+ page_content=' The essential idea is that a population where the entities have different rates of survival will be ‘purified’ so that, in the long run, surviving entities have must have properties conducive to survival.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
14
+ page_content=' Several works e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
15
+ page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
16
+ page_content=' [1, 2, 9], attempt to disentangle Darwinian selection from this ‘differential persistence’ (essentially a synonym for SBS).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
17
+ page_content=' [2] and [9] emphasise the importance of this selection principle acting on higher order phenomena like whole ecosystems, planetary scale biogeochemical cycles and the entire life-Earth coupled system i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
18
+ page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
19
+ page_content=' Gaia [10].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
20
+ page_content=' They argue that in systems of hereditary replicators Darwinian selection is more powerful, however for entities like populations, ecosystems or bio-geochemical cycles [11], which do not have ∗E-mail: R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
21
+ page_content='Arthur@exeter.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
22
+ page_content='ac.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
23
+ page_content='uk †E-mail: A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
24
+ page_content='E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
25
+ page_content='Nicholson@exeter.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
26
+ page_content='ac.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
27
+ page_content='uk 1 arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
28
+ page_content='02623v1 [q-bio.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
29
+ page_content='PE] 6 Jan 2023 strict heredity and reproduction, SBS will operate to favour certain macroscopic features.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
30
+ page_content=' Specific examples suggested by e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
31
+ page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
32
+ page_content=' [8] are sexual reproduction and macroevolutionary freezing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
33
+ page_content=' [3] defines another, related, selection principle called Sequential Selection (SS) [12].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
34
+ page_content=' This is a similar idea to SBS, but motivated by the frequent upheavals in the history of life on Earth and meant to account for life’s apparent stabilising effect on Earth’s habitability.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
35
+ page_content=' [3] propose a simple algorithm - evolutionary innovations have a stabilizing or destabilizing effect on the environment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
36
+ page_content=' If they have a destabilizing effect, habitability is reduced, eventually eliminating the destabilizing innovation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
37
+ page_content=' In this way destabilizing effects are eliminated by ‘near fatal resets’ while stabilizing innovations persist and accumulate.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
38
+ page_content=' In [13, 4] we argue for a refinement of this algorithm, emphasising that the resets are‘near fatal i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
39
+ page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
40
+ page_content=' the evolutionary innovations developed during the previous stable period are not completely lost.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
41
+ page_content=' The algorithm of [3] applies: destabilizing innovations lead to resets which greatly reduce species abundance but have a lesser effect on species diversity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
42
+ page_content=' The life-earth system which arises after the reset is selected from a larger ‘pool’, which has the potential to generate better, more stable ecosystems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
43
+ page_content=' Higher species and functional diversity give Gaia more tools to generate stability.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
44
+ page_content=' The process is completely blind, so unstable states can also be selected, however, by definition, these are short lived and eventually a long-lived stable state will arise.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
45
+ page_content=' During this stable period species diversity can increase again leading to a kind of ratcheting effect.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
46
+ page_content=' To emphasise this cumulative process, in contrast to the sequential selection algorithm of [3], we call this ‘Sequential Selection with Memory’ (SSM).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
47
+ page_content=' A variety of abstract models of varying complexity have been proposed to explore these selection principles e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
48
+ page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
49
+ page_content=' [1, 14, 15, 4, 16].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
50
+ page_content=' Though these models have great value, it can sometimes be unclear which of their features are programmed in (as alleged by [11] of the famous Daisyworld [17] model) and which are emergent.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
51
+ page_content=' It must also be said that the mathematical or computational complexity of these models can give them an air of mystery - especially to biologists not well versed in these methods.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
52
+ page_content=' Indeed, the very fact that the key model features are often emergent means that understanding how they emerge requires a detailed understanding of each model’s dynamics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
53
+ page_content=' Here we propose an extremely simple probability model as a setting to study selection princi- ples.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
54
+ page_content=' The aim is to strip out as much complexity as possible to understand the core meaning of these principles and their consequences.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
55
+ page_content=' A very loose analogy would be trying to understand the approximately Gaussian distribution of, say, human height.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
56
+ page_content=' This has some genetic and environ- mental causes which, with great difficultly, could be experimentally isolated and formulated into a mechanistic model of height, simulated and shown to result in a Gaussian distribution.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
57
+ page_content=' However, a much simpler, and in many ways more satisfactory, explanation is that a Gaussian distribution is the expected outcome for an observable which is a sum of independent effects.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
58
+ page_content=' Continuing the height analogy, by simulating sums of random variables and showing this results in a Gaussian distribution we might start to suspect that a more general principle is operating, one which isn’t affected by the particular details of our model, in this case, the Central Limit Theorem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
59
+ page_content=' This paper doesn’t propose anything as general as a statistical convergence theorem, what we do propose are models simple enough to be analytically solved but complex enough to see selection principles operating.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
60
+ page_content=' Theses models will be shown to exhibit interesting behaviour which is also observed in more complex models.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
61
+ page_content=' The aim is to provide some clarity on exactly what non-Darwinian selection principles can do in a clear and tractable setting.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
62
+ page_content=' 2 2 Introducing the Model Consider an M sided die with the rule that, once rolled, whatever number is showing on the top face gives the number of steps to wait before rolling again or finishing the game.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
63
+ page_content=' For r ≥ 1 dice we roll each one independently to get x1, x2, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
64
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
65
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
66
+ page_content=' , xr, and take the highest face value: max(x1, x2, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
67
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
68
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
69
+ page_content=' , xr).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
70
+ page_content=' Based on this consider the following dice games: 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
71
+ page_content=' Selection By Survival(SBS): Roll N (where N is a very large number) independent dice once each.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
72
+ page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
73
+ page_content=' Sequential Selection (SS): Roll one die repeatedly for T time steps.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
74
+ page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
75
+ page_content=' Sequential Selection with Memory (SSM): (a) Starting with r = 1, roll r dice repeatedly for T time steps.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
76
+ page_content=' Add a new die every time the top face shows the maximum value, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
77
+ page_content=' (b) Starting with M = 1, roll an M sided die repeatedly for T times steps.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
78
+ page_content=' Every time the top face shows the maximum value M, increase M by 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
79
+ page_content=' The SSM games are reminiscent of the Polya Urn model, though have not been studied before to our knowledge.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
80
+ page_content=' The quantity of interest will be the expected face value at time t.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
81
+ page_content=' The names chosen are based on the discussion in the Introduction and follow the conventions of [3] and [4].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
82
+ page_content=' Our version of Selection By Survival is much simpler than the (mostly verbal) models proposed by others e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
83
+ page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
84
+ page_content=' [9] and most closely follows the graphical model from [2].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
85
+ page_content=' As a rough mapping to reality - a die represents an inhabited ‘planet’.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
86
+ page_content=' Each roll is a period of stability for the planet’s biosphere.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
87
+ page_content=' The face value represents something akin to the ‘fitness’ of the biosphere on that planet, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
88
+ page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
89
+ page_content=' how long it will persist.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
90
+ page_content=' If we observe an inhabited planet at some random point in its history we may see a biosphere with properties conducive to long term stability (high face value) or only short term stability (low face value).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
91
+ page_content=' The question of interest for astrobiology is, if we were to survey a large catalogue of inhabited planets, what would be the average ‘fitness’?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
92
+ page_content=' For Earth history (or for the history of any inhabited planet) the equivalent question is, if we were to observe a planet at a random point in its history, what should we expect about the habitability properties of that planet?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
93
+ page_content=' 3 3 Selection by Survival t = 1 t = 2 t = 3 t = 4 t = 5 t = 6 Figure 1: One possible unfolding of the SBS game with N = 25 and M = 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
94
+ page_content=' At t = 1 we have our initial ensemble, at t = 2 we have removed all the 1s, at t = 3 we remove all the 2s etc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
95
+ page_content=' Figure 1 shows one realisation of the SBS game.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
96
+ page_content=' At t = 1 all of the dice are in play and the average face value (over very large N or many different realisations of the same game) is (1 + 2 + .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
97
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
98
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
99
+ page_content=' + M)/M at t = 2 all of the dice showing 1 on the top face are removed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
100
+ page_content=' Restricting our survey to inhabited planets, the average face value is now (2 + 3 + .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
101
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
102
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
103
+ page_content=' + M)/(M − 1) At time t ≤ M the average face value is (t + (t + 1) + .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
104
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
105
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
106
+ page_content=' + M) (M − t) = M + t 2 (1) So that average face value increases linearly with time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
107
+ page_content=' 4 : :围 880 围Figure 2: Average face value in the SBS game as a function of t for an M = 10 sided dice over N = 1000 dice rolls.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
108
+ page_content=' Figure 2 shows the result of simulations of the game compared to equation 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
109
+ page_content=' In terms of ‘planets’ this model is simply stating the (obvious) fact that planets which survive have properties (high face value) which allow them to survive!' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
110
+ page_content=' Looking at the catalogue of inhabited planets will necessarily yield planets with properties conducive to maintaining life, without the need for any additional mechanism.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
111
+ page_content=' This realisation has all planets are seeded with life at the same time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
112
+ page_content=' More complex games could be devised (say a constant rate of habitable planet generation) to study how the generateion rate interacts with this simple selection mechanism.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
113
+ page_content=' For this paper, SBS represents a basic null model - older inhabited planets must have features which have enabled them to remain inhabited.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
114
+ page_content=' The growth in fitness of the ‘surviving’ planets is simply a sampling artefact, the average fitness of an inhabited planet increases because we throw away more and more of the unfit planets from our average.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
115
+ page_content=' Considering our solar system according to SBS, the single inhabited planet we see is habitable because if it wasn’t, we wouldn’t be looking at it, or living on it.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
116
+ page_content=' Thus in this context, SBS is nothing more than an observer effect or anthropic principle.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
117
+ page_content=' 5 10 Exact Average of 1000 runs 8 Face Value 6 4 2 0 2 6 8 10 4 t4 Sequential Selection The Earth has experienced numerous mass extinction events, had very different planetary regulation mechanisms, atmospheric composition, levels of volcanic activity and life has persisted the entire time [18].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
118
+ page_content=' We seek to model these sequential resets with another simple game: repeatedly rolling a single die.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
119
+ page_content=' Figure 3: One possible unfolding of the SS game with T = 20 and M = 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
120
+ page_content=' At t = 1 we roll 2 which shows for 2 steps, we roll 1 which shows for 1 step, then 3 for 3 steps etc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
121
+ page_content=' The game is played a large number, N, of times as in figure 1 so we are interested in average behaviour.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
122
+ page_content=' When observing the die at a random time t, what should we expect the face of the die to show, on average?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
123
+ page_content=' The chance of the die showing k is proportional to the probability of rolling a k, p(k), times the number of ‘slots’ where the observation could occur e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
124
+ page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
125
+ page_content=' if the die is showing 3 this could be an observation of the die on the first, second or third step where it is face up.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
126
+ page_content=' We normalise this probability and compute the expected value for the top face as M � k=1 k kp(k) �m k=1 kp(k) (2) For one die p(k) = 1/M and this simplifies to 2M + 1 3 (3) Note this is larger than the expected value of a single dice roll, M+1 2 for M > 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
127
+ page_content=' 6 Figure 4: Average face value in the SS game as a function of t for an M = 10 sided dice, averaged over N = 1000 independent instances.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
128
+ page_content=' The expected value of a single 10 sided dice roll is 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
129
+ page_content='5 which is less than the expected face value in the sequential sampling game, 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
130
+ page_content=' Note the logged x-axis.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
131
+ page_content=' Figure 4 shows the results of simulations of the game compared to equation 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
132
+ page_content=' The results here imply that when observing a ‘planet’ at a random time, it is more likely to be in a state with stability enhancing properties (high face value).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
133
+ page_content=' Again this reflects the obvious fact that if we depict Earth’s history as a time line and pick a random point on the line we are more likely to pick a point in a long stable period.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
134
+ page_content=' In particular, our present time is most likely to be a stable period, without the need for any additional mechanism.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
135
+ page_content=' Like with SBS, Earth’s current stability is simply an observer effect.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
136
+ page_content=' One thing missing from this game (and from the algorithm of [3]) is the possibility of total extinc- tion, that is, finishing the game early.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
137
+ page_content=' We could implement an additional rule, say when we roll a 1, stop the game.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
138
+ page_content=' This would give a model where SBS and SS are both operating simultaneously.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
139
+ page_content=' Here, for simplicity and clarity, we don’t account for total extinction, so as not to mix the mechanisms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
140
+ page_content=' All of our SS games persist for the same amount of time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
141
+ page_content=' More complex models e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
142
+ page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
143
+ page_content=' [4, 19] do have the possibility of early stopping and come to very similar conclusions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
144
+ page_content=' 7 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
145
+ page_content='4 Average of 1000 runs Exact 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
146
+ page_content='2 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
147
+ page_content='0 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
148
+ page_content='8 Face Value 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
149
+ page_content='6 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
150
+ page_content='4 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
151
+ page_content='2 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
152
+ page_content='0 100 101 102 103 t5 Sequential Selection with Memory The continued inhabitance of Earth and the fact that biodiversity has increased over time motivates the final games.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
153
+ page_content=' Each reset does not start from scratch, but builds on evolutionary and ecological innovations that came before.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
154
+ page_content=' We propose 2 models with an extremely simple ‘memory’.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
155
+ page_content=' This memory is implemented in two ways, first by adding extra dice at fixed M, second by increasing M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
156
+ page_content=' 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
157
+ page_content='1 Game A: Adding dice The face value in this game is determined by rolling multiple dice and choosing the one with the maximum face value.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
158
+ page_content=' The idea is that stable biospheres outlast unstable ones.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
159
+ page_content=' One could imagine independent ecosystems co-existing with the final ‘reset’ only occurring when the most stable sub- system collapses.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
160
+ page_content=' If this seems contrived, in more complex models e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
161
+ page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
162
+ page_content=' [4], a similar feature emerges as a consequence of model dynamics rather than being enforced.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
163
+ page_content=' Figure 5: One possible unfolding of SSM game A with T = 20 and M = 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
164
+ page_content=' At t = 3 we roll 6 which adds an extra die.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
165
+ page_content=' At t = 13 we roll six again which adds a third die.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
166
+ page_content=' The bottom row is the observable, the other rows show dice rolls which are not observed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
167
+ page_content=' The analysis is a little more complex that the previous two games.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
168
+ page_content=' The first thing we need is the probability to get the face value f when rolling r dice and applying the rule f = max(x1, x2, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
169
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
170
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
171
+ page_content=' , xr).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
172
+ page_content=' This is pr(f) = r � i=1 �r i � p(f)ip(x < f)r−i (4) where p(f) = 1/M and p(x < f) = f−1 M .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
173
+ page_content=' This is just the probability to get at least one f and nothing higher, multiplied by a combinatoric factor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
174
+ page_content=' To simplify this, consider arranging all the possible outcomes of r rolls in an r-dimensional hypercube.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
175
+ page_content=' The number of ways to obtain f is given by the difference in volumes between an f and f − 1 sided hypercube so pr(f) = f r − (f − 1)r M r (5) It is shown in Appendix A that the expected face value for large M is E[f|r, M ≫ 1] = M r r + 1 + 1 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
176
+ page_content=' (6) In a game with r dice, the expected number of dice rolls before hitting the value M, where we add an extra die, is 1/pr(M).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
177
+ page_content=' Therefore, the expected time spent playing with exactly r dice is T(r) = �M i=1 ipr(i) pr(M) .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
178
+ page_content=' (7) 8 JFor large M (using the summation result from Appendix A) this is T(r) ≃ M 2 r + 1 (8) To calculate the expected face value at t, we first compute the expected number of dice at time t by solving r � k T(k) = t (9) for r.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
179
+ page_content=' Using the large M approximation r � k T(k) ≃ M 2 r � k 1 i + 1 = M 2(Hr − 1) (10) where Hr is the rth harmonic number.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
180
+ page_content=' This has a standard approximation, valid for large r but quite accurate even at r = 1: Hr ≃ γ +ln(r), where γ is the Euler-Mascheroni constant.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
181
+ page_content=' Substituting and solving for r gives r(t) = exp � t M 2 + 1 − γ � = A exp � t M 2 � (11) where A = exp(1 − γ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
182
+ page_content=' The number of dice grows exponentially, with growth rate 1/M 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
183
+ page_content=' The more faces a die has, the longer we have to wait to land on a specific one, for example the time to go from r dice to 2r dice is ≃ M 2 ln 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
184
+ page_content=' The expected face value at time t is the expected face value with r(t) dice.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
185
+ page_content=' Still working in the large M limit this is E[f : M ≫ 1](t) ≃ M r r + 1 = M A exp � t M 2 � 1 + A exp � t M 2 � (12) This is a sigmoid function in the variable t/M 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
186
+ page_content=' At large values of t the value is M, as expected, we have so many dice we are virtually guaranteed to roll at least one M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
187
+ page_content=' The small t behaviour is interesting, the function is roughly linear which means, despite the exponential growth in the number of dice suggested by equation 11, the expected face value grows much more slowly.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
188
+ page_content=' 9 Figure 6: Average face value in the SSM game as a function of t for an M = 10 sided dice, averaged over N = 1000 independent instances.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
189
+ page_content=' Figure 6 shows the results of 1000 simulations of the game with M = 10 compared to the ‘exact’ answer, equation 12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
190
+ page_content=' We observe convergence to the upper bound M at a rate that is roughly linear in log time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
191
+ page_content=' Such slow convergence is seen in more complex evolutionary models, especially the Tangled Nature Model [20, 21] and its variants [13, 4].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
192
+ page_content=' There, it arises from the simulated ecosystem successively crossing ‘entropic barriers’ [22].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
193
+ page_content=' Each time a barrier is crossed the system is likely to be in a more stable configuration with a higher barrier.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
194
+ page_content=' This behaviour has been discussed before in the language of record statistics and is also observed in physical systems like spin glasses, colloids and high temperature superconductors [23].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
195
+ page_content=' This model is simple enough for an approximate analytic solution.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
196
+ page_content=' This shows that there is competition between the growth in the number of dice over time against the growth in the time taken between trials.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
197
+ page_content=' There is also a trade-off between large values of M, leading to higher average rolls, versus time taken to add a new die.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
198
+ page_content=' What this model suggests is that selection plus accumulation leads to slow growth in stability.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
199
+ page_content=' This model implies that older inhabited planets should be more habitable, so our presence on Earth is not just an observer effect but a statistically more likely outcome.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
200
+ page_content=' 10 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
201
+ page_content='0 Average of 1000 runs Exact 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
202
+ page_content='5 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
203
+ page_content='0 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
204
+ page_content='5 Value 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
205
+ page_content='0 Face 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
206
+ page_content='5 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
207
+ page_content='0 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
208
+ page_content='5 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
209
+ page_content='0 100 101 102 103 t5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
210
+ page_content='2 Game B: Increasing M This game similar to the previous one, except instead of adding extra dice we have just one die and add a extra faces to it, which makes this harder to play with real dice!' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
211
+ page_content=' A rough analogy to a real ecosystem is to assume that species diversity is not lost after each collapse (dice roll) and that species persist at low abundance, in dormant states or isolated refugia.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
212
+ page_content=' Reaching a fitness peak (hitting the max value of M) generates more latent diversity and allows ecosystems to explore more of the so-called fitness landscape [24].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
213
+ page_content=' Thus each reset has the potential to find a more stable state because the space of possibilities is wider.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
214
+ page_content=' Figure 7: One possible unfolding of SSM game B with T = 20.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
215
+ page_content=' The bottom row shows the actual face values and the top row shows the number of sides of the die.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
216
+ page_content=' For example at t = 11 we roll a 4 on a 4-sided die, increasing the number of sides to 5 for the next roll.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
217
+ page_content=' If the die has M sides, the expected number of rolls required to hit the M face is just 1/M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
218
+ page_content=' Each roll is expected to last M+1 2 steps so the expected waiting time before increasing the number of faces is T(M) = M M + 1 2 (13) Summing up the wait times from each M gives the total duration of the experiment t = M � i=1 i(i + 1) 2 = 1 2 �M(M + 1)(2M + 1) 6 + M(M + 1) 2 � (14) Keeping only the terms of highest order in M and solving for t gives M = 3√ 6t (15) Substituting into equation 3 gives E[f](t) = 2 3√ 6t + 1 3 (16) 11 M=1 2 2 33 4 4 5 5 .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
219
+ page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
220
+ page_content='.r.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
221
+ page_content='Figure 8: Average face value in SSM game B as a function of t, averaged over N = 1000 independent instances.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
222
+ page_content=' Figure 8 shows the results of simulations of the game compared to the exact answer, equation 16.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
223
+ page_content=' Unlike game A there is no convergence and the expected face value grows without bound, though fairly slowly.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
224
+ page_content=' Again there is a trade off between increasing M by performing a large number of trials and the time it takes to complete those trials.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
225
+ page_content=' This is again reminiscent of Tangled Nature Model dynamics [22, 13, 4] and other physical systems which cross energetic or entropic barriers [23].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
226
+ page_content=' 6 Discussion These three mechanisms give three reasonable ideas about what to expect when surveying large catalogues of inhabited planets, or looking at an inhabited one at a random point in its history.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
227
+ page_content=' The first two (SBS, SS) have no role for life.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
228
+ page_content=' The stability properties of inhabited planets are down to observer effects - unless they had these properties we wouldn’t be looking at them.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
229
+ page_content=' The third mechanism is more interesting.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
230
+ page_content=' Once a planet is inhabited life can have a positive effect on habitability.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
231
+ page_content=' In particular - inhabited planets have properties conducive to stability because of their history of inhabitance.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
232
+ page_content=' This idea has appeared previously as ‘The inhabitance paradox’ in [25] and is closely related 12 16 Average of 1000 runs Exact 14 12 Face Value 10 8 6 4 2 100 101 102 103 tto the idea of the Gaian bottleneck [26].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
233
+ page_content=' This paradox says that for a planet to be habitable, it must be inhabited.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
234
+ page_content=' This means life must seize the reins and exert a stabilising effect early in a planet’s history or go extinct due to deteriorating geophysical conditions - an effect dubbed the Gaian bottleneck.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
235
+ page_content=' The SSM game shows a very simple mechanism by which this could occur, combining the sequential selection algorithm of [3] with some method of making cumulative improvements will tend to generate more stable systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
236
+ page_content=' We have argued previously [4] that such cumulative processes are widespread on Earth, for example: microbial seed banks, dormancy [27] and lateral gene transfer [28] all contribute to the maintenance of microbial diversity and therefore the stabilising effect of functional redundancy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
237
+ page_content=' We hope that this model and its analysis provides some clarity on selection principles as well as providing a sandbox for studying selection effects.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
238
+ page_content=' In particular, we believe that Sequential Selection with Memory provides a plausible way for a complex system, like an inhabited planet, to become more stable over time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
239
+ page_content=' We propose that Gaia - the stabilising and symbiotic feedback of life and the environment - can be born through this kind of natural, but non-Darwinian, selection.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
240
+ page_content=' A Expected value for the max of r, M-sided dice.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
241
+ page_content=' Rolling r, M sided dice gives the face value f with probability p(f) = kr − (k − 1)r M r as discussed in the text.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
242
+ page_content=' The expectation for the face value is therefore E[f] = M � k=1 k kr − (k − 1)r M r Writing out the sum explicitly 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
243
+ page_content='1r + 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
244
+ page_content='2r + 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
245
+ page_content='3r + .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
246
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
247
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
248
+ page_content=' + M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
249
+ page_content='M r −(1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
250
+ page_content='0r + 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
251
+ page_content='1r + 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
252
+ page_content='2r + .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
253
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
254
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
255
+ page_content=' + M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
256
+ page_content='(M − 1)r) Shows that we can regroup and rewrite as E[f] = 1 M r � M r+1 − M−1 � k=1 kr � The sum can be simplified using Faulhaber’s formula [29] M−1 � k=1 kr = (M − 1)r+1 r + 1 + (M − 1)r 2 + O(M r−1) where the lower order terms are fairly complex coefficients involving the Bernoulli numbers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
257
+ page_content=' Substi- tuting and taking the limit of large M we get E[f] = M r r + 1 + 1 2 as stated in the text.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
258
+ page_content=' 13 References [1] Pierrick Bourrat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
259
+ page_content=' From survivors to replicators: evolution by natural selection revisited.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
260
+ page_content=' Biology & Philosophy, 29(4):517–538, 2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
261
+ page_content=' [2] W Ford Doolittle.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
262
+ page_content=' Natural selection through survival alone, and the possibility of gaia.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
263
+ page_content=' Biology & Philosophy, 29(3):415–423, 2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
264
+ page_content=' [3] Timothy M Lenton, Stuart J Daines, James G Dyke, Arwen E Nicholson, David M Wilkinson, and Hywel TP Williams.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
265
+ page_content=' Selection for gaia across multiple scales.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
266
+ page_content=' Trends in Ecology & Evolution, 33(8):633–645, 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
267
+ page_content=' [4] Rudy Arthur and Arwen Nicholson.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
268
+ page_content=' Selection principles for gaia.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
269
+ page_content=' Journal of Theoretical Biology, 533:110940, 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
270
+ page_content=' [5] Ignas A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
271
+ page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
272
+ page_content=' Snellen, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
273
+ page_content=' Snik, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
274
+ page_content=' Kenworthy, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
275
+ page_content=' Albrecht, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
276
+ page_content=' Anglada-Escud´e, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
277
+ page_content=' Baraffe, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
278
+ page_content=' Bau- doz, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
279
+ page_content=' Benz, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
280
+ page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
281
+ page_content=' Beuzit, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
282
+ page_content=' Biller, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
283
+ page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
284
+ page_content=' Birkby, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
285
+ page_content=' Boccaletti, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
286
+ page_content=' van Boekel, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
287
+ page_content=' de Boer, Matteo Brogi, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
288
+ page_content=' Buchhave, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
289
+ page_content=' Carone, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
290
+ page_content=' Claire, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
291
+ page_content=' Claudi, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
292
+ page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
293
+ page_content=' Demory, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
294
+ page_content=' O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
295
+ page_content=' D´esert, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
296
+ page_content=' Desidera, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
297
+ page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
298
+ page_content=' Gaudi, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
299
+ page_content=' Gratton, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
300
+ page_content=' Gillon, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
301
+ page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
302
+ page_content=' Grenfell, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
303
+ page_content=' Guyon, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
304
+ page_content=' Henning, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
305
+ page_content=' Hink- ley, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
306
+ page_content=' Huby, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
307
+ page_content=' Janson, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
308
+ page_content=' Helling, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
309
+ page_content=' Heng, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
310
+ page_content=' Kasper, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
311
+ page_content=' U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
312
+ page_content=' Keller, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
313
+ page_content=' Krause, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
314
+ page_content=' Kreidberg, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
315
+ page_content=' Madhusudhan, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
316
+ page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
317
+ page_content=' Lagrange, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
318
+ page_content=' Launhardt, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
319
+ page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
320
+ page_content=' Lenton, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
321
+ page_content=' Lopez-Puertas, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
322
+ page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
323
+ page_content=' Maire, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
324
+ page_content=' Mayne, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
325
+ page_content=' Meadows, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
326
+ page_content=' Mennesson, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
327
+ page_content=' Micela, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
328
+ page_content=' Miguel, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
329
+ page_content=' Milli, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
330
+ page_content=' Min, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
331
+ page_content=' de Mooij, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
332
+ page_content=' Mouillet, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
333
+ page_content=' N’Diaye, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
334
+ page_content=' D’Orazi, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
335
+ page_content=' Palle, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
336
+ page_content=' Pagano, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
337
+ page_content=' Piotto, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
338
+ page_content=' Queloz, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
339
+ page_content=' Rauer, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
340
+ page_content=' Ribas, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
341
+ page_content=' Ruane, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
342
+ page_content=' Selsis, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
343
+ page_content=' Sozzetti, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
344
+ page_content=' Stam, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
345
+ page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
346
+ page_content=' Stark, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
347
+ page_content=' Vigan, and Pieter de Visser.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
348
+ page_content=' Detecting life outside our solar system with a large high-contrast-imaging mission.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
349
+ page_content=' Experimental Astronomy, October 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
350
+ page_content=' [6] Sascha P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
351
+ page_content=' Quanz, Olivier Absil, Willy Benz, Xavier Bonfils, Jean-Philippe Berger, Denis Defr`ere, Ewine van Dishoeck, David Ehrenreich, Jonathan Fortney, Adrian Glauser, John Lee Grenfell, Markus Janson, Stefan Kraus, Oliver Krause, Lucas Labadie, Sylvestre Lacour, Michael Line, Hendrik Linz, J´erˆome Loicq, Yamila Miguel, Enric Pall´e, Didier Queloz, Heike Rauer, Ignasi Ribas, Sarah Rugheimer, Franck Selsis, Ignas Snellen, Alessandro Sozzetti, Karl R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
352
+ page_content=' Stapelfeldt, Stephane Udry, and Mark Wyatt.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
353
+ page_content=' Atmospheric characterization of terrestrial exoplanets in the mid-infrared: biosignatures, habitability, and diversity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
354
+ page_content=' Experimental Astronomy, pages 1–25, September 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
355
+ page_content=' [7] A Lenardic, JW Crowley, AM Jellinek, and M Weller.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
356
+ page_content=' The solar system of forking paths: bifur- cations in planetary evolution and the search for life-bearing planets in our galaxy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
357
+ page_content=' Astrobiology, 16(7):551–559, 2016.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
358
+ page_content=' [8] Jan Toman and Jaroslav Flegr.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
359
+ page_content=' Stability-based sorting: The forgotten process behind (not only) biological evolution.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
360
+ page_content=' Journal of theoretical biology, 435:29–41, 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
361
+ page_content=' [9] Fr´ed´eric Bouchard.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
362
+ page_content=' Ecosystem evolution is about variation and persistence, not populations and reproduction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
363
+ page_content=' Biological Theory, 9(4):382–391, 2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
364
+ page_content=' [10] James E Lovelock and Lynn Margulis.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
365
+ page_content=' Atmospheric homeostasis by and for the biosphere: the gaia hypothesis.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
366
+ page_content=' Tellus, 26(1-2):2–10, 1974.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
367
+ page_content=' [11] W Ford Doolittle.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
368
+ page_content=' Darwinizing gaia.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
369
+ page_content=' Journal of Theoretical Biology, 434:11–19, 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
370
+ page_content=' [12] Richard A Betts and Timothy M Lenton.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
371
+ page_content=' Second chances for lucky gaia: a hypothesis of sequential selection.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
372
+ page_content=' Met Office, 2008.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
373
+ page_content=' 14 [13] Rudy Arthur and Arwen Nicholson.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
374
+ page_content=' An entropic model of gaia.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
375
+ page_content=' Journal of theoretical biology, 430:177–184, 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
376
+ page_content=' [14] Arwen E Nicholson, David M Wilkinson, Hywel TP Williams, and Timothy M Lenton.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
377
+ page_content=' Alter- native mechanisms for gaia.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
378
+ page_content=' Journal of theoretical biology, 457:249–257, 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
379
+ page_content=' [15] Timothy M Lenton, Timothy A Kohler, Pablo A Marquet, Richard A Boyle, Michel Crucifix, David M Wilkinson, and Marten Scheffer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
380
+ page_content=' Survival of the systems.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
381
+ page_content=' Trends in Ecology & Evolution, 36(4):333–344, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
382
+ page_content=' [16] Richard A Boyle and Timothy M Lenton.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
383
+ page_content=' The evolution of biogeochemical recycling by persistence-based selection.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
384
+ page_content=' Communications Earth & Environment, 3(1):1–14, 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
385
+ page_content=' [17] Andrew J Watson and James E Lovelock.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
386
+ page_content=' Biological homeostasis of the global environment: the parable of daisyworld.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
387
+ page_content=' Tellus B: Chemical and Physical Meteorology, 35(4):284–289, 1983.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
388
+ page_content=' [18] Tim Lenton and Andrew Watson.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
389
+ page_content=' Revolutions that made the Earth.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
390
+ page_content=' OUP Oxford, 2013.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
391
+ page_content=' [19] Rudy Arthur and Arwen Nicholson.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
392
+ page_content=' A gaian habitable zone, 2023.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
393
+ page_content=' [20] Kim Christensen, Simone A Di Collobiano, Matt Hall, and Henrik J Jensen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
394
+ page_content=' Tangled nature: a model of evolutionary ecology.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
395
+ page_content=' Journal of theoretical Biology, 216(1):73–84, 2002.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
396
+ page_content=' [21] Rudy Arthur, Arwen Nicholson, Paolo Sibani, and Michael Christensen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
397
+ page_content=' The tangled na- ture model for organizational ecology.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
398
+ page_content=' Computational and Mathematical Organization Theory, 23(1):1–31, 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
399
+ page_content=' [22] Nikolaj Becker and Paolo Sibani.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
400
+ page_content=' Evolution and non-equilibrium physics: A study of the tangled nature model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
401
+ page_content=' EPL (Europhysics Letters), 105(1):18005, 2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
402
+ page_content=' [23] Paolo Sibani, Stefan Boettcher, and Henrik Jeldtoft Jensen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
403
+ page_content=' Record dynamics of evolving metastable systems: theory and applications.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
404
+ page_content=' The European Physical Journal B, 94(1):1–23, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
405
+ page_content=' [24] Rudy Arthur and Paolo Sibani.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
406
+ page_content=' Decision making on fitness landscapes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
407
+ page_content=' Physica A: Statistical Mechanics and its Applications, 471:696–704, 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
408
+ page_content=' [25] Colin Goldblatt.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
409
+ page_content=' The inhabitance paradox: How habitability and inhabitancy are inseparable.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
410
+ page_content=' arXiv preprint arXiv:1603.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
411
+ page_content='00950, 2016.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
412
+ page_content=' [26] Aditya Chopra and Charles H Lineweaver.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
413
+ page_content=' The case for a gaian bottleneck: the biology of habitability.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
414
+ page_content=' Astrobiology, 16(1):7–22, 2016.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
415
+ page_content=' [27] Jay T Lennon and Stuart E Jones.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
416
+ page_content=' Microbial seed banks: the ecological and evolutionary implications of dormancy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
417
+ page_content=' Nature reviews microbiology, 9(2):119–130, 2011.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
418
+ page_content=' [28] Nigel Goldenfeld and Carl Woese.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
419
+ page_content=' Biology’s next revolution.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
420
+ page_content=' Nature, 445(7126):369–369, 2007.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
421
+ page_content=' [29] Eric W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
422
+ page_content=' Weisstein.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
423
+ page_content=' Faulhaber’s formula.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
424
+ page_content=' From MathWorld—A Wolfram Web Resource.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
425
+ page_content=' Last visited on 1/1/2023.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
426
+ page_content=' 15' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/6NE0T4oBgHgl3EQfvwGn/content/2301.02623v1.pdf'}
79E5T4oBgHgl3EQfQQ7U/vector_store/index.faiss ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ddbbafcea4d80fc1633d01159943e3b1128ea2252a1ed4a4b91ef1a46d391400
3
+ size 1376301
79E5T4oBgHgl3EQfQQ7U/vector_store/index.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:33c049e63d5c1eafc97ae9eec88553f906744b55807b515d4cf8b64eda449aaf
3
+ size 55275
79FLT4oBgHgl3EQfsi_P/content/2301.12148v1.pdf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7751bf140b5262b40c80b5a8eb576af937f96a2ce775a9d6303026fa86ae3171
3
+ size 980559
79FLT4oBgHgl3EQfsi_P/vector_store/index.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b4d892237174462f34b9c694147aaaa02825d91dc37ab91fc0129e4d141256bf
3
+ size 256529
7dA0T4oBgHgl3EQfOf8M/content/tmp_files/2301.02160v1.pdf.txt ADDED
@@ -0,0 +1,1003 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ANNA: Abstractive Text-to-Image Synthesis with Filtered News Captions
2
+ Aashish Anantha Ramakrishnan
3
+ Sharon X. Huang
4
+ Dongwon Lee
5
+ The Pennsylvania State University, State College, Pennsylvania, USA
6
+ {aza6352, suh972, dul13}@psu.edu
7
+ Abstract
8
+ Advancements in Text-to-Image synthesis over recent
9
+ years have focused more on improving the quality of gener-
10
+ ated samples on datasets with descriptive captions. How-
11
+ ever, real-world image-caption pairs present in domains
12
+ such as news data do not use simple and directly descrip-
13
+ tive captions. With captions containing information on both
14
+ the image content and underlying contextual cues, they be-
15
+ come abstractive in nature. In this paper, we launch ANNA,
16
+ an Abstractive News captioNs dAtaset extracted from on-
17
+ line news articles in a variety of different contexts.
18
+ We
19
+ explore the capabilities of current Text-to-Image synthesis
20
+ models to generate news domain-specific images using ab-
21
+ stractive captions by benchmarking them on ANNA, in both
22
+ standard training and transfer learning settings. The gen-
23
+ erated images are judged on the basis of contextual rele-
24
+ vance, visual quality, and perceptual similarity to ground-
25
+ truth image-caption pairs. Through our experiments, we
26
+ show that techniques such as transfer learning achieve lim-
27
+ ited success in understanding abstractive captions but still
28
+ fail to consistently learn the relationships between content
29
+ and context features. The ANNA Dataset is available at
30
+ https://github.com/aashish2000/ANNA.
31
+ 1. Introduction
32
+ Image Generation has been improving by leaps and
33
+ bounds over the last few years thanks to advancements in
34
+ Generative Modelling approaches and availability of higher
35
+ compute capacities [20]. Areas such as text-to-image syn-
36
+ thesis have grown in prominence due to the development
37
+ of model pre-training paradigms on vast image-text pairs
38
+ mined from the internet [17]. This has promoted the use of
39
+ these generators for a variety of applications such as online
40
+ content creation, art synthesis [14] and even more malicious
41
+ use-cases such as DeepFake generation [31]. With Internet
42
+ news media and social networking websites becoming the
43
+ more preferred forms of news distribution, the impact that
44
+ generative modelling, especially semantically-relevant im-
45
+ age generation can have on the news media industry is sig-
46
+ Figure 1. Example of descriptive captions from the COCO Cap-
47
+ tions dataset [2] (Above) and abstractive captions from the ANNA
48
+ (Below). In this case, the abstractive captions contain high-level
49
+ visual content information relevant to the type of room depicted
50
+ and contextual information explaining who are its inhabitants, who
51
+ sponsored it, etc.
52
+ nificant. Images accompanying news articles are primarily
53
+ used as supporting media to convey the key message of the
54
+ article along with complementary information to aid reader
55
+ understanding.
56
+ Commonly, text-to-image synthesis has made use of de-
57
+ scriptive captions, where only visual objects present within
58
+ each image are described in detail. However, news captions
59
+ also relay contextual information correlating the image’s
60
+ contents to the article. The captions are thus abstractive
61
+ (beyond being descriptive), containing both higher-level de-
62
+ scriptive information and contextual cues. Here, we define
63
+ context of a text caption as an attribute that does not have a
64
+ direct visual translation, but contributes towards modifying
65
+ an image’s appearance in relation to the situation in which
66
+ the image is referenced. Fig. 1 provides an example of this
67
+ where both images depict rooms within living spaces, but
68
+ there is a noticeable difference in the appearance of a room
69
+ within a house and that of a shelter. The study of pragmatic
70
+ reasoning in linguistics [5] typically deals with how the in-
71
+ 1
72
+ arXiv:2301.02160v1 [cs.CV] 5 Jan 2023
73
+
74
+ Caption: This room has a
75
+ bed with blue sheets and a
76
+ large bookcase
77
+ Caption: A room in a shelter
78
+ for victims of domestic
79
+ violence that was able to
80
+ reopen recently because of
81
+ a contribution from a donorformativeness of text is influenced by its relevance to con-
82
+ text. Past research has established the importance of prag-
83
+ matic factors in ascertaining the true meaning of context-
84
+ driven text information and how it affects accurate caption-
85
+ ing of images [22], [21] [11]. Directly descriptive captions
86
+ lack this contextual grounding, limiting their usefulness for
87
+ describing news images. To replicate the same types of im-
88
+ ages with contextual relevance using descriptive captions,
89
+ we require intensive caption engineering efforts. This com-
90
+ bination of image content information along with contex-
91
+ tual cues make abstractive captions much more challenging
92
+ to understand, directly impacting the relevance of generated
93
+ results.
94
+ Current datasets for Text-to-Image synthesis are either
95
+ focused on narrow domains with simple, descriptive cap-
96
+ tions or contain minimally filtered image-text pairs from a
97
+ multitude of online sources. There are not many domain-
98
+ specific datasets with image caption pairs containing con-
99
+ textual information in addition to image descriptions. Addi-
100
+ tionally, while most models use improved visual quality of
101
+ output images to be indicators of superior performance, not
102
+ much focus is placed on evaluating the correlation between
103
+ the output image and input text captions. This becomes
104
+ more important when dealing with captions whose features
105
+ are only partially aligned with the ground truth images due
106
+ to its non-descriptive nature. The task of abstractive text-to-
107
+ image synthesis aims to generate images from abstractive
108
+ captions with contextual cues. To evaluate this task, we de-
109
+ sign ANNA, a dataset containing abstractive captions per-
110
+ taining to news image-caption pairs. Abstractive captions
111
+ can motivate text-to-image synthesis models to effectively
112
+ identify these different feature types along with their rel-
113
+ ative importance and represent them appropriately in gen-
114
+ erated images.
115
+ With current Text-to-Image architectures
116
+ implicitly delineating content and context features, we pro-
117
+ vide detailed visualizations of both their success and failure
118
+ cases on ANNA and the need for better understanding of
119
+ sentence structures for generating image features.
120
+ Our contributions in this paper can be summarized as the
121
+ following:
122
+ • We introduce ANNA, a dataset containing approxi-
123
+ mately 30K abstractive image-caption pairs from pop-
124
+ ular media outlet The New York Times
125
+ • We show how current Text-to-Image architectures are
126
+ able to understand abstractive captions and transfer-
127
+ learned concepts from descriptive captions for abstrac-
128
+ tive text-to-image synthesis
129
+ • Using an exhaustive set of evaluation metrics, we
130
+ benchmark popular Text-to-Image architectures on the
131
+ basis of generated image quality, image similarity to
132
+ ground truth images and contextual relevance with ref-
133
+ erence captions
134
+ 2. Related Work
135
+ Text-to-Image Synthesis
136
+ Text-to-Image synthesis is a
137
+ multi-modal generation task that produces relevant images
138
+ conditioned on features described in a text caption. Ini-
139
+ tial approaches such as [16] found success by leveraging
140
+ Generative Adversarial Networks (GANS) [4] for this task.
141
+ Motivated by the success of GAN’s, StackGAN [30] uses
142
+ a stacked generator to simplify the generation pipeline into
143
+ stages: semantically relevant low-resolution image synthe-
144
+ sis followed by progressive up-scaling and defect correc-
145
+ tion. AttnGAN [26] integrates an attention mechanism to
146
+ capture sentence and word level features for increasing the
147
+ correlation between generated images and input text. At-
148
+ tnGAN proved to be a strong baseline based on which mul-
149
+ tiple advancements were developed. One such approach,
150
+ DMGAN [33] integrates a dynamic memory based refine-
151
+ ment module for improving image quality and key-word se-
152
+ lection from reference captions. [29] and [27] build on the
153
+ same model architecture by introducing Contrastive learn-
154
+ ing approaches to improve consistency between learned text
155
+ and image representations. In recent years, the success of
156
+ Vision-Language Pre-training (VLP) has prompted the de-
157
+ velopment of newer and more robust Text-to-Image synthe-
158
+ sis architectures. Contrastive Language-Image Pre-training
159
+ (CLIP) [13], is one of the largest open-source, pre-trained
160
+ models that uses raw text for supervising the learning pro-
161
+ cess of visual concepts. Using pre-trained encoders such
162
+ as CLIP for input text captions, [32], [15], [14] use differ-
163
+ ent generator architectures such as GANs, Auto-regressive
164
+ Transformers and Diffusion models respectively.
165
+ Datasets
166
+ Traditional datasets used as benchmarks for
167
+ measuring Text-to-Image synthesis include domain-specific
168
+ datasets Oxford-102 Flowers [12] and CUB-200 [23]. The
169
+ Oxford-102 Flowers contains images of 102 classes of flow-
170
+ ers along with 5 human-annotated descriptions per im-
171
+ age. Similarly the CUB-200 dataset contains 11,788 im-
172
+ ages of 200 subcategories belonging to different categories
173
+ of birds along with 5 captions per image.
174
+ The captions
175
+ for each of the images in CUB-200 and Oxford-102 were
176
+ collected and released by [16] as a part of their evalua-
177
+ tion. COCO Captions [2] is another popular dataset devel-
178
+ oped using images from the MS-COCO [9], a large-scale
179
+ object detection dataset. It contains over one and a half
180
+ million captions describing over 330,000 images contain-
181
+ ing 80 different classes of everyday objects. Some of the
182
+ other datasets used for this task include the Multi-Modal-
183
+ CelebA-HQ Dataset [25] which provides text-descriptions
184
+ of facial features for images sourced from the CelebA-HQ
185
+ dataset [6]. Conceptual Captions [18] consists of over 3
186
+ million image-caption pairs mined from the internet. In this
187
+ dataset, all the captions are hypernymized, i.e. all proper
188
+ 2
189
+
190
+ nouns and named-entities are replaced with their respective
191
+ hypernynms to make the captions simpler to learn and more
192
+ descriptive. [1] expands this dataset by increasing the num-
193
+ ber of image-caption pairs from 3 million to 12 million. All
194
+ the datasets discussed above focus on descriptive captions
195
+ for each image, where minimal or no contextual informa-
196
+ tion regarding the image is present. Our dataset is one of
197
+ the first to investigate the previously unexplored interaction
198
+ between content and context features for text-to-image syn-
199
+ thesis.
200
+ 3. Constructing Abstractive News Captions
201
+ Dataset: ANNA
202
+ The ANNA (Abstractive News captioNs dAtaset) has
203
+ been constructed to perform news image generation us-
204
+ ing abstractive captions. We source images from the NY-
205
+ Times800K dataset [19] which contains news articles and
206
+ associated image-caption pairs scraped from the news or-
207
+ ganization The New York Times (NYT). This dataset was
208
+ originally developed for News Image Captioning. Using
209
+ news image-caption pairs from a reputable media outlet
210
+ such as NYT helps ensure the dataset’s quality.
211
+ As we
212
+ aim to observe the relationship between content and con-
213
+ text features and how it translates to generated images, we
214
+ focus on selecting generalizable entities within our dataset.
215
+ News data contains a multitude of named-entities, often
216
+ with very low repetition and distinct physical appearances,
217
+ such as faces and geographic landmarks.
218
+ The inclusion
219
+ of named-entities from news images would drastically in-
220
+ crease the complexity of the generative task. The inabil-
221
+ ity to accurately generate named-entity attributes would fur-
222
+ ther hamper context feature representation due to their inter-
223
+ dependent nature. In order to combat the mentioned issues,
224
+ we carefully curate our dataset to include image-caption
225
+ pairs containing adequate contextual and content related in-
226
+ formation. We select Image-caption pairs with lesser de-
227
+ pendence on named-entities and more general visual com-
228
+ ponents to make the task feasible. The specific preprocess-
229
+ ing and filtering approaches utilized are detailed below.
230
+ 3.1. Preprocessing and Filtering Approaches
231
+ The original NYTimes800K dataset contains 445K news
232
+ articles accompanied by 793K image-caption pairs. It spans
233
+ 14 years of articles published on The NYT website. The
234
+ dataset has been provided as a MongoDB dump for public
235
+ access. The first step of preprocessing focuses on removing
236
+ image-caption pairs with explicit entities described both in
237
+ images and text. We use the provided NER tags for each
238
+ caption for filtering. We exclude all captions containing the
239
+ NER tags ’PERSON’, ’GPE’, ’LOC’, ’WORK OF ART’,
240
+ ’ORG’. This ensures any visually significant named-entity
241
+ without adequate description isn’t present in the dataset.
242
+ Subsequently, we also set bounds on the caption length be-
243
+ tween 4 to 70 words. Any captions lesser than 4 words
244
+ would not be informative enough for extracting usable fea-
245
+ tures and captions larger than 70 words cannot be handled
246
+ by the CLIP-based Text encoder [13] that we employ in our
247
+ experiments.
248
+ Following caption-based filtering, we also remove all
249
+ images where human faces are clearly visible in the fore-
250
+ ground. We accomplish this by using a RetinaFace-based
251
+ face detector [3], removing around 1000 additional im-
252
+ ages. Through these filtering techniques, we extract rel-
253
+ evant image-caption pairs and corresponding article head-
254
+ lines from the NYTimes800K Dataset. Data pre-processing
255
+ steps include uniformly scaling our news images to our tar-
256
+ get input resolution 256x256. To accomplish this, we rela-
257
+ tively scale the smaller dimension (height or width) of the
258
+ image to our target resolution and take its center crop. This
259
+ makes sure that we have minimal information loss and also
260
+ helps center the foreground objects in each image. Dis-
261
+ claimer: The dataset samples may use words or language
262
+ that is considered profane, vulgar, or offensive by some
263
+ readers as they are extracted from real-world news articles.
264
+ 3.2. Dataset Insights
265
+ The filtered and pre-processed version of the ANNA con-
266
+ tains 29625 image-text pairs. We split the dataset into Train,
267
+ Validation and Testing sets in the ratio of 80%:10%:10% re-
268
+ spectively. All metric scores reported have been calculated
269
+ on the Test set. To better understand the composition of the
270
+ dataset, we analyze various attributes of the image-text pairs
271
+ and the articles they have been selected from.
272
+ Dataset
273
+ Unique Tokens
274
+ Caption Length
275
+ Mean
276
+ StdDev
277
+ ANNA Train
278
+ 17897
279
+ 14.1
280
+ 7.75
281
+ ANNA Validation
282
+ 1622
283
+ 13.8
284
+ 7.60
285
+ ANNA Test
286
+ 1649
287
+ 14.1
288
+ 7.71
289
+ COCO Captions Train
290
+ 11046
291
+ 10.4
292
+ 1.75
293
+ COCO Captions Validation
294
+ 4758
295
+ 10.4
296
+ 1.74
297
+ Table 1. Dataset Statistics of ANNA and COCO-Captions
298
+ 3.2.1
299
+ Caption Statistics
300
+ In this section, we evaluate different statistical measures for
301
+ quantifying the distribution of captions across the dataset.
302
+ Table 1 shows the average caption length of captions present
303
+ in the dataset and across the train, validation and test sub-
304
+ sets.
305
+ We see that the average caption lengths are simi-
306
+ lar across the different data splits with the average caption
307
+ length being slightly greater than that of the COCO Cap-
308
+ tions dataset. We also show the standard deviation in cap-
309
+ tions sizes across different image-caption pairs. We also ex-
310
+ amine the words appearing in these captions by identifying
311
+ 3
312
+
313
+ Figure 2. Object Frequency Analysis using Treemaps
314
+ unique tokens present. To calculate the unique tokens, we
315
+ use the spaCy library for tokenizing and lemmatizing our
316
+ captions along with the removal of all stop words. Subse-
317
+ quently, we tag the different Parts of Speech (POS) present
318
+ and select tokens that belong to the classes [Common Noun,
319
+ Proper Noun, Adjectives and Verbs]. This provides a mini-
320
+ mum guarantee that the abstractive captions present are long
321
+ enough to contain adequate content and contextual features.
322
+ This analysis also ensures that the composition captions
323
+ present in the train, validation and test splits are consistent
324
+ with each other.
325
+ 3.2.2
326
+ News Image Analysis
327
+ Along with the captions, we also estimate image proper-
328
+ ties such as the number of recognizable objects present in
329
+ each image and average number of detected objects per
330
+ image. We use a YOLO-R based object detector [24] for
331
+ identifying the objects present in each image of our dataset.
332
+ The YOLO-R detector has been trained on the MS-COCO
333
+ dataset, containing 80 unique object classes of common-
334
+ place objects [2]. We test the pre-trained model with 0.4 as
335
+ the confidence threshold. We find that there are an average
336
+ of 2.57 objects per image in the ANNA. Fig. 2 shows the
337
+ most frequently appearing classes of objects in our dataset
338
+ using a treemap for visualization.
339
+ 3.2.3
340
+ Categories of News Articles Selected
341
+ In this section, we identify the different types of news ar-
342
+ ticles from which image-caption pairs were sourced for
343
+ dataset construction. In total, there exist 123 unique article
344
+ topics within our dataset. Only 13 of image-caption pairs do
345
+ not have accompanying article type information so we dis-
346
+ regard those pairs from our article topic analysis. From Fig.
347
+ 3, we see that there exists a good distribution across topics
348
+ such as Dining, Business, Real Estate, etc. This shows that
349
+ the news image-caption pairs are diverse and not limited to
350
+ only a particular type of news article.
351
+ 4. Experiments
352
+ In order to understand how different architectures learn
353
+ abstractive captions on the ANNA, we consider various text-
354
+ to-image synthesis models previously proposed in litera-
355
+ ture. The three model architectures we test as a part of our
356
+ evaluation are: Lafite [32], AttnGAN+CL [26] and DM-
357
+ GAN+CL [27]. These models are selected for comparison
358
+ as they are among the top-10 on the COCO Captions Text-
359
+ to-Image synthesis leaderboard and take significantly dif-
360
+ ferent approaches for tackling the same task. As all these
361
+ models have achieved State-of-the-Art scores on descrip-
362
+ tive caption datasets, we evaluate how they perform with
363
+ news domain-specific, abstractive captions in our experi-
364
+ ments and visualize our results.
365
+ Text-to-Image Synthesis Models
366
+ The Lafite model uti-
367
+ lizes a pre-trained CLIP encoder for translating text em-
368
+ beddings into the image feature space.
369
+ It adapts an un-
370
+ conditional StyleGAN2 generator [7] by injecting text-
371
+ conditional information through affine transformations.
372
+ Two Fully Connected Layers are utilized to transform the
373
+ input text features to be more semantically similar with
374
+ StyleGAN’s image Stylespace. In our experiments, we train
375
+ Lafite on ANNA in a fully-supervised setting. We train 2
376
+ variants of Lafite, with and without Transfer Learning. In
377
+ the non-transfer learning variant, we train it on the ANNA
378
+ 4
379
+
380
+ Object Frequency Analysis
381
+ person
382
+ chair
383
+ book
384
+ cake
385
+ spoon
386
+ couch
387
+ boat
388
+ 17,069
389
+ 5,252
390
+ 2,073
391
+ 1,179
392
+ 1,087
393
+ 1,064
394
+ 1,001
395
+ potted plant
396
+ carrot
397
+ bed
398
+ wine
399
+ fork
400
+ 1,964
401
+ 890
402
+ 702
403
+ 627
404
+ dining table
405
+ bench
406
+ 4,090
407
+ cup
408
+ 868
409
+ 6op
410
+ 1,717
411
+ clock
412
+ knife
413
+ 835
414
+ bird
415
+ bowl
416
+ 1,669
417
+ 3,779
418
+ cell
419
+ phone
420
+ cat
421
+ bus
422
+ car
423
+ orange
424
+ truck
425
+ tv
426
+ 6,155
427
+ 1,332
428
+ 780
429
+ airplane
430
+ bottle
431
+ 2,416
432
+ vase
433
+ donut
434
+ train
435
+ 1,183
436
+ 762
437
+ cowFigure 3. Visualizing Article Categories of image-caption pairs present in ANNA
438
+ Model
439
+ IS (↑)
440
+ FIDCLIP (↓)
441
+ LPIPS(↓)
442
+ CLIPScore (↑)
443
+ Lafite (Transfer Learning)
444
+ 16.49
445
+ 13.93
446
+ 0.7470
447
+ 0.7575
448
+ Lafite (Base)
449
+ 12.59
450
+ 20.48
451
+ 0.7432
452
+ 0.7277
453
+ DMGAN+CL (512 dim)
454
+ 14.07
455
+ 29.30
456
+ 0.7568
457
+ 0.5913
458
+ DMGAN+CL (256 dim)
459
+ 13.37
460
+ 29.87
461
+ 0.7581
462
+ 0.5861
463
+ AttnGAN+CL (512 dim)
464
+ 12.56
465
+ 41.00
466
+ 0.7623
467
+ 0.5695
468
+ AttnGAN+CL (256 dim)
469
+ 13.06
470
+ 37.41
471
+ 0.7616
472
+ 0.5748
473
+ Table 2. Results of Abstractive Text-to-Image synthesis on ANNA
474
+ until convergence for 4000 epochs. To perform Transfer
475
+ Learning, we initialize the model with pre-trained weights
476
+ from the Conceptual Captions (CC3M) dataset [18] and
477
+ continue training on the ANNA until convergence for 2000
478
+ epochs.
479
+ The AttnGAN+CL and DMGAN+CL models share sim-
480
+ ilar architectures, with both utilizing a Deep Attentional
481
+ Multimodal Similarity Model (DAMSM) for computing the
482
+ similarity between extracted images and text. These archi-
483
+ tectures have been supplemented with a Constrastive Learn-
484
+ ing Loss function along with their DAMSM loss to improve
485
+ pre-training performance. We first train the DAMSM mod-
486
+ ule on the Train and Validation sets of our dataset to con-
487
+ struct the mapping between image and text features. We
488
+ compare 2 different embedding sizes of the DAMSM mod-
489
+ ule for both models: 256 and 512. The default AttnGAN
490
+ and DMGAM models have 256 embedding feature vectors
491
+ by default, but the CLIP based model Lafite uses 512 em-
492
+ bedding feature vectors instead. Thus, we train the models
493
+ with both embedding sizes to ensure a fair comparison.
494
+ Evaluation Metrics
495
+ To evaluate the performance of these
496
+ architectures, we report 4 different metrics: Inception Score
497
+ (IS), Fr´echet Inception Distance (FID), Learned Perceptual
498
+ Image Patch Similarity (LPIPS) and CLIPScore. IS and FID
499
+ evaluate the quality and diversity of generated images. They
500
+ estimate probability distribution properties of the generated
501
+ images and how far it diverges from that of the reference im-
502
+ ages. For FID, we adapt the proposed FIDCLIP from [8]
503
+ due to its closer correspondence with human judgement on
504
+ real-world, diverse datasets. LPIPS judges the perceptual
505
+ similarity between the reference and generated images us-
506
+ ing deep features extracted across image patches instead of
507
+ measuring pixel-level similarity. We use LPIPS version 0.1
508
+ for our testing. Since LPIPS is an image-wise similarity
509
+ metric, we report the average of scores obtained by the gen-
510
+ erated test set images. CLIPScore is a reference-free metric
511
+ that can be employed to evaluate the relevance of input text
512
+ captions to the content of generated images. We selected
513
+ these 4 metrics as they provide a holistic evaluation of the
514
+ different key aspects involved in measuring text-to-image
515
+ model performance. We report our scores in Table 2.
516
+ 5
517
+
518
+ Article Categories
519
+ Metro
520
+ 3,646
521
+ Dining
522
+ 3,010
523
+ Business
524
+ RealEstate
525
+ 2,573
526
+ Science
527
+ 12,224
528
+ National
529
+ Foreign
530
+ 2,023
531
+ Travel
532
+ 1,432
533
+ Culture
534
+ 1,120
535
+ Styles
536
+ Sports
537
+ 959
538
+ Weekend
539
+ 805
540
+ Home
541
+ 601
542
+ Magazine
543
+ TStyle
544
+ 412
545
+ Automobiles
546
+ SundayBusiness
547
+ 398
548
+ Metropolitan
549
+ Arts&Leisure
550
+ 1342
551
+ OpEd
552
+ NYTNOW
553
+ 1247
554
+ BookReview
555
+ Escapes
556
+ 1199
557
+ SpecialSections
558
+ Learning
559
+ 158
560
+ CityWeekly
561
+ Express
562
+ 140
563
+ Washington
564
+ Upshot
565
+ 109
566
+ Regionals
567
+ 0
568
+ 200
569
+ 400
570
+ 600
571
+ 800
572
+ 1000
573
+ 1200 1400 1600
574
+ 1800
575
+ 2000
576
+ 2200 2400 2600 2800
577
+ 3000
578
+ 32003400
579
+ 3600 3800
580
+ Count =(a) Original Image
581
+ (b) Lafite (Transfer
582
+ Learning)
583
+ (c) Lafite (Base)
584
+ (d)
585
+ DMGAN
586
+ (512
587
+ dim)
588
+ (e)
589
+ DMGAN
590
+ (256
591
+ dim)
592
+ (f)
593
+ AttnGAN
594
+ (512
595
+ dim)
596
+ (g)
597
+ AttnGAN
598
+ (256
599
+ dim)
600
+ Figure 4. Result Visualization for Caption: The castle, draped with vines and adorned with bougainvillea, is set on 10 acres, with
601
+ gardens, a swimming pool and a private chapel.
602
+ (a) Original Image
603
+ (b) Lafite (Transfer
604
+ Learning)
605
+ (c) Lafite (Base)
606
+ (d)
607
+ DMGAN
608
+ (512
609
+ dim)
610
+ (e)
611
+ DMGAN
612
+ (256
613
+ dim)
614
+ (f)
615
+ AttnGAN
616
+ (512
617
+ dim)
618
+ (g)
619
+ AttnGAN
620
+ (256
621
+ dim)
622
+ Figure 5. Result Visualization for Caption: Pollutants in the Gowanus Canal include pesticides, heavy metals and carcinogens like
623
+ PCBs.
624
+ 4.1. Evaluation of Generated Samples
625
+ Image Quality
626
+ From the reported IS and FID scores, we
627
+ can clearly identify that Lafit with Transfer Learning out-
628
+ performs all other models. Although the IS score of the
629
+ baseline model is lower than that of DMGAN+CL, this
630
+ trend is reversed in FID scores. This result can be attributed
631
+ to the fact that the Inception model feature space is aligned
632
+ to the classes present in ImageNet, hence penalizing other
633
+ datasets that diverge from this distribution [8]. The updated
634
+ CLIP feature space used for computing FIDCLIP helps
635
+ mitigate this issue and makes the metric more resistant to
636
+ fluctuations caused by image preprocessing and distortions.
637
+ These results also correlate with observed image quality on
638
+ other benchmark datasets, such as COCO Captions. We
639
+ provide visualizations of generated outputs from the test set
640
+ for all the trained models in Figures 4, 5, 6, 7, 8, 9.
641
+ Delineation between Content and Context features
642
+ The Lafite (Transfer Learning) model benefits from learned
643
+ associations between visual concepts and text represen-
644
+ tations in the absence of extremely descriptive captions,
645
+ which corroborates its high CLIPScore.
646
+ Similarly, for
647
+ the other models trained without transfer learning on our
648
+ dataset, we observe that the LPIPS score and CLIP-
649
+ Score follow the same trajectory as FIDCLIP with the
650
+ Lafite (Base) model exhibiting the best correlation between
651
+ ground truth image similarity and relevance with reference
652
+ captions. These results show that the top performing models
653
+ do have an implicit understanding of what constitutes image
654
+ content and context information. But limitations still exist
655
+ for implicit delineation of captions features, as shown in
656
+ Fig. 9. With the reference image and descriptive section of
657
+ the caption dealing with the image of an animal tracking de-
658
+ vice, the Text-to-Image models incorrectly generate an an-
659
+ 6
660
+
661
+ (a) Original Image
662
+ (b) Lafite (Transfer
663
+ Learning)
664
+ (c) Lafite (Base)
665
+ (d)
666
+ DMGAN
667
+ (512
668
+ dim)
669
+ (e)
670
+ DMGAN
671
+ (256
672
+ dim)
673
+ (f)
674
+ AttnGAN
675
+ (512
676
+ dim)
677
+ (g)
678
+ AttnGAN
679
+ (256
680
+ dim)
681
+ Figure 6. Result Visualization for Caption: Left, the New Museum and the original adjacent building it purchased 12 years ago on
682
+ the Bowery, at right.
683
+ (a) Original Image
684
+ (b) Lafite (Transfer
685
+ Learning)
686
+ (c) Lafite (Base)
687
+ (d)
688
+ DMGAN
689
+ (512
690
+ dim)
691
+ (e)
692
+ DMGAN
693
+ (256
694
+ dim)
695
+ (f)
696
+ AttnGAN
697
+ (512
698
+ dim)
699
+ (g)
700
+ AttnGAN
701
+ (256
702
+ dim)
703
+ Figure 7. Result Visualization for Caption: The rooms at the Ace Hotel have high ceilings and oversized windows. Some of the larger
704
+ rooms and suites includes details like guitars, turntables and vinyl records.
705
+ imal as the image foreground rather than the tracker. Thus,
706
+ comprehension of caption structures and explicit feature de-
707
+ lineation must be improved.
708
+ These experiments demon-
709
+ strate the need for non-descriptive image-captions datasets,
710
+ such as ANNA for bridging the performance gap between
711
+ descriptive and abstractive captions.
712
+ 5. Discussion and Conclusion
713
+ Our experiments demonstrate how existing text-to-image
714
+ architectures understand abstractive captions present in
715
+ domain-specific data such as news media. We show that
716
+ implicit delineation between content and context features
717
+ have limitations, prompting the need for explicit feature de-
718
+ lineation and modified objective functions to better suit this
719
+ task. One major impact of understanding abstractive cap-
720
+ tions such as those present in ANNA is the reduction in re-
721
+ quirements for directly descriptive captioning. As the size
722
+ of datasets keep increasing, scaling up human annotation of
723
+ images to match demand adds a huge overhead. As descrip-
724
+ tive captions need to be tightly-coupled with the reference
725
+ image’s contents, there needs to be multiple rounds of eval-
726
+ uation and filtering, making it a manually tedious task. The
727
+ use of abstractive captions for images can greatly simplify
728
+ the human annotation process for datasets. Additionally,
729
+ ANNA motivates the development of journalism assistance
730
+ solutions. The use of keywords and descriptive prompts
731
+ with current image generators involves a lot of prompt en-
732
+ gineering to get relevant images for a specific topic [10].
733
+ High quality images are generated only when a particu-
734
+ larly restrictive sentence structure and vocabulary is used
735
+ in prompts. As models are trained to understand abstractive
736
+ captions, the requirements for intensive prompt engineering
737
+ would be significantly reduced. Similarly, achieving better
738
+ delineation between different feature types present in non-
739
+ 7
740
+
741
+ (a) Original Image
742
+ (b) Lafite (Transfer
743
+ Learning)
744
+ (c) Lafite (Base)
745
+ (d)
746
+ DMGAN
747
+ (512
748
+ dim)
749
+ (e)
750
+ DMGAN
751
+ (256
752
+ dim)
753
+ (f)
754
+ AttnGAN
755
+ (512
756
+ dim)
757
+ (g)
758
+ AttnGAN
759
+ (256
760
+ dim)
761
+ Figure 8. Result Visualization for Caption: The Full Orange: two all-beef patties, special sauce, lettuce.
762
+ (a) Original Image
763
+ (b) Lafite (Transfer
764
+ Learning)
765
+ (c) Lafite (Base)
766
+ (d)
767
+ DMGAN
768
+ (512
769
+ dim)
770
+ (e)
771
+ DMGAN
772
+ (256
773
+ dim)
774
+ (f)
775
+ AttnGAN
776
+ (512
777
+ dim)
778
+ (g)
779
+ AttnGAN
780
+ (256
781
+ dim)
782
+ Figure 9. Result Visualization for Caption: With the RoamEO base unit, left (which includes a collar), a dog owner can get radio
783
+ signals tracking the animal’s location, up to 1.5 miles away.
784
+ descriptive captions can also benefit related tasks such as
785
+ image retrieval. The addition of context can play a major
786
+ role in influencing the quality of retrievals.
787
+ Limitations
788
+ This paper aims at introducing the potential
789
+ of abstractive captions to motivate the development of more
790
+ contextually-grounded text-to-image synthesis models, par-
791
+ ticularly when synthesizing news-domain specific images.
792
+ Although news articles contain a lot of named-entities, we
793
+ choose to filter them out and instead focus on context fea-
794
+ tures that can be inferred from text captions and depicted by
795
+ general visual concepts. Developing text-to-image synthe-
796
+ sis architectures that can take advantage of named-entities
797
+ using external knowledge bases as reference would help
798
+ overcome this limitation. Large-scale human evaluation of
799
+ images generated by text-to-image architectures on abstrac-
800
+ tive captions is another important step towards measuring
801
+ their relative performance, which we aim to perform as a
802
+ part of our future research.
803
+ Potential negative societal impacts
804
+ Image generation ar-
805
+ chitectures have the potential to be misused for nefarious
806
+ use-cases such as spreading disinformation [31] and gen-
807
+ erating neural fake news [28]. Our current preprocessing
808
+ pipeline removes most images containing named-entities,
809
+ i.e. public figures and locations of national importance, con-
810
+ tributing towards risk mitigation. However, we recognize
811
+ the threat posed by contextually-relevant Deepfake images
812
+ when dealing with news media images. Future research di-
813
+ rections include understanding the extent up to which text-
814
+ to-image models can be used for neural fake news genera-
815
+ tion and identifying appropriate detection strategies.
816
+ 6. Acknowledgements
817
+ This research has been partially supported by NSF
818
+ Awards #1820609 and #2114824.
819
+ 8
820
+
821
+ References
822
+ [1] Soravit Changpinyo, Piyush Sharma, Nan Ding, and Radu
823
+ Soricut.
824
+ Conceptual 12m: Pushing web-scale image-text
825
+ pre-training to recognize long-tail visual concepts. In 2021
826
+ IEEE/CVF Conference on Computer Vision and Pattern
827
+ Recognition (CVPR), pages 3558–3568. IEEE, June 2021.
828
+ 3
829
+ [2] Xinlei Chen, Hao Fang, Tsung-Yi Lin, Ramakrishna Vedan-
830
+ tam, Saurabh Gupta, Piotr Dollar, and C Lawrence Zitnick.
831
+ Microsoft COCO captions: Data collection and evaluation
832
+ server. arXiv preprint arXiv:1504.00325, Apr. 2015. 1, 2, 4
833
+ [3] Jiankang Deng, Jia Guo, Evangelos Ververas, Irene Kot-
834
+ sia, and Stefanos Zafeiriou. Retinaface: Single-shot multi-
835
+ level face localisation in the wild.
836
+ In Proceedings of
837
+ the IEEE/CVF conference on computer vision and pattern
838
+ recognition, pages 5203–5212. IEEE, 2020. 3
839
+ [4] Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing
840
+ Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and
841
+ Yoshua Bengio. Generative adversarial nets. In Advances in
842
+ Neural Information Processing Systems, volume 27. Curran
843
+ Associates, Inc., 2014. 2
844
+ [5] Herbert P Grice. Logic and conversation. In Speech acts,
845
+ pages 41–58. Brill, 1975. 1
846
+ [6] Huaibo Huang, Zhihang Li, Ran He, Zhenan Sun, and Tieniu
847
+ Tan.
848
+ IntroVAE: introspective variational autoencoders for
849
+ photographic image synthesis. In Proceedings of the 32nd
850
+ International Conference on Neural Information Processing
851
+ Systems, NIPS’18, pages 52–63, Red Hook, NY, USA, Dec.
852
+ 2018. Curran Associates Inc. 2
853
+ [7] Tero Karras, Miika Aittala, Janne Hellsten, Samuli Laine,
854
+ Jaakko Lehtinen, and Timo Aila. Training generative adver-
855
+ sarial networks with limited data. In Proceedings of the 34th
856
+ International Conference on Neural Information Processing
857
+ Systems, number Article 1015 in NIPS’20, pages 12104–
858
+ 12114, Red Hook, NY, USA, Dec. 2020. Curran Associates
859
+ Inc. 4
860
+ [8] Tuomas Kynk¨a¨anniemi, Tero Karras, Miika Aittala, Timo
861
+ Aila, and Jaakko Lehtinen. The role of ImageNet classes
862
+ in fr´echet inception distance. Mar. 2022. 5, 6
863
+ [9] Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays,
864
+ Pietro Perona, Deva Ramanan, Piotr Doll´ar, and C Lawrence
865
+ Zitnick. Microsoft COCO: Common objects in context. In
866
+ Computer Vision – ECCV 2014, pages 740–755. Springer In-
867
+ ternational Publishing, 2014. 2
868
+ [10] Vivian Liu and Lydia B Chilton.
869
+ Design guidelines for
870
+ prompt engineering Text-to-Image generative models.
871
+ In
872
+ Proceedings of the 2022 CHI Conference on Human Fac-
873
+ tors in Computing Systems, number Article 384 in CHI ’22,
874
+ pages 1–23, New York, NY, USA, Apr. 2022. Association
875
+ for Computing Machinery. 7
876
+ [11] Allen Nie, Reuben Cohn-Gordon, and Christopher Potts.
877
+ Pragmatic Issue-Sensitive image captioning.
878
+ In Findings
879
+ of the Association for Computational Linguistics: EMNLP
880
+ 2020, pages 1924–1938, Online, Nov. 2020. Association for
881
+ Computational Linguistics. 2
882
+ [12] Maria-Elena Nilsback and Andrew Zisserman. Automated
883
+ flower classification over a large number of classes. In 2008
884
+ Sixth Indian Conference on Computer Vision, Graphics &
885
+ Image Processing, pages 722–729. IEEE, Dec. 2008. 2
886
+ [13] Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya
887
+ Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sas-
888
+ try,
889
+ Amanda
890
+ Askell,
891
+ Pamela
892
+ Mishkin,
893
+ Jack
894
+ Clark,
895
+ Gretchen Krueger, and Ilya Sutskever.
896
+ Learning trans-
897
+ ferable visual models from natural language supervision.
898
+ arXiv:2103.00020 [cs], Feb. 2021. 2, 3
899
+ [14] Aditya Ramesh, Prafulla Dhariwal, Alex Nichol, Casey Chu,
900
+ and Mark Chen. Hierarchical Text-Conditional image gener-
901
+ ation with CLIP latents. arXiv:2204.06125 [cs], Apr. 2022.
902
+ 1, 2
903
+ [15] Aditya Ramesh, Mikhail Pavlov, Gabriel Goh, Scott Gray,
904
+ Chelsea Voss, Alec Radford, Mark Chen, and Ilya Sutskever.
905
+ Zero-Shot Text-to-Image generation. Feb. 2021. 2
906
+ [16] Scott Reed, Zeynep Akata, Xinchen Yan, Lajanugen Lo-
907
+ geswaran, Bernt Schiele, and Honglak Lee. Generative ad-
908
+ versarial text to image synthesis.
909
+ In Proceedings of The
910
+ 33rd International Conference on Machine Learning, pages
911
+ 1060–1069. PMLR, June 2016. 2
912
+ [17] Christoph Schuhmann, Richard Vencu, Romain Beaumont,
913
+ Robert Kaczmarczyk, Clayton Mullis, Aarush Katta, Theo
914
+ Coombes, Jenia Jitsev, and Aran Komatsuzaki.
915
+ LAION-
916
+ 400M: Open dataset of CLIP-Filtered 400 million Image-
917
+ Text pairs. Nov. 2021. 1
918
+ [18] Piyush Sharma, Nan Ding, Sebastian Goodman, and Radu
919
+ Soricut. Conceptual captions: A cleaned, hypernymed, im-
920
+ age alt-text dataset for automatic image captioning. In Pro-
921
+ ceedings of the 56th Annual Meeting of the Association for
922
+ Computational Linguistics (Volume 1: Long Papers), pages
923
+ 2556–2565, Melbourne, Australia, July 2018. Association
924
+ for Computational Linguistics. 2, 5
925
+ [19] Alasdair Tran, Alexander Mathews, and Lexing Xie. Trans-
926
+ form and tell: Entity-aware news image captioning.
927
+ In
928
+ Proceedings of the IEEE/CVF Conference on Computer Vi-
929
+ sion and Pattern Recognition, pages 13035–13045. openac-
930
+ cess.thecvf.com, 2020. 3
931
+ [20] A Tsirikoglou, G Eilertsen, and J Unger. A survey of im-
932
+ age synthesis methods for visual machine learning. Comput.
933
+ Graph. Forum, 39(6):426–451, Sept. 2020. 1
934
+ [21] Emiel van Miltenburg, Roser Morante, and Desmond Elliott.
935
+ Pragmatic factors in image description: The case of nega-
936
+ tions.
937
+ In Proceedings of the 5th Workshop on Vision and
938
+ Language, pages 54–59, Berlin, Germany, Aug. 2016. Asso-
939
+ ciation for Computational Linguistics. 2
940
+ [22] Ramakrishna Vedantam, Samy Bengio, Kevin Murphy, Devi
941
+ Parikh, and Gal Chechik.
942
+ Context-Aware captions from
943
+ Context-Agnostic supervision.
944
+ In 2017 IEEE Conference
945
+ on Computer Vision and Pattern Recognition (CVPR), pages
946
+ 1070–1079. IEEE, July 2017. 2
947
+ [23] Catherine Wah, Steve Branson, Peter Welinder, Pietro Per-
948
+ ona, and Serge Belongie.
949
+ The Caltech-UCSD birds-200-
950
+ 2011 dataset, July 2011. 2
951
+ [24] Chien-Yao Wang, I-Hau Yeh, and Hong-Yuan Mark Liao.
952
+ You only learn one representation: Unified network for mul-
953
+ tiple tasks. May 2021. 4
954
+ [25] Weihao Xia, Yujiu Yang, Jing-Hao Xue, and Baoyuan Wu.
955
+ TediGAN: Text-Guided diverse face image generation and
956
+ 9
957
+
958
+ manipulation. In 2021 IEEE/CVF Conference on Computer
959
+ Vision and Pattern Recognition (CVPR), pages 2256–2265.
960
+ IEEE, June 2021. 2
961
+ [26] Tao Xu, Pengchuan Zhang, Qiuyuan Huang, Han Zhang, Zhe
962
+ Gan, Xiaolei Huang, and Xiaodong He.
963
+ AttnGAN: Fine-
964
+ Grained text to image generation with attentional genera-
965
+ tive adversarial networks. In 2018 IEEE/CVF Conference
966
+ on Computer Vision and Pattern Recognition, pages 1316–
967
+ 1324, Salt Lake City, UT, USA, 2018. IEEE. 2, 4
968
+ [27] Hui Ye, Xiulong Yang, Martin Takac, Rajshekhar Sunder-
969
+ raman, and Shihao Ji. Improving Text-to-Image synthesis
970
+ using contrastive learning. The 32nd British Machine Vision
971
+ Conference (BMVC), July 2021. 2, 4
972
+ [28] Rowan Zellers, Ari Holtzman, Hannah Rashkin, Yonatan
973
+ Bisk, Ali Farhadi, Franziska Roesner, and Yejin Choi. De-
974
+ fending against neural fake news. Adv. Neural Inf. Process.
975
+ Syst., 32, 2019. 8
976
+ [29] Han Zhang, Jing Yu Koh, Jason Baldridge, Honglak Lee, and
977
+ Yinfei Yang. Cross-modal contrastive learning for text-to-
978
+ image generation. In 2021 IEEE/CVF Conference on Com-
979
+ puter Vision and Pattern Recognition (CVPR), pages 833–
980
+ 842. IEEE, June 2021. 2
981
+ [30] Han Zhang, Tao Xu, Hongsheng Li, Shaoting Zhang, Xiao-
982
+ gang Wang, Xiaolei Huang, and Dimitris N Metaxas. Stack-
983
+ gan: Text to photo-realistic image synthesis with stacked
984
+ generative adversarial networks. In Proceedings of the IEEE
985
+ international conference on computer vision, pages 5907–
986
+ 5915, 2017. 2
987
+ [31] Tao Zhang.
988
+ Deepfake generation and detection, a survey.
989
+ Multimed. Tools Appl., 81(5):6259–6276, Feb. 2022. 1, 8
990
+ [32] Yufan Zhou, Ruiyi Zhang, Changyou Chen, Chunyuan Li,
991
+ Chris Tensmeyer, Tong Yu, Jiuxiang Gu, Jinhui Xu, and
992
+ Tong Sun.
993
+ Towards Language-Free training for Text-to-
994
+ Image generation. In Proceedings of the IEEE/CVF Con-
995
+ ference on Computer Vision and Pattern Recognition, pages
996
+ 17907–17917. openaccess.thecvf.com, 2022. 2, 4
997
+ [33] Minfeng Zhu, Pingbo Pan, Wei Chen, and Yi Yang. Dm-gan:
998
+ Dynamic memory generative adversarial networks for text-
999
+ to-image synthesis. In Proceedings of the IEEE/CVF con-
1000
+ ference on computer vision and pattern recognition, pages
1001
+ 5802–5810. openaccess.thecvf.com, 2019. 2
1002
+ 10
1003
+
7dA0T4oBgHgl3EQfOf8M/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
8tE1T4oBgHgl3EQfngT9/content/tmp_files/2301.03311v1.pdf.txt ADDED
@@ -0,0 +1,2053 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Astronomy & Astrophysics manuscript no. main_new
2
+ ©ESO 2023
3
+ January 10, 2023
4
+ Unbound stars hold the key to young star cluster history
5
+ Arunima Arunima1,2, Susanne Pfalzner1,2,3, and Amith Govind1,2
6
+ 1 Jülich Supercomputing Center, Forschungszentrum Jülich, 52428 Jülich, Germany
7
+ e-mail: [email protected]
8
+ 2 Physics Department, University of Cologne, Cologne, Germany
9
+ 3 Max Planck Institute for Radio Astronomy, Auf dem Hügel 69, 53121 Bonn, Germany
10
+ Received ...
11
+ ABSTRACT
12
+ Aims. GAIA delivers the positions and velocities of stars at an unprecedented precision. Therefore, for star clusters, there exists much
13
+ higher confidence in whether a specific star is a member of a particular cluster or not. However, membership determination is still
14
+ especially challenging for young star clusters. At ages 2–10 Myr, the gas is expelled, ending the star formation process and leading to
15
+ their expansion, while at the same time, many former members become unbound. As a first step, we aim to assess the accuracy of the
16
+ methods commonly used to distinguish between bound and unbound cluster members; after identifying the most suitable technique
17
+ for this task, we wish to understand which of the two populations is more suited to provide insights into the initial configuration and
18
+ the dynamical history of a cluster starting from its currently observed properties.
19
+ Methods. Here, we perform N-body simulations of the dynamics of such young star clusters. We investigate how cluster dynamics
20
+ and observational limitations affect the recovered information about the cluster from a theoretical perspective.
21
+ Results. We find that the much-used method of distance and velocity cutoffs for membership determination often leads to false
22
+ negatives and positives alike. Often observational studies focus on the stars remaining bound. However, bound stars quickly lose the
23
+ memory of the pre-gas expulsion phase due to their ongoing interaction with their fellow cluster members. Our study shows that it
24
+ is the unbound stars that hold the key to charting a cluster’s dynamic history. Backtracking unbound stars can provide the original
25
+ cluster size and determine the time of gas expulsion – two parameters that are currently still poorly constrained. This information
26
+ is lost in the bound population. In addition, former members are often better indicators for disc lifetimes or initial binary fractions.
27
+ We apply the backtracking analysis, with varying success, to the clusters: Upper Scorpius and NGC 6530. For highly substructured
28
+ clusters such as Upper Scorpius, backtracking to the individual subcluster centres will provide better results in future.
29
+ Key words. stars: formation – open clusters and associations: general – ISM: clouds – solar neighbourhood
30
+ 1. Introduction
31
+ Star clusters are the nurseries for most stars (Porras et al. 2003;
32
+ Lada & Lada 2003). As such, young star clusters play a vital role
33
+ in our understanding of how young stars form and develop. They
34
+ signify the starting point for all that happens later on, as they pro-
35
+ vide the initial stellar mass distribution (e.g. Kroupa 2002) and
36
+ the fraction of stars forming as a single-, binary-, or multiple-star
37
+ system (e.g. Duchêne et al. 2018). It is a standard procedure to
38
+ use properties of clusters of different ages to obtain information
39
+ on the dynamical development of young binary stars or the dis-
40
+ persal time of discs (e.g. Haisch et al. 2001; Ansdell et al. 2017;
41
+ Marks et al. 2014; Ribas et al. 2014; Richert et al. 2018; Michel
42
+ et al. 2021). Often the task of determining cluster membership
43
+ and deriving the temporal development of specific properties are
44
+ separate endeavours. While distinguishing members is a chal-
45
+ lenge in itself, any bias in membership determination (i.e. false
46
+ positives and false negatives) feeds through to the derived pa-
47
+ rameters used in other applications.
48
+ This study’s central aim is to utilise cluster dynamics simula-
49
+ tions to optimise the data used to determine a cluster’s past. Un-
50
+ til recently, the role of dynamics during the formation history of
51
+ young clusters was highly uncertain (e.g. Elmegreen 2000; Fujii
52
+ et al. 2012; Ward et al. 2012; Banerjee & Kroupa 2017; Dib et al.
53
+ 2018), mainly because observational limitations hampered pre-
54
+ cise velocity determination. The precision of data coming from
55
+ the Gaia satellite (Gaia Collaboration et al. 2016, 2018, 2021)
56
+ helped shed light on this issue since a complete understanding
57
+ of the dynamical evolution of present-day clusters has not been
58
+ attained yet. Investigating a sample of 28 clusters and associa-
59
+ tions with ages ≈ 1–5 Myr, Kuhn et al. (2019) found that at least
60
+ 75% of these systems are expanding at typical expansion veloc-
61
+ ities of the order of ≈ 0.5 km s−1. Cluster expansion was pre-
62
+ dicted by the gas expulsion scenario (Mathieu 1983; Lada et al.
63
+ 1984; Adams 2000; Kroupa et al. 2001; Baumgardt & Kroupa
64
+ 2007; Pelupessy & Portegies Zwart 2012; Pfalzner & Kaczmarek
65
+ 2013; Brinkmann et al. 2017; Pfalzner & Govind 2021). During
66
+ the star formation phase, the stars are embedded in the gas and
67
+ dust reservoir from which they are forming. However, after ap-
68
+ proximately 1–2 Myr, the gas starts to be expelled from the clus-
69
+ ters by various mechanisms (e.g. Krumholz et al. 2019; Fujii
70
+ et al. 2021). Due to loss in gas and dust mass, the system is no
71
+ longer in equilibrium. Therefore, a considerable portion of the
72
+ stars, bound in the embedded phase, become unbound in the gas
73
+ expulsion phase.
74
+ The three-dimensional information available from the Gaia
75
+ data has been a tremendous step forward in this field. Neverthe-
76
+ less, discriminating the members of star clusters and associations
77
+ from the foreground and background population is still challeng-
78
+ ing (Gagné et al. 2018). Many new methods have been devel-
79
+ oped for determining the members of open and globular clusters
80
+ (e.g. Sollima et al. 2019; Garro et al. 2021; Vitral 2021). Cluster
81
+ Article number, page 1 of 14
82
+ arXiv:2301.03311v1 [astro-ph.GA] 9 Jan 2023
83
+
84
+ A&A proofs: manuscript no. main_new
85
+ membership determination is challenging in the early expansion
86
+ phase (< 10 Myr), especially if a clear-cut distinction between
87
+ currently bound and formerly bound (i.e. unbound) members is
88
+ required. In this case, there are additional difficulties to over-
89
+ come compared to older clusters. First, the earliest stages of the
90
+ formation of star clusters are hidden from view by gas and dust.
91
+ Thus, at this young age, veiling is a severe problem. Second, the
92
+ young clusters’ expansion requires additional attention in mem-
93
+ bership determination. Third, short- and long-lived clusters co-
94
+ exist during a 10 Myr timespan (Lada & Lada 2003). They un-
95
+ dergo very different cluster dynamics (Pfalzner & Kaczmarek
96
+ 2013), and it is not always straightforward whether a specific
97
+ cluster will remain bound for a long time or not.
98
+ Here, we concentrate on these dynamical aspects of young
99
+ short-lived clusters1. Any cluster observation is just a snapshot
100
+ in time of the sequence of its dynamical evolution. Based on
101
+ simulations of the cluster dynamics, we show the importance of
102
+ cluster dynamics in membership determination. We investigate
103
+ the efficiency of backtracking cluster expansion and find that dis-
104
+ tinguishing between bound and unbound stars in the expansion
105
+ phase is vital. Finally, we show that the unbound stars hold the
106
+ key to determining a cluster’s past.
107
+ 2. Cluster observation techniques
108
+ Historically, star clusters have been identified visually as stel-
109
+ lar density enhancements (Dreyer 1888; Trumpler 1930; Bailey
110
+ 1908; Collinder 1931). Surveys like Hipparcos (Perryman et al.
111
+ 1997), 2MASS (Skrutskie et al. 2006), and Gaia have each in-
112
+ creased the samples by hundreds of candidate clusters. Due to
113
+ Gaia’s high-precision parallax measurements, the clustering of
114
+ stars can be analysed in a higher dimensional space by combin-
115
+ ing their positions in the sky, proper motions, parallaxes, and
116
+ radial velocities (when available). For studies which do auto-
117
+ mated blind searches with clustering algorithms, the youth of the
118
+ stars is used as a confirmation of membership. Such youth indi-
119
+ cators can be X-ray activity, infrared excess (Broos et al. 2013;
120
+ Feigelson et al. 2013; Getman et al. 2017), lithium abundance
121
+ (Soderblom 2010), and gravity-sensitive spectral indices such
122
+ as TiO molecular lines (Wilking et al. 2005), empirically con-
123
+ structed spectral indices (Damiani et al. 2014), or the shape of
124
+ the H-band peak (Scholz et al. 2009).
125
+ Among the clustering algorithms, one can distinguish differ-
126
+ ent classes: Density-based spatial clustering like DBSCAN (Es-
127
+ ter et al. 1996; Wilkinson et al. 2018; Zari et al. 2019; Castro-
128
+ Ginard et al. 2019, 2020, 2022; Hunt & Reffert 2021), HDB-
129
+ SCAN (Campello et al. 2013), and OPTICS (Ordering Points
130
+ To Identify the Clustering Structure; Ankerst et al. 1999), mul-
131
+ tidimensional Gaussian-based methods (Vasiliev 2019; Cantat-
132
+ Gaudin et al. 2019; Kuhn et al. 2020), k-means clustering (Mac-
133
+ Queen 1967; Hunt & Reffert 2021), and Friend of Friend algo-
134
+ rithm (FoF; Liu & Pang 2019). In addition, there exist several un-
135
+ supervised algorithms like UPMASK (Krone-Martins & Moit-
136
+ inho 2014; Cantat-Gaudin et al. 2018; Cantat-Gaudin & Anders
137
+ 2020), the nearest neighbour-based method by He et al. (2021),
138
+ and STARGO (Tang et al. 2019; Zhang et al. 2020; Pang et al.
139
+ 2020).
140
+ 1 The nomenclature of short-lived clusters is not unequivocal. While
141
+ referred to as clusters while embedded, they are often classified as as-
142
+ sociations when the gas is expelled, and most of their stars become un-
143
+ bound. Here, we refer to short-lived clusters as clusters and point out
144
+ expressly when talking about long-lived clusters, that is, open and glob-
145
+ ular clusters.
146
+ Young star clusters pose additional challenges compared to
147
+ open or globular clusters due to their highly dynamic nature
148
+ after gas expulsion. Although space velocity is used to iden-
149
+ tify clusters, algorithms rarely consider dynamics. Observations
150
+ only provide a snapshot in the dynamic evolution of the clus-
151
+ ter. Hence, even clustering in the velocity space at the present
152
+ moment might be a chance alignment as the velocity changes
153
+ rapidly in young star cluster members. More limitations in iden-
154
+ tifying clusters come from Gaia’s poor completeness in crowded
155
+ fields and no particular regard for binarity. Moreover, young
156
+ clusters are still embedded in natal gas and dust that can not be
157
+ penetrated by optical wavelengths, which presents another diffi-
158
+ culty in identifying and analysing young clusters.
159
+ Blaauw (1964) first gave the notion of linear expansion in
160
+ associations, assuming that all members move away from their
161
+ birthplace without any forces acting on them. Then, the recipro-
162
+ cal of the expansion coefficient can provide an estimate of the as-
163
+ sociation’s kinematic age. Alternatively, the individual motions
164
+ of the stars can be traced back until they reach the smallest con-
165
+ figuration at a past time, and the kinematic age, as well as the
166
+ initial configuration of the association, can be possibly obtained
167
+ (Blaauw 1978).
168
+ Most studies apply cutoffs to remove objects with low-
169
+ quality astrometry and outliers. The sigma-clipping method aims
170
+ to reduce the chances of contaminants or uninformative stars and
171
+ improve clusters’ signal-to-noise ratio (S/N). Alternatively, out-
172
+ liers can be modelled in the fitting procedure without rejecting
173
+ points a priori (see Hogg et al. 2010).
174
+ Before Gaia, the significant errors in astrometry and the low
175
+ number of confirmed members with available radial velocities
176
+ were the main hindrances in the analysis (Fernández et al. 2008).
177
+ The higher precision of the Gaia data allows for better trace-
178
+ back analysis. For example, recent studies by Heyl et al. (2022,
179
+ 2021) trace back the stars of clusters aged 40–200 Myr using
180
+ Gaia EDR3 data and determine their kinematic ages. Similarly,
181
+ Schoettler et al. (2022) trace back runaway (RW) and slower
182
+ walkaway (WW) stars within a distance of 100 pc of NGC 2264
183
+ to the three subclusters S Mon, IRS 1 and IRS 2. The study by
184
+ Ma et al. (2022) uses Gaia DR2 data to trace back (and extrapo-
185
+ late) the trajectories of members of the Scorpius-Centaurus (Sco-
186
+ Cen) association and find evidence of past and future close stellar
187
+ flybys.
188
+ Observational challenges like distinguishing the cluster pop-
189
+ ulation from the back and foreground stars, limiting magnitudes,
190
+ imprecision of derived properties like age and mass, etc., com-
191
+ plicate backtracking. Here we apply backtracking to snapshots
192
+ in the simulations of the cluster dynamics. Under these idealised
193
+ conditions, membership is certain, the exact positions and ve-
194
+ locities of the stars are known at all times, and last, but not least,
195
+ we know what the result should be. This certainty allows us to
196
+ determine the most expedient method and suggest measures to
197
+ optimise the backtracking technique.
198
+ 3. Cluster simulation method
199
+ We use a sub-set of simulations of the dynamics of clusters
200
+ containing N stars we performed recently (Pfalzner & Govind
201
+ 2021), using the simulation code NBODY6++GPU (Aarseth 2003).
202
+ The simulations try to represent the situation in real clusters as
203
+ closely as possible by adopting initial conditions backed by re-
204
+ cent observations and following the observed cluster expansion
205
+ derived from the sizes of clusters in the age range of 1–10 Myr.
206
+ Here we give only a summary of the assumptions, and the nu-
207
+ merical method we applied in Pfalzner & Govind (2021), as the
208
+ Article number, page 2 of 14
209
+
210
+ Arunima Arunima et al.: Unbound stars hold the key to star cluster history
211
+ actual choice of simulation parameters is uncritical for the gen-
212
+ eral challenges in membership determination and backtracking
213
+ of the cluster history.
214
+ We model the dynamics of the young clusters covering all
215
+ the phases: Starting from the embedded phase, we simulate the
216
+ subsequent gas expulsion that leaves the cluster in a super-virial
217
+ state and results in the cluster expanding until it reaches a new
218
+ equilibrium. It is assumed that all stars are already formed and
219
+ that the gas expulsion occurs at temb = 2 Myr. Observations in-
220
+ dicate that the entire gas expulsion process takes ≈ 1 – 2 Myr
221
+ (Kuhn et al. 2019). Simulations investigating the dependence of
222
+ the cluster dynamics on the gas expulsion time found that the
223
+ gas expulsion can be modelled as being instantaneous (Geyer &
224
+ Burkert 2001; Portegies Zwart et al. 2010). Stellar evolution has
225
+ not been included in this work as it has little influence on the
226
+ results.
227
+ We analyse the dynamics of clusters with different numbers
228
+ of cluster members N. The corresponding clusters’ masses Mc
229
+ and sizes, illustrated by their half-mass radius rhm, are given in
230
+ Table 1. Low-mass clusters are usually smaller than high-mass
231
+ clusters of the same age (Lada & Lada 2003; Adams 2010;
232
+ Pfalzner et al. 2016). This relation between the cluster’s mass
233
+ and its half-mass radius can be approximated by a power law:
234
+ Mc = Crhm
235
+ γ.
236
+ (1)
237
+ The values of the constant C and scaling exponent γ differ in
238
+ different observational studies due to the involved observational
239
+ uncertainties. The clusters’ sizes given in Table 1 are based on
240
+ the mass-radius relation by Pfalzner et al. (2016) where C =
241
+ 717.794 and γ = 1.7 ± 0.2. We assume that the star formation
242
+ efficiency in the system is 30 % (Lada & Lada 2003), which
243
+ sets the gas mass. The gas and dust component of the embedded
244
+ phase is implemented as a background potential.
245
+ In our simulations, a test particle represents a star with a
246
+ given mass, position, and velocity. The particles’ positions are
247
+ chosen so that the resulting stellar number density distribution
248
+ obeys a King profile with King parameter, W0 = 9 (King 1966a).
249
+ The King model is an empirical law that can not be defined ana-
250
+ lytically. It consists of an energy distribution function of the form
251
+ fK(E) =
252
+ �ρ1(2πσ2
253
+ K)−3/2(eE/σ2
254
+ K − 1)
255
+ : E > 0,
256
+ 0
257
+ : E ≤ 0,
258
+ (2)
259
+ with E = Ψ− 1
260
+ 2ν2 and Ψ = −Φ+Φ0 being the relative energy and
261
+ relative potential of a particle, respectively. Also, f(E) > 0 for
262
+ E > 0 and σK is the King velocity dispersion. The profiles are
263
+ characterised by the King parameter W0 = Ψ/σ2
264
+ K, an increase of
265
+ which signifies decrease in the relative size of the cluster core
266
+ Table 1. Initial cluster parameters for the simulation campaign using
267
+ mass-radius dependencies.
268
+ N
269
+ Nsim
270
+ Mc
271
+ [M⊙]
272
+ rhm
273
+ [pc]
274
+ Mt
275
+ [M⊙]
276
+ temb
277
+ [Myr]
278
+ 200
279
+ 1941
280
+ 117.99
281
+ 0.26
282
+ 393.31
283
+ 2.0
284
+ 1000
285
+ 497
286
+ 589.97
287
+ 0.67
288
+ 1966.57
289
+ 2.0
290
+ 4000
291
+ 127
292
+ 2359.88
293
+ 1.3
294
+ 7866.27
295
+ 2.0
296
+ Notes. Here N denotes the number of cluster members, Nsim the number
297
+ of simulations, temb the duration of the embedded phase, Mc the stellar
298
+ mass of the cluster, rhm the half-mass radius, and Mt the total cluster
299
+ mass (stars + gas).
300
+ rc/rhm. Observationally, determining the stellar density distribu-
301
+ tion of young star clusters can be challenging but it has been
302
+ found that young clusters are best represented by King model
303
+ with W0 ≥ 7 (Hillenbrand & Hartmann 1998; Nürnberger &
304
+ Petr-Gotzens 2002). The choice of W0 mainly affects the size
305
+ of the central high-density area. Hence, the number of expelled
306
+ stars also depends on the choice of W0. Even for a relatively steep
307
+ W0 = 9-potential, the number of escapers is < 1%. Therefore, the
308
+ conclusions about membership determination methodology are
309
+ unaffected by the choice of potential. The individual test par-
310
+ ticles are assigned masses following the initial mass function
311
+ (IMF) by Kroupa (2002), with the lower mass limit set to 0.08
312
+ M⊙ (hydrogen-burning limit) and an upper mass limit of 150
313
+ M⊙. Potentially existing initial mass segregation in the clusters
314
+ is neglected. The cluster members are given velocities following
315
+ a Maxwellian distribution. We assume that the cluster is initially
316
+ in virial equilibrium.
317
+ We perform (Nsim) simulations for every cluster mass, where
318
+ the actual distribution of the stars depends on the seed selected in
319
+ the randomised procedure. We analyse all the simulation results
320
+ in this statistical study. However, why a specific method works
321
+ or fails, we illustrate exemplarily for just one specific randomly
322
+ chosen realisation in Figs. 1 – 3. Figures 6 – 8 also show the
323
+ method applied to randomly chosen specific clusters for visual
324
+ understanding; however, statistical results are mentioned in the
325
+ text.
326
+ For simplicity, we exclude primordial binaries, modelling all
327
+ cluster stars as initially being single stars. The absence of pri-
328
+ mordial binaries can lead to underestimating ejections from the
329
+ cluster centre (Heggie 1975). However, in most clusters, ≪1%
330
+ of the stars are affected (Olczak et al. 2006).
331
+ 4. Results
332
+ Observations investigate one specific cluster at a snapshot of its
333
+ development. Mimicking this observational situation, we ran-
334
+ domly choose one of our sets of simulations and investigate it
335
+ at a specific time. However, unlike actual observations, we have
336
+ complete temporal information available. Hence, we know the
337
+ past and the future of this particular cluster down to the path of
338
+ each star. Equally, all other observational challenges, like mem-
339
+ bership uncertainty due to back and foreground populations and
340
+ limiting magnitudes, are removed. We even know each star’s ex-
341
+ act properties like its mass, position, and velocity. This informa-
342
+ tion allows us to investigate the fundamental and unavoidable
343
+ challenges in backtracking caused by the cluster dynamics that
344
+ exist even without the mentioned additional observational diffi-
345
+ culties.
346
+ 4.1. Bound and unbound stars
347
+ After gas expulsion, bound and unbound stars coexist in the
348
+ same spatial area for some time. Distinguishing the two popu-
349
+ lations is vital for some applications; it does not matter or is not
350
+ even desirable for others. An example of the latter is the use of
351
+ clusters in determining disc lifetimes (Haisch et al. 2001). Here,
352
+ it is best to identify all stars that once formed together in the clus-
353
+ ter. However, if one is interested in the long-term development
354
+ of clusters (≫ 20 Myr), one would be predominantly interested
355
+ in the portion of stars that remain bound. We subsequently see
356
+ here that using backtracking to distinguish between bound and
357
+ unbound stars after gas expulsion is the key to success in ob-
358
+ taining valuable information concerning a cluster’s past. At each
359
+ Article number, page 3 of 14
360
+
361
+ A&A proofs: manuscript no. main_new
362
+ 6
363
+ 4
364
+ 2
365
+ 0
366
+ 2
367
+ 4
368
+ 6
369
+ x[pc]
370
+ 6
371
+ 4
372
+ 2
373
+ 0
374
+ 2
375
+ 4
376
+ 6
377
+ y[pc]
378
+ (a)
379
+ 6
380
+ 4
381
+ 2
382
+ 0
383
+ 2
384
+ 4
385
+ 6
386
+ x[pc]
387
+ 6
388
+ 4
389
+ 2
390
+ 0
391
+ 2
392
+ 4
393
+ 6
394
+ y[pc]
395
+ (b)
396
+ 6
397
+ 4
398
+ 2
399
+ 0
400
+ 2
401
+ 4
402
+ 6
403
+ x [pc]
404
+ 6
405
+ 4
406
+ 2
407
+ 0
408
+ 2
409
+ 4
410
+ 6
411
+ y [pc]
412
+ (c)
413
+ 30
414
+ 20
415
+ 10
416
+ 0
417
+ 10
418
+ 20
419
+ 30
420
+ x [pc]
421
+ 30
422
+ 20
423
+ 10
424
+ 0
425
+ 10
426
+ 20
427
+ 30
428
+ y [pc]
429
+ (d)
430
+ 6
431
+ 4
432
+ 2
433
+ 0
434
+ 2
435
+ 4
436
+ 6
437
+ x [pc]
438
+ 6
439
+ 4
440
+ 2
441
+ 0
442
+ 2
443
+ 4
444
+ 6
445
+ y [pc]
446
+ (e)
447
+ 6
448
+ 4
449
+ 2
450
+ 0
451
+ 2
452
+ 4
453
+ 6
454
+ x [pc]
455
+ 6
456
+ 4
457
+ 2
458
+ 0
459
+ 2
460
+ 4
461
+ 6
462
+ y [pc]
463
+ (f)
464
+ 6
465
+ 4
466
+ 2
467
+ 0
468
+ 2
469
+ 4
470
+ 6
471
+ x [pc]
472
+ 6
473
+ 4
474
+ 2
475
+ 0
476
+ 2
477
+ 4
478
+ 6
479
+ y [pc]
480
+ (g)
481
+ 6
482
+ 4
483
+ 2
484
+ 0
485
+ 2
486
+ 4
487
+ 6
488
+ x [pc]
489
+ 6
490
+ 4
491
+ 2
492
+ 0
493
+ 2
494
+ 4
495
+ 6
496
+ y [pc]
497
+ (h)
498
+ Fig. 1. Snapshot of the positions and velocities of example simulations
499
+ with N = 200. Velocity vectors of bound stars are highlighted in blue,
500
+ and those of unbound stars in red. Counter-intuitive examples of (a)
501
+ outward-pointing distant bound stars and (b) inward-pointing central
502
+ unbound stars. Snapshot of the temporal development at (c) t=2 Myr
503
+ and (d) t=10 Myr. Backtracking from the results at 10 Myr to 2 Myr
504
+ considering only the stars within 6 pc from the cluster centre for (e)
505
+ bound stars only and (f) unbound stars only. Same backtracking con-
506
+ sidering all the (g) bound stars and (f) unbound stars of the cluster.
507
+ A film of the cluster dynamics and the backtracking can be found at
508
+ https://doi.org/10.5281/zenodo.6041920
509
+ snapshot of the simulations, bound and unbound stars are de-
510
+ fined as those having positive and negative total energy respec-
511
+ tively. However, in observations, distinguishing between these
512
+ two states is often not straightforward.
513
+ 4.1.1. Velocity vectors
514
+ Individual stars are sometimes classified as bound or unbound
515
+ simply because their velocity vectors point towards or away from
516
+ the cluster centre. In the past, doubts about this approach were
517
+ usually anchored on the fact that only two-dimensional informa-
518
+ tion was available. However, even with three-dimensional infor-
519
+ mation becoming more accurate, this method is not advisable
520
+ even for perfectly known 3D velocities for the following reason:
521
+ The top row of Fig. 1 shows a typical snapshot of a randomly
522
+ chosen example from our sample of simulated clusters. The clus-
523
+ ter centre is marked as a green dot as a reference point. As the
524
+ many outward-pointing velocity vectors indicate, this cluster is
525
+ in the expansion phase, with many former members becoming
526
+ unbound. Nevertheless, a considerable fraction of the outward-
527
+ pointing velocity vectors belongs to stars that remain bound in
528
+ the long term. Examples of such stars are shown in blue. Equally,
529
+ stars that point inwards and are close to the cluster centre can
530
+ nevertheless be unbound (shown in red). The dynamics of these
531
+ example stars can be seen better in the corresponding video
532
+ at https://doi.org/10.5281/zenodo.6041920. Especially
533
+ among the bound stars with outward-pointing velocity vectors,
534
+ quite a few are bound despite being located at relatively large
535
+ distances from the cluster centre. We find that there is a high
536
+ failure rate in this approach, not only for this specific cluster, but
537
+ for all clusters in our extensive sample. The situation improves
538
+ for clusters aged more than 15 Myr as many of the unbound stars
539
+ are better identifiable by their larger distances to the cluster cen-
540
+ tre.
541
+ 4.1.2. Advantage of using unbound stars for backtracking
542
+ The size of a cluster before expansion sets in is an essential pa-
543
+ rameter for constraining the cluster formation process. Besides
544
+ the density profile, the size of the cluster core and half-mass
545
+ radius are good indicators of the cluster density and, thus, the
546
+ importance of the environment in the star and planet formation
547
+ process. The environment’s influence includes close stellar fly-
548
+ bys and external photo-evaporation that can truncate protoplan-
549
+ etary discs or completely destroy them (Vincke et al. 2015; Win-
550
+ ter et al. 2018; Concha-Ramírez et al. 2019). These processes
551
+ influence the type and frequency of the formed planetary sys-
552
+ tems. Another example is binary capture and destruction pro-
553
+ cesses which can alter the binary fraction in clusters (Kaczmarek
554
+ et al. 2011; Marks et al. 2014; Guszejnov et al. 2022).
555
+ We find that using just the unbound stars gives the best re-
556
+ sult in determining the pre-expansion cluster size. As an exam-
557
+ ple, the second row in Fig. 1 illustrates the cluster expansion by
558
+ showing the bound and unbound stars, including their velocity
559
+ vectors, (a) shortly after gas expulsion and (b) at 10 Myr for a
560
+ cluster with N = 200. We note the different scales. Using only the
561
+ bound stars for backtracking (see Fig. 1g) results in a relatively
562
+ poor constraint on the pre-expansion size. The best performance
563
+ is obtained using only the unbound stars (see Fig. 1f). The rea-
564
+ son is twofold: First, the velocity vectors of the unbound stars
565
+ are rarely altered after gas expulsion. By contrast, bound stars
566
+ quickly lose the memory of the pre-gas expulsion phase due to
567
+ their ongoing interaction with their fellow cluster members. In
568
+ particular, close encounters hinder efficient backtracking for the
569
+ bound stars. Second, there is a more significant number of un-
570
+ bound than bound stars. Thus, statistical uncertainties are more
571
+ easily averaged out.
572
+ Figure 4 gives a more quantitative idea of the use of bound vs
573
+ unbound stars for backtracking and deriving the pre-expansion
574
+ Article number, page 4 of 14
575
+
576
+ Arunima Arunima et al.: Unbound stars hold the key to star cluster history
577
+ 0
578
+ 10
579
+ 20
580
+ 30
581
+ 40
582
+ 50
583
+ 60
584
+ 70
585
+ d [pc]
586
+ 0.0
587
+ 0.2
588
+ 0.4
589
+ 0.6
590
+ 0.8
591
+ Frequency
592
+ Time= 1.8 Myr
593
+ 0
594
+ 10
595
+ 20
596
+ 30
597
+ 40
598
+ 50
599
+ 60
600
+ 70
601
+ d [pc]
602
+ 0.0
603
+ 0.1
604
+ 0.2
605
+ 0.3
606
+ 0.4
607
+ 0.5
608
+ Frequency
609
+ Time= 2.3 Myr
610
+ 0
611
+ 10
612
+ 20
613
+ 30
614
+ 40
615
+ 50
616
+ 60
617
+ 70
618
+ d [pc]
619
+ 0.000
620
+ 0.025
621
+ 0.050
622
+ 0.075
623
+ 0.100
624
+ 0.125
625
+ 0.150
626
+ 0.175
627
+ 0.200
628
+ Frequency
629
+ Time= 5.0 Myr
630
+ 0
631
+ 10
632
+ 20
633
+ 30
634
+ 40
635
+ 50
636
+ 60
637
+ 70
638
+ d [pc]
639
+ 0.00
640
+ 0.02
641
+ 0.04
642
+ 0.06
643
+ 0.08
644
+ 0.10
645
+ 0.12
646
+ Frequency
647
+ Time= 10.0 Myr
648
+ 0
649
+ 10
650
+ 20
651
+ 30
652
+ 40
653
+ 50
654
+ 60
655
+ 70
656
+ d [pc]
657
+ 0.00
658
+ 0.02
659
+ 0.04
660
+ 0.06
661
+ 0.08
662
+ 0.10
663
+ Frequency
664
+ Time= 20.0 Myr
665
+ 0
666
+ 1
667
+ 2
668
+ 3
669
+ 4
670
+ 5
671
+ 6
672
+ v [km/s]
673
+ 0.00
674
+ 0.02
675
+ 0.04
676
+ 0.06
677
+ 0.08
678
+ Frequency
679
+ Time= 1.8 Myr
680
+ 0
681
+ 1
682
+ 2
683
+ 3
684
+ 4
685
+ 5
686
+ 6
687
+ v [km/s]
688
+ 0.00
689
+ 0.01
690
+ 0.02
691
+ 0.03
692
+ 0.04
693
+ 0.05
694
+ 0.06
695
+ 0.07
696
+ 0.08
697
+ Frequency
698
+ Time= 2.3 Myr
699
+ 0
700
+ 1
701
+ 2
702
+ 3
703
+ 4
704
+ 5
705
+ 6
706
+ v [km/s]
707
+ 0.00
708
+ 0.02
709
+ 0.04
710
+ 0.06
711
+ 0.08
712
+ 0.10
713
+ Frequency
714
+ Time= 5.0 Myr
715
+ 0
716
+ 1
717
+ 2
718
+ 3
719
+ 4
720
+ 5
721
+ 6
722
+ v [km/s]
723
+ 0.00
724
+ 0.02
725
+ 0.04
726
+ 0.06
727
+ 0.08
728
+ 0.10
729
+ 0.12
730
+ Frequency
731
+ Time= 10.0 Myr
732
+ 0
733
+ 1
734
+ 2
735
+ 3
736
+ 4
737
+ 5
738
+ 6
739
+ v [km/s]
740
+ 0.00
741
+ 0.02
742
+ 0.04
743
+ 0.06
744
+ 0.08
745
+ 0.10
746
+ Frequency
747
+ Time= 20.0 Myr
748
+ 0.3
749
+ 1.0
750
+ 2.0
751
+ 5.0 10.0 20.0 40.0 80.0
752
+ d [pc]
753
+ 0.1
754
+ 0.2
755
+ 0.5
756
+ 1.0
757
+ 2.0
758
+ 4.0
759
+ 8.0
760
+ v [km/s]
761
+ Time= 1.8 Myr
762
+ (a)
763
+ 0.3
764
+ 1.0
765
+ 2.0
766
+ 5.0
767
+ 10.0 20.0 40.0 80.0
768
+ d [pc]
769
+ 0.1
770
+ 0.2
771
+ 0.5
772
+ 1.0
773
+ 2.0
774
+ 4.0
775
+ 8.0
776
+ v [km/s]
777
+ Time= 2.3 Myr
778
+ (b)
779
+ 0.3
780
+ 1.0
781
+ 2.0
782
+ 5.0
783
+ 10.0 20.0 40.0 80.0
784
+ d [pc]
785
+ 0.1
786
+ 0.2
787
+ 0.5
788
+ 1.0
789
+ 2.0
790
+ 4.0
791
+ 8.0
792
+ v [km/s]
793
+ Time= 5.0 Myr
794
+ (c)
795
+ 0.3
796
+ 1.0
797
+ 2.0
798
+ 5.0
799
+ 10.0
800
+ 20.0
801
+ 40.0
802
+ 80.0
803
+ d [pc]
804
+ 0.1
805
+ 0.2
806
+ 0.5
807
+ 1.0
808
+ 2.0
809
+ 4.0
810
+ 8.0
811
+ v [km/s]
812
+ Time= 10.0 Myr
813
+ (d)
814
+ 0.3
815
+ 1.0
816
+ 2.0
817
+ 5.0
818
+ 10.0
819
+ 20.0
820
+ 40.0
821
+ 80.0
822
+ d [pc]
823
+ 0.1
824
+ 0.2
825
+ 0.5
826
+ 1.0
827
+ 2.0
828
+ 4.0
829
+ 8.0
830
+ v [km/s]
831
+ Time= 20.0 Myr
832
+ (e)
833
+ Fig. 2. Snapshot of distance (top) and velocity distribution (middle), and distance vs velocity scatter plot (bottom) (a) before gas expulsion (t =
834
+ 1.8 Myr), (b) just after gas expulsion (t = 2.3 Myr), (c) at t = 5 Myr, (d) at t = 10 Myr, and (e) at the end of our simulation (t =20 Myr). All plots
835
+ show the bound stars in blue and the unbound stars in red. A simulation of N = 1000 stars is used here.
836
+ .
837
+ 0.3
838
+ 1.0
839
+ 2.0
840
+ 5.0
841
+ 10.0 20.0 40.0
842
+ d [pc]
843
+ 0.1
844
+ 0.2
845
+ 0.5
846
+ 1.0
847
+ 2.0
848
+ 4.0
849
+ v [km/s]
850
+ Time= 10.0 Myr
851
+ Fig. 3. Phase space diagram for an N = 1000 star cluster simulation at
852
+ t = 10 Myr. The bound and unbound members are shown in blue and
853
+ red colours respectively. Vertical and horizontal red lines indicate dis-
854
+ tance and velocity cutoffs respectively for unbound stars. The light blue
855
+ line represents the analytical escape velocity dependence on distance
856
+ from the cluster centre derived assuming a Plummer distribution for the
857
+ members. The black crosses show the stars that underwent a strong en-
858
+ counter.
859
+ cluster size. All the simulations of N = 1000 cluster have been
860
+ used to obtain these distributions. It can be seen that the size
861
+ distribution obtained using unbound stars is closer to the real size
862
+ distribution than the size distribution obtained using bound stars.
863
+ Performing a t-test on the two size distributions with the null
864
+ hypothesis being that the distributions have the same mean—
865
+ while the alternative hypothesis is that bound stars have a larger
866
+ mean than unbound stars—results in a p-value much lower than
867
+ the significance level α = 0.01. Hence, unbound stars are clearly
868
+ better at recovering the size of the cluster before gas expulsion
869
+ than bound stars.
870
+ 4.1.3. Distance and velocity cutoffs for bound-unbound
871
+ classification
872
+ While distinguishing between the bound and unbound popula-
873
+ tion is straightforward in simulations, it is very challenging in
874
+ observations. Often a cut in the distance to the cluster centre or
875
+ the velocity is used to distinguish between bound and unbound
876
+ stars. Here we want to test when such a method is successful.
877
+ In our simulation, the relevant time frame starts at 2 Myr,
878
+ when the gas expulsion happens, and many stars become un-
879
+ bound. Figure 2 shows snapshots of the distributions of the stel-
880
+ lar distance to the cluster centre and velocity distribution before
881
+ (1.8 Myr), just after gas expulsion at 2.3 Myr, during the expan-
882
+ sion process (5 and 10 Myr) and towards the end (20 Myr) of the
883
+ expansion phase for an example cluster. The distributions for the
884
+ bound (blue) and unbound (red) stars are shown separately. As
885
+ we chose the cluster to be in virial equilibrium, very few stars
886
+ become unbound before gas expulsion (see Fig. 2a). The few
887
+ unbound stars during this phase result from close encounters
888
+ leading to ejections. However, after gas expulsion, many stars
889
+ become unbound. Bound and unbound stars share considerable
890
+ parts of the phase space for quite some time, as seen in the bot-
891
+ tom row of Fig. 2. This increases the complexity of making the
892
+ distinction.
893
+ In observations, usually, a velocity cutoff is chosen as a given
894
+ deviation from the mean for making this distinction (e.g. Luh-
895
+ man 2018; Bastian 2019; Esplin & Luhman 2019). However, the
896
+ location of these cutoffs is not apparent. Thus, there is some ele-
897
+ Article number, page 5 of 14
898
+
899
+ A&A proofs: manuscript no. main_new
900
+ 0
901
+ 1
902
+ 2
903
+ 3
904
+ 4
905
+ 5
906
+ Half-mass radius [pc]
907
+ 0.0
908
+ 0.5
909
+ 1.0
910
+ 1.5
911
+ 2.0
912
+ 2.5
913
+ 3.0
914
+ 3.5
915
+ Real:
916
+ r = 0.67,
917
+ r = 0.13
918
+ Unbound:
919
+ u = 1.47,
920
+ u = 0.12
921
+ Bound:
922
+ b = 2.70,
923
+ b = 0.61
924
+ Real
925
+ Unbound
926
+ Bound
927
+ 1
928
+ 2
929
+ 3
930
+ 4
931
+ 5
932
+ Half-mass radius [pc]
933
+ Fig. 4. Distributions of sizes derived using actual positions of all stars
934
+ (Real, shown in green), using backtraced positions of unbound stars
935
+ (Unbound, shown in orange), and using backtraced positions of bound
936
+ stars (Bound, shown in blue) shown with histograms (top) and boxplots
937
+ (bottom). The box extends from the lower to upper quartile values of
938
+ the data, with a line at the median while the whiskers reach 1.5 times
939
+ the interquartile range from the box.
940
+ ment of arbitrariness here, and this is even more so for distance
941
+ cutoffs. However, in our simulations, we are in the ideal situation
942
+ where we can determine where to apply the cutoff in distance
943
+ and velocity. These experiences can be used to provide guide-
944
+ lines for both types of cutoffs. Figure 5 shows suggestions for
945
+ the choice of distance and velocity cutoff for clusters older than
946
+ 5 Myr. These have been calculated to minimise the sum of the
947
+ false positive rate (FPR) and false negative rate (FNR) for all the
948
+ simulations.
949
+ It does not make much sense to make distance and velocity
950
+ cutoffs in clusters younger than at least 5 Myr to avoid substan-
951
+ tial errors in the classification of the members. However, even
952
+ at 5 Myr, the FPR and FNR introduced by a cutoff can be of the
953
+ order of 15% – 30%. Generally, the percentage of stars identified
954
+ as bound members while being unbound is higher than the oppo-
955
+ site situation. Only for clusters older than 10 Myr, this method is
956
+ relatively robust as the overlap in phase space is of the order of
957
+ 5% – 10%. Figure 3 shows the phase space diagram for a simu-
958
+ lated cluster of 1000 stars with red lines at a distance of 8.09 pc
959
+ and a velocity of 0.78 km/s representing the distance and veloc-
960
+ ity cutoffs shown in Fig. 5. Applying these to the distribution of
961
+ all simulations of 1000 stars leads to a median FNR of 9.7%. The
962
+ 25th and 75th percentile of the distribution of FNR are 7.5% and
963
+ 11.4%, respectively. We represent this as an FNR of 9.7+1.7
964
+ −2.2%.
965
+ Similarly, an FPR of 0 ± 0% is obtained. The percentage of cor-
966
+ rectly identified stars is found to be 94.1 ± 1.1%.
967
+ Combining distance and velocity cutoffs gives the best dis-
968
+ tinction. This can be done by analytically determining the de-
969
+ pendence of the escape velocity of the stars on the distance from
970
+ the cluster’s centre. Although the distribution of the stars in the
971
+ simulations follows a King (1966b) profile, we use an approxi-
972
+ mation of a Plummer (1911) profile to obtain an analytical solu-
973
+ tion. The escape velocity vesc(r) at any point in the cluster is then
974
+ described by
975
+ vesc(r) =
976
+
977
+ 2GMcl
978
+
979
+ a2 + r2 ,
980
+ (3)
981
+ where Mcl is the cluster mass, and a is the initial half-mass
982
+ radius. This analytical cutoff can be seen in Fig. 3 as the blue
983
+ curve. Applying this as the cutoff for bound-unbound star dis-
984
+ tinction leads to an FPR of 0.74+0.91
985
+ −0.47% and an FNR of 4.80+0.88
986
+ −0.10%.
987
+ The median of the distribution of the correctly identified stars’
988
+ percentage is found to be 96.7+0.5
989
+ −0.7%. Hence, this analytical cutoff
990
+ is an improvement over the distance and velocity cutoffs in the
991
+ case of our simulations.
992
+ 4.2. Backtracking
993
+ In the following, we use our simulations of the cluster dynam-
994
+ ics to develop guidelines for backtracking depending on cluster
995
+ type, age, and mass. We subsequently demonstrate that using the
996
+ right subset of stars for backtracking is the key to making the
997
+ most of the available information. Here, we employ the simplest
998
+ form of backtracking, namely, taking present-day positions and
999
+ velocities as constant values and just reversing the arrow of time
1000
+ (i.e. neglecting any source of acceleration acting upon the stars).
1001
+ The high quality of the recent Gaia data allows backtrack-
1002
+ ing from the observed present situation holding the promise to
1003
+ reveal information about a cluster’s past. So far, unbound stars
1004
+ are chiefly analysed as ‘runaway’ (v > 30 km/s) stars and ‘walk-
1005
+ away’ (5 km/s < v < 30 km/s) stars (Eldridge 2011; Schoettler
1006
+ et al. 2020). The idea is that both types of high-velocity stars
1007
+ have been ejected from their star-forming regions, and back-
1008
+ tracking will allow us to determine their origins and characterise
1009
+ their parent star cluster (e.g. Olczak et al. 2008; Farias et al.
1010
+ 2020; Schoettler et al. 2022). Schoettler et al. (2022) search for
1011
+ runaway and walkaway stars within 100 pc of the 3–5 Myr old
1012
+ cluster NGC 2264 using Gaia DR2. They compare the num-
1013
+ ber of the runaway and walkaway stars (17) to a range of N-
1014
+ body simulations with different initial conditions and find con-
1015
+ sistency with initial conditions with a high initial stellar density
1016
+ (≈ 10 000 M⊙ pc−3) and a high initial amount of spatial substruc-
1017
+ ture.
1018
+ However, our simulations find that high-velocity ejec-
1019
+ tions are rare for short-lived clusters. We found no ejections
1020
+ with v > 30 km/s and only a few with v > 5 km/s. Thus, back-
1021
+ tracking based on runaway and walkaway stars suffers from low-
1022
+ number statistics for young clusters (< 20 Myr) typical for the
1023
+ solar neighbourhood. As the ejection happens mainly from the
1024
+ highest-density regions of the cluster, the derived age at gas ex-
1025
+ pulsion is too short, and the cluster size is also too small. For the
1026
+ much denser clusters that turn into long-lived open clusters, the
1027
+ Article number, page 6 of 14
1028
+
1029
+ Arunima Arunima et al.: Unbound stars hold the key to star cluster history
1030
+ n200
1031
+ n1000
1032
+ n4000
1033
+ 0
1034
+ 2
1035
+ 4
1036
+ 6
1037
+ 8
1038
+ 10
1039
+ 12
1040
+ 14
1041
+ 16
1042
+ Distance cutoff [pc]
1043
+ n200
1044
+ n1000
1045
+ n4000
1046
+ 0.00
1047
+ 0.25
1048
+ 0.50
1049
+ 0.75
1050
+ 1.00
1051
+ 1.25
1052
+ 1.50
1053
+ 1.75
1054
+ 2.00
1055
+ Velocity cutoff [km/s]
1056
+ Fig. 5. Distance (top) and velocity (bottom) cutoffs for selection of un-
1057
+ bound members for clusters with different number of members: N =
1058
+ 200, 1000, 4000. The box extends from the lower to upper quartile val-
1059
+ ues of the data, with a line at the median while the whiskers reach 1.5
1060
+ times the interquartile range from the box.
1061
+ backtracking of cluster sizes is of higher quality as the number of
1062
+ ejected stars is higher and the ejection happens over larger areas
1063
+ of the cluster (Pfalzner & Kaczmarek 2013).
1064
+ 4.2.1. Pre-expansion cluster size
1065
+ Using our simulation results as a starting point for backtracking,
1066
+ we find that the restriction to the unbound stars gives the best
1067
+ result in determining the pre-expansion cluster size. This can be
1068
+ seen clearly in Fig. 6 (top panel), where backtracked half-mass
1069
+ radius has been plotted against time. Backtracking the bound
1070
+ members provides no information, whereas using just unbound
1071
+ members fares much better. It recovers the half-mass radius (rhm)
1072
+ of the cluster at the time of gas expulsion with a relative error of
1073
+ 121.4+16.3
1074
+ −15.0% to the relative error of 298.9+48.1
1075
+ −46.7% obtained using
1076
+ bound members.
1077
+ It is equally important to include the unbound stars from a
1078
+ sufficiently large area. Fig. 6 (bottom panel) shows a compari-
1079
+ son of the backtracked half-mass radius determined by consid-
1080
+ ering different areas for the member sampling. The horizontal
1081
+ lines show the derived pre-gas expulsion half-mass radii. It can
1082
+ be seen that the half-mass radius derived from the unbound stars
1083
+ sampled from a relatively small area (10 pc) results in a consider-
1084
+ ably larger error than those derived from including the unbound
1085
+ 0
1086
+ 2
1087
+ 4
1088
+ 6
1089
+ 8
1090
+ 10
1091
+ Time [Myr]
1092
+ 0.0
1093
+ 2.5
1094
+ 5.0
1095
+ 7.5
1096
+ 10.0
1097
+ 12.5
1098
+ 15.0
1099
+ 17.5
1100
+ 20.0
1101
+ Half mass radius [pc]
1102
+ Bound
1103
+ Unbound
1104
+ 1.80 Myr, 1.40 pc
1105
+ 2 Myr, 0.74 pc
1106
+ 0
1107
+ 2
1108
+ 4
1109
+ 6
1110
+ 8
1111
+ 10
1112
+ Time [Myr]
1113
+ 0
1114
+ 2
1115
+ 4
1116
+ 6
1117
+ 8
1118
+ 10
1119
+ 12
1120
+ Half mass radius [pc]
1121
+ 1.72 Myr, 1.19 pc
1122
+ 1.59 Myr, 1.41 pc
1123
+ 1.27 Myr, 1.49 pc
1124
+ 2 Myr, 0.82 pc
1125
+ Fig. 6. Backtracked half-mass radii for a simulation with 1000 stars,
1126
+ Top: using bound (blue) and unbound (red) members only. Red dashed
1127
+ lines show temb and rhm at the time of gas expulsion determined using
1128
+ unbound stars whereas black dashed lines show the actual values of the
1129
+ same. Bottom: using unbound stars within 10 pc (blue), 20 pc (red) and
1130
+ 40 pc (green) from the cluster centre. The actual values of temb and rhm
1131
+ at the time of gas expulsion are shown in cyan.
1132
+ stars from larger areas. In relative error terms, the error decreases
1133
+ from 248.9+41.6
1134
+ −27.4% to 149.1+16.1
1135
+ −14.2% to finally, 121.4+16.3
1136
+ −15.0% as the
1137
+ search area around the cluster centre increases from 10 pc to
1138
+ 20 pc to 40 pc. The actual size of the ideal backtracking area
1139
+ depends, among others, on the cluster’s mass. Details on this de-
1140
+ pendence can be found in Pfalzner et al. (in preparation).
1141
+ Our simulations work with the idealised situation, where the
1142
+ search areas are uncontaminated by the presence of a population
1143
+ of foreground and background stars. In an actual application,
1144
+ extending the field increases the contamination by these fore-
1145
+ ground and background stars. A more significant fraction of con-
1146
+ taminants yields a larger half-mass radius estimate and a shorter
1147
+ age estimate. As the ideal search radius increases as a function
1148
+ of cluster age, so do the errors due to the background population.
1149
+ However, the advent of Gaia again improved the situation; nev-
1150
+ ertheless, it is still a point to consider in real applications. While
1151
+ Rizzuto et al. (2012) found ten years ago that the disc fractions
1152
+ in Upper Sco depend very much on cluster membership proba-
1153
+ bility and distance to the cluster centre, nowadays, a search area
1154
+ Article number, page 7 of 14
1155
+
1156
+ A&A proofs: manuscript no. main_new
1157
+ of > 100 pc is regarded as giving reliable data (Luhman & Esplin
1158
+ 2020).
1159
+ 4.2.2. Time of gas expulsion
1160
+ Backtracking can also be used to obtain information concern-
1161
+ ing the time when gas expulsion happened. Here the same rules
1162
+ apply as for determining the pre-gas expulsion size: restricting
1163
+ to unbound stars and including sufficiently large sampling areas
1164
+ improve the results. In the example shown in Fig. 6, the sim-
1165
+ ulated and the backtracked time of gas expulsion are shown as
1166
+ vertical lines. The backtracking of unbound members determines
1167
+ temb to be 1.8 Myr, which is in excellent agreement with the ac-
1168
+ tual value from the simulations (2 Myr, see Fig. 6 top panel). The
1169
+ relative error in gas expulsion time derived using unbound stars
1170
+ is 40 ± 4% which is much better than that derived using bound
1171
+ stars (826+45
1172
+ −84%). Moreover, including only the unbound particles
1173
+ within 10 pc is not advisable with its relative error of 88+11
1174
+ −32%
1175
+ in the recovery of temb. The error is reduced to 63+11
1176
+ −8 % when
1177
+ the search area increases to 20 pc. Although the results derived
1178
+ by including the unbound particles within 20 pc and 40 pc of
1179
+ the cluster’s centre give nearly identical results for this example
1180
+ cluster (see Fig. 6 bottom panel), the relative error in the derived
1181
+ temb decreases significantly to 40 ± 4% when all the N = 1000
1182
+ simulations are considered for the 40 pc case. The derived gas
1183
+ expulsion times tend to underestimate the time of gas expulsion
1184
+ by a 32+7
1185
+ −6%. Given the general uncertainty of cluster ages, this
1186
+ can be considered a minimal error. Again, it is the stars that un-
1187
+ derwent close encounters that are responsible for the derived too
1188
+ short times.
1189
+ 4.2.3. Further improvements
1190
+ We saw that using the unbound stars from a sufficiently large
1191
+ area gives the best backtracking results for the pre-gas expul-
1192
+ sion half-mass radius. However, the value can still be a factor of
1193
+ two too large. One reason is that even some of the unbound stars
1194
+ have a relatively strong encounter before leaving the cluster (see
1195
+ Fig. 3). However, the main reason is that backtracking the un-
1196
+ bound stars gives the half-mass radius of the unbound, not that
1197
+ of the entire cluster sample. The stars that become unbound are
1198
+ predominantly located at the outskirts of the cluster at the mo-
1199
+ ment of gas expulsion. Therefore, backtracking them, one ob-
1200
+ tains a value that is larger than the complete half-mass radius.
1201
+ The actual pre-gas expulsion half-mass radius includes the un-
1202
+ bound stars. However, simply multiplying the determined value
1203
+ by a factor of 0.5 recovers the half-mass radius in our case quite
1204
+ well. For our simulations, the empirical scaling factor has a value
1205
+ of 0.46+0.06
1206
+ −0.04. There does not seem to be any correlation between
1207
+ the cluster mass and the scaling factor. Although the Spearman
1208
+ correlation coefficient is calculated to be −0.0133, the p-value
1209
+ for the hypothesis test of their correlation is found to be 0.48
1210
+ which is greater than the significance level α = 0.05. Hence, the
1211
+ null hypothesis that the cluster mass and the scaling factor are
1212
+ unrelated can not be rejected. To some degree, the actual cor-
1213
+ rection value might depend on the star formation efficiency in
1214
+ the clusters, however, new sets of simulations with varying star
1215
+ formation efficiencies need to be analysed to establish the depen-
1216
+ dence. The gas dispersion timescale, on the other hand, should
1217
+ not affect the factor.
1218
+ 0.0
1219
+ 0.5
1220
+ 1.0
1221
+ 1.5
1222
+ 2.0
1223
+ 2.5
1224
+ Time [Myr]
1225
+ 0
1226
+ 1
1227
+ 2
1228
+ 3
1229
+ 4
1230
+ 5
1231
+ Half mass radius [pc]
1232
+ 0.2 M : 1.54 Myr, 2.98 pc
1233
+ 0.3 M : 1.54 Myr, 2.78 pc
1234
+ 0.5 M : 1.55 Myr, 2.57 pc
1235
+ all stars: 1.57 Myr, 2.52 pc
1236
+ 2 Myr, 1.32 pc
1237
+ 0
1238
+ 2
1239
+ 4
1240
+ 6
1241
+ 8
1242
+ 10
1243
+ Time [Myr]
1244
+ 0
1245
+ 5
1246
+ 10
1247
+ 15
1248
+ 20
1249
+ 25
1250
+ Half mass radius [pc]
1251
+ 1.51 Myr, 2.54 pc
1252
+ 1.40 Myr, 9.22 pc
1253
+ 1.58 Myr, 3.44 pc
1254
+ 2.68 Myr, 7.67 pc
1255
+ 2 Myr, 1.19 pc
1256
+ Fig. 7. Backtracked half-mass radii for a simulation with 4000 stars,
1257
+ Top: calculated using actual masses (green), 0.2 M⊙ (red), 0.3 M⊙ (blue)
1258
+ and 0.5 M⊙ (yellow). Bottom: calculated using exact velocity values
1259
+ (green), using vz = 0 (red), using velocities values with systematic er-
1260
+ rors as well as different levels of statistical uncertainty (blue: 0.27 km/s
1261
+ & yellow: 1 km/s). The actual values of temb and rhm at the time of gas
1262
+ expulsion from the simulation are shown in cyan.
1263
+ 4.2.4. Mass of stars
1264
+ When we determine bound and unbound stars in a cluster, the
1265
+ mass of the stars plays a role. However, in observations, the stel-
1266
+ lar classification is often known but not the actual mass of the
1267
+ stars. Especially for young clusters, there are large uncertainties
1268
+ between these two properties, and the assumption of different
1269
+ evolutionary models leads to significant differences. Here, we
1270
+ test to what extent this uncertainty in classification as bound or
1271
+ bound due to missing mass information influences backtracking.
1272
+ To mimic this problem, we assign the same mass to all stars,
1273
+ determine the bound and unbound stars and then perform the
1274
+ same backtracking procedure as before. Figure 7 (top) shows the
1275
+ result of backtracking with the fully known IMF (green) and with
1276
+ the assumption that all stars have the same mass (Ms = 0.2 M⊙,
1277
+ 0.3 M⊙ and 0.5 M⊙). It can be seen that not knowing the actual
1278
+ masses of the stars does not influence the derived time of gas
1279
+ expulsion. In all cases, it is too low. The relative error for the
1280
+ derived temb is 46+3
1281
+ −2% for the case of using actual stellar masses
1282
+ Article number, page 8 of 14
1283
+
1284
+ Arunima Arunima et al.: Unbound stars hold the key to star cluster history
1285
+ (green curve). Using the same stellar mass for all stars increases
1286
+ this error only marginally to 52+3
1287
+ −4%, 49+3
1288
+ −2%, and 47+3
1289
+ −2% for the
1290
+ case of Ms = 0.2 M⊙ (red), 0.3 M⊙ (blue), and 0.5 M⊙ (yellow)
1291
+ respectively. The situation is different for the cluster size at the
1292
+ moment of gas expulsion. Here, assuming that all stars have the
1293
+ same mass leads to up to a factor of 1.2 larger sizes than using
1294
+ the actual stellar masses in the case shown in Fig. 7 (top). The
1295
+ smaller the assumed mass, the error is larger. The relative error
1296
+ for the derived rhm is 124.4+8.5
1297
+ −5.3% for the case of using actual
1298
+ stellar masses (green curve). This error increases to 130.6+9.6
1299
+ −7.0%
1300
+ when using stellar mass as 0.5 M⊙ (yellow), to 155.6+11.2
1301
+ −9.4 % for
1302
+ 0.3 M⊙ (blue), and to 180.1+15.0
1303
+ −12.1% for 0.2 M⊙ (red).2 We find that
1304
+ assuming all stars to have a mass of 0.5 M⊙, which corresponds
1305
+ to the mean stellar mass in the cluster, is the best alternative to
1306
+ knowing the actual stellar masses.
1307
+ 4.2.5. Velocity in the z direction
1308
+ We also consider the effects of errors in the vz values on the back-
1309
+ tracking in Fig. 7 (bottom). The velocity component along the z
1310
+ axis, corresponding with close approximation to the radial ve-
1311
+ locity component, constitutes the main source of uncertainty in
1312
+ the total velocity vector (Krolikowski et al. 2021). As a starting
1313
+ point, we consider the effect induced by the existence of non-null
1314
+ proper motion uncertainties; the error on radial velocity is for the
1315
+ moment assumed to be null. Gaia DR2 data have systematic un-
1316
+ certainties in the measurement of parallax and proper motions
1317
+ (Lindegren et al. 2018; Vasiliev 2019). The 2D random error is
1318
+ considered to be of the order of 0.27 km/s, equivalent to the er-
1319
+ ror in 2D proper motion (0.28 mas yr−1) for sources with G = 17
1320
+ mag at a distance of 200 pc in Gaia DR2. Using this error, blue
1321
+ curve is obtained for backtracked radii. The pre-expansion size is
1322
+ derived to be about 1.5 times the size obtained compared to the
1323
+ velocities having no error (green curve in Fig. 7, bottom). The
1324
+ relative error distributions (with respect to the actual rhm) are de-
1325
+ termined for rhm obtained using velocities with no error (green)
1326
+ and using velocities with error (blue). The relative error in rhm
1327
+ goes from 124.4+8.5
1328
+ −5.3% for the green curve to 213.7+12.7
1329
+ −10.5% for the
1330
+ blue curve. An accuracy improvement is seen for the value of
1331
+ the cluster’s age at the time of gas expulsion. The relative er-
1332
+ ror decreases from 46+3
1333
+ −2% for the green curve to 35+4
1334
+ −5% for the
1335
+ blue curve. However, this improvement is less due to recovering
1336
+ more information about the cluster’s past, but more with a gen-
1337
+ eral move of the curve towards the right on the time axis with an
1338
+ increase in the standard deviation in random errors.
1339
+ The impact of radial velocity errors results in an even shorter
1340
+ estimate of the expansion timescale. Krolikowski et al. (2021)
1341
+ point out that the radial velocity (RV) uncertainty is roughly an
1342
+ order of magnitude larger than the reported projected proper mo-
1343
+ tion uncertainty, even when collecting RV measurements from
1344
+ more precise catalogues than Gaia.Ma et al. (2022) also point
1345
+ out that even with future Gaia releases, the precision of RV
1346
+ would be ∼ 1 km/s. The yellow curve in Fig. 7 (bottom) cor-
1347
+ responds to the backtracked radii determined using the same
1348
+ systematic error but a random error of 1 km/s. This increases
1349
+ the relative error in temb and rhm at the time of gas expulsion to
1350
+ 60+8
1351
+ −13.5% and 639.0+35.9
1352
+ −41.1% respectively.
1353
+ Only 0.54% of the sources with astrometric data have the RV
1354
+ measurements available in Gaia DR2. For the extreme situation
1355
+ of zero information on vz, the red curve in Fig. 7 (bottom) is
1356
+ obtained. The relative error for the determined size in this case
1357
+ 2 The distributions of sizes and gas expulsion times derived using dif-
1358
+ ferent masses can be seen in Appendix A.
1359
+ is the highest of all previously discussed cases at 821.6+47.6
1360
+ −55.5%
1361
+ whereas the relative error in derived time of gas expulsion is
1362
+ 40+10
1363
+ −12%3. In reality, for Gaia DR2, the deviation from the actual
1364
+ parameter values will be somewhere between the cases of vz = 0
1365
+ and the added systematic error along with statistical uncertainty.
1366
+ 5. Application to observational data
1367
+ So far, we have dealt exclusively with the idealised situation that
1368
+ simulations provide. In the following, we want to show two ex-
1369
+ amples of applying backtracking procedures to observed clus-
1370
+ ters. The aim is not so much the age and initial size determination
1371
+ of these specific clusters, but to show which additional problems
1372
+ can be expected in real applications. Therefore, we choose two
1373
+ clusters that differ considerably in age and geometry. When re-
1374
+ ferring to the age of the cluster, we quote the time elapsed since
1375
+ the gas started to be expelled and refer to the cluster age as the
1376
+ median age of all the stars in the cluster. This differs from the
1377
+ time elapsed since the molecular cloud started producing stars
1378
+ (Pecaut & Mamajek 2016; Kim et al. 2021; Fujii et al. 2021).
1379
+ 5.1. NGC 6530
1380
+ We first apply the before-described backtracking method to NGC
1381
+ 6530, which is a young cluster within Lagoon Nebula. Its age
1382
+ has been estimated to be 1–2.3 Myr (Prisinzano, L. et al. 2005;
1383
+ Mayne et al. 2007; Bell et al. 2013) and its distance to be 1326+77
1384
+ −69
1385
+ pc (Wright et al. 2019; Damiani et al. 2019). We use the cat-
1386
+ alogue of members provided by Wright et al. (2019), who use
1387
+ GES spectroscopy, Gaia DR2 astrometry, and ancillary member-
1388
+ ship information from X-ray, infrared, and Hα surveys to com-
1389
+ pile the said catalogue. 691 of these cluster members have Gaia
1390
+ DR2 data and have been used in the following analyses. We as-
1391
+ sume that all the stars have a mass of 0.5 M⊙. Using the radial
1392
+ velocity for individual sources when available and assuming it to
1393
+ be equal to the bulk radial velocity of the cluster when not, 3D
1394
+ positions and velocities of the stars are calculated in the stan-
1395
+ dard right-handed Cartesian Galactic frame using the conversion
1396
+ equations prescribed by the Gaia DR2 documentation. These are
1397
+ then used to determine the bound and unbound members of the
1398
+ cluster.
1399
+ For backtracking the stars’ trajectories, we backtrack the po-
1400
+ sitions in the plane of the sky using the velocities along α and
1401
+ δ. Radial velocity is used to backtrack along the line-of-sight
1402
+ and change the distance of the stars which is assumed to be the
1403
+ same for all stars at the present time (1326 pc). Although indi-
1404
+ vidual distances are available for all the stars (Bailer-Jones et al.
1405
+ 2018), the uncertainty is extremely high (fractional uncertainty
1406
+ is 0.20+0.43
1407
+ −0.09 as compared to 0.02 ± 0.01 for the distance data-
1408
+ set of member stars of Upper Sco in Sec.5.2) and leads to very
1409
+ high half-mass radius along with loss of most information about
1410
+ the cluster. The calculated coordinates are then converted to the
1411
+ Cartesian coordinates to calculate the half-mass radii. The result
1412
+ of this procedure is shown in Fig. 8 (left panel). However, for
1413
+ considering the uncertainty in astrometry of the member stars,
1414
+ we run 1000 Monte Carlo simulations, that is to say repeat the
1415
+ entire procedure while varying astrometric information in a ran-
1416
+ dom, normal manner according to the uncertainties associated
1417
+ with each Gaia DR2 source’s parameters. For the distance value
1418
+ for all the stars, the uncertainty is taken as 73 pc (Wright et al.
1419
+ 2019). The results of these simulations are fitted with a Gaussian
1420
+ 3 The distributions of values of size and gas expulsion time obtained
1421
+ for all the cases discussed here can be seen in Appendix B
1422
+ Article number, page 9 of 14
1423
+
1424
+ A&A proofs: manuscript no. main_new
1425
+ 10.0
1426
+ 7.5
1427
+ 5.0
1428
+ 2.5
1429
+ 0.0
1430
+ 2.5
1431
+ 5.0
1432
+ 7.5
1433
+ 10.0
1434
+ Time [Myr]
1435
+ 0
1436
+ 10
1437
+ 20
1438
+ 30
1439
+ 40
1440
+ Half mass radius [pc]
1441
+ 0.04 Myr, 4.01 pc
1442
+ 10.0
1443
+ 7.5
1444
+ 5.0
1445
+ 2.5
1446
+ 0.0
1447
+ 2.5
1448
+ 5.0
1449
+ 7.5
1450
+ 10.0
1451
+ Time [Myr]
1452
+ 10
1453
+ 15
1454
+ 20
1455
+ 25
1456
+ 30
1457
+ Half mass radius [pc]
1458
+ -0.54 Myr, 13.14 pc
1459
+ 10.0
1460
+ 7.5
1461
+ 5.0
1462
+ 2.5
1463
+ 0.0
1464
+ 2.5
1465
+ 5.0
1466
+ 7.5
1467
+ 10.0
1468
+ Time [Myr]
1469
+ 10
1470
+ 12
1471
+ 14
1472
+ 16
1473
+ 18
1474
+ 20
1475
+ 22
1476
+ 24
1477
+ 26
1478
+ Half mass radius [pc]
1479
+ -0.25 Myr, 12.16 pc
1480
+ -1.04 Myr, 10.29 pc
1481
+ Fig. 8. Backtracked (and extrapolated) half-mass radii determined for bound (blue) and unbound (orange) stars 10 Myr into the past and into the
1482
+ future. The green dashed lines show the minima of the backtracked half-mass radius for unbound stars. Left: For NGC 6530 members. Middle: For
1483
+ Upper Sco members. Right: Backtracked (and extrapolated) half-mass radii determined for the unbound members of subclusters of Upper Sco.
1484
+ to obtain the parameters of the cluster along with their errors.
1485
+ Hence, we find the gas expulsion to have happened 0.03 ± 0.03
1486
+ Myr ago and the size of the cluster at the time of gas expulsion
1487
+ is found to be 4.16 ± 0.23 pc. This agrees well with the current
1488
+ age estimate of the cluster. However, the half-mass radius might
1489
+ be underestimated by the assumption of a fixed distance of the
1490
+ stars. A more realistic estimate might be obtained by multiplying
1491
+ it by a factor √3/2, which would yield a limit of 5.09 pc on the
1492
+ cluster size at the time of gas expulsion.
1493
+ Despite obtaining a reasonable fit, the reservations pointed
1494
+ out in Section 4.2.5 also hold here. The median uncertainty in
1495
+ proper motion amount to 2 km/s (Wright et al. 2019). Any un-
1496
+ certainty added to the true velocity acts to reduce the best fit.
1497
+ This uncertainty is the most problematic issue in applying the
1498
+ backtracking method for determining the age of NGC 6530.
1499
+ 5.2. Upper Scorpius
1500
+ Upper Sco is a sub-group of Sco-Cen that has been widely stud-
1501
+ ied with the Gaia data, identifying the cluster’s members (Galli
1502
+ et al. 2018; Wilkinson et al. 2018; Luhman & Esplin 2020;
1503
+ Damiani et al. 2019; Žerjal et al. 2021; Squicciarini et al. 2021;
1504
+ Kerr et al. 2021) and an isochronal age of around 10 Myr has
1505
+ been recently accepted (Feiden 2016; David et al. 2019; Luh-
1506
+ man & Esplin 2020; Sullivan & Kraus 2021). We test the quality
1507
+ of the backtracking for clusters with a more complex morphol-
1508
+ ogy using Upper Sco as an example. We use the list of mem-
1509
+ bers compiled by Luhman & Esplin (2020) using optical and IR
1510
+ spectra to confirm the stars’ youth while parallax and proper mo-
1511
+ tion offsets to get the kinematic criteria for these candidates. The
1512
+ list contains 1761 member candidates, 1682 of which have Gaia
1513
+ DR2 data available and have been used in the following analy-
1514
+ ses. We apply the same method described for NGC 6530 with
1515
+ the exception of considering individual distances for the stars in
1516
+ this case as the uncertainty in distance is much lower.
1517
+ Despite its complex morphology, we first work with the as-
1518
+ sumption that Upper Sco was a centrally condensed spherical
1519
+ structure in the past. In this case, we find that the cluster went
1520
+ through gas expulsion 0.54 Myr ago and had a half-mass radius
1521
+ of 13.14 pc at this time as shown in Fig. 8 (middle panel). How-
1522
+ ever, the Monte Carlo simulations for error propagation estima-
1523
+ tion provide the gas expulsion time to be 0.80 ± 0.21 Myr ago
1524
+ while the cluster size is found to be 13.11 ± 0.11 pc.
1525
+ This value agrees with other backtracking results for Upper
1526
+ Sco. For example, Žerjal et al. (2021) determine the kinematic
1527
+ age of the population in the Upper Sco region as 4 ± 4 Myr,
1528
+ whereas Squicciarini et al. (2021) find 8 subclusters with kine-
1529
+ matic ages varying from 0.0 ± 0.1 Myr to 3.8 ± 0.4 Myr. How-
1530
+ ever, this cluster age deviates considerably from that of 10 Myr
1531
+ obtained by applying corrections, for undetected binaries (Sulli-
1532
+ van & Kraus 2021) or strong magnetic fields impeding convec-
1533
+ tion in low-mass stars (Feiden 2016; David et al. 2019), to the
1534
+ isochronal age determination of Upper Sco. One possible expla-
1535
+ nation for this discrepancy would be that the backtracking yields
1536
+ the time elapsed since gas was expelled and refers to the age of
1537
+ the youngest stars in the association. Taking into account a star
1538
+ formation history lasting 6-7 Myr, most stars might be about 11
1539
+ Myr old and the median age of the association ≈ 7 Myr. These
1540
+ values are more similar to the ones obtained through stellar evo-
1541
+ lution models.
1542
+ Additional complications arise from Upper Sco, unlike NGC
1543
+ 6530, being highly substructured (Kerr et al. 2021; Squicciarini
1544
+ et al. 2021). Likely, star formation did not happen as a single
1545
+ burst, but was rather characterised by several formation episodes
1546
+ (Galli et al. 2018). Thus, the assumption of a centrally condensed
1547
+ spherical structure in the past is oversimplifying the situation.
1548
+ Hence, we try to improve our analysis by considering Upper Sco
1549
+ to consist of subclusters. A density distribution of the cluster
1550
+ members on the plane of the sky at the present time is plotted
1551
+ (see Appendix C for more details and plots). Two dense areas
1552
+ seem to emerge and we consider two rectangles in these areas.
1553
+ The members’ positions are traced back using the same method
1554
+ as described above. When a member star enters one of the said
1555
+ rectangles, it is assigned to the corresponding subcluster. After
1556
+ the assignment of subcluster membership using this simplified
1557
+ method, the backtracked and extrapolated half-mass radii are de-
1558
+ termined using unbound stars for both subclusters. The result is
1559
+ shown in Fig. 8 (right). To determine the errors, the Monte Carlo
1560
+ simulations are used which provide the time of gas expulsion in
1561
+ the two subclusters as −1.09±0.29 Myr and −0.25±0.17 Myr ago
1562
+ respectively. Similarly, the half-mass radii at the time of gas ex-
1563
+ pulsion is found to be 10.15±0.20 pc and 12.10±0.23 pc. Various
1564
+ characterisations of the subclusters are summarised in Table C.1.
1565
+ There is a slight improvement in the determination of the size
1566
+ and time of gas expulsion when considering Upper Sco to have
1567
+ subclusters rather than being one coeval population. However, it
1568
+ must be reiterated that ours is a simplified method. More robust
1569
+ clustering methods can be used in the future to get better results
1570
+ on the subcluster membership and hence, their parameters. For
1571
+ example, Kerr et al. (2021) use HDBSCAN clustering algorithm
1572
+ on Gaia DR2 data and find 9 subclusters in the Upper Sco re-
1573
+ gion. Two of these (Group H and Group I) have more than 100
1574
+ members. We analyse these subclusters and find the time of gas
1575
+ expulsion and their sizes at that time. According to our results,
1576
+ Article number, page 10 of 14
1577
+
1578
+ Arunima Arunima et al.: Unbound stars hold the key to star cluster history
1579
+ gas expulsion in Group H happened 3.40 ± 0.42 Myr ago and its
1580
+ half-mass radius was 3.96 ± 0.215 pc at the time. For Group I,
1581
+ the gas expulsion happened 0.78±0.91 Myr ago and its size was
1582
+ 3.73 ± 0.37 pc. The age found by Kerr et al. (2021), using Gaia
1583
+ DR2’s photometric data, for the groups is 10.2 ± 0.7 Myr and
1584
+ 5.7 ± 0.4 Myr respectively. So, even though there is an improve-
1585
+ ment in the age and size estimates when using a more robust
1586
+ clustering algorithm, the kinematic age estimates still show con-
1587
+ siderable deviation from the photometric estimates. Availability
1588
+ of accurate radial velocities and distances for the member candi-
1589
+ dates to use in the subclustering analysis in future would improve
1590
+ the situation further.
1591
+ 6. Discussion
1592
+ The improvement in the cluster size, when considering sub-
1593
+ clusters, already shows that backtracking is more complex for
1594
+ substructured clusters like Upper Sco. Thus, the less substruc-
1595
+ tured a cluster is, the more straightforward the backtracking. The
1596
+ substructured clusters require backtracking to multiple centres,
1597
+ which is the more complex the more subcluster centres exist.
1598
+ Another potential difficulty could be the presence of multi-
1599
+ ple differently aged populations in the Upper Sco region leading
1600
+ to the miscalculation of the cluster’s age (Wright & Mamajek
1601
+ 2018; Žerjal et al. 2021; Squicciarini et al. 2021). However, this
1602
+ would require large subgroups to be well over 15 Myr to intro-
1603
+ duce such a substantial error. This seems unlikely as an expla-
1604
+ nation. We suspect that the real reason is a different one. The
1605
+ arguments based on kinematic analysis of a cluster for its his-
1606
+ tory can not be considered on their own due to the significant
1607
+ errors in radial velocity and its unavailability for most stars in
1608
+ Gaia. Large uncertainty in the velocities of the stars can lead to
1609
+ a significant loss of information about the past of the cluster (see
1610
+ Fig. 7, bottom panel). This might be the reason for underesti-
1611
+ mating the cluster age and overestimating the size at the time of
1612
+ gas expulsion. Furthermore, the assumptions in the backtrack-
1613
+ ing analysis are numerous. The exact masses of the stars are
1614
+ unknown, so the distinction between bound and unbound stars
1615
+ could be highly inaccurate when combined with astrometric un-
1616
+ certainties and incomplete or inaccurate membership of the clus-
1617
+ ter. In conclusion, the determination of a much younger age, of
1618
+ the Upper Sco region, by kinematic analysis than the more accu-
1619
+ rate isochronal determination could be affected by multiple, dif-
1620
+ ferently aged and kinematically distinct populations; however,
1621
+ precise radial velocity measurements are needed to rule out the
1622
+ possibility that the discrepancy in age determination is due to
1623
+ astrometric errors.
1624
+ 7. Summary and conclusion
1625
+ Young star clusters (< 10 Myr) are highly dynamical entities.
1626
+ Therefore, observations provide only snapshots of this highly
1627
+ dynamic cluster evolution sequence. Nevertheless, in light of the
1628
+ unprecedented precision of Gaia position and velocity data, it
1629
+ should be possible to obtain information about a young cluster’s
1630
+ past using backtracking techniques. In this work, we used simu-
1631
+ lations of the cluster dynamics as an idealised version to suggest
1632
+ how to optimise the backtracking method. Under ideal observa-
1633
+ tional conditions, the following statements should hold:
1634
+ – For backtracking to be successful, it is essential to distin-
1635
+ guish between bound and unbound cluster members. Under
1636
+ ideal conditions, backtracking the unbound members exclu-
1637
+ sively, the time of gas expulsion can be determined with only
1638
+ a 32% error. However, the quality of the backtracking de-
1639
+ pends on the number of cluster stars, with the best results
1640
+ obtained for clusters containing a few thousand stars.
1641
+ – While still the best result, the sizes backtracked from un-
1642
+ bound members are about a factor of two larger than the ac-
1643
+ tual value. However, this error is systematic and reflects that
1644
+ unbound members are primarily located at the cluster out-
1645
+ skirts at the time of gas expulsion. Thus, applying a correc-
1646
+ tion factor of 0.46 approximates the actual value very well.
1647
+ – For obtaining this accuracy, it is essential to determine all the
1648
+ unbound members to > 20 – 40 pc from the cluster centre.
1649
+ – The classification of bound and unbound stars based on the
1650
+ direction of their velocity vectors, or ad hoc distance or ve-
1651
+ locity cutoffs is highly error-prone. We provide analytical
1652
+ cutoffs based on the escape velocity and the number of clus-
1653
+ ter members with a success rate of 96% – 97% for distin-
1654
+ guishing between bound and unbound stars.
1655
+ – Runaway and walkaway stars are less suitable to determine
1656
+ past cluster properties because of their low number and their
1657
+ production by dynamical ejection. Ejection traces only past
1658
+ locations of high stellar density regions but not actual cluster
1659
+ sizes or the time of gas expulsion.
1660
+ Uncertainty in membership and stellar properties provide
1661
+ additional challenges. Modelling these uncertainties, we find
1662
+ that the lack of information about the line-of-sight velocity can
1663
+ severely affect the determination of the pre-expansion size of the
1664
+ cluster. Nevertheless, the time of gas expulsion can still be esti-
1665
+ mated with an error of 40% − 60% due to the unavailability of
1666
+ radial velocities and uncertainty in the value even when avail-
1667
+ able. The uncertainty in the mass of the members seems to af-
1668
+ fect the results much less. Similarly, larger search areas often
1669
+ struggle with higher false-positive and -negative rates in mem-
1670
+ bership. Applying our results to observational data, the method
1671
+ works reasonably for centrally concentrated clusters, but less for
1672
+ very substructured clusters like Upper Sco. For such substruc-
1673
+ tured clusters, backtracking to the individual subcluster centres
1674
+ would be the next step to pursue.
1675
+ In summary, restricting backtracking to the unbound stars al-
1676
+ lows deducing the times of gas expulsion and the pre-expansion
1677
+ cluster size values with relatively high accuracy. Analysing a
1678
+ large number of clusters with the presented method will allow
1679
+ drawing valuable conclusions about the clustered star formation
1680
+ process in the future.
1681
+ Acknowledgements. We thank the referee for a very detailed report that
1682
+ made this article significantly better. This work has made use of data
1683
+ from the European Space Agency (ESA) mission Gaia (https://www.
1684
+ cosmos.esa.int/gaia), processed by the Gaia Data Processing and Anal-
1685
+ ysis Consortium (DPAC, https://www.cosmos.esa.int/web/gaia/dpac/
1686
+ consortium). Funding for the DPAC has been provided by national institutions,
1687
+ in particular the institutions participating in the Gaia Multilateral Agreement.
1688
+ References
1689
+ Aarseth, S. J. 2003, Gravitational N-Body Simulations
1690
+ Adams, F. C. 2000, ApJ, 542, 964
1691
+ Adams, F. C. 2010, ARA&A, 48, 47
1692
+ Ankerst, M., Breunig, M. M., Kriegel, H.-P., & Sander, J. 1999, ACM SIGMOD
1693
+ Record, 28, 49
1694
+ Ansdell, M., Williams, J. P., Manara, C. F., et al. 2017, AJ, 153, 240
1695
+ Bailer-Jones, C. A. L., Rybizki, J., Fouesneau, M., Mantelet, G., & Andrae, R.
1696
+ 2018, AJ, 156, 58
1697
+ Bailey, S. I. 1908, Annals of Harvard College Observatory, 60, 199
1698
+ Banerjee, S. & Kroupa, P. 2017, A&A, 597, A28
1699
+ Bastian, U. 2019, A&A, 630, L8
1700
+ Baumgardt, H. & Kroupa, P. 2007, MNRAS, 380, 1589
1701
+ Article number, page 11 of 14
1702
+
1703
+ A&A proofs: manuscript no. main_new
1704
+ Bell, C. P. M., Naylor, T., Mayne, N. J., Jeffries, R. D., & Littlefair, S. P. 2013,
1705
+ MNRAS, 434, 806
1706
+ Blaauw, A. 1964, ARA&A, 2, 213
1707
+ Blaauw, A. 1978, in Problems of Physics and Evolution of the Universe, ed. L. V.
1708
+ Mirzoyan, 101
1709
+ Brinkmann, N., Banerjee, S., Motwani, B., & Kroupa, P. 2017, A&A, 600, A49
1710
+ Broos, P. S., Getman, K. V., Povich, M. S., et al. 2013, ApJS, 209, 32
1711
+ Campello, R. J., Moulavi, D., & Sander, J. 2013, in Pacific-Asia conference on
1712
+ knowledge discovery and data mining, Springer, 160–172
1713
+ Cantat-Gaudin, T. & Anders, F. 2020, A&A, 633, A99
1714
+ Cantat-Gaudin, T., Jordi, C., Vallenari, A., et al. 2018, A&A, 618, A93
1715
+ Cantat-Gaudin, T., Krone-Martins, A., Sedaghat, N., et al. 2019, A&A, 624,
1716
+ A126
1717
+ Castro-Ginard, A., Jordi, C., Luri, X., et al. 2020, A&A, 635, A45
1718
+ Castro-Ginard, A., Jordi, C., Luri, X., Cantat-Gaudin, T., & Balaguer-Núñez, L.
1719
+ 2019, A&A, 627, A35
1720
+ Castro-Ginard, A., Jordi, C., Luri, X., et al. 2022, A&A, 661, A118
1721
+ Collinder, P. 1931, Annals of the Observatory of Lund, 2, B1
1722
+ Concha-Ramírez, F., Wilhelm, M. J. C., Portegies Zwart, S., & Haworth, T. J.
1723
+ 2019, MNRAS, 490, 5678
1724
+ Damiani, F., Prisinzano, L., Micela, G., et al. 2014, A&A, 566, A50
1725
+ Damiani, F., Prisinzano, L., Micela, G., & Sciortino, S. 2019, A&A, 623, 25
1726
+ Damiani, F., Prisinzano, L., Pillitteri, I., Micela, G., & Sciortino, S. 2019, A&A,
1727
+ 623, A112
1728
+ David, T. J., Hillenbrand, L. A., Gillen, E., et al. 2019, ApJ, 872, 161
1729
+ Dib, S., Schmeja, S., & Parker, R. J. 2018, MNRAS, 473, 849
1730
+ Dreyer, J. L. E. 1888, MmRAS, 49, 1
1731
+ Duchêne, G., Lacour, S., Moraux, E., Goodwin, S., & Bouvier, J. 2018, MNRAS,
1732
+ 478, 1825
1733
+ Eldridge, J. J. 2011, in Astronomical Society of the Pacific Conference Series,
1734
+ Vol. 440, UP2010: Have Observations Revealed a Variable Upper End of the
1735
+ Initial Mass Function?, ed. M. Treyer, T. Wyder, J. Neill, M. Seibert, & J. Lee,
1736
+ 217
1737
+ Elmegreen, B. G. 2000, ApJ, 530, 277
1738
+ Esplin, T. L. & Luhman, K. L. 2019, AJ, 158, 54
1739
+ Ester, M., Kriegel, H.-P., Sander, J., & Xu, X. 1996, in (AAAI Press), 226–231
1740
+ Farias, J. P., Tan, J. C., & Eyer, L. 2020, ApJ, 900, 14
1741
+ Feiden, G. A. 2016, A&A, 593, A99
1742
+ Feigelson, E. D., Townsley, L. K., Broos, P. S., et al. 2013, ApJS, 209, 26
1743
+ Fernández, D., Figueras, F., & Torra, J. 2008, A&A, 480, 735
1744
+ Fujii, M. S., Saitoh, T. R., Hirai, Y., & Wang, L. 2021, PASJ, 73, 1074
1745
+ Fujii, M. S., Saitoh, T. R., & Portegies Zwart, S. F. 2012, ApJ, 753, 85
1746
+ Gagné, J., Mamajek, E. E., Malo, L., et al. 2018, ApJ, 856, 23
1747
+ Gaia Collaboration, Brown, A. G. A., Vallenari, A., et al. 2021, A&A, 649, A1
1748
+ Gaia Collaboration, Katz, D., Antoja, T., et al. 2018, A&A, 616, A11
1749
+ Gaia Collaboration, Prusti, T., de Bruijne, J. H. J., et al. 2016, A&A, 595, A1
1750
+ Galli, P. A. B., Joncour, I., & Moraux, E. 2018, MNRAS, 477, L50
1751
+ Garro, E. R., Minniti, D., Gómez, M., et al. 2021, A&A, 649, A86
1752
+ Getman, K. V., Broos, P. S., Kuhn, M. A., et al. 2017, ApJS, 229, 28
1753
+ Geyer, M. P. & Burkert, A. 2001, MNRAS, 323, 988
1754
+ Guszejnov,
1755
+ D.,
1756
+ Raju,
1757
+ A.
1758
+ N.,
1759
+ Offner,
1760
+ S.
1761
+ S.
1762
+ R.,
1763
+ et
1764
+ al.
1765
+ 2022,
1766
+ MN-
1767
+ RAS[arXiv:2208.02844]
1768
+ Haisch, Karl E., J., Lada, E. A., & Lada, C. J. 2001, ApJ, 553, L153
1769
+ He, Z.-H., Xu, Y., Hao, C.-J., Wu, Z.-Y., & Li, J.-J. 2021, RAA, 21, 093
1770
+ Heggie, D. C. 1975, MNRAS, 173, 729
1771
+ Heyl, J., Caiazzo, I., Richer, H., & Miller, D. R. 2021, arXiv e-prints,
1772
+ arXiv:2110.04296
1773
+ Heyl, J., Caiazzo, I., & Richer, H. B. 2022, ApJ, 926, 132
1774
+ Hillenbrand, L. A. & Hartmann, L. W. 1998, ApJ, 492, 540
1775
+ Hogg, D. W., Bovy, J., & Lang, D. 2010, arXiv e-prints, arXiv:1008.4686
1776
+ Hunt, E. L. & Reffert, S. 2021, A&A, 646, A104
1777
+ Kaczmarek, T., Olczak, C., & Pfalzner, S. 2011, A&A, 528, A144
1778
+ Kerr, R. M. P., Rizzuto, A. C., Kraus, A. L., & Offner, S. S. R. 2021, ApJ, 917,
1779
+ 23
1780
+ Kim, J., Chevance, M., Kruijssen, J. M. D., et al. 2021, MNRAS, 504, 487
1781
+ King, I. R. 1966a, AJ, 71, 64
1782
+ King, I. R. 1966b, AJ, 71, 64
1783
+ Krolikowski, D. M., Kraus, A. L., & Rizzuto, A. C. 2021, AJ, 162, 110
1784
+ Krone-Martins, A. & Moitinho, A. 2014, A&A, 561, A57
1785
+ Kroupa, P. 2002, Science, 295, 82
1786
+ Kroupa, P., Aarseth, S., & Hurley, J. 2001, MNRAS, 321, 699
1787
+ Krumholz, M. R., McKee, C. F., & Bland-Hawthorn, J. 2019, ARA&A, 57, 227
1788
+ Kuhn, M. A., Hillenbrand, L. A., Carpenter, J. M., & Avelar Menendez, A. R.
1789
+ 2020, ApJ, 899, 128
1790
+ Kuhn, M. A., Hillenbrand, L. A., Sills, A., Feigelson, E. D., & Getman, K. V.
1791
+ 2019, ApJ, 870, 32
1792
+ Lada, C. J. & Lada, E. A. 2003, ARA&A, 41, 57
1793
+ Lada, C. J., Margulis, M., & Dearborn, D. 1984, ApJ, 285, 141
1794
+ Lindegren, L., Hernández, J., Bombrun, A., et al. 2018, A&A, 616, A2
1795
+ Liu, L. & Pang, X. 2019, ApJS, 245, 32
1796
+ Luhman, K. L. 2018, AJ, 156, 271
1797
+ Luhman, K. L. & Esplin, T. L. 2020, AJ, 160, 44
1798
+ Ma, Y., De Rosa, R. J., & Kalas, P. 2022, AJ, 163, 219
1799
+ MacQueen, J. 1967, in 5th Berkeley Symp. Math. Statist. Probability, 281–297
1800
+ Marks, M., Leigh, N., Giersz, M., et al. 2014, MNRAS, 441, 3503
1801
+ Mathieu, R. D. 1983, ApJ, 267, L97
1802
+ Mayne, N. J., Naylor, T., Littlefair, S. P., Saunders, E. S., & Jeffries, R. D. 2007,
1803
+ MNRAS, 375, 1220
1804
+ Michel, A., van der Marel, N., & Matthews, B. C. 2021, ApJ, 921, 72
1805
+ Nürnberger, D. & Petr-Gotzens, M. 2002, A&A, 382, 537
1806
+ Olczak, C., Pfalzner, S., & Eckart, A. 2008, A&A, 488, 191
1807
+ Olczak, C., Pfalzner, S., & Spurzem, R. 2006, ApJ, 642, 1140
1808
+ Pang, X., Li, Y., Tang, S.-Y., Pasquato, M., & Kouwenhoven, M. B. N. 2020,
1809
+ ApJ, 900, L4
1810
+ Pecaut, M. J. & Mamajek, E. E. 2016, MNRAS, 461, 794
1811
+ Pelupessy, F. I. & Portegies Zwart, S. 2012, MNRAS, 420, 1503
1812
+ Perryman, M. A. C., Lindegren, L., Kovalevsky, J., et al. 1997, A&A, 323, L49
1813
+ Pfalzner, S. & Govind, A. 2021, ApJ, 921, 90
1814
+ Pfalzner, S. & Kaczmarek, T. 2013, A&A, 559, A38
1815
+ Pfalzner, S., Kirk, H., Sills, A., et al. 2016, A&A, 586, A68
1816
+ Plummer, H. C. 1911, MNRAS, 71, 460
1817
+ Porras, A., Christopher, M., Allen, L., et al. 2003, AJ, 126, 1916
1818
+ Portegies Zwart, S. F., McMillan, S. L. W., & Gieles, M. 2010, ARA&A, 48, 431
1819
+ Prisinzano, L., Damiani, F., Micela, G., & Sciortino, S. 2005, A&A, 430, 941
1820
+ Ribas, Á., Merín, B., Bouy, H., & Maud, L. T. 2014, A&A, 561, A54
1821
+ Richert, A. J. W., Getman, K. V., Feigelson, E. D., et al. 2018, MNRAS, 477,
1822
+ 5191
1823
+ Rizzuto, A. C., Ireland, M. J., & Zucker, D. B. 2012, MNRAS, 421, L97
1824
+ Schoettler, C., de Bruijne, J., Vaher, E., & Parker, R. J. 2020, MNRAS, 495,
1825
+ 3104
1826
+ Schoettler, C., Parker, R. J., & de Bruijne, J. 2022, MNRAS, 510, 3178
1827
+ Scholz, A., Geers, V., Jayawardhana, R., et al. 2009, ApJ, 702, 805
1828
+ Skrutskie, M. F., Cutri, R. M., Stiening, R., et al. 2006, AJ, 131, 1163
1829
+ Soderblom, D. R. 2010, ARA&A, 48, 581
1830
+ Sollima, A., Baumgardt, H., & Hilker, M. 2019, MNRAS, 485, 1460
1831
+ Squicciarini, V., Gratton, R., Bonavita, M., & Mesa, D. 2021, MNRAS, 507,
1832
+ 1381
1833
+ Sullivan, K. & Kraus, A. L. 2021, ApJ, 912, 137
1834
+ Tang, S.-Y., Pang, X., Yuan, Z., et al. 2019, ApJ, 877, 12
1835
+ Trumpler, R. J. 1930, Lick Observatory Bulletin, 420, 154
1836
+ Vasiliev, E. 2019, MNRAS, 489, 623
1837
+ Vincke, K., Breslau, A., & Pfalzner, S. 2015, A&A, 577, A115
1838
+ Vitral, E. 2021, MNRAS, 504, 1355
1839
+ Ward, R. L., Wadsley, J., Sills, A., & Petitclerc, N. 2012, ApJ, 756, 119
1840
+ Wilking, B. A., Meyer, M. R., Robinson, J. G., & Greene, T. P. 2005, AJ, 130,
1841
+ 1733
1842
+ Wilkinson, S., Merín, B., & Riviere-Marichalar, P. 2018, A&A, 618, A12
1843
+ Winter, A. J., Booth, R. A., & Clarke, C. J. 2018, MNRAS, 479, 5522
1844
+ Wright, N. J., Jeffries, R. D., Jackson, R. J., et al. 2019, MNRAS, 486, 2477
1845
+ Wright, N. J. & Mamajek, E. E. 2018, MNRAS, 476, 381
1846
+ Zari, E., Brown, A. G. A., & de Zeeuw, P. T. 2019, A&A, 628, A123
1847
+ Zhang, Y., Tang, S.-Y., Chen, W. P., Pang, X., & Liu, J. Z. 2020, ApJ, 889, 99
1848
+ Žerjal, M., Ireland, M. J., Crundall, T. D., Krumholz, M. R., & Rains, A. D. 2021,
1849
+ in Star Clusters: the Gaia Revolution; Online Workshop, 29
1850
+ Article number, page 12 of 14
1851
+
1852
+ Arunima Arunima et al.: Unbound stars hold the key to star cluster history
1853
+ Appendix A: Mass of stars
1854
+ We discussed how the unavailability of the mass of stars in ob-
1855
+ servations affects the determination of gas expulsion time and
1856
+ cluster size at the time of gas expulsion using backtracking anal-
1857
+ ysis. Here, we provide the distributions of the derived sizes and
1858
+ gas expulsion time (Fig. A.1) for all the cases discussed in Sec.
1859
+ 4.2.4.
1860
+ 0.5
1861
+ 1.0
1862
+ 1.5
1863
+ 2.0
1864
+ 2.5
1865
+ 3.0
1866
+ 3.5
1867
+ 4.0
1868
+ Half-mass radius [pc]
1869
+ 0
1870
+ 1
1871
+ 2
1872
+ 3
1873
+ 4
1874
+ = 1.14, = 0.09
1875
+ = 2.56, = 0.10
1876
+ = 2.92, = 0.16
1877
+ = 2.64, = 0.13
1878
+ = 3.21, = 0.18
1879
+ 1.3
1880
+ 1.4
1881
+ 1.5
1882
+ 1.6
1883
+ 1.7
1884
+ 1.8
1885
+ 1.9
1886
+ 2.0
1887
+ Time [Myr]
1888
+ 0
1889
+ 2
1890
+ 4
1891
+ 6
1892
+ 8
1893
+ 10
1894
+ 12
1895
+ green: = 1.54, = 0.04
1896
+ blue: = 1.51, = 0.04
1897
+ red: = 1.48, = 0.05
1898
+ yellow: = 1.53, = 0.04
1899
+ actual value of gas expulsion time
1900
+ Fig. A.1. Distributions of the backtracked half-mass radii (top) and the
1901
+ time of gas expulsion (bottom) obtained using actual masses (green),
1902
+ 0.2 M⊙ (red), 0.3 M⊙ (blue) and 0.5 M⊙ (yellow). The actual values of
1903
+ rhm at the time of gas expulsion (as a distribution) and temb from all the
1904
+ simulations (of N=4000 clusters) are shown in cyan.
1905
+ Appendix B: Velocity in the z direction
1906
+ Similarly, we provide the distributions of the derived sizes and
1907
+ gas expulsion time for all the cases in Sec. 4.2.5 to supplement
1908
+ the discussion of the effects of errors in the vz values on the back-
1909
+ tracking analysis and derived parameters (Fig. B.1).
1910
+ Appendix C: Upper Sco subclusters
1911
+ The density distribution of the Upper Sco members is shown in
1912
+ Fig. C.1 (top) along with the rectangles showing the subclus-
1913
+ ter areas used for the subcluster membership assignment. Figure
1914
+ C.1 (bottom) shows the scatter plot of the member stars with the
1915
+ 0
1916
+ 2
1917
+ 4
1918
+ 6
1919
+ 8
1920
+ 10
1921
+ 12
1922
+ Half-mass radius [pc]
1923
+ 0
1924
+ 1
1925
+ 2
1926
+ 3
1927
+ 4
1928
+ cyan: = 1.14, = 0.09
1929
+ green: = 2.56, = 0.10
1930
+ blue: = 3.57, = 0.10
1931
+ yellow: = 8.37, = 0.17
1932
+ red: = 10.44, = 0.42
1933
+ 1.00
1934
+ 1.25
1935
+ 1.50
1936
+ 1.75
1937
+ 2.00
1938
+ 2.25
1939
+ 2.50
1940
+ 2.75
1941
+ 3.00
1942
+ Time [Myr]
1943
+ 0
1944
+ 2
1945
+ 4
1946
+ 6
1947
+ 8
1948
+ 10
1949
+ 12
1950
+ green: = 1.54, = 0.04
1951
+ blue: = 1.66, = 0.06
1952
+ red: = 1.62, = 0.18
1953
+ yellow: = 2.58, = 0.17
1954
+ actual value of gas expulsion time
1955
+ Fig. B.1. Distributions of the backtracked half-mass radii (top) and the
1956
+ time of gas expulsion (bottom) obtained using exact velocity values
1957
+ (green), using vz = 0 (red), using velocities values with systematic er-
1958
+ rors as well as different levels of statistical uncertainty (blue: 0.27 km/s
1959
+ & yellow: 1 km/s). The actual values of rhm at the time of gas expulsion
1960
+ (as a distribution) and temb from all the simulations (of N=4000 clusters)
1961
+ are shown in cyan.
1962
+ same rectangles and the members of the two subclusters in red
1963
+ and green. The purple points represent the few members which
1964
+ did not enter any of the rectangles in the 10 Myr up to which the
1965
+ positions were backtracked and hence, are not assigned to any
1966
+ subcluster. Furthermore, Table C.1 provides characteristic infor-
1967
+ mation about the subclusters identified in this work as well as
1968
+ about Group H and I from Kerr et al. (2021).
1969
+ Article number, page 13 of 14
1970
+
1971
+ A&A proofs: manuscript no. main_new
1972
+ Fig. C.1. Density distribution (top) and scatter plot (bottom) of the Up-
1973
+ per Sco members at the present time. The two rectangles show the area
1974
+ selected for the clustering process. Green and red points in the bottom
1975
+ plot show the members of Group 1 and Group 2, respectively. Purple
1976
+ points are the ones which were not assigned to any group.
1977
+ Table C.1. Information about the subclusters identified in this work (ID:
1978
+ 1,2) and the groups from Kerr et al. (2021) (ID: H, I).
1979
+ ID
1980
+ N
1981
+ RA
1982
+ Dec
1983
+ tK
1984
+ rhm
1985
+ [deg]
1986
+ [deg]
1987
+ [Myr]
1988
+ [pc]
1989
+ 1
1990
+ 1102
1991
+ 241.60
1992
+ -21.93
1993
+ −1.09 ± 0.29
1994
+ 10.15 ± 0.20
1995
+ 2
1996
+ 454
1997
+ 245.68
1998
+ -25.12
1999
+ −0.25 ± 0.17
2000
+ 12.10 ± 0.23
2001
+ H
2002
+ 102
2003
+ 240.6
2004
+ -22.4
2005
+ −3.40 ± 0.42
2006
+ 3.96 ± 0.21
2007
+ I
2008
+ 110
2009
+ 246.4
2010
+ -23.9
2011
+ −0.78 ± 0.91
2012
+ 3.73 ± 0.37
2013
+ Notes. Number of stars (N) and mean positions (RA, Dec) are provided
2014
+ along with the time of gas expulsion (tK, kinematic age) and half-mass
2015
+ radius of subcluster at the time of gas expulsion (rhm).
2016
+ Article number, page 14 of 14
2017
+
2018
+ 20.0
2019
+ -18
2020
+ 17.5
2021
+ -20
2022
+ 15.0
2023
+ Number
2024
+ -22
2025
+ 12.5
2026
+ (。)9
2027
+ -24
2028
+ 10.0
2029
+ of
2030
+ sources
2031
+ -26
2032
+ 7.5
2033
+ -28
2034
+ 5.0
2035
+ -30
2036
+ 2.5
2037
+ 0.0
2038
+ 235
2039
+ 240
2040
+ 245
2041
+ 250
2042
+ α(°)-16
2043
+ -18
2044
+ -20
2045
+ -22
2046
+
2047
+ -24
2048
+ -26
2049
+ -28
2050
+ -30
2051
+ -32
2052
+ 232.5 235.0 237.5 240.0 242.5 245.0 247.5 250.0 252.5
2053
+ α(°
8tE1T4oBgHgl3EQfngT9/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
BNE3T4oBgHgl3EQfswvl/vector_store/index.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:278cc7c7e04552cdb5625e3a1611804cbbf6297f449b5e7fbe62c473ac0620c9
3
+ size 193224
D9AzT4oBgHgl3EQfwv5m/content/tmp_files/2301.01727v1.pdf.txt ADDED
@@ -0,0 +1,1790 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ arXiv:2301.01727v1 [nlin.SI] 4 Jan 2023
2
+ Classical Solutions of the Degenerate
3
+ Fifth Painlev´e Equation
4
+ Peter A. Clarkson
5
+ School of Mathematics, Statistics and Actuarial Science,
6
+ University of Kent, Canterbury, CT2 7FS, UK
7
8
+ January 5, 2023
9
+ Abstract
10
+ In this paper classical solutions of the degenerate fifth Painlev´e equation are classified, which
11
+ include hierarchies of algebraic solutions and solutions expressible in terms of Bessel functions. Solu-
12
+ tions of the degenerate fifth Painlev´e equation are known to expressible in terms of the third Painlev´e
13
+ equation. Two applications of these classical solutions are discussed, deriving exact solutions of the
14
+ complex sine-Gordon equation and of the coefficients in the three-term recurrence relation associated
15
+ with generalised Charlier polynomials.
16
+ 1
17
+ Introduction
18
+ In this paper we are concerned with solutions of the equation
19
+ d2w
20
+ dz2 =
21
+ � 1
22
+ 2w +
23
+ 1
24
+ w − 1
25
+ � �dw
26
+ dz
27
+ �2
28
+ − 1
29
+ z
30
+ dw
31
+ dz + (w − 1)2(αw2 + β)
32
+ z2w
33
+ + γw
34
+ z ,
35
+ (1.1)
36
+ with α, β and γ constants. Equation (1.1) is the special case of the fifth Painlev´e equation (PV)
37
+ d2w
38
+ dz2 =
39
+ � 1
40
+ 2w +
41
+ 1
42
+ w − 1
43
+ � �dw
44
+ dz
45
+ �2
46
+ − 1
47
+ z
48
+ dw
49
+ dz + (w − 1)2(αw2 + β)
50
+ z2w
51
+ + γw
52
+ z + δw(w + 1)
53
+ w − 1
54
+ .
55
+ (1.2)
56
+ with α, β, γ and δ constants, when δ = 0 and is known as the degenerate fifth Painlev´e equation (deg-
57
+ PV), cf. [42].
58
+ The six Painlev´e equations (PI–PVI), were discovered by Painlev´e, Gambier and their colleagues
59
+ whilst studying second order ordinary differential equations of the form
60
+ d2w
61
+ dz2 = F
62
+
63
+ z, w, dw
64
+ dz
65
+
66
+ ,
67
+ (1.3)
68
+ where F is rational in dw/dz and w and analytic in z. The Painlev´e equations can be thought of as
69
+ nonlinear analogues of the classical special functions. The general solutions of the Painlev´e equations
70
+ are transcendental in the sense that they cannot be expressed in terms of known elementary functions
71
+ and so require the introduction of a new transcendental function to describe their solution. However,
72
+ it is well known that PII–PVI possess rational solutions, algebraic solutions and solutions expressed in
73
+ terms of the classical special functions — Airy, Bessel, parabolic cylinder, Kummer and hypergeometric
74
+ functions, respectively — for special values of the parameters, see, e.g. [11, 22] and the references
75
+ therein. These hierarchies are usually generated from “seed solutions” using the associated B¨acklund
76
+ transformations and frequently can be expressed in the form of determinants. These solutions of the
77
+ Painlev´e equations are often called “classical solutions”, cf. [53, 54].
78
+ It is well known that solutions of deg-PV (1.1) are related to solutions of the third Painlev´e equation
79
+ d2q
80
+ dx2 = 1
81
+ q
82
+ � dq
83
+ dx
84
+ �2
85
+ − 1
86
+ x
87
+ dq
88
+ dx + Aq2 + B
89
+ x
90
+ + Cq3 + D
91
+ q ,
92
+ (1.4)
93
+ 1
94
+
95
+ with A, B, C and D constants, a result originally due to Gromak [21]; see also [22, §34]. The purpose
96
+ of this paper is to give a classification and description of the classical solutions of deg-PV (1.1) directly,
97
+ rather than indirectly through (1.4).
98
+ In §2, the relationship between deg-PV (1.1) and the third Painlev´e equation (1.4) is discussed. In
99
+ §3, classical solutions of the third Painlev´e equation (1.4) are reviewed, the rational solutions in §3.1
100
+ and the Bessel function solutions in §3.2. In §4, B¨acklund transformations of deg-PV (1.1) are given,
101
+ which can be used to derive a hierarchy of solutions from a “seed solution”. In §5, classical solutions
102
+ of deg-PV (1.1) are classified, the algebraic solutions in §5.1 and the Bessel function solutions in §5.2.
103
+ In §6, two applications of classical solutions of deg-PV (1.1) are given to derive exact solutions of the
104
+ complex sine-Gordon equation, which is equivalent to the Pohlmeyer-Lund-Regge model, and to derive
105
+ explicit representations of the coefficients in the three-term recurrence relation satisfied by generalised
106
+ Charlier polynomials, which are discrete orthogonal polynonials.
107
+ 2
108
+ The relationship between deg-PV and PIII
109
+ In the generic case when CD ̸= 0 in the third Painlev´e equation (1.4), we set C = 1 and D = −1,
110
+ without loss of generality (by rescaling the variables if necessary), and so consider the equation
111
+ d2q
112
+ dx2 = 1
113
+ q
114
+ � dq
115
+ dx
116
+ �2
117
+ − 1
118
+ x
119
+ dq
120
+ dx + Aq2 + B
121
+ x
122
+ + q3 − 1
123
+ q .
124
+ (2.1)
125
+ In the sequel, we shall refer to this equation as PIII since it is the generic case.
126
+ Consider the Hamiltonian associated with PIII (2.1) given by
127
+ HIII(q, p, x; a, b, ε) = q2p2 − xq2p − (2a + 2b + 1)qp + εxp + 2bxq,
128
+ (2.2)
129
+ with a and b parameters and ε = ±1, see [28, 46]. Then p(x) and q(x) satisfy the Hamiltonian system
130
+ x dq
131
+ dx = ∂HIII
132
+ ∂p
133
+ = 2q2p − xq2 − (2a + 2b + 1)q + εx,
134
+ (2.3a)
135
+ x dp
136
+ dx = −∂HIII
137
+ ∂q
138
+ = −2qp2 + 2xqp + (2a + 2b + 1)p − 2bx.
139
+ (2.3b)
140
+ Solving (2.3a) for p(x) gives
141
+ p(x) = 1
142
+ 2q
143
+
144
+ x dq
145
+ dx + xq2 + (2a + 2b + 1)q − εx
146
+
147
+ ,
148
+ and then substituting this in (2.3b) gives
149
+ d2q
150
+ dx2 = 1
151
+ q
152
+ � dq
153
+ dx
154
+ �2
155
+ − 1
156
+ x
157
+ dq
158
+ dx + 2(a ��� b)q2
159
+ x
160
+ + 2ε(a + b + 1)
161
+ x
162
+ + q3 − 1
163
+ q .
164
+ (2.4)
165
+ which is PIII (2.1), with parameters
166
+ A = 2(a − b),
167
+ B = 2ε(a + b + 1).
168
+ (2.5)
169
+ Solving (2.3a) for q(x) gives
170
+ q(x) =
171
+ 1
172
+ 2p(x − p)
173
+
174
+ x dp
175
+ dx − (2a + 2b + 1) + 2bx
176
+
177
+ ,
178
+ and then substituting this in (2.3a) gives
179
+ d2p
180
+ dx2 = 1
181
+ 2
182
+ �1
183
+ p +
184
+ 1
185
+ p − x
186
+ � � dp
187
+ dx
188
+ �2
189
+
190
+ p
191
+ x(p − x)
192
+ dp
193
+ dx + 2εp − 2b2
194
+ p − 4a2 − 1
195
+ 2(p − x) + 1 − 4(a2 − b2) − 4εp2
196
+ 2x
197
+ .
198
+ (2.6)
199
+ Then making the transformation
200
+ p(x) = 2√z w(z)
201
+ w(z) − 1 ,
202
+ x = 2√z,
203
+ (2.7)
204
+ 2
205
+
206
+ in (2.6) gives
207
+ d2w
208
+ dz2 =
209
+ � 1
210
+ 2w +
211
+ 1
212
+ w − 1
213
+ � �dw
214
+ dz
215
+ �2
216
+ − 1
217
+ z
218
+ dw
219
+ dz + (w − 1)2(a2w2 − b2)
220
+ 2z2w
221
+ + εw
222
+ z ,
223
+ (2.8)
224
+ which is deg-PV (1.1) with parameters
225
+ α = 1
226
+ 2a2,
227
+ β = − 1
228
+ 2b2,
229
+ γ = ε.
230
+ (2.9)
231
+ Hence we have the following result; see also [22, Theorem 34.2].
232
+ Lemma 2.1. If q(x) is a solution of (2.4) then
233
+ w(z) = xq′(x) + xq2(x) + (2a + 2b + 1)q(x) − εx
234
+ xq′(x) − xq2(x) + (2a + 2b + 1)q(x) − εx,
235
+ z = 1
236
+ 2x2,
237
+ (2.10)
238
+ with ′ ≡ d/dx is a solution of (2.8), provided that
239
+ x dq
240
+ dx − xq2 + (2a + 2b + 1)q − εx ̸= 0.
241
+ Conversely, if w(z) is a solution of (2.8), then
242
+ q(x) =
243
+ 1
244
+ 2√z w
245
+
246
+ z dw
247
+ dz + (w − 1)(aw + b)
248
+
249
+ ,
250
+ x =
251
+
252
+ 2z,
253
+ (2.11)
254
+ is a solution of (2.4).
255
+ Proof. Solving (2.3a) for p(x), substituting in (2.7) and solving for w(z) gives (2.10). Also solving (2.3b)
256
+ for q(x) and substituting (2.7) gives (2.11).
257
+ An alternative method of deriving solutions of (2.8) involves the second-order, second-degree equa-
258
+ tion satisfied associated with the Hamiltonian (2.2), due to Jimbo and Miwa [28] and Okamoto [46],
259
+ which is often called the “σ-equation”.
260
+ Theorem 2.2. If HIII(q, p, x; a, b, ε) is given by (2.2), then
261
+ σ(x; a, b, ε) = HIII(q, p, x; a, b, ε) + qp − 1
262
+ 2εx2 + (a + b)2,
263
+ (2.12)
264
+ where q(x) and p(x) satisfy the system (2.3), satisfies the second-order, second-degree equation (SIII)
265
+
266
+ xd2σ
267
+ dx2 − dσ
268
+ dx
269
+ �2
270
+ + 2
271
+ ��dσ
272
+ dx
273
+ �2
274
+ − x2
275
+ � �
276
+ xdσ
277
+ dx − 2σ
278
+
279
+ − 8ε(a2 − b2)xdσ
280
+ dx = 8(a2 + b2)x2.
281
+ (2.13)
282
+ Conversely, if σ(x; a, b, ε) satisfies (2.13) then the solution of the Hamiltonian system (2.3) is given by
283
+ q(x) = εxσ′′(x) − ε(2a + 2b + 1)σ′(x) − 2(a − b)x
284
+ x2 − [σ′(x)]2
285
+ ,
286
+ p(x) = 1
287
+ 2εσ′(x) + 1
288
+ 2x.
289
+ (2.14)
290
+ Proof. See Jimbo and Miwa [28] and Okamoto [46].
291
+ Consequently solutions of (2.8) can be expressed in terms of solutions of SIII (2.13).
292
+ Corollary 2.3. If σ(x; a, b, ε) is a solution of SIII (2.13), then
293
+ w(z; a, b, ε) = σ′(x; a, b, ε) + εx
294
+ σ′(x; a, b, ε) − εx,
295
+ z = 1
296
+ 2x2,
297
+ (2.15)
298
+ is a solution of (2.8).
299
+ Proof. This immediately follows from (2.7) and (2.14).
300
+ 3
301
+
302
+ 3
303
+ Classical solutions of PIII and SIII
304
+ 3.1
305
+ Rational solutions of PIII and SIII
306
+ Rational solutions of PIII (2.1) are classified in the following theorem.
307
+ Theorem 3.1. Equation (2.1) has a rational solution if and only if
308
+ ε1A + ε2B = 4n,
309
+ with n ∈ Z and ε2
310
+ 1 = 1, ε2
311
+ 2 = 1, independently.
312
+ Proof. For details see Lukashevich [32]; see also [39, 40].
313
+ Umemura [55]1 derived special polynomials associated with rational solutions of (2.1), which we
314
+ now define; see also [9, 29, 30].
315
+ Definition 3.2. The Umemura polynomial Sn(x; µ) is given by the recursion relation
316
+ Sn+1Sn−1 = −x
317
+
318
+ Sn
319
+ d2Sn
320
+ dx2 −
321
+ �dSn
322
+ dx
323
+ �2�
324
+ − Sn
325
+ dSn
326
+ dx + (x + µ)S2
327
+ n,
328
+ (3.1)
329
+ where S−1(x; µ) = S0(x; µ) = 1, with µ an arbitrary parameter.
330
+ Remark 3.3. The Umemura polynomial Sn(x; µ) has the Wronskian representation
331
+ Sn(x; µ) = cnW (ϕ1, ϕ3, . . . , ϕ2n−1) ,
332
+ cn =
333
+ n
334
+
335
+ k=0
336
+ (2k + 1)n−k,
337
+ (3.2a)
338
+ where
339
+ ϕm(x; µ) = L(µ−2m+1)
340
+ 2m−1
341
+ (−x),
342
+ (3.2b)
343
+ with L(α)
344
+ k (x) the Laguerre polynomial, for details see Kajiwara and Masuda [30]; see also [9, 29].
345
+ Theorem 3.4. The rational function solution of SIII (2.13) is given by
346
+ σn(x; µ, ε) = 2x d
347
+ dx {ln Sn(x; µ)} − 1
348
+ 2x2 − 2µx − 1
349
+ 4,
350
+ n ≥ 0,
351
+ (3.3a)
352
+ with Sn(x; µ) the Umemura polynomial, for the parameters
353
+ a = n + 1
354
+ 2,
355
+ b = µ,
356
+ ε = 1.
357
+ (3.3b)
358
+ Proof. See Clarkson [9].
359
+ 3.2
360
+ Special function solutions of PIII and SIII
361
+ Special function solutions of PIII (2.1), which are expressed in terms of Bessel functions and are classi-
362
+ fied in the following Theorem.
363
+ Theorem 3.5. Equation (2.1) has solutions expressible in terms of the Riccati equation
364
+ x dq
365
+ dx = ε1xq2 + (Aε1 − 1)q + ε2x,
366
+ (3.4)
367
+ if and only if
368
+ ε1A + ε2B = 4n + 2,
369
+ (3.5)
370
+ with n ∈ Z and ε2
371
+ 1 = 1, ε2
372
+ 2 = 1, independently. Further, the Riccati equation (3.4) has the solution
373
+ q(x) = −ε1
374
+ d
375
+ dx ln ψν(x),
376
+ (3.6)
377
+ 1The original manuscript was written by Umemura in 1996 for the proceedings of the conference “Theory of nonlinear special
378
+ functions: the Painlev´e transcendents” in Montreal, which were not published; for further details see [47].
379
+ 4
380
+
381
+ where ψν(x) satisfies
382
+ xd2ψν
383
+ dx2 + (1 − 2ε1ν)dψν
384
+ dx + ε1ε2xψν = 0,
385
+ (3.7)
386
+ which has solution
387
+ ψν(x) =
388
+
389
+
390
+
391
+
392
+
393
+
394
+
395
+
396
+
397
+ xν {C1Jν(x) + C2Yν(x)} ,
398
+ if
399
+ ε1 = 1,
400
+ ε2 = 1,
401
+ x−ν {C1Jν(x) + C2Yν(x)} ,
402
+ if
403
+ ε1 = −1, ε2 = −1,
404
+ xν {C1Iν(x) + C2Kν(x)} ,
405
+ if
406
+ ε1 = 1,
407
+ ε2 = −1,
408
+ x−ν {C1Iν(x) + C2Kν(x)} ,
409
+ if
410
+ ε1 = −1, ε2 = 1,
411
+ (3.8)
412
+ with C1, C2 arbitrary constants, and Jν(x), Yν(x), Iν(x), Kν(x) Bessel functions.
413
+ Proof. For details see Okamoto [46]; see also [11, 22, 36, 39, 40].
414
+ Determinantal representations of special function solutions of PIII (2.1) were given by Okamoto
415
+ [46]; see also [19, 38].
416
+ Theorem 3.6. Suppose τn(x; µ, ε) is the determinant given by
417
+ τn(x; µ, ε) = det
418
+ ��
419
+ x d
420
+ dx
421
+ �j+k
422
+ ϕµ(x; ε)
423
+ �n−1
424
+ j,k=0
425
+ ,
426
+ (3.9a)
427
+ where
428
+ ϕµ(x; ε) =
429
+
430
+ c1Jµ(x) + c2Yµ(x),
431
+ if
432
+ ε = 1,
433
+ c1Iµ(x) + c2Kµ(x),
434
+ if
435
+ ε = −1,
436
+ (3.9b)
437
+ with c1, c2 arbitrary constants, and Jµ(z), Yµ(z), Iµ(z), Kµ(z) Bessel functions.
438
+ The Bessel function solution of SIII (2.13) is given by
439
+ σn(x; µ, ε) = 2x d
440
+ dx {ln τn(x; µ, ε)} + 1
441
+ 2εx2 + µ2 − n2 + 2n,
442
+ (3.10a)
443
+ for the parameters
444
+ a = n,
445
+ b = µ.
446
+ (3.10b)
447
+ Lemma 3.7. The determinant τn(x; µ, ε) given by (3.9) satisfies the equation
448
+ x2
449
+
450
+ τn
451
+ d2τn
452
+ dx2 −
453
+ �dτn
454
+ dx
455
+ �2�
456
+ + xτn
457
+ dτn
458
+ dx = τn+1τn−1,
459
+ (3.11)
460
+ or equivalently
461
+
462
+ x d
463
+ dx
464
+ �2
465
+ ln τn = τn+1τn−1
466
+ τ 2n
467
+ .
468
+ (3.12)
469
+ Proof. See Okamoto [46, Theorem 2].
470
+ 4
471
+ B¨acklund transformations
472
+ We note that deg-PV (1.1) has the symmetries
473
+ S1 :
474
+ w1(z; α1, β1, γ1) = w(−z; α, β, γ),
475
+ (α1, β1, γ1) = (α, β, −γ),
476
+ (4.1)
477
+ S2 :
478
+ w2(z; α2, β2, γ2) = 1/w(z; α, β, γ),
479
+ (α2, β2, γ2) = (−β, −α, −γ),
480
+ (4.2)
481
+ where w(z; α, β, γ) is a solution of (1.1).
482
+ 5
483
+
484
+ Theorem 4.1. Suppose that w = w(z; α, β, γ) satisfies (1.1) with parameters
485
+ α = 1
486
+ 2a2,
487
+ β = − 1
488
+ 2b2,
489
+ γ = c.
490
+ Then wj = w(z; αj, βj, γj) given by
491
+ W1 :
492
+ w1 =
493
+ {zw′ + (w − 1)(aw − b)} {zw′ + (w − 1)(aw + b)}
494
+ z2(w′)2 + 2azw(w − 1)w′ + 2cz2w(w − 1) + (w − 1)2(a2w2 − b2),
495
+ (4.3a)
496
+ W2 :
497
+ w2 =
498
+ {zw′ − (w − 1)(aw − b)} {zw′ − (w − 1)(aw + b)}
499
+ z2(w′)2 − 2azw(w − 1)w′ + 2cz2w(w − 1) + (w − 1)2(a2w2 − b2),
500
+ (4.3b)
501
+ W3 :
502
+ w3 = z2(w′)2 + 2bz(w − 1)w′ + 2cz2w2(w − 1) − (w − 1)2(a2w2 − b2)
503
+ {zw′ − (w − 1)(aw − b)} {zw′ + (w − 1)(aw + b)}
504
+ ,
505
+ (4.3c)
506
+ W4 :
507
+ w4 = z2(w′)2 − 2bz(w − 1)w′ + 2cz2w2(w − 1) − (w − 1)2(a2w2 − b2)
508
+ {zw′ − (w − 1)(aw − b)} {zw′ + (w − 1)(aw + b)}
509
+ ,
510
+ (4.3d)
511
+ satisfy (1.1) with parameters
512
+ α1 = 1
513
+ 2(a + 1)2,
514
+ β1 = − 1
515
+ 2b2,
516
+ γ1 = c,
517
+ α2 = 1
518
+ 2(a − 1)2,
519
+ β2 = − 1
520
+ 2b2,
521
+ γ2 = c,
522
+ α3 = 1
523
+ 2a2,
524
+ β3 = − 1
525
+ 2(b + 1)2,
526
+ γ3 = c,
527
+ α4 = 1
528
+ 2a2,
529
+ β4 = − 1
530
+ 2(b − 1)2,
531
+ γ4 = c,
532
+ respectively.
533
+ Proof. See Adler [2]; also Filipuk and Van Assche [18].
534
+ 5
535
+ Classical solutions of deg-PV
536
+ To discuss classical solutions of deg-PV (1.1), it is convenient to make the transformation
537
+ w(z) = u(x),
538
+ z = 1
539
+ 2x2,
540
+ (5.1)
541
+ in (1.1), which gives
542
+ d2u
543
+ dx2 =
544
+ � 1
545
+ 2u +
546
+ 1
547
+ u − 1
548
+ � �du
549
+ dx
550
+ �2
551
+ − 1
552
+ x
553
+ du
554
+ dx + 4(u − 1)2(αu2 + β)
555
+ x2u
556
+ + 2γu.
557
+ (5.2)
558
+ We could fix the parameter γ in (5.2), by rescaling x if necessary, but it is more convenient not to do so.
559
+ Instead classical solutions will be classified for γ = ±1. From Corollary 2.3 and (5.1), we have that if
560
+ σ(x; a, b, ε) is a solution of SIII (2.13), then
561
+ u(x; a, b, ε) = σ′(x; a, b, ε) + εx
562
+ σ′(x; a, b, ε) − εx,
563
+ (5.3)
564
+ is a solution of (5.2) with γ = ε.
565
+ Theorem 5.1. Supppose that u = u(x; α, β, γ) satisfies (5.2) with parameters
566
+ α = 1
567
+ 2a2,
568
+ β = − 1
569
+ 2b2,
570
+ γ = c.
571
+ Then uj = u(x; αj, βj, γj) given by
572
+ U1 :
573
+ u1 =
574
+ {xu′ + 2(u − 1)(au − b)} {xu′ + 2(u − 1)(au + b)}
575
+ x2(u′)2 + 4axu(u − 1)u′ + 4cu(u − 1)x2 + 4(u − 1)2(a2u2 − b2),
576
+ (5.4a)
577
+ U2 :
578
+ u2 =
579
+ {xu′ − 2(u − 1)(au − b)} {xu′ − 2(u − 1)(au + b)}
580
+ x2(u′)2 − 4axu(u − 1)u′ + 4cu(u − 1)x2 + 4(u − 1)2(a2u2 − b2),
581
+ (5.4b)
582
+ U3 :
583
+ u3 = x2(u′)2 + 4bx(u − 1)u′ + 4cx2u2(u − 1) − 4(u − 1)2(a2u2 − b2)
584
+ {xu′ − 2(u − 1)(au − b)} {xu′ + 2(u − 1)(au + b)}
585
+ ,
586
+ (5.4c)
587
+ U4 :
588
+ u4 = x2(u′)2 − 4bx(u − 1)u′ + 4cx2u2(u − 1) − 4(u − 1)2(a2u2 − b2)
589
+ {xu′ − 2(u − 1)(au − b)} {xu′ + 2(u − 1)(au + b)}
590
+ ,
591
+ (5.4d)
592
+ 6
593
+
594
+ satisfy (5.2) with parameters
595
+ α1 = 1
596
+ 2(a + 1)2,
597
+ β1 = − 1
598
+ 2b2,
599
+ γ1 = c,
600
+ α2 = 1
601
+ 2(a − 1)2,
602
+ β2 = − 1
603
+ 2b2,
604
+ γ2 = c,
605
+ α3 = 1
606
+ 2a2,
607
+ β3 = − 1
608
+ 2(b + 1)2,
609
+ γ3 = c,
610
+ α4 = 1
611
+ 2a2,
612
+ β4 = − 1
613
+ 2(b − 1)2,
614
+ γ4 = c,
615
+ respectively.
616
+ Proof. This is easily proved by applying (5.1) to B¨acklund transformations in Theorem 4.1.
617
+ 5.1
618
+ Algebraic solutions
619
+ Algebraic solutions of (1.1) are equivalent to rational solutions of (5.2) and so we discuss rational
620
+ solutions of (5.2), which are classified in the following Theorem.
621
+ Theorem 5.2. Necessary and sufficient conditions for the existence of rational solutions of (5.2) are
622
+ either
623
+ (α, β, γ) =
624
+ � 1
625
+ 2(n + 1
626
+ 2), − 1
627
+ 2µ2, 1
628
+
629
+ ,
630
+ (5.5)
631
+ or
632
+ (α, β, γ) =
633
+ � 1
634
+ 2µ2, − 1
635
+ 2(n + 1
636
+ 2), −1
637
+
638
+ ,
639
+ (5.6)
640
+ where n ∈ Z and µ is an arbitrary constant.
641
+ Proof. For details see Gromak, Laine and Shimomura [22, §38]; see also [39, 40].
642
+ We remark that the solutions of (5.2) satisfying (5.5) are related to those satisfying (5.6) through
643
+ the analog of the symmetry (4.2). Consequently we shall be concerned only with rational solutions of
644
+ (5.2) for the parameters given by (5.5).
645
+ Theorem 5.3. The rational solution of (5.2) for the parameters (5.5) is given by
646
+ un(x; µ) = 1 −
647
+ xS2
648
+ n(x; µ)
649
+ Sn+1(x; µ)Sn−1(x; µ),
650
+ n ≥ 0,
651
+ (5.7)
652
+ where Sn(x; µ) is the Umemura polynomial (3.2).
653
+ Proof. Substituting the rational solution of SIII (2.13) given by (3.3) into (5.3) and then using the
654
+ reccurence relation (3.1) gives the result.
655
+ Remark 5.4. The Umemura polynomial Sn(x; µ) satisfies the difference equation
656
+ Sn+1(x; µ)Sn−1(x; µ) = xS2
657
+ n(x; µ) + µSn(x; µ + 1) Sn(x; µ − 1).
658
+ (5.8)
659
+ Hence from (5.7) there are two alternative representations of the rational solution
660
+ un(x; µ) =
661
+ µSn(x; µ + 1) Sn(x; µ − 1)
662
+ µSn(x; µ + 1) Sn(x; µ − 1) + xS2n(x; µ),
663
+ un(x; µ) = µSn(x; µ + 1) Sn(x; µ − 1)
664
+ Sn+1(x; µ)Sn−1(x; µ)
665
+ .
666
+ 5.2
667
+ Bessel function solutions
668
+ Theorem 5.5. Necessary and sufficient conditions for the existence of Bessel function solutions of (5.2)
669
+ are either
670
+ (α, β, γ) =
671
+ � 1
672
+ 2n2, − 1
673
+ 2µ2, ε
674
+
675
+ ,
676
+ (5.9)
677
+ or
678
+ (α, β, γ) =
679
+ � 1
680
+ 2µ2, − 1
681
+ 2n2, −ε
682
+
683
+ ,
684
+ (5.10)
685
+ with ε = ±1, and where n ∈ Z+ and µ is an arbitrary constant.
686
+ 7
687
+
688
+ Proof. From (2.5) and (2.9), the parameters in PIII (2.1) and deg-PV (5.2) are given by
689
+ (A, B) =
690
+
691
+ 2(a − b), 2ε(a + b + 1)
692
+
693
+ ,
694
+ (α, β, γ) = ( 1
695
+ 2a2, − 1
696
+ 2b2, ε),
697
+ respectively, for parameters a, b and ε. The result then follows from Theorem 3.5.
698
+ Theorem 5.6. The Bessel function solution of (5.2) for the parameters
699
+ (α, β, γ) =
700
+ � 1
701
+ 2n2, − 1
702
+ 2µ2, ε
703
+
704
+ ,
705
+ is given by
706
+ un(x; µ, ε) = 1 +
707
+ εx2τ 2
708
+ n(x; µ, ε)
709
+ τn+1(x; µ, ε) τn−1(x; µ, ε),
710
+ n ≥ 1,
711
+ (5.11)
712
+ where
713
+ τn(x; µ, ε) = det
714
+ ��
715
+ x d
716
+ dx
717
+ �j+k
718
+ ϕµ(x; ε)
719
+ �n−1
720
+ j,k=0
721
+ ,
722
+ (5.12)
723
+ and τ0(x; µ, ε) = 1, with
724
+ ϕµ(x; ε) =
725
+
726
+ c1Jµ(x) + c2Yµ(x),
727
+ if
728
+ ε = 1,
729
+ c1Iµ(x) + c2Kµ(x),
730
+ if
731
+ ε = −1,
732
+ (5.13)
733
+ c1 and c2 arbitrary constants, and Jµ(x), Yµ(x), Iµ(x) and Kµ(x) Bessel functions.
734
+ Proof. Substituting the Bessel function solution of SIII (2.13) given by (3.10) into (5.3) and then using
735
+ (3.11) gives the result.
736
+ Corollary 5.7. The Bessel function solution of (5.2) for the parameters
737
+ (α, β, γ) =
738
+ � 1
739
+ 2n2, − 1
740
+ 2µ2, 2ε
741
+
742
+ ,
743
+ is given by
744
+ wn(z; µ, ε) = 1 +
745
+ εzT 2
746
+ n (z; µ, ε)
747
+ Tn+1(z; µ, ε) Tn−1(z; µ, ε),
748
+ n ≥ 1,
749
+ (5.14)
750
+ where
751
+ Tn(z; µ, ε) = det
752
+ ��
753
+ z d
754
+ dz
755
+ �j+k
756
+ ψµ(z; ε)
757
+ �n−1
758
+ j,k=0
759
+ ,
760
+ (5.15)
761
+ and T0(z; µ, ε) = 1, with
762
+ ϕµ(z; ε) =
763
+
764
+ c1Jµ(2√z) + c2Yµ(2√z),
765
+ if
766
+ ε = 1,
767
+ c1Iµ(2√z) + c2Kµ(2√z),
768
+ if
769
+ ε = −1,
770
+ (5.16)
771
+ c1 and c2 arbitrary constants, and Jµ(x), Yµ(x), Iµ(x) and Kµ(x) Bessel functions.
772
+ In the next Lemma, it is shown that the first solution u1(x; µ, ε), the “seed solution”, satisfies a
773
+ first-order, second-degree equation.
774
+ Lemma 5.8. The solution of (5.2) for the parameters
775
+ (α, β, γ) =
776
+ � 1
777
+ 2, − 1
778
+ 2µ2, ε
779
+
780
+ ,
781
+ is
782
+ u1(x; µ, ε) =
783
+ ϕµ+1(x; ε) [xϕµ+1(x; ε) − 2εµϕµ(x; ε)]
784
+ xϕ2
785
+ µ+1(x; ε) − 2εµϕµ+1(x; ε)ϕµ(x; ε) + εxϕ2µ(x; ε),
786
+ (5.17)
787
+ where
788
+ ϕµ(x; ε) =
789
+
790
+ c1Jµ(x) + c2Yµ(x),
791
+ if
792
+ ε = 1,
793
+ c1Iµ(x) + c2Kµ(x),
794
+ if
795
+ ε = −1,
796
+ with c1 and c2 constants, satisfies the first-order, second-degree equation
797
+ x2
798
+ �du
799
+ dx
800
+ �2
801
+ − 4xu(u − 1)du
802
+ dx + 4εx2u(u − 1) + 4(u − 1)2(u2 − µ2) = 0.
803
+ (5.18)
804
+ 8
805
+
806
+ Proof. Define
807
+ Φµ(x; ε) = ϕµ+1(x; ε)
808
+ ϕµ(x; ε) ,
809
+ then from (5.17)
810
+ u1(x; µ, ε) = 1 −
811
+ x
812
+ εxΦ2µ − 2µΦµ + x,
813
+ (5.19)
814
+ and Φµ(x; ε) satisfies the Riccati equation
815
+ xdΦµ
816
+ dx = εxΦ2
817
+ µ − (2µ + 1)Φµ + x.
818
+ (5.20)
819
+ Next we assume that u1(x; µ, ε) satisfies a first-order, second-degree equation of the form
820
+ x2
821
+ �du
822
+ dx
823
+ �2
824
+ + x
825
+
826
+ f2(x, µ, ε)u2 + f1(x, µ, ε)u + f0(x, µ, ε)
827
+ � du
828
+ dx +
829
+ 4
830
+
831
+ j=0
832
+ gj(x, µ, ε)uj = 0,
833
+ (5.21)
834
+ where {fj(x, µ, ε)}2
835
+ j=0 and {gj(x, µ, ε)}4
836
+ j=0 are to be determined. Then substituting (5.19) into (5.21),
837
+ using the fact that Φµ(x; ε) satisfies (5.20) and equating coefficients of powers of Φµ yields
838
+ f2 = −4,
839
+ f1 = 4,
840
+ f0 = 0,
841
+ g4 = 4,
842
+ g3 = −8,
843
+ g2 = 4εx2 − 4µ2 + 4,
844
+ g1 = −4εx2 + 8µ2,
845
+ g0 = −4µ2.
846
+ Hence we obtain equation (5.18), as required.
847
+ This demonstrates that special function solutions of (5.2), and hence also deg-PV (1.1) , are different
848
+ from special function solutions of PII–PVI where the “seed solution” satisfies a Riccati equation, a first-
849
+ order, first-degree equation.
850
+ 6
851
+ Applications
852
+ 6.1
853
+ Complex sine-Gordon equation
854
+ Consider the two-dimensional complex sine-Gordon equation
855
+ ∇2ψ + (∇ψ)2ψ
856
+ 1 − |ψ|2 + ψ
857
+
858
+ 1 − |ψ|2�
859
+ = 0,
860
+ (6.1)
861
+ where ∇ψ = (ψx, ψy). Making the transformation
862
+ ψ(x, y) = cos(ϕ(x, y)) exp{iη(x, y)},
863
+ ψ(x, y) = cos(ϕ(x, y)) exp{−iη(x, y)},
864
+ in the complex sine-Gordon equation (6.1) yields
865
+ ∇2ϕ + cos ϕ
866
+ sin3 ϕ(∇η)2 − 1
867
+ 2 sin(2ϕ) = 0,
868
+ sin(2ϕ) ∇2η = 4∇ϕ •∇η,
869
+ which is the Pohlmeyer-Lund-Regge model [33, 34, 50].
870
+ The complex sine-Gordon equation (6.1) has a separable solution in polar coordinates given by
871
+ ψ(r, θ) = Rn(r) einθ, where Rn(r) satisfies
872
+ d2Rn
873
+ dr2
874
+ + 1
875
+ r
876
+ dRn
877
+ dr
878
+ +
879
+ Rn
880
+ 1 − R2n
881
+ ��dRn
882
+ dr
883
+ �2
884
+ − n2
885
+ r2
886
+
887
+ + Rn
888
+
889
+ 1 − R2
890
+ n
891
+
892
+ = 0,
893
+ (6.2)
894
+ We remark that this equation also arises in extended quantum systems [4, 5, 6], in relativity [20] and
895
+ in coefficients in the three-term recurrence relation for orthogonal polynomials with respect to the
896
+ weight w(θ) = et cos θ on the unit circle, see [56, equation (3.13)]. The orthogonal polynomials for this
897
+ weight on the unit circle are related to unitary random matrices [49].
898
+ Equation (6.2) can be shown to possess the Painlev´e property, though is not in the list of 50 equa-
899
+ tions given in [25, Chapter 14]. Equation (6.2) can be transformed to the fifth Painlev´e equation (1.2)
900
+ in two different ways.
901
+ 9
902
+
903
+ (i) If Rn(r) satisfies (6.2) then making the transformation
904
+ Rn(r) = 1 + un(z)
905
+ 1 − un(z),
906
+ r = 1
907
+ 2z,
908
+ (6.3)
909
+ yields
910
+ d2un
911
+ dz2 =
912
+ � 1
913
+ 2un
914
+ +
915
+ 1
916
+ un − 1
917
+ � �dun
918
+ dz
919
+ �2
920
+ − 1
921
+ z
922
+ dun
923
+ dz + n2(un − 1)2(u2
924
+ n − 1)
925
+ 8z2un
926
+ − un(un + 1)
927
+ 2(un − 1) ,
928
+ (6.4)
929
+ which is PV (1.2) with α = 1
930
+ 8n2, β = − 1
931
+ 8n2, γ = 0 and δ = − 1
932
+ 2.
933
+ (ii) If Rn(r) satisfies (6.2) then making the transformation
934
+ Rn(r) =
935
+ 1
936
+
937
+ 1 − vn(x)
938
+ ,
939
+ r = √x,
940
+ (6.5)
941
+ yields
942
+ d2vn
943
+ dx2 =
944
+ � 1
945
+ 2vn
946
+ +
947
+ 1
948
+ vn − 1
949
+ � �dvn
950
+ dx
951
+ �2
952
+ − 1
953
+ x
954
+ dvn
955
+ dx − n2(vn − 1)2
956
+ 2x2vn
957
+ + vn
958
+ 2x,
959
+ (6.6)
960
+ which is deg-PV (1.1) with α = 0, β = − 1
961
+ 2n2 and γ = 1
962
+ 2 so is equivalent to PIII (2.1), as mentioned
963
+ above.
964
+ This shows that solutions of equations (6.4) and (6.6) are related by
965
+ vn(x) =
966
+ 4un(z)
967
+ 1 + u2n(z),
968
+ x = 1
969
+ 4z2.
970
+ The function Rn(r) satisfies the ordinary differential equation (6.2), the differential-difference equa-
971
+ tions
972
+ dRn
973
+ dr
974
+ + n
975
+ r Rn −
976
+
977
+ 1 − R2
978
+ n
979
+
980
+ Rn−1 = 0,
981
+ (6.7a)
982
+ dRn−1
983
+ dr
984
+ − n − 1
985
+ r
986
+ Rn−1 +
987
+
988
+ 1 − R2
989
+ n−1
990
+
991
+ Rn = 0,
992
+ (6.7b)
993
+ since solving (6.7a) for Rn−1(r) and substituting in (6.7b) yields equation (6.2). Also eliminating the
994
+ derivatives in (6.7), after letting n → n + 1 in (6.7b), yields the difference equation
995
+ Rn+1 + Rn−1 = 2n
996
+ r
997
+ Rn
998
+ 1 − R2n
999
+ ,
1000
+ (6.8)
1001
+ which is known as the discrete Painlev´e II equation [41, 49].
1002
+ If n = 1 then equations (6.7) have the solution
1003
+ R0(r) = 1,
1004
+ R1(r) = C1I1(r) − C2K1(r)
1005
+ C1I0(r) + C2K0(r),
1006
+ where I0(r), K0(r), I1(r) and K1(r) are the imaginary Bessel functions and C1 and C2 are arbitrary
1007
+ constants. For solutions which are bounded at r = 0 then necesssarily C2 = 0 and so
1008
+ R0(r) = 1,
1009
+ R1(r) = I1(r)
1010
+ I0(r).
1011
+ (6.9)
1012
+ Hence one can use the difference equation (6.8) to determine Rn(r), for n ≥ 2, which yields
1013
+ R2(r) = −rR2
1014
+ 1(r) + 2R1(r) − r
1015
+ r [R2
1016
+ 1(r) − 1]
1017
+ ,
1018
+ R3(r) = R3
1019
+ 1(r) − rR2
1020
+ 1(r) − 2R1(r) + r
1021
+ R1(r) [rR2
1022
+ 1(r) + R1(r) − r] ,
1023
+ R4(r) =
1024
+ r(r2 + 5)R4
1025
+ 1(r) + 4R3
1026
+ 1(r) − 2r(r2 + 3)R2
1027
+ 1(r) + r3
1028
+ r [(r2 − 1)R4
1029
+ 1(r) + 4rR3
1030
+ 1(r) − 2(r2 + 2)R2
1031
+ 1(r) − 4rR1(r) + r2].
1032
+ These results suggest that (6.2) should be solvable in terms of PIII (2.1), which is illustrated in the
1033
+ following theorem.
1034
+ 10
1035
+
1036
+ Theorem 6.1. If Rn(r) satisfies (6.2) then wn(r) = Rn+1(r)/Rn(r) satisfies
1037
+ d2wn
1038
+ dr2
1039
+ = 1
1040
+ wn
1041
+ �dwn
1042
+ dr
1043
+ �2
1044
+ − 1
1045
+ r
1046
+ dwn
1047
+ dr
1048
+ − 2n
1049
+ r w2
1050
+ n + 2(n + 1)
1051
+ r
1052
+ + w3
1053
+ n − 1
1054
+ wn
1055
+ ,
1056
+ (6.10)
1057
+ which is PIII (2.1) with parameters α = −2n and β = 2(n + 1).
1058
+ Proof. See Hisakado [23] and Tracy & Widom [52]; see also [56, §3.1].
1059
+ We note that since the parameters in (6.10) satisfy −α + β = 4n + 2, with n ∈ Z+, then the equation
1060
+ has solutions expressible in terms of the modified Bessel functions I0(r) and I1(r) (as well as K0(r) and
1061
+ K1(r), but these are not needed here).
1062
+ Theorem 6.2. Let τn(r; ν) be the n × n determinant
1063
+ τn(r; ν) = det
1064
+ ��
1065
+ r d
1066
+ dr
1067
+ �j+k
1068
+ Iν(r)
1069
+ �n−1
1070
+ j,k=0
1071
+ ,
1072
+ (6.11)
1073
+ with Iν(r) the modified Bessel function, then
1074
+ wn(r; ν) = τn+1(r; ν + 1) τn(r; ν)
1075
+ τn+1(r; ν) τn(r; ν + 1) ≡ d
1076
+ dz
1077
+
1078
+ ln τn+1(z; ν)
1079
+ τn(z; ν + 1)
1080
+
1081
+ − n + ν
1082
+ z
1083
+ ,
1084
+ n ≥ 0,
1085
+ (6.12)
1086
+ satisfies PIII (2.1) with α = 2(ν − n) and β = 2(ν + n + 1).
1087
+ Proof. See, for example, [19, 38].
1088
+ Theorem 6.3. Equation (6.2) has the solution
1089
+ Rn(r) = τn(r; 1)
1090
+ τn(r; 0),
1091
+ (6.13)
1092
+ where τn(r; ν) is the determinant given by (6.11).
1093
+ Proof. The proof is straightforward using induction. From (6.9) we have
1094
+ R1(r) = I1(r)
1095
+ I0(r) = τ1(r; 1)
1096
+ τ1(r; 0),
1097
+ so (6.13) is true if n = 1. Assuming (6.13) holds then from Theorems 6.1 and 6.2
1098
+ Rn+1(r) = wn(r; 0)Rn(r) = τn+1(r; 1) τn(r; 0)
1099
+ τn+1(r; 0) τn(r; 1) × τn(r; 1)
1100
+ τn(r; 0) = τn+1(r; 1)
1101
+ τn+1(r; 0),
1102
+ as required, and so the result follows by induction.
1103
+ Corollary 6.4. Equations (6.4) and (6.6) have the Bessel function solutions
1104
+ un(z) = τn( 1
1105
+ 2z; 1) + τn( 1
1106
+ 2z; 0)
1107
+ τn( 1
1108
+ 2z; 1) − τn( 1
1109
+ 2z; 0),
1110
+ vn(x) = 1 − τ 2
1111
+ n(√x; 0)
1112
+ τ 2n(√x; 1),
1113
+ respectively, with τn(r; ν) the determinant given by (6.11).
1114
+ Lemma 6.5. The formal asymptotic behaviour of the vortex solution Rn(r) is given by
1115
+ Rn(r) =
1116
+ rn
1117
+ 2n n!
1118
+
1119
+ 1 −
1120
+ r2
1121
+ 4(n + 1) + O
1122
+
1123
+ r4��
1124
+ ,
1125
+ as
1126
+ r → 0,
1127
+ (6.14)
1128
+ Rn(r) = 1 − n
1129
+ 2r − n2
1130
+ 8r2 − n(n2 + 1)
1131
+ 16r3
1132
+ + O(r−4),
1133
+ as
1134
+ r → ∞.
1135
+ (6.15)
1136
+ Proof. These are determined from (6.8) and (6.9).
1137
+ 11
1138
+
1139
+ 6.2
1140
+ Generalised Charlier polynomials
1141
+ The Charlier polynomials Cn(k; z) are a family of orthogonal polynomials introduced in 1905 by Char-
1142
+ lier [7] given by
1143
+ Cn(k; z) = 2F0 (−n, −k; ; −1/z) = (−1)nn!L(−1−k)
1144
+ n
1145
+ (−1/z) ,
1146
+ z > 0,
1147
+ (6.16)
1148
+ where 2F0(a, b; ; z) is the hypergeometric function and L(α)
1149
+ n (z) is the associated Laguerre polynomial,
1150
+ see, for example, [48, §18.19]. The Charlier polynomials are orthogonal on the lattice N with respect to
1151
+ the Poisson distribution
1152
+ ω(k) = zk
1153
+ k! ,
1154
+ z > 0,
1155
+ (6.17)
1156
+ and satisfy the orthogonality condition
1157
+
1158
+
1159
+ k=0
1160
+ Cm(k; z)Cn(k; z)zk
1161
+ k! = n! ez
1162
+ zn δm,n.
1163
+ Smet and Van Assche [51] generalized the Charlier weight (6.17) with one additional parameter
1164
+ through the weight function
1165
+ ω(k; ν) =
1166
+ Γ(ν + 1) zk
1167
+ Γ(ν + k + 1) Γ(k + 1),
1168
+ z > 0,
1169
+ with ν a parameter such that ν > −1. This gives the discrete weight
1170
+ ω(k; ν) =
1171
+ zk
1172
+ (ν + 1)k k!,
1173
+ z > 0,
1174
+ (6.18)
1175
+ where (ν + 1)k = Γ(ν + 1 + k)/Γ(ν + 1) is the Pochhammer symbol, on the lattice N. Discrete orthogonal
1176
+ polynomials are characterized by the discrete Pearson equation
1177
+
1178
+
1179
+ σ(k)ω(k)
1180
+
1181
+ = τ(k)ω(k),
1182
+ (6.19)
1183
+ where ∆ is the forward difference operator
1184
+ ∆f(k) = f(k + 1) − f(k).
1185
+ The weight (6.18) satisfies the discrete Pearson equation (6.19) with
1186
+ σ(k) = k(k + ν),
1187
+ τ(k) = −k2 − νk + z,
1188
+ and so the generalised Charlier polynomials are semi-classical orthogonal polynomials since τ(k) is a
1189
+ polynomial with deg(τ) > 1. The special case β = 0 was first considered by Hounkonnou, Hounga and
1190
+ Ronveaux [24] and later studied by Van Assche and Foupouagnigni [57].
1191
+ For the generalised Charlier weight (6.18), the orthonormal polynomials pn(k; z) satisfy the orthog-
1192
+ onality condition
1193
+
1194
+
1195
+ k=0
1196
+ pm(k; z)pn(k; z)
1197
+ zk
1198
+ (ν + 1)k k! = δm,n,
1199
+ and the three-term recurrence relation
1200
+ kpn(k; z) = an+1(z)pn+1(k; z) + bn(z)pn(k; z) + an(z)pn−1(k; z),
1201
+ (6.20)
1202
+ with p−1(k; z) = 0 and p0(k; z) = 1. Our interest is in the coefficients an(z) and bn(z) in the recurrence
1203
+ relation (6.20).
1204
+ Smet and Van Assche [51, Theorem 2.1] proved the following theorem for recurrence coefficients
1205
+ associated with the generalised Charlier weight (6.18).
1206
+ 12
1207
+
1208
+ Theorem 6.6. The recurrence coefficients an(z) and bn(z) for orthonormal polynomials associated with
1209
+ the generalised Charlier weight (6.18) on the lattice N satisfy the discrete system
1210
+ (a2
1211
+ n+1 − z)(a2
1212
+ n − z) = z(bn − n)(bn − n + ν),
1213
+ bn + bn−1 − n + ν + 1 = nz/a2
1214
+ n,
1215
+ (6.21)
1216
+ with initial conditions
1217
+ a2
1218
+ 0 = 0,
1219
+ b0 =
1220
+ √z Iν+1(2√z)
1221
+ Iν(2√z)
1222
+ = z d
1223
+ dz
1224
+
1225
+ ln Iν(2√z)
1226
+
1227
+ − ν
1228
+ 2 ,
1229
+ (6.22)
1230
+ with Iν(k) the modified Bessel function.
1231
+ Remark 6.7. The discrete system such as (6.21) for recurrence coefficients is sometimes known as the
1232
+ Laguerre-Freud equations, cf. [3, 24, 35].
1233
+ The recurrence coefficients an(z) and bn(z) also satisfy the Toda lattice, cf. [56, Theorem 3.8]
1234
+ z d
1235
+ dz a2
1236
+ n = a2
1237
+ n(bn − bn−1),
1238
+ (6.23a)
1239
+ z d
1240
+ dz bn = a2
1241
+ n+1 − a2
1242
+ n.
1243
+ (6.23b)
1244
+ Letting a2
1245
+ n(z) = xn(z) and bn(z) = yn(z) in (6.21) and (6.23) yields
1246
+ (xn+1 − z)(xn − z) = t(yn − n)(yn − n + ν),
1247
+ z dxn
1248
+ dt
1249
+ = xn(yn − yn−1),
1250
+ yn + yn−1 − n + ν + 1 = nz
1251
+ xn
1252
+ ,
1253
+ z dyn
1254
+ dz = xn+1 − xn.
1255
+ Eliminating xn+1 and yn−1 in these equations yields the differential system
1256
+ z dxn
1257
+ dz = xn(2yn + ν − n + 1) − nz,
1258
+ (6.24a)
1259
+ z dyn
1260
+ dz = −xn + z + (yn − n)(yn − n + ν)z
1261
+ xn − z
1262
+ .
1263
+ (6.24b)
1264
+ Solving (6.24a) for yn gives
1265
+ yn =
1266
+ z
1267
+ 2xn
1268
+ dxn
1269
+ dz + nz
1270
+ 2xn
1271
+ + n − ν − 1
1272
+ 2
1273
+ ,
1274
+ and substituting this into (6.24b) yields
1275
+ d2xn
1276
+ dz2 = 1
1277
+ 2
1278
+ � 1
1279
+ xn
1280
+ +
1281
+ 1
1282
+ xn − z
1283
+
1284
+
1285
+ xn
1286
+ z(xn − z)
1287
+ dxn
1288
+ dz − 2x2
1289
+ n
1290
+ z2 + 4xn + n2 − ν2 + 1
1291
+ 2z
1292
+ − n2
1293
+ 2xn
1294
+ +
1295
+ 1 − ν2
1296
+ 2(xn − z).
1297
+ (6.25)
1298
+ Making the transformation
1299
+ xn(z) =
1300
+ z
1301
+ 1 − wn(z).
1302
+ (6.26)
1303
+ in (6.25) yields
1304
+ d2wn
1305
+ dz2
1306
+ =
1307
+ � 1
1308
+ 2wn
1309
+ +
1310
+ 1
1311
+ wn − 1
1312
+ ��dwn
1313
+ dz
1314
+ �2
1315
+ − 1
1316
+ z
1317
+ dwn
1318
+ dz
1319
+ + (wn − 1)2(n2w2
1320
+ n − ν2)
1321
+ 2wnz2
1322
+ − 2wn
1323
+ z ,
1324
+ (6.27)
1325
+ which is deg-PV (1.1) with parameters α = 1
1326
+ 2n2, β = − 1
1327
+ 2ν2 and γ = −2.
1328
+ Solving (6.24b) for xn gives
1329
+ xn = − 1
1330
+ 2z dyn
1331
+ dz + z + 1
1332
+ 2Xn,
1333
+ (6.28)
1334
+ where
1335
+ X2
1336
+ n = z2
1337
+ �dyn
1338
+ dz
1339
+ �2
1340
+ + 4z(yn − n)(yn − n + ν).
1341
+ (6.29)
1342
+ 13
1343
+
1344
+ From (6.29) we get
1345
+ dXn
1346
+ dz
1347
+ = z2
1348
+ Xn
1349
+ d2yn
1350
+ dz2
1351
+ dyn
1352
+ dz + z
1353
+ Xn
1354
+ �dyn
1355
+ dz
1356
+ �2
1357
+ + 2z(2yn − 2n + ν)
1358
+ Xn
1359
+ dyn
1360
+ dz + 2(yn − n)(yn − n + ν)
1361
+ Xn
1362
+ (6.30)
1363
+ Substituting (6.28) into (6.24a), then using (6.30), solving for Xn, and substituting into (6.29) yields
1364
+ the second-order, second-degree equation
1365
+
1366
+ 2z d2yn
1367
+ dz2 + dyn
1368
+ dz + 8yn − 8n + 4ν
1369
+ �2
1370
+ = (4yn − 2n + 2ν + 1)2
1371
+ z
1372
+
1373
+ z
1374
+ �dyn
1375
+ dz
1376
+ �2
1377
+ + 4(yn − n)(yn − n + ν)
1378
+
1379
+ . (6.31)
1380
+ Making the transformation
1381
+ yn(z) = 1
1382
+ 2vn(x) + 1
1383
+ 2n − 1
1384
+ 2ν − 1
1385
+ 4,
1386
+ x = 2√z,
1387
+ in (6.31) yields
1388
+ �d2vn
1389
+ dx2 + 4vn − 4n − 2
1390
+ �2
1391
+ = 4v2
1392
+ n
1393
+ x2
1394
+ ��dvn
1395
+ dx
1396
+ �2
1397
+ + 4v2
1398
+ n − 4(2n + 1)vn + (2n + 1)2 − 4ν2
1399
+
1400
+ .
1401
+ (6.32)
1402
+ Equation (A.5) in [14] is
1403
+ �d2v
1404
+ dx2 − av − b
1405
+ �2
1406
+ = 4v2
1407
+ x2
1408
+ ��dv
1409
+ dx
1410
+ �2
1411
+ − av2 − 2bv − c
1412
+
1413
+ ,
1414
+ (6.33)
1415
+ with a, b and c parameters, an equation derived by Chazy [8], and is the primed version of equation
1416
+ SD-III in [15]. Hence equation (6.32) is the special case of equation (6.33) with
1417
+ a = −4,
1418
+ b = 4n + 2,
1419
+ c = 4ν2 − (2n + 1)2.
1420
+ Cosgrove [14] showed that equation (6.33) is solvable in terms of solutions of PIII (2.1). Consequently,
1421
+ the solution of (6.32) is given by
1422
+ vn(x) = x
1423
+ 2q
1424
+ � dq
1425
+ dx + q2 + 1
1426
+
1427
+ ,
1428
+ where q(x) satisfies PIII (2.1) for the parameters A = 2ν − 2n − 2 and B = 2ν + 2n.
1429
+ Theorem 6.8. The recurrence relations an(z) and bn(z) are given by
1430
+ a2
1431
+ n(z) = xn(z) = Tn+1(z; ν)Tn−1(z; ν)
1432
+ T 2
1433
+ n (z; ν)
1434
+ ,
1435
+ (6.34a)
1436
+ bn(z) = yn(z) = z d
1437
+ dz
1438
+
1439
+ ln Tn+1(z; ν)
1440
+ Tn(z; ν)
1441
+
1442
+ − ν
1443
+ 2 ,
1444
+ (6.34b)
1445
+ where
1446
+ Tn(z; ν) = det
1447
+ ��
1448
+ z d
1449
+ dz
1450
+ �j+k
1451
+
1452
+
1453
+ 2√z
1454
+
1455
+ �n−1
1456
+ j,k=0
1457
+ ,
1458
+ with T0(z; ν) = 1, and Iν(x) is the modified Bessel function.
1459
+ Proof. The expression (6.34a) for a2
1460
+ n(z) follows immediately by substituting (5.14) in (6.26). To prove
1461
+ the result (6.34b) for bn(z) we use induction and the factor that from equation (6.23b), a2
1462
+ n(z) = xn(z)
1463
+ and bn(z) = yn(z) are related by
1464
+ z dxn
1465
+ dt
1466
+ = xn(yn − yn−1),
1467
+ and initially
1468
+ y0(z) = z d
1469
+ dz
1470
+
1471
+ ln T1(z; ν)
1472
+
1473
+ } − ν
1474
+ 2 .
1475
+ 14
1476
+
1477
+ Hence
1478
+ y1(z) = z d
1479
+ dz
1480
+
1481
+ ln x1(z)
1482
+
1483
+ + y0(z)
1484
+ = z d
1485
+ dz
1486
+
1487
+ ln T2(z; ν)T0(z; ν)
1488
+ T 2
1489
+ 1 (z; ν)
1490
+
1491
+ + z d
1492
+ dz {ln T1(z; ν)} − ν
1493
+ 2
1494
+ = z d
1495
+ dz
1496
+
1497
+ ln T2(z; ν)
1498
+ T1(z; ν)
1499
+
1500
+ − ν
1501
+ 2 ,
1502
+ so (6.34b) is true for n = 1. Now suppose that (6.34b) is true, then
1503
+ yn+1(z) = z d
1504
+ dz
1505
+
1506
+ ln xn(z)
1507
+
1508
+ + yn(z)
1509
+ = z d
1510
+ dz
1511
+
1512
+ ln Tn+2(z; ν)Tn(z; ν)
1513
+ T 2
1514
+ n+1(z; ν)
1515
+
1516
+ + z d
1517
+ dz
1518
+
1519
+ ln Tn+1(z; ν)
1520
+ Tn(z; ν)
1521
+
1522
+ − ν
1523
+ 2
1524
+ = z d
1525
+ dz
1526
+
1527
+ ln Tn+2(z; ν)
1528
+ Tn+1(z; ν)
1529
+
1530
+ − ν
1531
+ 2 ,
1532
+ as required, and so the result follows by induction. We remark that equation (6.23a) is identically
1533
+ satisfied by a2
1534
+ n(z) and bn(z) given by (6.34).
1535
+ In a recent paper, Fern´andez-Irisarri and Ma˜nas [17, §2] discuss the generalised Charlier weight
1536
+ (6.18), in particular properties of the coefficients in the recurrence relation. The relationship between
1537
+ the notations in [17] and those here are xn(z) = γn(η) and yn(z) = βn(η). Fern´andez-Irisarri and Ma˜nas
1538
+ [17] relate xn(z) and yn(z) to Okamoto’s Hamiltonian for PIII′ [46] and derive two ordinary differential
1539
+ equations for xn(z).
1540
+ 1. Equation (45) in [17, Theorem 4] is the third order equation
1541
+ δz
1542
+ �xn
1543
+ z
1544
+
1545
+ δ2
1546
+ z(ln xn) + 2xn
1547
+
1548
+ + n2z
1549
+ xn
1550
+
1551
+ = 2xn,
1552
+ δz(f) = z df
1553
+ dz ,
1554
+ i.e.
1555
+ d3xn
1556
+ dz3 =
1557
+ 1
1558
+ zx2n
1559
+
1560
+ z dxn
1561
+ dz − xn
1562
+ � �
1563
+ 2xn
1564
+ d2xn
1565
+ dz2 −
1566
+ �dxn
1567
+ dz
1568
+ �2
1569
+ + n2
1570
+
1571
+ − 4xn
1572
+ z2
1573
+ dxn
1574
+ dz + 2xn(xn + z)
1575
+ z3
1576
+ ,
1577
+ (6.35)
1578
+ and the state that this equation “should have the Painlev´e property”. Equation (6.35) can be
1579
+ integrate to give equation (6.25), with ν2 as the constant of integration. Since equation (6.25) is
1580
+ equivalent to deg-PV (5.2) then equation (6.35) does have the Painlev´e property.
1581
+ 2. Equation (60) in [17, Theorem 5] is the second order equation
1582
+
1583
+ 1 − xn
1584
+ z
1585
+ � �
1586
+ δz
1587
+ �δz(xn) + nz
1588
+ xn
1589
+
1590
+ + 2xn
1591
+
1592
+ + 2{xn − z + (n − b)n}
1593
+ = − 1
1594
+ 2
1595
+ �δz(xn) + nz
1596
+ xn
1597
+ �2
1598
+ + (n + 1)
1599
+ �δz(xn) + nz
1600
+ xn
1601
+
1602
+ + (n − b − 1)(3n − b + 1),
1603
+ which is equation (6.25) with
1604
+ ν2 = 2(b − n)2 + n2 − 2n − 1.
1605
+ 7
1606
+ Discussion
1607
+ In this paper the classical solutions of deg-PV (5.2) have been classified. Ohyama and Okumura [43,
1608
+ Theorem 2.1] give a list of classical solutions of PI to PV and state that “deg-P5 with α = 1
1609
+ 2a2, β = − 1
1610
+ 8,
1611
+ γ = −2 has the algebraic solution w(z) = 1 + 2√z/a”2 and “deg-P5 with β = 0 has the Riccati type
1612
+ 2As noted in [1], there is typo in [43] who say β = −8 rather than β = − 1
1613
+ 8.
1614
+ 15
1615
+
1616
+ solutions”. The results in this paper show that there are more classical solutions of deg-PV (1.1). The
1617
+ algebraic solution is equivalent to the “seed solution” obtained by setting n = 0 in (5.7), i.e.
1618
+ u0(x; µ) =
1619
+ µ
1620
+ x + µ,
1621
+ and there is a more general hierarchy of “Riccati type solutions” which are described in Theorem 5.6.
1622
+ All solutions of PII–PVI that are expressible in terms of special functions satisfy a first-order equa-
1623
+ tion of the form
1624
+ �du
1625
+ dx
1626
+ �n
1627
+ =
1628
+ n−1
1629
+
1630
+ j=0
1631
+ Fj(u, x)
1632
+ �du
1633
+ dx
1634
+ �j
1635
+ ,
1636
+ (7.1)
1637
+ where Fj(u, x) is polynomial in u with coefficients that are rational functions of x. It can be shown
1638
+ that the Bessel function solutions of PIII (2.1) satisfy a first-order equation of the form (7.1) for n odd,
1639
+ whereas the Bessel function solutions of deg-PV (5.2) satisfy a first-order equation of the form (7.1) for
1640
+ n even.
1641
+ The relationship between PIII (2.1) and deg-PV (1.1) is similar to that between the second Painlev´e
1642
+ equation (PII)
1643
+ d2q
1644
+ dx2 = 2q3 + xq,
1645
+ (7.2)
1646
+ with α a parameter, and Painlev´e XXXIV equation (P34)
1647
+ d2p
1648
+ dx2 = 1
1649
+ 2p
1650
+ � dp
1651
+ dx
1652
+ �2
1653
+ + 2p2 − xp − (α + 1
1654
+ 2)2
1655
+ 2p
1656
+ ,
1657
+ (7.3)
1658
+ which is equivalent to equation XXXIV of Chapter 14 in [25], in that both pairs of equations arise from
1659
+ a Hamiltonian. The Hamiltonian associated with PII (7.2) and P34 (7.3) is
1660
+ HII(q, p, z; α) = 1
1661
+ 2p2 − (q2 + 1
1662
+ 2z)p − (α + 1
1663
+ 2)q
1664
+ (7.4)
1665
+ and so
1666
+ dq
1667
+ dz = p − q2 − 1
1668
+ 2z,
1669
+ dp
1670
+ dz = 2qp + α + 1
1671
+ 2,
1672
+ (7.5)
1673
+ see [28, 44]. It is known that PII (7.2) and P34 (7.3) have special function solutions in terms of Airy
1674
+ functions, cf. [13].
1675
+ It can be shown that the Airy function solutions of PII (7.2) satisfy first-order
1676
+ equation of the form (7.1) for n odd, whereas the Airy function solutions of P34 (7.3) satisfy a first-order
1677
+ equation of the form (7.1) for n even.
1678
+ Acknowledgements
1679
+ I thank Clare Dunning and Steffen Krusch for helpful comments and illuminating discussions.
1680
+ References
1681
+ [1] P.B. Acosta-Hum´anez, M. van der Put and J. Top, Variations for some Painlev´e equations, SIGMA, 15 (2019)
1682
+ 088.
1683
+ [2] V.E. Adler, Nonlinear chains and Painlev´e equations, Physica, D73 (1994) 335–351.
1684
+ [3] S. Belmehdi and A. Ronveaux, Laguerre-Freud’s Equations for the recurrence coefficients of semi-classical
1685
+ orthogonal polynomials, J. Approx. Theory, 76 (1994) 351–368.
1686
+ [4] H. Casini, C.D. Fosco and M. Huerta, Entanglement and alpha entropies for a massive Dirac field in two
1687
+ dimensions., J. Stat. Mech. (2005) P07 007.
1688
+ [5] H. Casini and M. Huerta, Entanglement and alpha entropies for a massive scalar field in two dimensions, J.
1689
+ Stat. Mech. (2005) P12 012.
1690
+ [6] H. Casini and M. Huerta, Analytic results on the geometric entropy for free fields, J. Stat. Mech. (2008)
1691
+ P01 012.
1692
+ [7] C.V.L. Charlier, ¨Uber die Darstellung willk¨urlicher Funktionen, Ark. Mat. Astr. och Fysic, 2 (1905-6) 1–9.
1693
+ [8] J. Chazy, Sur les ´equations diff´erentielles dont l’int´egrale g´en´erale est uniforme et admet des singularities
1694
+ essentielles mobiles, C.R. Acad. Sc. Paris, 149 (1909) 563–565.
1695
+ 16
1696
+
1697
+ [9] P.A. Clarkson, The third Painlev´e equation and associated special polynomials, J. Phys. A, 36 (2003) 9507–
1698
+ 9532.
1699
+ [10] P.A. Clarkson, Special polynomials associated with rational solutions of the fifth Painlev´e equation, J. Comp.
1700
+ Appl. Math., 178 (2005) 111–129.
1701
+ [11] P.A. Clarkson, Painlev´e equations — non-linear special functions, in: Orthogonal Polynomials and Special
1702
+ Functions: Computation and Application, F. Marcell`an and W. Van Assche (Editors) Lect. Notes Math.,
1703
+ vol. 1883, pp. 331–411, Springer-Verlag, Berlin, 2006.
1704
+ [12] P.A. Clarkson, Special polynomials associated with rational solutions of the Painlev´e equations and applica-
1705
+ tions to soliton equations, Comput. Methods Funct. Theory, 6 (2006) 329–401.
1706
+ [13] P.A. Clarkson, On Airy solutions of the second Painlev´e equation, Stud. Appl. Math., 137 (2016) 93–109.
1707
+ [14] C.M. Cosgrove, Chazy’s second-degree Painlev´e equations, J. Phys. A, 39 (2006) 11955–11971.
1708
+ [15] C.M. Cosgrove and G. Scoufis, Painlev´e classification of a class of differential equations of the second order
1709
+ and second-degree, Stud. Appl. Math., 88 (1993) 25–87.
1710
+ [16] K. Diarra and F. Loray, Classification of algebraic solutions of irregular Garnier systems, Compos. Math.,
1711
+ 156 (2020) 881–907.
1712
+ [17] I. Fern´andez-Irisarri and M. Ma˜nas, Pearson equations for discrete orthogonal polynomials: II. Generalized
1713
+ Charlier, Meixner and Hahn of type I cases, arXiv:2107.02177 [math.CA].
1714
+ [18] G. Filipuk and W. Van Assche, Recurrence coefficients of generalized Charlier polynomials and the fifth
1715
+ Painlev´e equation, Proc. Amer. Math. Soc., 141 (2013) 551–562.
1716
+ [19] P.J. Forrester and N.S. Witte, Application of the τ-function theory of Painlev´e equations to random matrices:
1717
+ PV, PIII, the LUE, JUE, and CUE, Comm. Pure Appl. Math, 55 (2002) 679–727.
1718
+ [20] J. Gariel, G. Marcilhacy and N.O. Santos, Parametrization of solutions of the Lewis metric by a Painlev´e
1719
+ transcendent III, J. Math. Phys., 47 (2006) 062502.
1720
+ [21] V.I. Gromak, On the theory of Painlev´e’s equations, Diff. Eqns., 11 (1975) 285–287.
1721
+ [22] V.I. Gromak, I. Laine and S. Shimomura, Painlev´e Differential Equations in the Complex Plane, Studies in
1722
+ Math., vol. 28, de Gruyter, Berlin, New York, 2002.
1723
+ [23] M. Hisakado, Unitary matrix models and Painlev´e III, Mod. Phys. Lett., A11 (1996) 3001–3010.
1724
+ [24] M.N. Hounkonnou, C. Hounga and A. Ronveaux, Discrete semi-classical orthogonal polynomials: General-
1725
+ ized Charlier, J. Comput. Appl. Math., 114 (2000) 361–366.
1726
+ [25] E.L. Ince, Ordinary Differential Equations, Dover, New York, 1956.
1727
+ [26] M.E.H. Ismail, Classical and Quantum Orthogonal Polynomials in One Variable, Encyclopedia of Mathemat-
1728
+ ics and its Applications, vol. 98, Cambridge University Press, Cambridge, 2005.
1729
+ [27] K. Iwasaki, H. Kimura, S. Shimomura and M. Yoshida, From Gauss to Painlev´e: a Modern Theory of Special
1730
+ Functions, Aspects of Mathematics E, vol. 16. Braunschweig. Friedr. Vieweg and Sohn, 1991.
1731
+ [28] M. Jimbo and T. Miwa, Monodromy preserving deformations of linear ordinary differential equations with
1732
+ rational coefficients. II, Physica, D2 (1981) 407–448.
1733
+ [29] K. Kajiwara, On a q-difference Painlev´e III equation: II. Rational solutions, J. Nonl. Math. Phys., 10 (2003)
1734
+ 282–303.
1735
+ [30] K. Kajiwara and T. Masuda, On the Umemura polynomials for the Painlev´e III equation, Phys. Lett., 260
1736
+ (1999) 462–467.
1737
+ [31] A.V. Kitaev, C.K. Law and J.B. McLeod, Rational solutions of the fifth Painlev´e equation, Diff. Int. Eqns., 7
1738
+ (1994) 967–1000.
1739
+ [32] N.A. Lukashevich, On the theory of Painlev´e’s third equation, Diff. Eqns., 3 (1967) 994–999.
1740
+ [33] F. Lund, Example of a relativistic, completely integrable, Hamiltonian system, Phys. Rev. Lett., 38 (1977)
1741
+ 1175–1178.
1742
+ [34] F. Lund and T. Regge, Unified approach to strings and vortices with soliton solutions, Phys. Rev. D, 14 (1976)
1743
+ 1524–1535.
1744
+ [35] A.P. Magnus, On Freud’s equations for exponential weights, J. Approx. Theory, 46 (1986) 65–99.
1745
+ [36] E.L. Mansfield and H.N. Webster, On one-parameter families of Painlev´e III, Stud. Appl. Math., 101 (321–
1746
+ 341) 1998.
1747
+ [37] T. Masuda, Classical transcendental solutions of the Painlev´e equations and their degeneration, Tohoku
1748
+ Math. J., 56 (2004) 467–490.
1749
+ [38] T. Masuda, The anti-self-dual Yang-Mills equation and the Painlev´e III equation, J. Phys. A, 40 (2007) 14433–
1750
+ 14445.
1751
+ [39] A.E. Milne, P.A. Clarkson and A.P. Bassom, B¨acklund transformationsand solution hierarchies for the third
1752
+ Painlev´e equation, Stud. Appl. Math., 98 (1997) 139–194.
1753
+ [40] Y. Murata, Classical solutions of the third Painlev´e equations, Nagoya Math. J., 139 (1995) 37–65.
1754
+ 17
1755
+
1756
+ [41] F.W. Nijhoff and V.G. Papageorgiou, Similarity reductions of integrable lattices and discrete analogues of the
1757
+ Painlev´e II equation, Phys. Lett., 153A (1991) 337–344.
1758
+ [42] Y. Ohyama and S. Okumura, A coalescent diagram of the Painlev´e equations from the viewpoint of isomon-
1759
+ odromic deformations, J. Phys. A, 39 (2006) 12129–12151.
1760
+ [43] Y. Ohyama and S. Okumura, Fuchs’ problem of the Painlev´e equations from the first to the fifth, in: Algebraic
1761
+ and geometric aspects of integrable systems and random matrices, A. Dzhamay, K. Muruno and V.U. Pierce
1762
+ (Editors) Contemp. Math., vol. 593, pp. 163–178, Amer. Math. Soc., Providence, RI, 2013. [math.CA/0512243].
1763
+ [44] K. Okamoto, Polynomial Hamiltonians associated with Painlev´e equations. I & II, Proc. Japan Acad. Ser. A
1764
+ Math. Sci., 56 (1980) 264–268, 367–371.
1765
+ [45] K. Okamoto, Studies on the Painlev´e equations. II. Fifth Painlev´e equation PV, Japan. J. Math., 13 (1987)
1766
+ 47–76.
1767
+ [46] K. Okamoto, Studies on the Painlev´e equations IV. Third Painlev´e equation PIII, Funkcial. Ekvac., 30 (1987)
1768
+ 305–332.
1769
+ [47] K. Okamoto and Y. Ohyama, Mathematical works of Hiroshi Umemura, Ann. Fac. Sci. Toulouse Math. (6),
1770
+ 29 (2020) 1053–1062.
1771
+ [48] F.W.J. Olver, A.B. Olde Daalhuis, D.W. Lozier, B.I. Schneider, R.F. Boisvert, C.W. Clark, B.R. Miller,
1772
+ B.V. Saunders, H.S. Cohl, and M.A. McClain (Editors), NIST Digital Library of Mathematical Functions,
1773
+ http://dlmf.nist.gov/, Release 1.1.8 (December 15, 2022).
1774
+ [49] V. Periwal and D. Shevitz, Unitary-matrix models as exactly solvable string theories, Phys. Rev. Lett., 64
1775
+ (1990) 1326–1329.
1776
+ [50] K. Pohlmeyer, Integrable Hamiltonian systems and interactions through quadratic constraints, Commun.
1777
+ Math. Phys., 46 (1976) 207–221.
1778
+ [51] C. Smet and W. Van Assche, Orthogonal polynomials on a bi-lattice, Constr. Approx., 36 (2012) 215–242.
1779
+ [52] C.A. Tracy and H. Widom, Random unitary matrices, permutations and Painlev´e, Commun. Math. Phys.,
1780
+ 207 (1999) 665–685.
1781
+ [53] H. Umemura, Painlev´e equations and classical functions, Sugaku Expositions, 11 (1998) 77–100.
1782
+ [54] H. Umemura, Painlev´e equations in the past 100 Years, A.M.S. Translations, 204 (2001) 81–110.
1783
+ [55] H. Umemura, Special polynomials associated with the Painlev´e equations I, Ann. Fac. Sci. Toulouse Math.
1784
+ (6), 29 (2020) 1063–1089.
1785
+ [56] W. Van Assche, Orthogonal Polynomials and Painlev´e Equations, Australian Mathematical Society Lecture
1786
+ Series. Cambridge. Cambridge University Press, 2018.
1787
+ [57] W. Van Assche and M. Foupouagnigni, Analysis of non-linear recurrence relations for the recurrence coeffi-
1788
+ cients of generalized Charlier polynomials, J. Nonlinear Math. Phys., 10(suppl. 2) (2003) 231–237.
1789
+ 18
1790
+
D9AzT4oBgHgl3EQfwv5m/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
D9E2T4oBgHgl3EQfSgcI/content/tmp_files/2301.03792v1.pdf.txt ADDED
@@ -0,0 +1,775 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ arXiv:2301.03792v1 [math.GT] 10 Jan 2023
2
+ A G-FAMILY OF SINGQUANDLES AND INVARIANTS OF DICHROMATIC
3
+ SINGULAR LINKS
4
+ MOHD IBRAHIM SHEIKH, MOHAMED ELHAMDADI, AND DANISH ALI
5
+ ABSTRACT. We introduce and investigate dichromatic singular links. We also construct G-Family
6
+ of singquandles and use them to define counting invariants for unoriented dichromatic singular links.
7
+ We provide some examples to show that these invariants distinguish some dichromatic singular links.
8
+ CONTENTS
9
+ 1.
10
+ Introduction
11
+ 1
12
+ 2.
13
+ Singular links, Singquandles and Dichromatic Links
14
+ 2
15
+ 3.
16
+ Dichromatic Singular Links
17
+ 5
18
+ 4.
19
+ G-Family of Singquandles (Disingquandles)
20
+ 6
21
+ 5.
22
+ Computable Invariants for Unoriented Dichromatic Singular Links
23
+ 10
24
+ Acknowledgement
25
+ 14
26
+ References
27
+ 15
28
+ Mathematics Subject Classifications (2020): 57M25, 57M27.
29
+ Key words and Phrases: Knot; Link; Singular knot; Singular link; Dichromatic link; Dichromatic
30
+ singular link; Quandle; Singquandle; Disingquandle; Disingquandle counting invariant.
31
+ 1. INTRODUCTION
32
+ A knot is a simple closed curve in three dimensional space R3 and a disjoint union of two or
33
+ more knots forms a link with two or more components [8]. Knots and links are categorised in many
34
+ ways. One way is to use the crossing type as a tool to define a knot or link type. Classical, virtual
35
+ and singular knots and links serve as examples as they are all recognised by the type of crossing
36
+ they contain. The other way to define link types is by labelling the components of a classical link.
37
+ Dichromatic links are defined by using this technique as their components are either labelled by
38
+ “1” or “2” [1, 2, 10, 11, 14]. A singular link is a link with at least one singular crossing. In this
39
+ paper we use such labelling technique for singular links and define a new type of links which we
40
+ call dichromatic singular links.
41
+ A quandle is an algebraic structure satisfying some axioms that result from the Reidemeister
42
+ moves for oriented classical knots and links. If furthermore all right multiplications by fixed ele-
43
+ ments of the quandle are involutions then such structures are called involutory quandles or Kei’s
44
+ They are used to investigate unoriented knots and links. Quandles were independently introduced
45
+ by Joyce and Matveev [13, 16]. Since then they have been used to construct invariants of knots
46
+ and links [4, 6, 17]. Quandles have been also used to define new algebraic systems by taking a
47
+ 1
48
+
49
+ family of quandles at a time. Such systems are called G-Family of quandles and this notion was
50
+ introduced in 2013 by Ishii, Iwakiri, Jang and Oshiro [12]. A G-Family of quandles were used to
51
+ define invariants for handlebody-knots. Also in [15] Lee and Sheikh used Z2-Family of quandles
52
+ to construct algebraic invariants for oriented dichromatic links.
53
+ In this paper, we introduce the notions of G-Family of singquandles and dichromatic singular
54
+ links. A dichromatic singular link is an n component singular link with each of its component
55
+ labelled as “1” or “2”. A singquandle is an algebraic system whose axioms are motivated by
56
+ Reidemeister moves of unoriented singular knots. By taking a family of such algebaraic systems
57
+ (Singquandles), we define a new algebraic system which we call G-Family of singquandles or
58
+ disingquandle. The axioms of the latter are motivated by generalized Reidemeister moves of un-
59
+ oriented dichromatic singular links. We discuss various examples and some properties of G-Family
60
+ of singquandles, and also show that a G-Family of singquandles X enables us to distinguish unori-
61
+ ented dichromatic singular links by computing their sets of all X-colorings and proving that these
62
+ sets are different when their arcs are colored by the elements of X.
63
+ This paper is organized as follows. Section 2 reviews some preliminaries about singular links,
64
+ singquandles as well as dichromatic links and their generalized Reidemeister moves. In Section 3
65
+ we introduce the notion of dichromatic singular links with some typical examples of unoriented
66
+ dichromatic singular links. Section 4 introduces the notion of G-Family of singquandles (dis-
67
+ ingquandles) with some typical examples of G-Family of singquandles. Section 5 discusses how
68
+ G-Family of singquandles is related to unoriented dichromatic singular links and develop com-
69
+ putable invariants for unoriented dichromatic singular links. We discuss some examples which
70
+ show how the invariants distinguish unoriented dichromatic singular links, and especially how
71
+ they detect the change of component labelings.
72
+ 2. SINGULAR LINKS, SINGQUANDLES AND DICHROMATIC LINKS
73
+ In this section we review some preliminaries about singular links, singquandles and dichromatic
74
+ links. Most of the terminologies of this section can be found in [5, 9, 15]. We begin with the
75
+ definition of a singular link.
76
+ Definition 2.1. A singular link in S3 is the image of a smooth immersion of n circles in S3 that has
77
+ finitely many double points, called singular points.
78
+ A singular link in R3 is represented by a singular link diagram in the plane R2, which is a
79
+ classical link diagram with one or more singularities. A singularity is a rigid vertex where a link is
80
+ glued to itself. Figure 1 gives two examples of singular links.
81
+ FIGURE 1. Singular Links
82
+ 2
83
+
84
+ Two singular links L� and L� are isotopy equivalent if one can be obtained from the other by a
85
+ finite sequence generalized Reidemeister moves for singular links as shown in the following figure.
86
+ Let D� and D� be two singular link diagrams in R2 representing L� and L�, respectively. Then
87
+ L� and L� are equivalent if and only if D� and D� can be transformed into each other by a finite
88
+ sequence of classical and singular Reidemeister moves shown in Figure 2.
89
+ FIGURE 2. Classical and Singular Reidemeister Moves
90
+ Definition 2.2. [5] Let (X, ∗) be an involutive quandle. Let R1 and R2 be two maps from X × X
91
+ to X. The quadruple (X, ∗, R1, R2) is called a singquandle if the following axioms are satisfied
92
+ (2.2.1)
93
+ x = R1(y, R2(x, y)) = R2(R2(x, y), R1(x, y)),
94
+ (2.2.2)
95
+ y = R2(R2(x, y), x) = R1(R2(x, y), R1(x, y)),
96
+ (2.2.3)
97
+ R(x, y) = (R1(y, R2(x, y)), R2(R2(x, y), x)),
98
+ (2.2.4)
99
+ (y ∗ z) ∗ R2(x, z) = (y ∗ x) ∗ R1(x, z),
100
+ (2.2.5)
101
+ R1(x, y) = R2(y ∗ x, x),
102
+ (2.2.6)
103
+ R2(x, y) = R1(y ∗ x, x) ∗ R2(y ∗ x, x),
104
+ (2.2.7)
105
+ R1(x ∗ y, z) ∗ y = R1(x, z ∗ y),
106
+ 3
107
+
108
+ (2.2.8)
109
+ R2(x ∗ y, z) = R2(x, z ∗ y) ∗ y.
110
+ We remind the reader that the singquandle axioms come from the generalized Reidemeister
111
+ moves for unoriented singular knots. Singquandles were introduced as a ramification of quandles
112
+ with the purpose of studying singular links, see for example [4,5,17].
113
+ The following are few typical examples of singquandles.
114
+ • For an involutive quandle (X, ∗) with x ∗ y = 2y − x and X = Zn, the quadruple
115
+ (X, ∗, R1, R2) forms a singquandle if and only if the following conditions are satisfied:
116
+ (1) R2(x, y) = R1(x, y) + y − x,
117
+ (2) R1(x, y) = R1(2x − y, x) + y − x,
118
+ (3) R1(x, 2y − z) = 2y − R1(2y − x, z),
119
+ (4) R2(2y − x, z) = 2y − R2(x, 2x − z).
120
+ • For an involutive quandle (X, ∗) where X is a group G and x ∗ y = yx−1y, the quadruple
121
+ (X, ∗, R1, R2) forms a singquandle if and only if the following conditions are satisfied:
122
+ (1) R2(x, z)z−1yz−1R2(x, z) = R1(x, z)x−1yx−1R1(x, z),
123
+ (2) R1(x, y) = R2(xy−1x, x),
124
+ (3) R2(x, y) = R2(xy−1x, x)[R1(xy−1x, x)]−1R2(xy−1x, x),
125
+ (4) y[R1(yx−1y, z)]−1y = R1(x, yz−1y),
126
+ (5) R1(yx−1y, z) = y[R2(x, yz−1y)]−1y.
127
+ Definition 2.3. For a positive integer n ≥ 1. A dichromatic link is a smooth imbedding of n circles
128
+ in R3 such that each component is labeled as “1” or “2”.
129
+ In R2 every dichromatic link is represented by a dichromatic link diagram which is a classical link
130
+ diagram with each component labelled either “1” or “2”. For example, see Figure 3.
131
+ 1
132
+ 1
133
+ 2
134
+ 2
135
+ FIGURE 3. Dichromatic Links
136
+ Two dichromatic links L� and L� are isotopy equivalent if one can be obtained from the other
137
+ by a finite sequence of generalized Reidemeister moves for the dichromatic links as shown in
138
+ the figure 4. Let D� and D� be two dichromatic link diagrams in R2 representing L� and L�,
139
+ respectively. Then L� and L� are equivalent if and only D� and D� can be transformed into each
140
+ other by a finite sequence of generalized Reidemeister moves shown in the following Figure 4.
141
+ 4
142
+
143
+ i
144
+ i
145
+ i
146
+ j
147
+ j
148
+ i
149
+ i
150
+ k
151
+ j
152
+ i
153
+ k
154
+ j
155
+ FIGURE 4. Generalized Reidemeister Moves for Dichromatic Links
156
+ 3. DICHROMATIC SINGULAR LINKS
157
+ This section is devoted to dichromatic singular links which is a generalization of singular links.
158
+ To generate a dichromatic singular link we label a singular link’s components with “1” or “2”.
159
+ Thus We have the following definition.
160
+ Definition 3.1. A singular link L in R3 whose each component is colored (labelled) by either “1”
161
+ or “2” is called a dichromatic singular link.
162
+ A dichromatic singular link L in R3 is represented by a dichromatic singular link diagram D in
163
+ R2 in which each component is labelled “1” or “2”. Figure 5 shows two examples of unoriented
164
+ dichromatic singular link diagrams.
165
+ 1
166
+ 2
167
+ 1
168
+ 2
169
+ FIGURE 5. Dichromatic Singular Links
170
+ Two dichromatic singular links L� and L� in R3 are ambient isotopic if there exists a self home-
171
+ omorphism h : R3 → R3 that takes one link to the other and preserves the singularities as well
172
+ as the labels “1”, “2” such that h(L�) = L�. Thus two singular dichromatic links L� and L� are
173
+ equivalent if one can be obtained from the other by a finite sequence of generalized dichromatic
174
+ singular Reidemeister moves preserving the label of each component as shown in the Figure 6. Let
175
+ D� and D� be two dichromatic singular link diagrams in R2 representing L� and L�, respectively.
176
+ Then L� and L� are equivalent if and only if D� and D� can be transformed into each other by
177
+ a finite sequence of generalized dichromatic singular Reidemeister moves shown in the following
178
+ Figure 6 where i, j, k ∈ {1, 2}.
179
+ A dichromatic singular link with n components is called as an n-component dichromatic singular
180
+ link. Thus an n-component dichromatic singular link in R3 can be defined as L = K1 ∪ · · · ∪ Kn.
181
+ 5
182
+
183
+ i
184
+ k
185
+ k
186
+ j
187
+ i
188
+ i
189
+ j
190
+ j
191
+ i
192
+ j
193
+ i
194
+ k
195
+ k
196
+ j
197
+ i
198
+ j
199
+ i
200
+ i
201
+ i
202
+ j
203
+ j
204
+ i
205
+ i
206
+ k
207
+ j
208
+ i
209
+ k
210
+ j
211
+ FIGURE 6. Regular Dichromatic Reidemeister Moves RI, RII and RIII on the
212
+ top and Dichromatic Singular Reidemeister Moves RIV a, RIV b and RV in the
213
+ middle and on the bottom.
214
+ Taking n = 2, we obtain 2-component dichromatic singular links. Some 2-component unoriented
215
+ dichromatic singular link diagrams (see p 814 of [18]) are shown in Figure 12.
216
+ Proposition 3.2. Let L� and L� be two unoriented dichromatic singular links in R3 and let D�
217
+ and D� be two unoriented dichromatic singular link diagrams in R2 representing L� and L�,
218
+ respectively. Then L� and L� are equivalent if and only if D� and D� are transformed into each
219
+ other by a finite sequence of generalized Reidemeister moves for unoriented dichromatic singular
220
+ links which preserve the singularities and the label of each component as shown in the Fig. 6
221
+ where i, j, k ∈ {1, 2} and ambient isotopies of R2.
222
+ 4. G-FAMILY OF SINGQUANDLES (DISINGQUANDLES)
223
+ Before introducing the notion of G-Family of Singquandles, we first recall the definition of
224
+ G-family of quandles from [12].
225
+ Definition 4.1. Given a group G and a set X, a G-family of quandles, denoted by (G, X), is
226
+ a choice of quandle operation ∗g on the set X for each element g ∈ G such that the following
227
+ axioms are satisfied
228
+ (1) For all g ∈ G and for all x ∈ X, x ∗g x = x,
229
+ (2) For all g, h ∈ G and for all x, y ∈ X, (x ∗g y) ∗h y = x ∗gh y,
230
+ (3) For all x, y ∈ X, x ∗e x = x, where e is the identity element of G,
231
+ 6
232
+
233
+ (4) For all x, y, z ∈ X, (x ∗g y) ∗h z = (x ∗h z) ∗h−1gh (y ∗h z)
234
+ The following are two examples of G-families of quandles.
235
+ • For any group G and any set X, defining x ∗g y = x for all x, y ∈ X and all g ∈ G. This
236
+ gives a G-family of quandles called the trivial G-family of quandles.
237
+ • Let (X, ∗) be a quandle of cyclic type [19] with cardinality n. Let Rx denotes the right
238
+ multiplication by x, thus by definition Rx
239
+ (n−1) is the identity map. Then define x ∗i y =
240
+ Ry
241
+ i(x) then it is shown in Proposition 2.3 of [12] that (Z, X) is a Z-family of quandles and
242
+ also Z(n−1)-family of quandles.
243
+ A G-family of quandles (G, X) induces a quandle operation on the set G × X by
244
+ (g, x) ∗ (h, y) = (h−1gx, x ∗h y).
245
+ The notion of G-family of quandles was introduced by Ishii, Iwakiri, Jang and Oshiro in 2013
246
+ in [12] in order to produce invariants of handlebody knots. They defined coloring invariants and
247
+ cocycle invariants of handlebody knots. They used these invariants to detect chirality of some han-
248
+ dlebody knots. Later in 2015, Ishii independently studied the notion of G-family of quandles in
249
+ connection with the multiple conjugation quandle and showed that the later one can be obtained
250
+ from the first one. In 2017 and 2018 Ishii, Nelson and Ishii, Iwakiri, Kamada, Kim, Matsuzaki, Os-
251
+ hiro respectively, used this work and introduced the notions of partially multiplicative biquandles
252
+ and multiple conjugation biquandle. In 2021 Lee and Sheikh jointly used G-family of quandles
253
+ to construct algebraic invariants for oriented dichromatic links [15]. We introduce the following
254
+ definition.
255
+ Definition 4.2. Let X be a set equipped with two binary operations ∗1 and ∗2 such that both
256
+ (X, ∗1), (X, ∗2) are involutive quandles. Let R1, R2 be two maps from X × X to X such that the
257
+ quadruples (X, ∗1, R1, R2) and (X, ∗2, R1, R2) are singquandles. Then the quintuple (X, ∗1, ∗2, R1, R2)
258
+ is called a disingquandle or Z2-family of singquandles if the following axioms are satisfied
259
+ (4.2.1)
260
+ (x ∗1 y) ∗2 z = (x ∗2 z) ∗1 (y ∗2 z),
261
+ (4.2.2)
262
+ (x ∗2 y) ∗1 z = (x ∗1 z) ∗2 (y ∗1 z),
263
+ (4.2.3)
264
+ (y ∗1 z) ∗2 R2(x, z) = (y ∗2 x) ∗1 R1(x, z),
265
+ (4.2.4)
266
+ (y ∗2 z) ∗1 R2(x, z) = (y ∗1 x) ∗2 R1(x, z),
267
+ (4.2.5)
268
+ R2(x, y) = R1(y ∗1 x, x) ∗2 R2(y ∗1 x, x),
269
+ (4.2.6)
270
+ R2(x, y) = R1(y ∗2 x, x) ∗1 R2(y ∗2 x, x),
271
+ The above axioms of a disingquandle come from the generalized dichromatic singular Reide-
272
+ meister moves shown in Figure 6 when we take the coloring rule shown in Figure 7 under consid-
273
+ eration.
274
+ 7
275
+
276
+ i
277
+ i
278
+ j
279
+ i/j
280
+ j/i
281
+ j
282
+ i
283
+ y
284
+ x
285
+ y
286
+ x
287
+ y
288
+ x
289
+ x
290
+ x
291
+ y
292
+ x
293
+ *
294
+ j x
295
+ y*
296
+ R ( )
297
+ x
298
+ 1
299
+ 2
300
+ y,
301
+ R ( )
302
+ x y,
303
+ FIGURE 7. Coloring by a disingquandle
304
+ The following lemma is motivated by the above construction.
305
+ Lemma 4.3. The set of colorings of a dichromatic singular link by a disingquandle does not change
306
+ by the dichromatic singular Reidemeister moves shown in Figure 6.
307
+ Proof. As in the case of classical and singular knot theories, there is one to one correspondence
308
+ between colorings before and after each of the dichromatic singular Reidemeister moves. The
309
+ invariance follows directly from the equations 4.2.1, 4.2.2, 4.2.3, 4.2.4, 4.2.5 and 4.2.6 given in
310
+ Definition 4.2.
311
+
312
+ Example 4.4. Let (X, ∗1, R1, R2) and (X, ∗2, R1, R2) be two singquandles such that such that
313
+ x ∗1 y = x = x ∗2 y and R1(x, y) = R2(x, y), then (X, ∗1, ∗2, R1, R2) forms a disingquandle.
314
+ Example 4.5. Let (X, ∗1, R1, R2) and (X, ∗2, R1, R2) be two singquandles. If for all x, y ∈ X
315
+ we have x ∗1 y = x ∗2 y then (X, ∗1, ∗2, R1, R2) forms a disingquandle.
316
+ Now Example 4.5 combined with Proposition 3.6 in [5] gives the following example.
317
+ Example 4.6. Let Λ = Z[t, B]/(t2 − 1, B(1 + t), t − (1 − B)2) and X be an Λ-module. Define
318
+ x ∗1 y = tx + (1 − t)y, R1(x, y) = (1 − t − b)x + (t + b)y and R2(x, y) = (1 − B)x + By, then
319
+ by setting ∗2 = ∗1, then one obtains that (X, ∗1, ∗2, R1, R2) forms a disingquandle.
320
+ Example 4.7. Let X be a module over Λ = Z[t]. Define x∗1y = x∗2y = tx+(1−t)y, R1(x, y) =
321
+ (1 − t − B)x + (t + B)y and R2(x, y) = (1 − B)x + By. Setting t = −1 and X = Z7, then the
322
+ quintuple (X, ∗1, ∗2, R1, R2) forms a disingquandle if B = 4 or if B = 5.
323
+ This example can be generalized to Zp, where p is a prime as follows.
324
+ Example 4.8. Let p be an odd prime and let B ∈ Zp. Consider Zp with x ∗1 y = x ∗2 y =
325
+ −x + 2y, R1(x, y) = (2 − B)x + (−1 + B)y and R2(x, y) = (1 − B)x + By. Let ζ be a
326
+ primitive root of unity in Zp so that ζ
327
+ p−1
328
+ 2
329
+ = −1. By choosing 1 − B = ζ
330
+ p−1
331
+ 2
332
+ we obtain that
333
+ (Zp, ∗1, ∗2, R1, R2) forms a disingquandle.
334
+ Example 4.9. Let X = G be a multiplicative group with the involutive quandle operations
335
+ x ∗1 y = x ∗2 y = yx−1y (core quandle on G), then a direct computation gives the fact that
336
+ the quintuple (X, ∗1, ∗2, R1, R2) forms a disingquandle if and only if R1 and R2 satisfies the
337
+ following equations:
338
+ (5.1)
339
+ R2(x, z)z−1yz−1R2(x, z) = R1(x, z)x−1yx−1R1(x, z),
340
+ (5.2)
341
+ R2(x, y) = R2(xy−1x, x)[R1(xy−1x, x)]−1R2(xy−1x, x),
342
+ 8
343
+
344
+ A straightforward computation gives the following solution
345
+ R1(x, y) = x and R2(x, y) = y, for all x, y, z ∈ G.
346
+ Now assume that G is an abelian group without 2-torsion, so that x ∗ y = −x + 2y, then
347
+ (X, ∗1, ∗2, R1, R2) forms a disingquandle if and only if R2(x, y) = R1(x, y) + y − x, where R1
348
+ satisfies the identity R1(x, y) = R1(−x + 2y, x) + y − x. For example for any integer m, the map
349
+ R1(x, y) = mx + (2m + 1)y give a solution. Thus we have a family of solutions parametrized by
350
+ the integer m:
351
+ R1(x, y) = mx + (2m + 1)y, R2(x, y) = (m − 1)x + 2(m + 1)y.
352
+ Definition 4.10. A map f : X → Y is called a homomorphism of disingquandle (X, ∗1, ∗2, R1, R2)
353
+ and (Y, ∗′
354
+ 1, ∗′
355
+ 2, R′
356
+ 1, R′
357
+ 2) if the following conditions are satisfied for all x, y, z ∈ X
358
+ (i) f(x ∗1 y) = f(x) ∗′
359
+ 1 f(y),
360
+ (ii) f(x ∗2 y) = f(x) ∗′
361
+ 2 f(y),
362
+ (iii) f(R1(x, y)) = R′
363
+ 1(f(x), f(y)),
364
+ (iv) f(R2(x, y)) = R′
365
+ 1(f(x), f(y)).
366
+ If a homomorphism of disingquandle is bijective, then it is called an isomorphism of disingquan-
367
+ dle. We say that two Z2-families of singquandles are isomorphic if there exists an ismorphism of
368
+ disingquandle between them.
369
+ Definition 4.11. Let (X, ∗1, ∗2, R1, R2) be a disingquandle. A subset Y ⊂ X is called a sub-
370
+ disingquandle if (Y, ∗1, ∗2, R1, R2) is itself a disingquandle.
371
+ Example 4.12. We use Example 4.9 to get the following 2 examples:
372
+ • Let X = Z9 be the dihedral quandle with x ∗ y = −x + 2y, R1(x, y) = mx + (2m +
373
+ 1)y, R2(x, y) = (m−1)x+ 2(m+ 1)y.. Then (Y, ∗1, ∗2, R1, R2) is itself a disingquandle
374
+ with Y = Z3.
375
+ • Let X = Z25 be the dihedral quandle with x ∗ y = −x + 2y, R1(x, y) = mx + (2m +
376
+ 1)y, R2(x, y) = (m−1)x+ 2(m+ 1)y.. Then (Y, ∗1, ∗2, R1, R2) is itself a disingquandle
377
+ with Y = Z5.
378
+ Given a homomorphism of disingquandles, we obtain the following lemma.
379
+ Lemma 4.13. The image Im(f) of any homomorphism of disingquandle f defined from (X, ∗1, ∗2, R1, R2)
380
+ to (Y, ∗′
381
+ 1, ∗′
382
+ 2, R′
383
+ 1, R′
384
+ 2) is always a sub-disingquandle.
385
+ Proof. Given that f : (X, ∗1, ∗2, R1, R2) → (Y, ∗′
386
+ 1, ∗′
387
+ 2, R′
388
+ 1, R′
389
+ 2) is a homomorphism. Then the
390
+ equations (i), (ii), (iii) and (iv) of Definition 4.10 imply that Im(f) is closed under ∗1, ∗2, R1 and
391
+ R2. Then the axioms of disingquandle are satisfied in Y . Hence they are automatically satisfied in
392
+ Im(f). This ends the proof of the lemma.
393
+
394
+ Now we introduce the notion of fundamental disingquandle of an unoriented dichromatic sin-
395
+ gular link and provide an illustrative example. Let D be a diagram of an unoriented dichromatic
396
+ singular link L in R2. We define the fundamental disingquandle of D, denoted by DSQ(D), as
397
+ the set of equivalence classes of disingquandle words W-DSQ(D) under the equivalence relation
398
+ generated by the axioms of disingquandle and the crossing relations shown in Figure 7, where W-
399
+ DSQ(D) are defined by taking a set of generators X = {x1, x2, x3, ....., xn} which corresponds
400
+ bijectively with the semi arcs in D, recursively by the following two rules:
401
+ 9
402
+
403
+ (1) X ⊂ W-DSQ(D),
404
+ (2) If x, y ∈ W-DSQ(D), then
405
+ x ∗1 y, x ∗2 y, R1(x, y), R2(x, y) ∈ W-DSQ(D).
406
+ Example 4.14. Consider the following unoriented dichromatic singular link L.
407
+ 1
408
+ 2
409
+ x
410
+ z
411
+ u
412
+ v
413
+ y
414
+ FIGURE 8. Fundamental Disingquandle of Unoriented Dichromatic Singular Links
415
+ The fundamental disingquandle of L is given by
416
+ DSQ(L) = ⟨x, y, z, u, v| z = x ∗2 y; u = y ∗1 z; v = z ∗2 u; x = R1(u, v); y = R2(u, v)⟩.
417
+ This presentation can be simplified to the following presentation of DSQ(L)
418
+ ⟨x, y| x = R1(y∗1(x∗2y), (x∗2y)∗2(y∗1(x∗2y))); y = R2(y∗1(x∗2y), (x∗2y)∗2(y∗1(x∗2y)))⟩.
419
+ 5. COMPUTABLE INVARIANTS FOR UNORIENTED DICHROMATIC SINGULAR LINKS
420
+ Let D be an unoriented dichromatic singular link diagram and let A(D) denote the set of all
421
+ arcs of D. Let (X, ∗1, ∗2, R1, R2) be a disingquandle. A disingquandle coloring of D by X, or
422
+ simply disingquandle X-coloring of D, is a map C : A(D) → X such that at every classical and
423
+ singular crossing, the relations depicted in Figure 7 hold. The disingquandle element C(s) is called
424
+ a color of the arc s and the pair (D, C) is called the X-colored unoriented dichromatic singular
425
+ link diagram by C. The set of all disingquandle X-colorings of D is denoted by Coldsq
426
+ X (D). Then
427
+ we have the following:
428
+ Lemma 5.1. Let D and D′ be two unoriented dichromatic singular link diagramss in R2 that
429
+ can be transformed into each other by unoriented generalized dichromatic singular Reidemeis-
430
+ ter moves as shown in the Figure 6. Then for any finite disingquandle X, there is a one-to-one
431
+ correspondence between Coldsq
432
+ X (D) and Coldsq
433
+ X (D′).
434
+ 10
435
+
436
+ Proof. It suffices to prove the assertion for the case that D′ is obtained from D by a single an
437
+ unoriented generalized dichromatic singular Reidemeister move. Let E be an open disk in R2
438
+ where the unoriented generalized dichromatic singular Reidemeister move under consideration is
439
+ applied. Then D∩(R2−E) = D′∩(R2−E). Now let C be a disingquandle X-coloring of D. Since
440
+ (X, ∗1, R1, R2) and (X, ∗2, R1, R2) are both singquandles by the disingquandle definition 4.2, it
441
+ is obviously seen from the Figure 6 that the restriction of C to D ∩(R2 −E)(= D′ ∩(R2 −E)) can
442
+ be extended to a unique disingquandle X-coloring of D′ for unoriented generalized dichromatic
443
+ singular Reidemeister moves RI, RII and RIII. Also, using the disingquandle axioms 4.2.1 to
444
+ 4.2.6, it is easily seen from the Figure 6 that the restriction of C to D ∩ (R2 − E)(= D′ ∩ (R2 −
445
+ E)) can be extended to a unique disingquandle X-coloring of D′ for an unoriented generalized
446
+ dichromatic Reidemeister moves RIV a, RIV b and RV . This completes the proof.
447
+
448
+ In an X-colored unoriented dichromatic singular link diagram (D, C), we think of elements
449
+ of a disingquandle X as labels for the arcs in D with different operations at crossings as shown
450
+ in Figure 7. Then it is seen from Lemma 5.1 that the disingquandle axioms of Definition 4.2
451
+ are transcriptions of a generating set of unoriented generalized Reidemeister moves for unoriented
452
+ dichromatic singular links which are sufficient to generate any other unoriented generalized dichro-
453
+ matic singular Reidemeister moves. That is, the axioms 4.2.1 and 4.2.2 come from the unoriented
454
+ generalized dichromatic singular Reidemeister move RIV a, the axioms 2.2.3 and 2.2.4 come from
455
+ the unoriented generalized dichromatic singular Reidemeister move RIV b and the axioms 2.2.5
456
+ and 2.2.6 come from the unoriented generalized dichromatic singular Reidemeister move RV as
457
+ seen in Figure 6.
458
+ Theorem 5.2. Let L be an unoriented dichromatic singular link in R3 and let D be a diagram of
459
+ L. Then for any finite disingquandle X, the cardinality ♯Coldsq
460
+ X (L) is an invariant of L.
461
+ Proof. Let D′ be any other unoriented dichromatic singular link diagram of L obtained from D
462
+ by applying a finite number of unoriented generalized dichromatic singular Reidemeister moves.
463
+ Then it is direct from Lemma 5.1 that ♯Coldsq
464
+ X (D′) = ♯Coldsq
465
+ X (D). This completes the proof.
466
+
467
+ If X is a finite disingquandle, we call the cardinality ♯Coldsq
468
+ X (D) the disingquandle X-coloring
469
+ number or the disingquandle counting invariant of L, and denote it by Zdsq
470
+ X (L), i.e., Zdsq
471
+ X (L) =
472
+ ♯Coldsq
473
+ X (D).
474
+ Theorem 5.3. Let L be an unoriented dichromatic singular link and let X be a disingquandle.
475
+ Then there is a one-to-one correspondence between Coldsq
476
+ X (L) and Hom(DSQ(L), X). Conse-
477
+ quently, Zdsq
478
+ X (L) = ♯Hom(DSQ(L), X).
479
+ Proof. Since the disingquandle X-colorings of L generate the fundamental disingquandle DSQ(L)
480
+ of a link L which is generated by its arc labels. Also each arc of L is assigned an element of X,
481
+ for a disingquandle X-coloring of L, so we can associate each coloring a map f : DSQ(L) → X
482
+ where if an arc is labelled a in the fundamental disingquandle and is assigned the color x ∈ X,
483
+ then f(a) = x. This completes the proof.
484
+
485
+ Now we give an example.
486
+ Example 5.4. Now, we give an explicit example of three unoriented dichromatic singular links L1,
487
+ L2 and L3 and show that the coloring invariant distinguishes them from each other. Consider the
488
+ singquandle (X, ∗, R1, R2), where X = Z6, x∗1 y = x∗2 y = −x+2y = x∗y, R1(x, y) = x+3,
489
+ 11
490
+
491
+ and R2(x, y) = 3x2 + 3x + y + 3 (see page 9 of [7]). By checking directly that the equations of
492
+ Definition 4.2 hold we obtain that the quintuple (X, ∗1, ∗2, R1, R2) form a disingquandle. Now
493
+ coloring the two top arcs of link L1 by x and y as in the figure 9 below gives that the coloring
494
+ equations are:
495
+ x = R1(R1(x, y), R2(x, y))
496
+ and
497
+ y = R2(R1(x, y), R2(x, y)).
498
+ One then gets the system,
499
+
500
+ x = 3 + 3 + x,
501
+ y = 3 + 3(3 + x) + 3(3 + x)2 + (3 + 3x + 3x2).
502
+ R ( )
503
+ x
504
+ 1
505
+ y,
506
+ 2
507
+ R ( )
508
+ x y,
509
+ x
510
+ y
511
+ 1
512
+ 2
513
+ FIGURE 9. Unoriented Dichromatic Singular Link(L1)
514
+ Any pair (x, y) gives a solution to this system over Z6 and thus the set Coldsq
515
+ X (L1) is equal to Z2
516
+ 6.
517
+ Now coloring the link L2 as in the figure 10 below gives that the coloring equations are:
518
+ R1(R1(x, y), x ∗ R1(x, y)) = R2(x, y) ∗ y
519
+ and
520
+ R2(R1(x, y), x ∗ R1(x, y)) = y.
521
+ One then obtain that the solution is given by y = 3x2 + 4x + 3, thus the Coldsq
522
+ X (L2) is
523
+ R ( )
524
+ x
525
+ 1
526
+ y,
527
+ R ( )
528
+ x
529
+ 1
530
+ y,
531
+ 2
532
+ R ( )
533
+ x y,
534
+ x
535
+ x
536
+ y
537
+ *
538
+ 1
539
+ 2
540
+ FIGURE 10. Unoriented Dichromatic Singular Link(L2)
541
+ 12
542
+
543
+ {(0, 3), (1, 4), (2, 5), (3, 0), (4, 1), (5, 2)}.
544
+ Now we consider the link L3 (dichromatic singular Whitehead) as in the following figure 11.
545
+ R ( )
546
+ x
547
+ 1
548
+ y,
549
+ R ( )
550
+ x
551
+ 1
552
+ y,
553
+ 2
554
+ R ( )
555
+ x y,
556
+ x
557
+ y
558
+ y
559
+ y
560
+ *
561
+ 2
562
+ R (
563
+ u x u)
564
+ ,
565
+ *
566
+ x u
567
+ *
568
+ u :=
569
+ 1
570
+ R (
571
+ u x u)
572
+ ,
573
+ *
574
+ 2
575
+ R ( )
576
+ x y,
577
+ *
578
+ y
579
+ 1
580
+ 2
581
+ FIGURE 11. Unoriented Dichromatic Singular Link(L3)
582
+ The coloring equations are:
583
+ R2(y ∗ R1(x, y), x ∗ (y ∗ R1(x, y))) = R1(x, y),
584
+ and
585
+ R1(y ∗ R1(x, y), x ∗ (y ∗ R1(x, y))) = R2(x, y) ∗ y.
586
+ The system of these two equations reduces to
587
+
588
+ 0 = 3y2 + y + 2x,
589
+ 0 = 3x2 + x + 2y,
590
+ and thus we obtain that 2(y − x) = 0 giving x = y or y = x + 3.
591
+ Then Coldsq
592
+ X (L3) = {(x, x), x ∈ X} ∪ {(x, x + 3), x ∈ X}.
593
+ Thus the three links L1, L2 and L3 are pairwise distinct.
594
+ Example 5.5. Let 12
595
+ 1, 32
596
+ 1, 42
597
+ 1, 52
598
+ 1, 52
599
+ 2, 52
600
+ 3, 62
601
+ 1, 62
602
+ 2, 62
603
+ 3, 62
604
+ 4, 62
605
+ 5, 62
606
+ 6, 62
607
+ 7, 62
608
+ 8, 62
609
+ 9, 62
610
+ 10, 62
611
+ 11, and 62
612
+ 12 be the eigh-
613
+ teen unoriented dichromatic singular links in Figure 12 and let X be the disingquandle in Example
614
+ 5.4. By similar calculations as in the example, we obtain the following table:
615
+ L
616
+ #Coldsq
617
+ X (L)
618
+ 62
619
+ 2
620
+ 0
621
+ 62
622
+ 6
623
+ 2
624
+ 42
625
+ 1, 62
626
+ 12
627
+ 18
628
+ 12
629
+ 1, 32
630
+ 1, 52
631
+ 1, 52
632
+ 2, 52
633
+ 3, 62
634
+ 1, 62
635
+ 3, 62
636
+ 4, 62
637
+ 5, 62
638
+ 7, 62
639
+ 8, 62
640
+ 9, 62
641
+ 10, 62
642
+ 11
643
+ 6
644
+ This table shows that the disingquandle counting invariant Zdsq
645
+ X (L) distinguishes some of these
646
+ eighteen unoriented dichromatic singular links.
647
+ 13
648
+
649
+ 11
650
+ 31
651
+ 41
652
+ 51
653
+ 1
654
+ 2
655
+ 1
656
+ 2
657
+ 1
658
+ 2
659
+ 1
660
+ 2
661
+ 1
662
+ 2
663
+ 1
664
+ 2
665
+ 1
666
+ 2
667
+ 1
668
+ 2
669
+ 1
670
+ 2
671
+ 1
672
+ 2
673
+ 1
674
+ 2
675
+ 1
676
+ 2
677
+ 1
678
+ 2
679
+ 1
680
+ 2
681
+ 1
682
+ 2
683
+ 1
684
+ 2
685
+ 1
686
+ 2
687
+ 1
688
+ 2
689
+ 2
690
+ 2
691
+ 2
692
+ 2
693
+ 52
694
+ 53
695
+ 61
696
+ 62
697
+ 2
698
+ 2
699
+ 2
700
+ 2
701
+ 63
702
+ 64
703
+ 65
704
+ 66
705
+ 2
706
+ 2
707
+ 2
708
+ 2
709
+ 67
710
+ 68
711
+ 69
712
+ 610
713
+ 2
714
+ 2
715
+ 611
716
+ 612
717
+ 2
718
+ 2
719
+ 2
720
+ 2
721
+ FIGURE 12. Table of Unoriented Dichromatic Singular Links
722
+ ACKNOWLEDGEMENT
723
+ Mohamed Elhamdadi was partially supported by Simons Foundation collaboration grant 712462.
724
+ 14
725
+
726
+ REFERENCES
727
+ [1] K. Bataineh, On skein theory of dichromatic links and invariants of finite type, Journal of Knot Theory and Its
728
+ Ramifications. 26(13) (2017) 1750092.
729
+ [2] K. Bataineh and I. Saidi, Involutory quandles and dichromatic links, Symmetry. 12(1) (2020) 111.
730
+ [3] Madeline Brown and Sam Nelson, G-family polynomials, J. Knot Theory Ramifications 30 (2021), no. 9, Paper
731
+ No. 2150070, 15, doi: 10.1142/S021821652150070X. MR4358333
732
+ [4] Jose Ceniceros, Indu R. Churchill, and Mohamed Elhamdadi, Singquandle shadows and singular knot invariants,
733
+ Canad. Math. Bull. 65 (2022), no. 3, 770–787, doi: 10.4153/S0008439521000837. MR4472501
734
+ [5] Indu R. U. Churchill, Mohamed Elhamdadi, Mustafa Hajij, and Sam Nelson, Singular knots and involutive
735
+ quandles, J. Knot Theory Ramifications 26 (2017), no. 14, 1750099, 14, doi: 10.1142/S0218216517500997.
736
+ MR3735400
737
+ [6] Jose Ceniceros, Indu R. Churchill, Mohamed Elhamdadi, and Mustafa Hajij, Singquandles, psyquandles,
738
+ and singular knots: a survey, J. Knot Theory Ramifications 30 (2021), no. 12, Paper No. 2141003, 22,
739
+ doi: 10.1142/S0218216521410030. MR4394062
740
+ [7]
741
+ , Cocycle invariants and oriented singular knots, Mediterr. J. Math. 18 (2021), no. 5, Paper No. 217, 17,
742
+ doi: 10.1007/s00009-021-01867-6. MR4309525
743
+ [8] Mohamed Elhamdadi and Sam Nelson, Quandles—an introduction to the algebra of knots, Student Mathematical
744
+ Library, vol. 74, American Mathematical Society, Providence, RI, 2015. MR3379534
745
+ [9] Allison Henrich and Sam Nelson, Semiquandles and flat virtual knots, Pacific J. Math. 248 (2010), no. 1, 155–
746
+ 170, doi: 10.2140/pjm.2010.248.155. MR2734169
747
+ [10] Jim Hoste and Mark E. Kidwell, Dichromatic link invariants, Trans. Amer. Math. Soc. 321 (1990), no. 1, 197–
748
+ 229, doi: 10.2307/2001598. MR961623
749
+ [11] Jim Hoste and J´ozef H. Przytycki, An invariant of dichromatic links, Proc. Amer. Math. Soc. 105 (1989), no. 4,
750
+ 1003–1007, doi: 10.2307/2047066. MR989100
751
+ [12] Atsushi Ishii, Masahide Iwakiri, Yeonhee Jang, and Kanako Oshiro, A G-family of quandles and handlebody-
752
+ knots, Illinois J. Math. 57 (2013), no. 3, 817–838. MR3275740
753
+ [13] David Joyce, A classifying invariant of knots, the knot quandle, J. Pure Appl. Algebra 23 (1982), no. 1, 37–65,
754
+ doi: 10.1016/0022-4049(82)90077-9. MR638121
755
+ [14] Uwe Kaiser, Strong band sum and dichromatic invariants, Manuscripta Math. 74 (1992), no. 3, 237–251,
756
+ doi: 10.1007/BF02567670. MR1149761
757
+ [15] Sang Youl Lee and Mohd Ibrahim Sheikh, Diquandles and invariants for oriented dichromatic links, J. Knot
758
+ Theory Ramifications 30 (2021), no. 7, Paper No. 2150049, 19, doi: 10.1142/S0218216521500498. MR4321932
759
+ [16] S. V. Matveev, Distributive groupoids in knot theory, Mat. Sb. (N.S.) 119(161) (1982), no. 1, 78–88, 160
760
+ (Russian). MR672410
761
+ [17] Sam Nelson, Natsumi Oyamaguchi, and Radmila Sazdanovic, Psyquandles, singular knots and pseudoknots,
762
+ Tokyo J. Math. 42 (2019), no. 2, 405–429, doi: 10.3836/tjm/1502179287. MR4106586
763
+ [18] Natsumi Oyamaguchi, Enumeration of spatial 2-bouquet graphs up to flat vertex isotopy. part B, Topology Appl.
764
+ 196 (2015), no. part B, 805–814, doi: 10.1016/j.topol.2015.05.049. MR3431017
765
+ [19] Hiroshi Tamaru, Two-point homogeneous quandles with prime cardinality, J. Math. Soc. Japan 65 (2013), no. 4,
766
+ 1117–1134, doi: 10.2969/jmsj/06541117. MR3127819
767
+ DEPARTMENT OF MATHEMATICS, GRADUATE SCHOOL OF NATURAL SCIENCES PUSAN NATIONAL UNIVER-
768
+ SITY, BUSAN 46241, REPUBLIC OF KOREA
769
+ Email address: [email protected]
770
+ UNIVERSITY OF SOUTH FLORIDA, TAMPA, FLORIDA, USA
771
+ Email address: [email protected]
772
+ DEPARTMENT OF MATHEMATICS, DALIAN UNIVERSITY OF TECHNOLOGY, CHINA
773
+ Email address: [email protected]
774
+ 15
775
+
D9E2T4oBgHgl3EQfSgcI/content/tmp_files/load_file.txt ADDED
@@ -0,0 +1,486 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf,len=485
2
+ page_content='arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
3
+ page_content='03792v1 [math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
4
+ page_content='GT] 10 Jan 2023 A G-FAMILY OF SINGQUANDLES AND INVARIANTS OF DICHROMATIC SINGULAR LINKS MOHD IBRAHIM SHEIKH, MOHAMED ELHAMDADI, AND DANISH ALI ABSTRACT.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
5
+ page_content=' We introduce and investigate dichromatic singular links.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
6
+ page_content=' We also construct G-Family of singquandles and use them to define counting invariants for unoriented dichromatic singular links.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
7
+ page_content=' We provide some examples to show that these invariants distinguish some dichromatic singular links.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
8
+ page_content=' CONTENTS 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
9
+ page_content=' Introduction 1 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
10
+ page_content=' Singular links, Singquandles and Dichromatic Links 2 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
11
+ page_content=' Dichromatic Singular Links 5 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
12
+ page_content=' G-Family of Singquandles (Disingquandles) 6 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
13
+ page_content=' Computable Invariants for Unoriented Dichromatic Singular Links 10 Acknowledgement 14 References 15 Mathematics Subject Classifications (2020): 57M25, 57M27.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
14
+ page_content=' Key words and Phrases: Knot;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
15
+ page_content=' Link;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
16
+ page_content=' Singular knot;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
17
+ page_content=' Singular link;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
18
+ page_content=' Dichromatic link;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
19
+ page_content=' Dichromatic singular link;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
20
+ page_content=' Quandle;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
21
+ page_content=' Singquandle;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
22
+ page_content=' Disingquandle;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
23
+ page_content=' Disingquandle counting invariant.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
24
+ page_content=' 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
25
+ page_content=' INTRODUCTION A knot is a simple closed curve in three dimensional space R3 and a disjoint union of two or more knots forms a link with two or more components [8].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
26
+ page_content=' Knots and links are categorised in many ways.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
27
+ page_content=' One way is to use the crossing type as a tool to define a knot or link type.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
28
+ page_content=' Classical, virtual and singular knots and links serve as examples as they are all recognised by the type of crossing they contain.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
29
+ page_content=' The other way to define link types is by labelling the components of a classical link.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
30
+ page_content=' Dichromatic links are defined by using this technique as their components are either labelled by “1” or “2” [1, 2, 10, 11, 14].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
31
+ page_content=' A singular link is a link with at least one singular crossing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
32
+ page_content=' In this paper we use such labelling technique for singular links and define a new type of links which we call dichromatic singular links.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
33
+ page_content=' A quandle is an algebraic structure satisfying some axioms that result from the Reidemeister moves for oriented classical knots and links.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
34
+ page_content=' If furthermore all right multiplications by fixed ele- ments of the quandle are involutions then such structures are called involutory quandles or Kei’s They are used to investigate unoriented knots and links.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
35
+ page_content=' Quandles were independently introduced by Joyce and Matveev [13, 16].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
36
+ page_content=' Since then they have been used to construct invariants of knots and links [4, 6, 17].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
37
+ page_content=' Quandles have been also used to define new algebraic systems by taking a 1 family of quandles at a time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
38
+ page_content=' Such systems are called G-Family of quandles and this notion was introduced in 2013 by Ishii, Iwakiri, Jang and Oshiro [12].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
39
+ page_content=' A G-Family of quandles were used to define invariants for handlebody-knots.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
40
+ page_content=' Also in [15] Lee and Sheikh used Z2-Family of quandles to construct algebraic invariants for oriented dichromatic links.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
41
+ page_content=' In this paper, we introduce the notions of G-Family of singquandles and dichromatic singular links.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
42
+ page_content=' A dichromatic singular link is an n component singular link with each of its component labelled as “1” or “2”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
43
+ page_content=' A singquandle is an algebraic system whose axioms are motivated by Reidemeister moves of unoriented singular knots.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
44
+ page_content=' By taking a family of such algebaraic systems (Singquandles), we define a new algebraic system which we call G-Family of singquandles or disingquandle.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
45
+ page_content=' The axioms of the latter are motivated by generalized Reidemeister moves of un- oriented dichromatic singular links.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
46
+ page_content=' We discuss various examples and some properties of G-Family of singquandles, and also show that a G-Family of singquandles X enables us to distinguish unori- ented dichromatic singular links by computing their sets of all X-colorings and proving that these sets are different when their arcs are colored by the elements of X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
47
+ page_content=' This paper is organized as follows.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
48
+ page_content=' Section 2 reviews some preliminaries about singular links, singquandles as well as dichromatic links and their generalized Reidemeister moves.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
49
+ page_content=' In Section 3 we introduce the notion of dichromatic singular links with some typical examples of unoriented dichromatic singular links.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
50
+ page_content=' Section 4 introduces the notion of G-Family of singquandles (dis- ingquandles) with some typical examples of G-Family of singquandles.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
51
+ page_content=' Section 5 discusses how G-Family of singquandles is related to unoriented dichromatic singular links and develop com- putable invariants for unoriented dichromatic singular links.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
52
+ page_content=' We discuss some examples which show how the invariants distinguish unoriented dichromatic singular links, and especially how they detect the change of component labelings.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
53
+ page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
54
+ page_content=' SINGULAR LINKS, SINGQUANDLES AND DICHROMATIC LINKS In this section we review some preliminaries about singular links, singquandles and dichromatic links.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
55
+ page_content=' Most of the terminologies of this section can be found in [5, 9, 15].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
56
+ page_content=' We begin with the definition of a singular link.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
57
+ page_content=' Definition 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
58
+ page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
59
+ page_content=' A singular link in S3 is the image of a smooth immersion of n circles in S3 that has finitely many double points, called singular points.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
60
+ page_content=' A singular link in R3 is represented by a singular link diagram in the plane R2, which is a classical link diagram with one or more singularities.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
61
+ page_content=' A singularity is a rigid vertex where a link is glued to itself.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
62
+ page_content=' Figure 1 gives two examples of singular links.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
63
+ page_content=' FIGURE 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
64
+ page_content=' Singular Links 2 Two singular links L� and L� are isotopy equivalent if one can be obtained from the other by a finite sequence generalized Reidemeister moves for singular links as shown in the following figure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
65
+ page_content=' Let D� and D� be two singular link diagrams in R2 representing L� and L�, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
66
+ page_content=' Then L� and L� are equivalent if and only if D� and D� can be transformed into each other by a finite sequence of classical and singular Reidemeister moves shown in Figure 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
67
+ page_content=' FIGURE 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
68
+ page_content=' Classical and Singular Reidemeister Moves Definition 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
69
+ page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
70
+ page_content=' [5] Let (X, ∗) be an involutive quandle.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
71
+ page_content=' Let R1 and R2 be two maps from X × X to X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
72
+ page_content=' The quadruple (X, ∗, R1, R2) is called a singquandle if the following axioms are satisfied (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
73
+ page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
74
+ page_content='1) x = R1(y, R2(x, y)) = R2(R2(x, y), R1(x, y)), (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
75
+ page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
76
+ page_content='2) y = R2(R2(x, y), x) = R1(R2(x, y), R1(x, y)), (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
77
+ page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
78
+ page_content='3) R(x, y) = (R1(y, R2(x, y)), R2(R2(x, y), x)), (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
79
+ page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
80
+ page_content='4) (y ∗ z) ∗ R2(x, z) = (y ∗ x) ∗ R1(x, z), (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
81
+ page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
82
+ page_content='5) R1(x, y) = R2(y ∗ x, x), (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
83
+ page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
84
+ page_content='6) R2(x, y) = R1(y ∗ x, x) ∗ R2(y ∗ x, x), (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
85
+ page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
86
+ page_content='7) R1(x ∗ y, z) ∗ y = R1(x, z ∗ y), 3 (2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
87
+ page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
88
+ page_content='8) R2(x ∗ y, z) = R2(x, z ∗ y) ∗ y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
89
+ page_content=' We remind the reader that the singquandle axioms come from the generalized Reidemeister moves for unoriented singular knots.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
90
+ page_content=' Singquandles were introduced as a ramification of quandles with the purpose of studying singular links, see for example [4,5,17].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
91
+ page_content=' The following are few typical examples of singquandles.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
92
+ page_content=' For an involutive quandle (X, ∗) with x ∗ y = 2y − x and X = Zn, the quadruple (X, ∗, R1, R2) forms a singquandle if and only if the following conditions are satisfied: (1) R2(x, y) = R1(x, y) + y − x, (2) R1(x, y) = R1(2x − y, x) + y − x, (3) R1(x, 2y − z) = 2y − R1(2y − x, z), (4) R2(2y − x, z) = 2y − R2(x, 2x − z).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
93
+ page_content=' For an involutive quandle (X, ∗) where X is a group G and x ∗ y = yx−1y, the quadruple (X, ∗, R1, R2) forms a singquandle if and only if the following conditions are satisfied: (1) R2(x, z)z−1yz−1R2(x, z) = R1(x, z)x−1yx−1R1(x, z), (2) R1(x, y) = R2(xy−1x, x), (3) R2(x, y) = R2(xy−1x, x)[R1(xy−1x, x)]−1R2(xy−1x, x), (4) y[R1(yx−1y, z)]−1y = R1(x, yz−1y), (5) R1(yx−1y, z) = y[R2(x, yz−1y)]−1y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
94
+ page_content=' Definition 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
95
+ page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
96
+ page_content=' For a positive integer n ≥ 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
97
+ page_content=' A dichromatic link is a smooth imbedding of n circles in R3 such that each component is labeled as “1” or “2”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
98
+ page_content=' In R2 every dichromatic link is represented by a dichromatic link diagram which is a classical link diagram with each component labelled either “1” or “2”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
99
+ page_content=' For example, see Figure 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
100
+ page_content=' 1 1 2 2 FIGURE 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
101
+ page_content=' Dichromatic Links Two dichromatic links L� and L� are isotopy equivalent if one can be obtained from the other by a finite sequence of generalized Reidemeister moves for the dichromatic links as shown in the figure 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
102
+ page_content=' Let D� and D� be two dichromatic link diagrams in R2 representing L� and L�, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
103
+ page_content=' Then L� and L� are equivalent if and only D� and D� can be transformed into each other by a finite sequence of generalized Reidemeister moves shown in the following Figure 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
104
+ page_content=' 4 i i i j j i i k j i k j FIGURE 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
105
+ page_content=' Generalized Reidemeister Moves for Dichromatic Links 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
106
+ page_content=' DICHROMATIC SINGULAR LINKS This section is devoted to dichromatic singular links which is a generalization of singular links.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
107
+ page_content=' To generate a dichromatic singular link we label a singular link’s components with “1” or “2”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
108
+ page_content=' Thus We have the following definition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
109
+ page_content=' Definition 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
110
+ page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
111
+ page_content=' A singular link L in R3 whose each component is colored (labelled) by either “1” or “2” is called a dichromatic singular link.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
112
+ page_content=' A dichromatic singular link L in R3 is represented by a dichromatic singular link diagram D in R2 in which each component is labelled “1” or “2”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
113
+ page_content=' Figure 5 shows two examples of unoriented dichromatic singular link diagrams.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
114
+ page_content=' 1 2 1 2 FIGURE 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
115
+ page_content=' Dichromatic Singular Links Two dichromatic singular links L� and L� in R3 are ambient isotopic if there exists a self home- omorphism h : R3 → R3 that takes one link to the other and preserves the singularities as well as the labels “1”, “2” such that h(L�) = L�.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
116
+ page_content=' Thus two singular dichromatic links L� and L� are equivalent if one can be obtained from the other by a finite sequence of generalized dichromatic singular Reidemeister moves preserving the label of each component as shown in the Figure 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
117
+ page_content=' Let D� and D� be two dichromatic singular link diagrams in R2 representing L� and L�, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
118
+ page_content=' Then L� and L� are equivalent if and only if D� and D� can be transformed into each other by a finite sequence of generalized dichromatic singular Reidemeister moves shown in the following Figure 6 where i, j, k ∈ {1, 2}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
119
+ page_content=' A dichromatic singular link with n components is called as an n-component dichromatic singular link.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
120
+ page_content=' Thus an n-component dichromatic singular link in R3 can be defined as L = K1 ∪ · · · ∪ Kn.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
121
+ page_content=' 5 i k k j i i j j i j i k k j i j i i i j j i i k j i k j FIGURE 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
122
+ page_content=' Regular Dichromatic Reidemeister Moves RI, RII and RIII on the top and Dichromatic Singular Reidemeister Moves RIV a, RIV b and RV in the middle and on the bottom.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
123
+ page_content=' Taking n = 2, we obtain 2-component dichromatic singular links.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
124
+ page_content=' Some 2-component unoriented dichromatic singular link diagrams (see p 814 of [18]) are shown in Figure 12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
125
+ page_content=' Proposition 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
126
+ page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
127
+ page_content=' Let L� and L� be two unoriented dichromatic singular links in R3 and let D� and D� be two unoriented dichromatic singular link diagrams in R2 representing L� and L�, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
128
+ page_content=' Then L� and L� are equivalent if and only if D� and D� are transformed into each other by a finite sequence of generalized Reidemeister moves for unoriented dichromatic singular links which preserve the singularities and the label of each component as shown in the Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
129
+ page_content=' 6 where i, j, k ∈ {1, 2} and ambient isotopies of R2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
130
+ page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
131
+ page_content=' G-FAMILY OF SINGQUANDLES (DISINGQUANDLES) Before introducing the notion of G-Family of Singquandles, we first recall the definition of G-family of quandles from [12].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
132
+ page_content=' Definition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
133
+ page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
134
+ page_content=' Given a group G and a set X,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
135
+ page_content=' a G-family of quandles,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
136
+ page_content=' denoted by (G,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
137
+ page_content=' X),' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
138
+ page_content=' is a choice of quandle operation ∗g on the set X for each element g ∈ G such that the following axioms are satisfied (1) For all g ∈ G and for all x ∈ X,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
139
+ page_content=' x ∗g x = x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
140
+ page_content=' (2) For all g,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
141
+ page_content=' h ∈ G and for all x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
142
+ page_content=' y ∈ X,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
143
+ page_content=' (x ∗g y) ∗h y = x ∗gh y,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
144
+ page_content=' (3) For all x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
145
+ page_content=' y ∈ X,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
146
+ page_content=' x ∗e x = x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
147
+ page_content=' where e is the identity element of G,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
148
+ page_content=' 6 (4) For all x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
149
+ page_content=' y,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
150
+ page_content=' z ∈ X,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
151
+ page_content=' (x ∗g y) ∗h z = (x ∗h z) ∗h−1gh (y ∗h z) The following are two examples of G-families of quandles.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
152
+ page_content=' For any group G and any set X, defining x ∗g y = x for all x, y ∈ X and all g ∈ G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
153
+ page_content=' This gives a G-family of quandles called the trivial G-family of quandles.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
154
+ page_content=' Let (X, ∗) be a quandle of cyclic type [19] with cardinality n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
155
+ page_content=' Let Rx denotes the right multiplication by x, thus by definition Rx (n−1) is the identity map.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
156
+ page_content=' Then define x ∗i y = Ry i(x) then it is shown in Proposition 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
157
+ page_content='3 of [12] that (Z, X) is a Z-family of quandles and also Z(n−1)-family of quandles.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
158
+ page_content=' A G-family of quandles (G, X) induces a quandle operation on the set G × X by (g, x) ∗ (h, y) = (h−1gx, x ∗h y).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
159
+ page_content=' The notion of G-family of quandles was introduced by Ishii, Iwakiri, Jang and Oshiro in 2013 in [12] in order to produce invariants of handlebody knots.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
160
+ page_content=' They defined coloring invariants and cocycle invariants of handlebody knots.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
161
+ page_content=' They used these invariants to detect chirality of some han- dlebody knots.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
162
+ page_content=' Later in 2015, Ishii independently studied the notion of G-family of quandles in connection with the multiple conjugation quandle and showed that the later one can be obtained from the first one.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
163
+ page_content=' In 2017 and 2018 Ishii, Nelson and Ishii, Iwakiri, Kamada, Kim, Matsuzaki, Os- hiro respectively, used this work and introduced the notions of partially multiplicative biquandles and multiple conjugation biquandle.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
164
+ page_content=' In 2021 Lee and Sheikh jointly used G-family of quandles to construct algebraic invariants for oriented dichromatic links [15].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
165
+ page_content=' We introduce the following definition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
166
+ page_content=' Definition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
167
+ page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
168
+ page_content=' Let X be a set equipped with two binary operations ∗1 and ∗2 such that both (X, ∗1), (X, ∗2) are involutive quandles.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
169
+ page_content=' Let R1, R2 be two maps from X × X to X such that the quadruples (X, ∗1, R1, R2) and (X, ∗2, R1, R2) are singquandles.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
170
+ page_content=' Then the quintuple (X, ∗1, ∗2, R1, R2) is called a disingquandle or Z2-family of singquandles if the following axioms are satisfied (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
171
+ page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
172
+ page_content='1) (x ∗1 y) ∗2 z = (x ∗2 z) ∗1 (y ∗2 z), (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
173
+ page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
174
+ page_content='2) (x ∗2 y) ∗1 z = (x ∗1 z) ∗2 (y ∗1 z), (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
175
+ page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
176
+ page_content='3) (y ∗1 z) ∗2 R2(x, z) = (y ∗2 x) ∗1 R1(x, z), (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
177
+ page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
178
+ page_content='4) (y ∗2 z) ∗1 R2(x, z) = (y ∗1 x) ∗2 R1(x, z), (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
179
+ page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
180
+ page_content='5) R2(x, y) = R1(y ∗1 x, x) ∗2 R2(y ∗1 x, x), (4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
181
+ page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
182
+ page_content='6) R2(x, y) = R1(y ∗2 x, x) ∗1 R2(y ∗2 x, x), The above axioms of a disingquandle come from the generalized dichromatic singular Reide- meister moves shown in Figure 6 when we take the coloring rule shown in Figure 7 under consid- eration.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
183
+ page_content=' 7 i i j i/j j/i j i y x y x y x x x y x j x y* R ( ) x 1 2 y, R ( ) x y, FIGURE 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
184
+ page_content=' Coloring by a disingquandle The following lemma is motivated by the above construction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
185
+ page_content=' Lemma 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
186
+ page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
187
+ page_content=' The set of colorings of a dichromatic singular link by a disingquandle does not change by the dichromatic singular Reidemeister moves shown in Figure 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
188
+ page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
189
+ page_content=' As in the case of classical and singular knot theories, there is one to one correspondence between colorings before and after each of the dichromatic singular Reidemeister moves.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
190
+ page_content=' The invariance follows directly from the equations 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
191
+ page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
192
+ page_content='1, 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
193
+ page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
194
+ page_content='2, 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
195
+ page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
196
+ page_content='3, 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
197
+ page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
198
+ page_content='4, 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
199
+ page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
200
+ page_content='5 and 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
201
+ page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
202
+ page_content='6 given in Definition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
203
+ page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
204
+ page_content=' □ Example 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
205
+ page_content='4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
206
+ page_content=' Let (X, ∗1, R1, R2) and (X, ∗2, R1, R2) be two singquandles such that such that x ∗1 y = x = x ∗2 y and R1(x, y) = R2(x, y), then (X, ∗1, ∗2, R1, R2) forms a disingquandle.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
207
+ page_content=' Example 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
208
+ page_content='5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
209
+ page_content=' Let (X, ∗1, R1, R2) and (X, ∗2, R1, R2) be two singquandles.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
210
+ page_content=' If for all x, y ∈ X we have x ∗1 y = x ∗2 y then (X, ∗1, ∗2, R1, R2) forms a disingquandle.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
211
+ page_content=' Now Example 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
212
+ page_content='5 combined with Proposition 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
213
+ page_content='6 in [5] gives the following example.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
214
+ page_content=' Example 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
215
+ page_content='6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
216
+ page_content=' Let Λ = Z[t, B]/(t2 − 1, B(1 + t), t − (1 − B)2) and X be an Λ-module.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
217
+ page_content=' Define x ∗1 y = tx + (1 − t)y, R1(x, y) = (1 − t − b)x + (t + b)y and R2(x, y) = (1 − B)x + By, then by setting ∗2 = ∗1, then one obtains that (X, ∗1, ∗2, R1, R2) forms a disingquandle.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
218
+ page_content=' Example 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
219
+ page_content='7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
220
+ page_content=' Let X be a module over Λ = Z[t].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
221
+ page_content=' Define x∗1y = x∗2y = tx+(1−t)y, R1(x, y) = (1 − t − B)x + (t + B)y and R2(x, y) = (1 − B)x + By.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
222
+ page_content=' Setting t = −1 and X = Z7, then the quintuple (X, ∗1, ∗2, R1, R2) forms a disingquandle if B = 4 or if B = 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
223
+ page_content=' This example can be generalized to Zp, where p is a prime as follows.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
224
+ page_content=' Example 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
225
+ page_content='8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
226
+ page_content=' Let p be an odd prime and let B ∈ Zp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
227
+ page_content=' Consider Zp with x ∗1 y = x ∗2 y = −x + 2y, R1(x, y) = (2 − B)x + (−1 + B)y and R2(x, y) = (1 − B)x + By.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
228
+ page_content=' Let ζ be a primitive root of unity in Zp so that ζ p−1 2 = −1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
229
+ page_content=' By choosing 1 − B = ζ p−1 2 we obtain that (Zp, ∗1, ∗2, R1, R2) forms a disingquandle.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
230
+ page_content=' Example 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
231
+ page_content='9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
232
+ page_content=' Let X = G be a multiplicative group with the involutive quandle operations x ∗1 y = x ∗2 y = yx−1y (core quandle on G), then a direct computation gives the fact that the quintuple (X, ∗1, ∗2, R1, R2) forms a disingquandle if and only if R1 and R2 satisfies the following equations: (5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
233
+ page_content='1) R2(x, z)z−1yz−1R2(x, z) = R1(x, z)x−1yx−1R1(x, z), (5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
234
+ page_content='2) R2(x, y) = R2(xy−1x, x)[R1(xy−1x, x)]−1R2(xy−1x, x), 8 A straightforward computation gives the following solution R1(x, y) = x and R2(x, y) = y, for all x, y, z ∈ G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
235
+ page_content=' Now assume that G is an abelian group without 2-torsion, so that x ∗ y = −x + 2y, then (X, ∗1, ∗2, R1, R2) forms a disingquandle if and only if R2(x, y) = R1(x, y) + y − x, where R1 satisfies the identity R1(x, y) = R1(−x + 2y, x) + y − x.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
236
+ page_content=' For example for any integer m, the map R1(x, y) = mx + (2m + 1)y give a solution.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
237
+ page_content=' Thus we have a family of solutions parametrized by the integer m: R1(x, y) = mx + (2m + 1)y, R2(x, y) = (m − 1)x + 2(m + 1)y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
238
+ page_content=' Definition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
239
+ page_content='10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
240
+ page_content=' A map f : X → Y is called a homomorphism of disingquandle (X, ∗1, ∗2, R1, R2) and (Y, ∗′ 1, ∗′ 2, R′ 1, R′ 2) if the following conditions are satisfied for all x, y, z ∈ X (i) f(x ∗1 y) = f(x) ∗′ 1 f(y), (ii) f(x ∗2 y) = f(x) ∗′ 2 f(y), (iii) f(R1(x, y)) = R′ 1(f(x), f(y)), (iv) f(R2(x, y)) = R′ 1(f(x), f(y)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
241
+ page_content=' If a homomorphism of disingquandle is bijective, then it is called an isomorphism of disingquan- dle.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
242
+ page_content=' We say that two Z2-families of singquandles are isomorphic if there exists an ismorphism of disingquandle between them.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
243
+ page_content=' Definition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
244
+ page_content='11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
245
+ page_content=' Let (X, ∗1, ∗2, R1, R2) be a disingquandle.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
246
+ page_content=' A subset Y ⊂ X is called a sub- disingquandle if (Y, ∗1, ∗2, R1, R2) is itself a disingquandle.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
247
+ page_content=' Example 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
248
+ page_content='12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
249
+ page_content=' We use Example 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
250
+ page_content='9 to get the following 2 examples: Let X = Z9 be the dihedral quandle with x ∗ y = −x + 2y, R1(x, y) = mx + (2m + 1)y, R2(x, y) = (m−1)x+ 2(m+ 1)y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
251
+ page_content='. Then (Y, ∗1, ∗2, R1, R2) is itself a disingquandle with Y = Z3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
252
+ page_content=' Let X = Z25 be the dihedral quandle with x ∗ y = −x + 2y, R1(x, y) = mx + (2m + 1)y, R2(x, y) = (m−1)x+ 2(m+ 1)y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
253
+ page_content='. Then (Y, ∗1, ∗2, R1, R2) is itself a disingquandle with Y = Z5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
254
+ page_content=' Given a homomorphism of disingquandles, we obtain the following lemma.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
255
+ page_content=' Lemma 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
256
+ page_content='13.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
257
+ page_content=' The image Im(f) of any homomorphism of disingquandle f defined from (X, ∗1, ∗2, R1, R2) to (Y, ∗′ 1, ∗′ 2, R′ 1, R′ 2) is always a sub-disingquandle.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
258
+ page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
259
+ page_content=' Given that f : (X, ∗1, ∗2, R1, R2) → (Y, ∗′ 1, ∗′ 2, R′ 1, R′ 2) is a homomorphism.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
260
+ page_content=' Then the equations (i), (ii), (iii) and (iv) of Definition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
261
+ page_content='10 imply that Im(f) is closed under ∗1, ∗2, R1 and R2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
262
+ page_content=' Then the axioms of disingquandle are satisfied in Y .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
263
+ page_content=' Hence they are automatically satisfied in Im(f).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
264
+ page_content=' This ends the proof of the lemma.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
265
+ page_content=' □ Now we introduce the notion of fundamental disingquandle of an unoriented dichromatic sin- gular link and provide an illustrative example.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
266
+ page_content=' Let D be a diagram of an unoriented dichromatic singular link L in R2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
267
+ page_content=' We define the fundamental disingquandle of D, denoted by DSQ(D), as the set of equivalence classes of disingquandle words W-DSQ(D) under the equivalence relation generated by the axioms of disingquandle and the crossing relations shown in Figure 7, where W- DSQ(D) are defined by taking a set of generators X = {x1, x2, x3, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
268
+ page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
269
+ page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
270
+ page_content=', xn} which corresponds bijectively with the semi arcs in D, recursively by the following two rules: 9 (1) X ⊂ W-DSQ(D), (2) If x, y ∈ W-DSQ(D), then x ∗1 y, x ∗2 y, R1(x, y), R2(x, y) ∈ W-DSQ(D).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
271
+ page_content=' Example 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
272
+ page_content='14.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
273
+ page_content=' Consider the following unoriented dichromatic singular link L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
274
+ page_content=' 1 2 x z u v y FIGURE 8.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
275
+ page_content=' Fundamental Disingquandle of Unoriented Dichromatic Singular Links The fundamental disingquandle of L is given by DSQ(L) = ⟨x, y, z, u, v| z = x ∗2 y;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
276
+ page_content=' u = y ∗1 z;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
277
+ page_content=' v = z ∗2 u;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
278
+ page_content=' x = R1(u, v);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
279
+ page_content=' y = R2(u, v)⟩.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
280
+ page_content=' This presentation can be simplified to the following presentation of DSQ(L) ⟨x, y| x = R1(y∗1(x∗2y), (x∗2y)∗2(y∗1(x∗2y)));' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
281
+ page_content=' y = R2(y∗1(x∗2y), (x∗2y)∗2(y∗1(x∗2y)))⟩.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
282
+ page_content=' 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
283
+ page_content=' COMPUTABLE INVARIANTS FOR UNORIENTED DICHROMATIC SINGULAR LINKS Let D be an unoriented dichromatic singular link diagram and let A(D) denote the set of all arcs of D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
284
+ page_content=' Let (X, ∗1, ∗2, R1, R2) be a disingquandle.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
285
+ page_content=' A disingquandle coloring of D by X, or simply disingquandle X-coloring of D, is a map C : A(D) → X such that at every classical and singular crossing, the relations depicted in Figure 7 hold.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
286
+ page_content=' The disingquandle element C(s) is called a color of the arc s and the pair (D, C) is called the X-colored unoriented dichromatic singular link diagram by C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
287
+ page_content=' The set of all disingquandle X-colorings of D is denoted by Coldsq X (D).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
288
+ page_content=' Then we have the following: Lemma 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
289
+ page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
290
+ page_content=' Let D and D′ be two unoriented dichromatic singular link diagramss in R2 that can be transformed into each other by unoriented generalized dichromatic singular Reidemeis- ter moves as shown in the Figure 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
291
+ page_content=' Then for any finite disingquandle X, there is a one-to-one correspondence between Coldsq X (D) and Coldsq X (D′).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
292
+ page_content=' 10 Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
293
+ page_content=' It suffices to prove the assertion for the case that D′ is obtained from D by a single an unoriented generalized dichromatic singular Reidemeister move.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
294
+ page_content=' Let E be an open disk in R2 where the unoriented generalized dichromatic singular Reidemeister move under consideration is applied.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
295
+ page_content=' Then D∩(R2−E) = D′∩(R2−E).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
296
+ page_content=' Now let C be a disingquandle X-coloring of D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
297
+ page_content=' Since (X, ∗1, R1, R2) and (X, ∗2, R1, R2) are both singquandles by the disingquandle definition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
298
+ page_content='2, it is obviously seen from the Figure 6 that the restriction of C to D ∩(R2 −E)(= D′ ∩(R2 −E)) can be extended to a unique disingquandle X-coloring of D′ for unoriented generalized dichromatic singular Reidemeister moves RI, RII and RIII.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
299
+ page_content=' Also, using the disingquandle axioms 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
300
+ page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
301
+ page_content='1 to 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
302
+ page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
303
+ page_content='6, it is easily seen from the Figure 6 that the restriction of C to D ∩ (R2 − E)(= D′ ∩ (R2 − E)) can be extended to a unique disingquandle X-coloring of D′ for an unoriented generalized dichromatic Reidemeister moves RIV a, RIV b and RV .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
304
+ page_content=' This completes the proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
305
+ page_content=' □ In an X-colored unoriented dichromatic singular link diagram (D, C), we think of elements of a disingquandle X as labels for the arcs in D with different operations at crossings as shown in Figure 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
306
+ page_content=' Then it is seen from Lemma 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
307
+ page_content='1 that the disingquandle axioms of Definition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
308
+ page_content='2 are transcriptions of a generating set of unoriented generalized Reidemeister moves for unoriented dichromatic singular links which are sufficient to generate any other unoriented generalized dichro- matic singular Reidemeister moves.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
309
+ page_content=' That is, the axioms 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
310
+ page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
311
+ page_content='1 and 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
312
+ page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
313
+ page_content='2 come from the unoriented generalized dichromatic singular Reidemeister move RIV a, the axioms 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
314
+ page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
315
+ page_content='3 and 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
316
+ page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
317
+ page_content='4 come from the unoriented generalized dichromatic singular Reidemeister move RIV b and the axioms 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
318
+ page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
319
+ page_content='5 and 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
320
+ page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
321
+ page_content='6 come from the unoriented generalized dichromatic singular Reidemeister move RV as seen in Figure 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
322
+ page_content=' Theorem 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
323
+ page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
324
+ page_content=' Let L be an unoriented dichromatic singular link in R3 and let D be a diagram of L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
325
+ page_content=' Then for any finite disingquandle X, the cardinality ♯Coldsq X (L) is an invariant of L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
326
+ page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
327
+ page_content=' Let D′ be any other unoriented dichromatic singular link diagram of L obtained from D by applying a finite number of unoriented generalized dichromatic singular Reidemeister moves.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
328
+ page_content=' Then it is direct from Lemma 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
329
+ page_content='1 that ♯Coldsq X (D′) = ♯Coldsq X (D).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
330
+ page_content=' This completes the proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
331
+ page_content=' □ If X is a finite disingquandle, we call the cardinality ♯Coldsq X (D) the disingquandle X-coloring number or the disingquandle counting invariant of L, and denote it by Zdsq X (L), i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
332
+ page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
333
+ page_content=', Zdsq X (L) = ♯Coldsq X (D).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
334
+ page_content=' Theorem 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
335
+ page_content='3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
336
+ page_content=' Let L be an unoriented dichromatic singular link and let X be a disingquandle.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
337
+ page_content=' Then there is a one-to-one correspondence between Coldsq X (L) and Hom(DSQ(L), X).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
338
+ page_content=' Conse- quently, Zdsq X (L) = ♯Hom(DSQ(L), X).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
339
+ page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
340
+ page_content=' Since the disingquandle X-colorings of L generate the fundamental disingquandle DSQ(L) of a link L which is generated by its arc labels.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
341
+ page_content=' Also each arc of L is assigned an element of X, for a disingquandle X-coloring of L, so we can associate each coloring a map f : DSQ(L) → X where if an arc is labelled a in the fundamental disingquandle and is assigned the color x ∈ X, then f(a) = x.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
342
+ page_content=' This completes the proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
343
+ page_content=' □ Now we give an example.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
344
+ page_content=' Example 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
345
+ page_content='4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
346
+ page_content=' Now, we give an explicit example of three unoriented dichromatic singular links L1, L2 and L3 and show that the coloring invariant distinguishes them from each other.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
347
+ page_content=' Consider the singquandle (X, ∗, R1, R2), where X = Z6, x∗1 y = x∗2 y = −x+2y = x∗y, R1(x, y) = x+3, 11 and R2(x, y) = 3x2 + 3x + y + 3 (see page 9 of [7]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
348
+ page_content=' By checking directly that the equations of Definition 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
349
+ page_content='2 hold we obtain that the quintuple (X, ∗1, ∗2, R1, R2) form a disingquandle.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
350
+ page_content=' Now coloring the two top arcs of link L1 by x and y as in the figure 9 below gives that the coloring equations are: x = R1(R1(x, y), R2(x, y)) and y = R2(R1(x, y), R2(x, y)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
351
+ page_content=' One then gets the system, � x = 3 + 3 + x, y = 3 + 3(3 + x) + 3(3 + x)2 + (3 + 3x + 3x2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
352
+ page_content=' R ( ) x 1 y, 2 R ( ) x y, x y 1 2 FIGURE 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
353
+ page_content=' Unoriented Dichromatic Singular Link(L1) Any pair (x, y) gives a solution to this system over Z6 and thus the set Coldsq X (L1) is equal to Z2 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
354
+ page_content=' Now coloring the link L2 as in the figure 10 below gives that the coloring equations are: R1(R1(x, y), x ∗ R1(x, y)) = R2(x, y) ∗ y and R2(R1(x, y), x ∗ R1(x, y)) = y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
355
+ page_content=' One then obtain that the solution is given by y = 3x2 + 4x + 3, thus the Coldsq X (L2) is R ( ) x 1 y, R ( ) x 1 y, 2 R ( ) x y, x x y 1 2 FIGURE 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
356
+ page_content=' Unoriented Dichromatic Singular Link(L2) 12 {(0, 3), (1, 4), (2, 5), (3, 0), (4, 1), (5, 2)}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
357
+ page_content=' Now we consider the link L3 (dichromatic singular Whitehead) as in the following figure 11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
358
+ page_content=' R ( ) x 1 y, R ( ) x 1 y, 2 R ( ) x y, x y y y 2 R ( u x u) , x u u := 1 R ( u x u) , 2 R ( ) x y, y 1 2 FIGURE 11.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
359
+ page_content=' Unoriented Dichromatic Singular Link(L3) The coloring equations are: R2(y ∗ R1(x, y), x ∗ (y ∗ R1(x, y))) = R1(x, y), and R1(y ∗ R1(x, y), x ∗ (y ∗ R1(x, y))) = R2(x, y) ∗ y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
360
+ page_content=' The system of these two equations reduces to � 0 = 3y2 + y + 2x, 0 = 3x2 + x + 2y, and thus we obtain that 2(y − x) = 0 giving x = y or y = x + 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
361
+ page_content=' Then Coldsq X (L3) = {(x, x), x ∈ X} ∪ {(x, x + 3), x ∈ X}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
362
+ page_content=' Thus the three links L1, L2 and L3 are pairwise distinct.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
363
+ page_content=' Example 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
364
+ page_content='5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
365
+ page_content=' Let 12 1, 32 1, 42 1, 52 1, 52 2, 52 3, 62 1, 62 2, 62 3, 62 4, 62 5, 62 6, 62 7, 62 8, 62 9, 62 10, 62 11, and 62 12 be the eigh- teen unoriented dichromatic singular links in Figure 12 and let X be the disingquandle in Example 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
366
+ page_content='4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
367
+ page_content=' By similar calculations as in the example, we obtain the following table: L #Coldsq X (L) 62 2 0 62 6 2 42 1, 62 12 18 12 1, 32 1, 52 1, 52 2, 52 3, 62 1, 62 3, 62 4, 62 5, 62 7, 62 8, 62 9, 62 10, 62 11 6 This table shows that the disingquandle counting invariant Zdsq X (L) distinguishes some of these eighteen unoriented dichromatic singular links.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
368
+ page_content=' 13 11 31 41 51 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 2 2 2 2 52 53 61 62 2 2 2 2 63 64 65 66 2 2 2 2 67 68 69 610 2 2 611 612 2 2 2 2 FIGURE 12.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
369
+ page_content=' Table of Unoriented Dichromatic Singular Links ACKNOWLEDGEMENT Mohamed Elhamdadi was partially supported by Simons Foundation collaboration grant 712462.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
370
+ page_content=' 14 REFERENCES [1] K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
371
+ page_content=' Bataineh, On skein theory of dichromatic links and invariants of finite type, Journal of Knot Theory and Its Ramifications.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
372
+ page_content=' 26(13) (2017) 1750092.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
373
+ page_content=' [2] K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
374
+ page_content=' Bataineh and I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
375
+ page_content=' Saidi, Involutory quandles and dichromatic links, Symmetry.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
376
+ page_content=' 12(1) (2020) 111.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
377
+ page_content=' [3] Madeline Brown and Sam Nelson, G-family polynomials, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
378
+ page_content=' Knot Theory Ramifications 30 (2021), no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
379
+ page_content=' 9, Paper No.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
380
+ page_content=' 2150070, 15, doi: 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
381
+ page_content='1142/S021821652150070X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
382
+ page_content=' MR4358333 [4] Jose Ceniceros, Indu R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
383
+ page_content=' Churchill, and Mohamed Elhamdadi, Singquandle shadows and singular knot invariants, Canad.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
384
+ page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
385
+ page_content=' Bull.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
386
+ page_content=' 65 (2022), no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
387
+ page_content=' 3, 770–787, doi: 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
388
+ page_content='4153/S0008439521000837.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
389
+ page_content=' MR4472501 [5] Indu R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
390
+ page_content=' U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
391
+ page_content=' Churchill, Mohamed Elhamdadi, Mustafa Hajij, and Sam Nelson, Singular knots and involutive quandles, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
392
+ page_content=' Knot Theory Ramifications 26 (2017), no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
393
+ page_content=' 14, 1750099, 14, doi: 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
394
+ page_content='1142/S0218216517500997.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
395
+ page_content=' MR3735400 [6] Jose Ceniceros, Indu R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
396
+ page_content=' Churchill, Mohamed Elhamdadi, and Mustafa Hajij, Singquandles, psyquandles, and singular knots: a survey, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
397
+ page_content=' Knot Theory Ramifications 30 (2021), no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
398
+ page_content=' 12, Paper No.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
399
+ page_content=' 2141003, 22, doi: 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
400
+ page_content='1142/S0218216521410030.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
401
+ page_content=' MR4394062 [7] , Cocycle invariants and oriented singular knots, Mediterr.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
402
+ page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
403
+ page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
404
+ page_content=' 18 (2021), no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
405
+ page_content=' 5, Paper No.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
406
+ page_content=' 217, 17, doi: 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
407
+ page_content='1007/s00009-021-01867-6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
408
+ page_content=' MR4309525 [8] Mohamed Elhamdadi and Sam Nelson, Quandles—an introduction to the algebra of knots, Student Mathematical Library, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
409
+ page_content=' 74, American Mathematical Society, Providence, RI, 2015.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
410
+ page_content=' MR3379534 [9] Allison Henrich and Sam Nelson, Semiquandles and flat virtual knots, Pacific J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
411
+ page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
412
+ page_content=' 248 (2010), no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
413
+ page_content=' 1, 155– 170, doi: 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
414
+ page_content='2140/pjm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
415
+ page_content='2010.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
416
+ page_content='248.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
417
+ page_content='155.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
418
+ page_content=' MR2734169 [10] Jim Hoste and Mark E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
419
+ page_content=' Kidwell, Dichromatic link invariants, Trans.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
420
+ page_content=' Amer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
421
+ page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
422
+ page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
423
+ page_content=' 321 (1990), no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
424
+ page_content=' 1, 197– 229, doi: 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
425
+ page_content='2307/2001598.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
426
+ page_content=' MR961623 [11] Jim Hoste and J´ozef H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
427
+ page_content=' Przytycki, An invariant of dichromatic links, Proc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
428
+ page_content=' Amer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
429
+ page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
430
+ page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
431
+ page_content=' 105 (1989), no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
432
+ page_content=' 4, 1003–1007, doi: 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
433
+ page_content='2307/2047066.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
434
+ page_content=' MR989100 [12] Atsushi Ishii, Masahide Iwakiri, Yeonhee Jang, and Kanako Oshiro, A G-family of quandles and handlebody- knots, Illinois J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
435
+ page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
436
+ page_content=' 57 (2013), no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
437
+ page_content=' 3, 817–838.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
438
+ page_content=' MR3275740 [13] David Joyce, A classifying invariant of knots, the knot quandle, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
439
+ page_content=' Pure Appl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
440
+ page_content=' Algebra 23 (1982), no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
441
+ page_content=' 1, 37–65, doi: 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
442
+ page_content='1016/0022-4049(82)90077-9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
443
+ page_content=' MR638121 [14] Uwe Kaiser, Strong band sum and dichromatic invariants, Manuscripta Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
444
+ page_content=' 74 (1992), no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
445
+ page_content=' 3, 237–251, doi: 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
446
+ page_content='1007/BF02567670.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
447
+ page_content=' MR1149761 [15] Sang Youl Lee and Mohd Ibrahim Sheikh, Diquandles and invariants for oriented dichromatic links, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
448
+ page_content=' Knot Theory Ramifications 30 (2021), no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
449
+ page_content=' 7, Paper No.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
450
+ page_content=' 2150049, 19, doi: 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
451
+ page_content='1142/S0218216521500498.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
452
+ page_content=' MR4321932 [16] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
453
+ page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
454
+ page_content=' Matveev, Distributive groupoids in knot theory, Mat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
455
+ page_content=' Sb.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
456
+ page_content=' (N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
457
+ page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
458
+ page_content=') 119(161) (1982), no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
459
+ page_content=' 1, 78–88, 160 (Russian).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
460
+ page_content=' MR672410 [17] Sam Nelson, Natsumi Oyamaguchi, and Radmila Sazdanovic, Psyquandles, singular knots and pseudoknots, Tokyo J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
461
+ page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
462
+ page_content=' 42 (2019), no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
463
+ page_content=' 2, 405–429, doi: 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
464
+ page_content='3836/tjm/1502179287.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
465
+ page_content=' MR4106586 [18] Natsumi Oyamaguchi, Enumeration of spatial 2-bouquet graphs up to flat vertex isotopy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
466
+ page_content=' part B, Topology Appl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
467
+ page_content=' 196 (2015), no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
468
+ page_content=' part B, 805–814, doi: 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
469
+ page_content='1016/j.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
470
+ page_content='topol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
471
+ page_content='2015.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
472
+ page_content='05.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
473
+ page_content='049.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
474
+ page_content=' MR3431017 [19] Hiroshi Tamaru, Two-point homogeneous quandles with prime cardinality, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
475
+ page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
476
+ page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
477
+ page_content=' Japan 65 (2013), no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
478
+ page_content=' 4, 1117–1134, doi: 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
479
+ page_content='2969/jmsj/06541117.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
480
+ page_content=' MR3127819 DEPARTMENT OF MATHEMATICS, GRADUATE SCHOOL OF NATURAL SCIENCES PUSAN NATIONAL UNIVER- SITY, BUSAN 46241, REPUBLIC OF KOREA Email address: ibrahimsheikh@pusan.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
481
+ page_content='ac.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
482
+ page_content='kr UNIVERSITY OF SOUTH FLORIDA, TAMPA, FLORIDA, USA Email address: emohamed@usf.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
483
+ page_content='edu DEPARTMENT OF MATHEMATICS, DALIAN UNIVERSITY OF TECHNOLOGY, CHINA Email address: danishali@mail.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
484
+ page_content='dlut.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
485
+ page_content='edu.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
486
+ page_content='cn 15' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/D9E2T4oBgHgl3EQfSgcI/content/2301.03792v1.pdf'}
DNFRT4oBgHgl3EQfxDge/content/tmp_files/2301.13640v1.pdf.txt ADDED
@@ -0,0 +1,894 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Vacuum enhanced charging of a quantum battery
2
+ Tiago F. F. Santos,1 Yohan Vianna de Almeida,1 and Marcelo F. Santos1, ∗
3
+ 1Instituto de F´ısica, Universidade Federal do Rio de Janeiro,
4
+ CP68528, Rio de Janeiro, Rio de Janeiro 21941-972, Brazil
5
+ (Dated: February 1, 2023)
6
+ Quantum batteries are quantum systems that store energy which can then be used for quantum
7
+ tasks. One relevant question about such systems concerns the differences and eventual advantages
8
+ over their classical counterparts, whether in the efficiency of the energy transference, input power,
9
+ total stored energy or other relevant physical quantities. Here, we show how a purely quantum
10
+ effect related to the vacuum of the electromagnetic field can enhance the charging of a quantum
11
+ battery. In particular, we demonstrate how an anti-Jaynes Cummings interaction derived from an
12
+ off-resonant Raman configuration can be used to increase the stored energy of an effective two-level
13
+ atom when compared to its classically driven counterpart, eventually achieving full charging of the
14
+ battery with zero entropic cost.
15
+ The quest for advanced quantum technologies or the ir-
16
+ reversible role of measurements in quantum dynamics are
17
+ examples of subjects that have stimulated the study of
18
+ thermodynamics in the microscopic world. An important
19
+ recent topic of investigation involves the role played by
20
+ quantum resources in the storage and use of energy by
21
+ quantized systems [1–19]. For example, coherence and
22
+ entanglement have been proven useful to speed up or
23
+ to super-extend the charging of quantum batteries [20–
24
+ 27]. Experimental results have also shown advances to-
25
+ wards the production of microscopic quantum thermal
26
+ machines and quantum batteries [28–33]. Most results re-
27
+ garding quantum properties influencing the performance
28
+ of quantum batteries, however, focus on increasing the
29
+ power of the process rather than enhancing the charg-
30
+ ing capacity. That is because the latter usually requires
31
+ entropy producing mechanisms [7–12, 18] that have dele-
32
+ terious effects in properties such as coherence and entan-
33
+ glement.
34
+ In this work we investigate how the quantized nature
35
+ of part of an entropy preserving charging circuit can in-
36
+ fluence the charging of a quantum battery. The circuit
37
+ comprises a classical power source (p.s.) and an auxiliary
38
+ frequency changer (f.c.). We compare the variation of the
39
+ internal energy stored in the battery and the efficiency of
40
+ the work extraction from the p.s., both for a classical and
41
+ quantum version of the f.c. component. In both cases,
42
+ the overall dynamics is unitary and, therefore, comes at
43
+ zero entropic cost.
44
+ In the classical scenario, both p.s.
45
+ and f.c. are connected to the battery for a fixed amount
46
+ of time, τc (“c” for classical), unitarily charging its ini-
47
+ tially thermal state: ρB(τc) = Uc(τc)ρT
48
+ BU −1
49
+ c
50
+ (τc), where
51
+ Uc(τc) is derived from the coupling Hamiltonian Hc =
52
+ HB0 + Vp.s.(t) + Vf.c.(t), Vj(t) is the potential created
53
+ by the circuit component j and HB0 is the free Hamilto-
54
+ nian of the battery. Thermal states are free resources in
55
+ thermodynamics [34–36] and, therefore, ideal to establish
56
+ the classical benchmark to be challenged by the quantum
57
+ version. The charging is measured by the variation ∆U
58
+ of internal energy of the battery, where U = Tr[ρBHB0].
59
+ In the quantized version, Vf.c.(t) is replaced by the inter-
60
+ action Hamiltonian HB−f.c.(t) and the initial state must
61
+ include the f.c. system which is also in a thermal state:
62
+ ρ(0) = ρT
63
+ B ⊗ ρT
64
+ f.c.. The variation of energy of the bat-
65
+ tery is now given by ∆U = Tr{[ρB(τq) − ρT
66
+ B]HB0} ( “q”
67
+ for quantum) where ρB(τq) = Trf.c.Uq(τq)ρ(0)U −1(τq)
68
+ and Uq is the time evolution operator obtained from
69
+ Hq = HB0 + Hf.c.0 + HB−f.c(t) + Vp.s.(t). Note that,
70
+ in both cases we assume isolation from the environment
71
+ and the charging does not produce any entropy. For com-
72
+ pleteness, we later add dissipative non-unitary terms to
73
+ the dynamics to verify how our results are affected by
74
+ the heat exchanged with surrounding reservoirs.
75
+ We investigate the classical protocol in a particular
76
+ setup where the battery is an oscillating two-level sys-
77
+ tem of frequency ωeg, the p.s.
78
+ generates an oscillat-
79
+ ing potential of frequency ωL > ωeg and the f.c.. gen-
80
+ erates another potential of frequency ωq = ωL − ωeg.
81
+ This situation is commonly found in many different
82
+ quantum optical experiments [37–43], where the battery
83
+ consists of two non-degenerate ground states {|g⟩, |e⟩}
84
+ (ωeg ≡ ωe − ωg > 0) of a real or artificial atom and
85
+ two modes of the electromagnetic field play the role of
86
+ power supply and f.c..
87
+ The couplings are intermedi-
88
+ ated by a third atomic level |m⟩ working as an ancilla
89
+ as depicted in Fig.
90
+ (1a).
91
+ Level |m⟩ should only con-
92
+ tribute virtually to the transference of energy and has
93
+ to be adiabatically eliminated from the dynamics. This
94
+ is achieved when each of p.s.
95
+ and f.c.
96
+ couples off-
97
+ resonantly one of the lower levels of the battery to |m⟩ in
98
+ a Raman configuration, where HB0 = ℏ �
99
+ j=g,e,m ωjσjj
100
+ (σjk ≡ |j⟩⟨k|), Vf.c.(t) = ℏΩq(σemeiωqt + σmee−iωqt)
101
+ and Vp.s.(t) = ℏΩL(σgmeiωLt + σmge−iωLt).
102
+ If ∆ =
103
+ ωmg − ωL = ωme − ωq ≫ ΩL, Ωq, the corresponding
104
+ time evolution Uc(t) induces Rabi oscillations between
105
+ levels |g⟩ and |e⟩ that are equivalent to directly cou-
106
+ pling them through one effective classical field of cou-
107
+ pling strength ¯Ω =
108
+ ΩLΩq
109
+
110
+ [44].
111
+ The optimal charg-
112
+ ing of the battery is then obtained for a full Rabi flip
113
+ that swaps the populations pT
114
+ g,e in the original ther-
115
+ arXiv:2301.13640v1 [quant-ph] 31 Jan 2023
116
+
117
+ 2
118
+ mal state ρT
119
+ B = �
120
+ j=g,e pT
121
+ j σjj, where pT
122
+ j =
123
+ e
124
+ ℏωj
125
+ KBT
126
+ ZB
127
+ and
128
+ ZB = �
129
+ j e
130
+ ℏωj
131
+ KBT .
132
+ In this case, ∆Uc = ℏωeg[pT
133
+ g − pT
134
+ e ].
135
+ Note that this is the most that a unitary transforma-
136
+ tion can charge an initially thermalized two-level bat-
137
+ tery and corresponds to the ergotropy Ec of the resulting
138
+ state, ρB(τc) = pT
139
+ e σgg + pT
140
+ g σee.
141
+ Ergotropy is defined
142
+ as Eρ(τ) = �
143
+ k,j rkEj(|⟨rk|Ej⟩|2 − δkj), where Ej are the
144
+ eigenenergies of H0 in increasing magnitude, i.e., Ei ≥ Ej
145
+ for i > j, and rk are the eigenvalues of ρ(τ) in decreasing
146
+ order, i.e., ri ≤ rj for i > j [45].
147
+ |m〉
148
+ |e〉
149
+ |g〉
150
+ Δ
151
+ 𝛀L
152
+ gq, 𝛀q
153
+ 𝚪mg
154
+ 𝚪gm
155
+ 𝚪em
156
+ 𝚪me
157
+ |e〉
158
+ |g〉
159
+ N = 1
160
+ |e, 0〉 |e, 1〉
161
+ |g, 0〉
162
+ N = 2
163
+ N = k
164
+ |e, 2〉
165
+ |e, k〉
166
+ |g, 1〉
167
+ |g, k-1〉
168
+ N = 1
169
+ N = 2
170
+ N = k
171
+ (a)
172
+ (b)
173
+ FIG. 1. (a) off-resonant Raman configuration: the battery is
174
+ a two-level atom ({|g⟩, |e⟩}); the p.s. is a laser of frequency
175
+ ωL (coupling ΩL), the f.c. is another harmonic oscillator of
176
+ frequency ωq and couplings Ωq (classical) and gq (quantum).
177
+ Level |m⟩ is an ancilla that intermediates both couplings.
178
+ Each channel can also exchange heat with the surrounding
179
+ reservoirs.the battery. (b) Selective scheme to charge the bat-
180
+ tery: in each step N, a selective Rabi flip transfers energy
181
+ from |g, N − 1⟩ to |e, N⟩.
182
+ If, now, the classical f.c. is replaced by a quantized
183
+ field, we need to add its free energy Hf.c.0 = ℏωqˆb†ˆb
184
+ to the Hamiltonian, where ˆb† creates an excitation,
185
+ and replace Vf.c.(t) by the interaction term HB−f.c. =
186
+ ℏgq(σemˆb†+σmeˆb). Once again, for ∆ ≫ ΩL, gq, we elim-
187
+ inate level |m⟩ and, as shown in [46–48], the dynamics of
188
+ the Battery-f.c. system becomes approximately given by
189
+ the effective Hamiltonian (ℏ = 1)
190
+ Heff = −g2
191
+ qN
192
+ ∆ σgg − g2
193
+ qˆb†ˆb
194
+
195
+ σee + ΩLgq
196
+
197
+ (σgeˆb + σegˆb†),
198
+ (1)
199
+ Note that Heff also includes a small correction to the
200
+ energy difference between levels |g⟩ and |e⟩, given by
201
+ ℏ∆N
202
+ eg = ℏ
203
+ Ω2
204
+ L−g2
205
+ qN
206
+
207
+ . This term, of the same order of Heff,
208
+ does not affect the conditions for eliminating |m⟩ and can
209
+ be physically implemented by applying a d.c. Stark shift
210
+ to the atom.
211
+ There are a few aspects of Heff useful for us: first,
212
+ the a.c. Stark shift correction to level |e⟩ depends on the
213
+ number of excitations of the f.c. and |e, 0⟩ is an eigen-
214
+ state of Heff with eigenvalue 0; second, the Rabi oscil-
215
+ lations occur in the joint Hilbert space of atom and f.c.,
216
+ splitting it into doublets {|g, n⟩, |e, n + 1⟩}. This corre-
217
+ sponds to the anti-Jaynes-Cummings (anti-JC) configu-
218
+ ration where the p.s. excites both the battery and the
219
+ f.c. at the same time. Third, each doublet oscillates at its
220
+ own Rabi frequency given by Ωn =
221
+
222
+ ∆2n/4 + G2n, where
223
+ ∆n =
224
+ r2Ω2
225
+ L(n+1−N)
226
+
227
+ , Gn =
228
+ rΩ2
229
+ L
230
+ √n+1
231
+
232
+ and r ≡
233
+ gq
234
+ ΩL , i.e.
235
+ each doublet is detuned from resonance by an amount
236
+ ∆n proportional to the number of excitations of the f.c..
237
+ Such Hamiltonians were predicted and implemented
238
+ in trapped ions, cavity QED and superconducting cir-
239
+ cuits, and for r ≫ 1, they operate in a selective regime
240
+ where ∆n ≫ Gn and the Rabi oscillation in all the dou-
241
+ blets is highly detuned except if n = N − 1.
242
+ In this
243
+ case, {|g, N −1⟩, |e, N⟩} oscillates resonantly (∆N−1 = 0,
244
+ ΩN−1 = rΩ2
245
+ L
246
+
247
+ N
248
+
249
+ ). Therefore, by properly choosing ∆N
250
+ eg
251
+ the battery population exchange is conditioned on the
252
+ number of excitations of the f.c. field as shown in [46, 47].
253
+ For example, for N
254
+ = 1, after an interaction time
255
+ τq =
256
+ π∆
257
+ 2rΩ2
258
+ L , the population in the {|g, 0⟩, |e, 1⟩} subspace
259
+ swaps while all other states only gain number dependent
260
+ phases. That takes the initial state ρ(0) = ρT
261
+ B ⊗ ρT
262
+ f.c. to
263
+ ρ(τq) = pT
264
+ e pT
265
+ 0 |e, 0⟩⟨e, 0| + pT
266
+ g pT
267
+ 0 |e, 1⟩⟨e, 1| + pT
268
+ e pT
269
+ 1 |g, 0⟩⟨g, 0|
270
+ + pT
271
+ g pT
272
+ 1 |g, 1⟩⟨g, 1| + (
273
+
274
+ n>1
275
+ pT
276
+ n|n⟩⟨n|) ⊗ ρT
277
+ B.
278
+ (2)
279
+ Here, ρT
280
+ f.c. = �
281
+ n pT
282
+ nσnn, pT
283
+ n = e−
284
+ nℏωq
285
+ KBT (1 − e−
286
+ ℏωq
287
+ KBT ). A
288
+ simple algebraic manipulation shows that this swap in-
289
+ creases the charge of the battery by ∆Uq = (pT
290
+ 0 pT
291
+ g −
292
+ pT
293
+ e pT
294
+ 1 )ℏωeg. In this case, there is an advantage over ∆Uc
295
+ if pT
296
+ e
297
+ pT
298
+ g > 1−pT
299
+ 0
300
+ 1−pT
301
+ 1 . We can better understand this condition
302
+ at low temperatures. When KBT ≪ ℏωq, ℏωm, the prob-
303
+ abilities pT
304
+ n are negligible for n > 1 and so is pT
305
+ m and we
306
+ can approximate 1 − pT
307
+ 1 ≈ pT
308
+ 0 and pT
309
+ e ≈ 1 − pT
310
+ g , mean-
311
+ ing that ∆Uq > ∆Uc if pT
312
+ e pT
313
+ 0
314
+ pT
315
+ g pT
316
+ 1 ≈ e
317
+ ℏωeg(ξ−1)
318
+ KBT
319
+ > 1, where
320
+ ξ =
321
+ ωq
322
+ ωeg . This happens whenever ξ > 1, i.e. whenever
323
+ the battery’s gap is smaller than one excitation of field ˆb.
324
+ In principle, the larger the value of ξ, the more accentu-
325
+ ated the enhancement due to the vacuum of field ˆb. This
326
+ is a purely quantum effect due solely to the vacuum of
327
+ the f.c. component.
328
+ Note, however, that the quantum protocol allows for
329
+ the relaxation of the ξ > 1 condition and an even more
330
+ enhanced charging, which is a much more powerful result,
331
+ due to the selectivity of Heff. In fact, similar Rabi flips
332
+ can be sequentially applied, each one tuned to resonance
333
+ by adjusting ∆N
334
+ eg in consecutive subspaces (N = 2, 3, ...)
335
+ as pictorially shown in Fig. (1b). In principle, this se-
336
+ quence must be infinite to maximize the charging of the
337
+ battery but, in practice, pT
338
+ n tends rapidly to zero unless
339
+ T is very high, and only a few cycles are required to ap-
340
+ proach maximum charging. After the sequence, the final
341
+
342
+ 3
343
+ state reads ρ(�
344
+ j τqj) ≈ [pT
345
+ e (1 − pT
346
+ 0 )σgg + (pT
347
+ g + pT
348
+ e pT
349
+ 0 )σee
350
+ and the variation of internal energy is ∆Uq = ∆Uc +
351
+ pT
352
+ e pT
353
+ 0 ℏωeg ≥ ∆Uc. This shows an advantage for any pos-
354
+ itive temperature and independent of ξ. More than that,
355
+ in the limit of ℏωq ≫ KBT, pT
356
+ 0 → 1 and the quantized
357
+ protocol fully charges the battery, independent of its ini-
358
+ tial state. This is a purely quantum effect due to the
359
+ vacuum of the f.c. and consists in the main result of this
360
+ paper. Not that similar charging can be obtained with
361
+ open system entropy producing dynamics, such as opti-
362
+ cal pumping. Here, we match it in an entropy preserving
363
+ protocol.
364
+ This sequence of cycles, however, can be cumbersome
365
+ and, in practice, escape from the isentropic condition of
366
+ no heat exchanged with external reservoirs. Furthermore,
367
+ the classical protocol is much faster, only requiring one
368
+ Rabi flip. One may wonder, then, if the quantized ad-
369
+ vantage still holds under equivalent restrictions. To an-
370
+ alyze this, we compute, from now on, single shot scenar-
371
+ ios designed with a sole detuning adjustment. The en-
372
+ ergy variation is obtained by solving the Von-Neumann
373
+ equation with Heff. The separation of Heff in doublets
374
+ makes it easy to derive the time evolution of the eigen-
375
+ states of HB0 + Hf.c.0. The anti-JC dynamics is similar
376
+ to the JC and it is simple to show that an initial state
377
+ |Ψ(0)⟩ = |g, n⟩ evolves to |Ψ(t)⟩ = e−i∆nt/2[(cos Ωnt +
378
+ i∆n
379
+ 2Ωn sin Ωnt)|g, n⟩ − iGn
380
+ Ωn sin Ωnt|e, n + 1⟩]. A similar ex-
381
+ pression can be found for the initial state |e, n+1⟩. There-
382
+ fore, after evolving for τq, the state of the battery changes
383
+ to ρB(τq) = Trf.c.[e−iHeff τq/ℏ(ρT
384
+ B ⊗ ρT
385
+ f.c.)eiHeff τq/ℏ] =
386
+ � pjσjj where pg = pT
387
+ g − S(τq), pe = pT
388
+ e + S(τq) and
389
+ pm = pT
390
+ m (due to the elimination of level |m⟩).
391
+ Here,
392
+ S(τq) = �∞
393
+ n=0 An[pT
394
+ g pT
395
+ n(0) − pT
396
+ e pT
397
+ n+1] sin2 �
398
+ Ωnτq
399
+ 2
400
+
401
+ , An =
402
+ 1
403
+ 1+ r2(n+1−N0)2
404
+ 4(n+1)
405
+ and r = gq
406
+ ΩL (see Sup. Mat. for full deriva-
407
+ tion). In this case, ∆Uq = ℏωegS(τq) and the battery’s
408
+ ergotropy reads Eq = ℏωeg[pT
409
+ e −pT
410
+ g +2S(τq)] = 2∆Uq−Ec.
411
+ The quantized version will be advantageous whenever
412
+ ∆Uq > Ec.
413
+ A quick inspection of S(τq) shows that, for single shots
414
+ (ss), it is the non-selective regime of r ≪ 1 that optimizes
415
+ the charging of the atom. In this case, all the doublets
416
+ evolve almost resonantly, each of them contributing to
417
+ enhance the charge. Because they oscillate at different
418
+ Rabi frequencies, it is impossible to choose a τq,ss that
419
+ simultaneously maximizes the energy transfer in all of
420
+ them. The optimal interaction time, which depends on
421
+ T, has to be numerically extracted by maximizing S(t)
422
+ and, because higher excited states oscillate faster, it gets
423
+ shorter for higher temperatures. In Fig. (2) we plot the
424
+ relative gain Kq ≡ ∆Uq
425
+ ss−∆Uc
426
+ ∆Uc
427
+ = ∆Uq
428
+ ss
429
+ ∆Uc − 1 induced by the
430
+ single shot quantized protocol as a function of ξ and for
431
+ two temperatures. Note that, similar to the single shot
432
+ selective case, Kq increases with T and requires ξ > 1 to
433
+ represent positive gain over the classical counterpart.
434
+ We also plot in the same figure the efficiency of the
435
+ work extraction, defined as η ≡
436
+ Eq
437
+ WL , where WL is the
438
+ work injected by the power supply.
439
+ The first law of
440
+ thermodynamics says that WL = ∆Uq + ∆Ufc where
441
+ ∆Ufc = ℏωqS(τq) is the energy variation of the f.c..
442
+ Therefore, the efficiency assumes the very simple formula
443
+ η =
444
+ 1
445
+ 1+ξ
446
+ 1+2Kq
447
+ 1+Kq . For a fixed value of ξ, the best efficiency,
448
+ η =
449
+ 2
450
+ 1+ξ, is achieved when Kq ≫ 1. On the other hand,
451
+ because ξ > 1 is a necessary condition for the advan-
452
+ tage of the single shot quantum protocol and because Kq
453
+ increases for larger values of ξ, it is clear that the best
454
+ gains are achieved at lower efficiencies. This should be
455
+ expected since ξ ≫ 1 means that most of the energy in-
456
+ jected by the power supply is actually going to the f.c..
457
+ Note that for each temperature, there is an ideal value of
458
+ ξ if one wishes for the best gain at a given efficiency.
459
+ FIG. 2. Relative gain Kq =
460
+ ∆Uq−∆Uc
461
+ ∆Uc
462
+ (blue, straight) and
463
+ efficiency η =
464
+ Eq
465
+ WL (red, curved) as a function of parameter
466
+ ξ =
467
+ ωq
468
+ ωeg for different values of the adimensional temperature
469
+ ¯T = KBT
470
+ ℏωm (≈ 0.1 for solid and ≈ 0.4 for dashed lines).
471
+
472
+ 2π = 1
473
+ MHz, gq =
474
+
475
+ 600, ΩL = ∆
476
+ 20.
477
+ So far, we have considered the isentropic injection of
478
+ energy by the external source.
479
+ However, neither the
480
+ battery nor the f.c.
481
+ are ever fully isolated from their
482
+ environment and there will always be heat exchanged
483
+ with the external reservoir. From the battery’s perspec-
484
+ tive, if both |g⟩ → |m⟩ and |e⟩ → |m⟩ transitions are
485
+ dipole coupled, levels |g⟩ and |e⟩ must be of the same
486
+ parity and, therefore, cannot be dipole coupled them-
487
+ selves. That means that the time scale for direct energy
488
+ exchange between them is usually much slower than any
489
+ other time scale of the problem and, in general, the cor-
490
+ responding heat channel can be ignored.
491
+ Considering
492
+ the standard weak coupling to thermal reservoirs, the
493
+ overall dynamics of the system is, then, governed by a
494
+ master equation of the form ˙ρ = − i
495
+ ℏ[Hq, ρ] + L(ρ) [49],
496
+ where L(ρ) = �
497
+ s Γs[2LsρL†
498
+ s − {L†
499
+ sLs, ρ}], with s =
500
+ gm, mg, em, me, +, −. The rates of the non-unitary parts
501
+ are given by Γjm = γ0j(¯nj + 1), Γmj = γ0j¯nj, Γ− =
502
+
503
+ 0
504
+ 20
505
+ 40
506
+ 60
507
+ 80
508
+ 100
509
+ 35 6
510
+ 0.6
511
+ Kq
512
+ n
513
+ 30 E
514
+ 0.5
515
+ 25
516
+ 0.4
517
+ 20
518
+ 0.3
519
+ 15
520
+ 0.2
521
+ 10 E
522
+ 0.1
523
+ 0.0
524
+ 0
525
+ 20
526
+ 40
527
+ 60
528
+ 80
529
+ 100
530
+ 54
531
+ γ0q(¯nq + 1), and Γ+ = γ0q¯nq.
532
+ Here, the γ0’s indicate
533
+ the spontaneous decay rates and ¯n’s the average number
534
+ of photons of the thermal reservoir at frequencies ωmj
535
+ and ωq. The respective jump operators are Ljk = σjk,
536
+ L− = ˆb and L+ = ˆb†.
537
+ 0.0
538
+ 0.5
539
+ 1.0
540
+ 1.5
541
+ 2.0
542
+ T
543
+ 0
544
+ 10
545
+ 20
546
+ 30
547
+ 40
548
+ 50
549
+ 60
550
+ Kq
551
+ 0 = 0
552
+ 0 = 0.01
553
+ gq
554
+ L
555
+ 0 = 0.1
556
+ gq
557
+ L
558
+ 0 =
559
+ gq
560
+ L
561
+ 0 = 10
562
+ gq
563
+ L
564
+ FIG. 3. Relative gain as a function of the adimensional tem-
565
+ perature ¯T ≡ kBT
566
+ ℏωm for different values of spontaneous decay
567
+ rates γ0 and for ξ = 99,
568
+
569
+ 2π = 1 MHz, gq =
570
+
571
+ 600, ΩL = ∆
572
+ 20. The
573
+ solid curve is obtained from the unitary evolution with Heff.
574
+ The dotted curves are numerical solutions of the open system
575
+ dynamics (master equation) with full Hamiltonian Hq.
576
+ The couplings to the thermal reservoirs establish at
577
+ least four typical regimes to the problem, depending on
578
+ their strength. The first one, already addressed, corre-
579
+ sponds to γ0’s much smaller than the effective coupling
580
+ gqΩL
581
+
582
+ and kBT ≪ ℏωeg, ℏωq. This is well approximated
583
+ by the isentropic dynamics considered so far. However,
584
+ we saw that the higher the temperature, the more advan-
585
+ tageous the quantum protocol is. This may not hold true
586
+ when we take into consideration the heat exchanges with
587
+ the reservoir. As the spontaneous decay rates increase,
588
+ a combination of effects begin to affect the charging of
589
+ the battery and may even create optimal temperatures
590
+ for better quantum gain.
591
+ In Fig. (3) we present Kq as a function of the adimen-
592
+ sional temperature ¯T ≡ kBT
593
+ ℏωm for different values of γ0. ¯T
594
+ is relevant to the problem because it regulates the pop-
595
+ ulation of level |m⟩. Although each reservoir has its own
596
+ spontaneous decay rate, they all produce similar effects
597
+ on both Kq and η, therefore we have considered a single
598
+ γ0 for all of them. The result was obtained by solving
599
+ the full dynamics of the open quantum system and choos-
600
+ ing the best τq,ss for each temperature. In these plots,
601
+ ωm
602
+ 2π = 1012Hz, ∆ = 2πMHz = 600g = 20ΩL, ξ = 99,
603
+ r = 1/30. As previously discussed, for γ0 ≪
604
+ gqΩL
605
+
606
+ we
607
+ reach the unitary regime calculated with Hamiltonian (1)
608
+ (solid curve), except for very high temperatures ( ¯T ∼ 2)
609
+ when the population of level |m⟩ becomes too significant
610
+ and start to affect the protocol as a whole. As we in-
611
+ crease γ0, effects such as decoherence of the f.c.
612
+ field
613
+ and the augmented relaxation rates Γj begin to limit the
614
+ quantum advantage. These effects become particularly
615
+ relevant when Γ’s rates approach the effective battery-
616
+ f.c. coupling gqΩL/∆. Note, however, that even for such
617
+ values of dissipation, the quantum protocol can still pro-
618
+ duce gains 30 times larger than its classical counterparts
619
+ for ξ = 99. Finally, a fourth effect takes place for higher
620
+ values of γ0 and at much higher temperatures: when Γ’s
621
+ become of the order of ∆ the heat exchange eventually
622
+ brings the transitions back into resonance in which case
623
+ level |m⟩ cannot be adiabatically eliminated anymore and
624
+ the charging scheme breaks down.
625
+ 0.00
626
+ 0.25
627
+ 0.50
628
+ 0.75
629
+ 1.00
630
+ 1.25
631
+ 1.50
632
+ 1.75
633
+ 2.00
634
+ T
635
+ 0.0000
636
+ 0.0025
637
+ 0.0050
638
+ 0.0075
639
+ 0.0100
640
+ 0.0125
641
+ 0.0150
642
+ 0.0175
643
+ 0.0200
644
+ 0 = 0
645
+ 0 = 0.01
646
+ gq
647
+ L
648
+ 0 = 0.1
649
+ gq
650
+ L
651
+ 0 =
652
+ gq
653
+ L
654
+ 0 = 10
655
+ gq
656
+ L
657
+ FIG. 4. Efficiency as a function of the adimensional tempera-
658
+ ture ¯T ≡ kBT
659
+ ℏωm for different values of spontaneous decay rates
660
+ γ0 and for ξ = 99,
661
+
662
+ 2π = 1 MHz, gq =
663
+
664
+ 600, ΩL =
665
+
666
+ 20. The
667
+ solid curve is obtained from the unitary evolution with Heff.
668
+ The dotted curves are numerical solutions of the open system
669
+ dynamics (master equation) with full Hamiltonian Hq.
670
+ In Fig.
671
+ (4) we repeat the numerical calculations of
672
+ the open system dynamics (same parameters), this time
673
+ for the efficiency. Again, we see that very low γ0’s are
674
+ consistent with the isentropic hypothesis, whereas higher
675
+ values of the spontaneous decay rates severely affect the
676
+ efficiency, specially for higher values of ¯T. Note that for
677
+ some parameters, the plotted efficiency is corrected to
678
+ η =
679
+ Eq
680
+ WL+Qem to adjust for the fact that the |e⟩ → |m⟩
681
+ reservoir may also inject energy in the system in the form
682
+ of heat Qem.
683
+ The correction takes place whenever we
684
+ obtain Qem > 0.
685
+ To conclude, we have shown that the quantized nature
686
+ of a component of a charging circuit can significantly
687
+ enhance the isentropic charging of a quantum battery
688
+ when benchmarked against its classical counterpart. This
689
+ is a purely quantum effect due to the vacuum state of
690
+ the quantized component and the ability to selectively
691
+ manipulate quantum states in the Hilbert space.
692
+ We
693
+ have also shown that our protocol can achieve the same
694
+ full charging capacity of open system entropy producing
695
+ equivalent schemes. We have demonstrated the effect in
696
+
697
+ 5
698
+ a typical setup of off-resonant Raman population transfer
699
+ in three-level λ−configuration where the power supply is
700
+ an external laser field and the quantized component is a
701
+ harmonic oscillator. This example is particularly useful
702
+ due to its broad presence in a variety of quantum opti-
703
+ cal setups such as trapped ions and atoms, cavity QED,
704
+ superconducting qubits, quantum dots and many other
705
+ equivalent experiments.
706
+ This
707
+ work
708
+ was
709
+ supported
710
+ by
711
+ CNPq
712
+ Projects
713
+ 302872/2019-1, INCT-IQ 465469/2014-0, and FAPERJ
714
+ project E-26/202.576/2019.
715
+ TFFS and YVA thank
716
+ Capes for financial support.
717
+ ∗ Corresponding author: [email protected]
718
+ [1] K. V. Hovhannisyan, M. Perarnau-Llobet, M.s Huber,
719
+ and A. Ac´ın, Entanglement Generation is Not Neces-
720
+ sary for Optimal Work Extraction, Phys. Rev. Lett. 111,
721
+ 240401 (2013).
722
+ [2] F. C. Binder, S. Vinjanampathy, K. Modi & J. Goold,
723
+ Quantacell: power charging of quantum batteries, New
724
+ Journal of Physics 17, 075015 (2015).
725
+ [3] D. Ferraro, M. Campisi, G. M. Andolina, V. Pellegrini &
726
+ M. Polini, High-power collective charging of a solid-state
727
+ quantum battery, Phys. Rev. Lett. 120, 117702 (2018).
728
+ [4] G. M. Andolina, M. Keck, A. Mari, M. Campisi, V. Gio-
729
+ vannetti and M. Polini, Extractable Work, the Role of
730
+ Correlations, and Asymptotic Freedom in Quantum Bat-
731
+ teries, Phys. Rev. Lett. 122, 047702 (2019).
732
+ [5] G. M. Andolina, D. Farina, A. Mari, V. Pellegrini, V.
733
+ Giovannetti, M. Polini, Charger-mediated energy trans-
734
+ fer in exactly solvable models for quantum batteries,
735
+ Phys.Rev. B 98, 205423 (2018).
736
+ [6] D. Farina, G. M. Andolina, A. Mari, M. Polini and
737
+ Vittorio Giovannetti, Charger-mediated energy transfer
738
+ for quantum batteries: An open-system approach, Phys.
739
+ Rev. B 99, 035421 (2019).
740
+ [7] F. Barra, Dissipative Charging of a Quantum Battery,
741
+ Phys. Rev. Lett. 122, 210601 (2019).
742
+ [8] A.C. Santos, B. C¸akmak, S. Campbell and N.T. Zinner,
743
+ Stable adiabatic quantum batteries, Phys. Rev. E 100,
744
+ 032107 (2019).
745
+ [9] A. Crescente, M. Carrega, M. Sassetti and D. Ferraro,
746
+ Charging and energy fluctuations of a driven quantum
747
+ battery, New J. Phys. 22 063057 (2020).
748
+ [10] Sergi Julia-F`arr´e, Tymoteusz Salamon, Arnau Riera,
749
+ Manabendra N. Bera, and Maciej Lewenstein, Bounds
750
+ on the capacity and power of quantum batteries, Phys.
751
+ Rev. Research 2, 023113 (2020).
752
+ [11] F. Tacchino, T. F. F. Santos, D. Gerace, M. Campisi, and
753
+ M. F. Santos, Charging a quantum battery via nonequi-
754
+ librium heat current, Phys. Rev. E 102, 062133 (2020).
755
+ [12] M. Carrega, A. Crescente, D. Ferraro and M. Sassetti,
756
+ Dissipative dynamics of an open quantum battery, New
757
+ J. Phys. 22 083085 (2020).
758
+ [13] F. H. Kamin, F. T. Tabesh, S. Salimi, and Alan C. San-
759
+ tos, Entanglement, coherence, and charging process of
760
+ quantum batteries, Phys. Rev. E 102, 052109 (2020).
761
+ [14] T. F. F. Santos, F. Tacchino, D. Gerace, M. Campisi
762
+ and M. F. Santos, Maximally efficient quantum thermal
763
+ machines fueled by nonequilibrium steady states, Phys.
764
+ Rev. A 103, 062225 (2021).
765
+ [15] Thiago R. de Oliveira and Daniel Jonathan, Efficiency
766
+ gain and bidirectional operation of quantum engines with
767
+ decoupled internal levels, Phys. Rev. E 104, 044133
768
+ (2021).
769
+ [16] Javier Carrasco, Jer´onimo R. Maze, Carla Hermann-
770
+ Avigliano, and Felipe Barra, Collective enhancement in
771
+ dissipative quantum batteries, Phys. Rev. E 105, 064119
772
+ (2022).
773
+ [17] I. Maillette de Buy Wenniger, S. E. Thomas, M. Maffei,
774
+ S. C. Wein, M. Pont, A. Harouri, A. Lemaˆıtre, I. Sagnes,
775
+ N. Somaschi,
776
+ A. Auff`eves,
777
+ P. Senellart. Coherence-
778
+ powered work exchanges between a solid-state qubit and
779
+ light fields, arXiv:2202.01109 (2022).
780
+ [18] T. F. F. Santos and Marcelo F. Santos, Efficiency of opti-
781
+ cally pumping a quantum battery and a two-stroke heat
782
+ engine, arXiv:2208.02373 (2022), accepted in Phys.
783
+ Rev. A.
784
+ [19] B. Ahmadi, P. Mazurek, R. R. Rodr´ıguez, S. Barzan-
785
+ jeh, R. Alicki, and P. Horodecki, Catalysis in Charging
786
+ Quantum Batteries, arXiv:2205.05018 (2022).
787
+ [20] Francesco Campaioli, Felix A. Pollock, Felix C. Binder,
788
+ Lucas C´eleri, John Goold, Sai Vinjanampathy, and Ka-
789
+ van Modi, Enhancing the Charging Power of Quantum
790
+ Batteries, Phys. Rev. Lett. 118, 150601 (2017).
791
+ [21] T. P. Le, J. Levinsen, K. Modi, M. M. Parish, and F.
792
+ A. Pollock, Spin-chain model of a many-body quantum
793
+ battery, Phys. Rev. A 97, 022106 (2018).
794
+ [22] YY. Zhang, TR. Lang, L. Fu and X. Wang, Powerful
795
+ harmonic charging in a quantum battery, Phys. Rev. E
796
+ 99, 052106.
797
+ [23] S. Zakavati, F. T. Tabesh, and S. Salimi, Bounds on
798
+ charging power of open quantum batteries, Phys. Rev.
799
+ E 104, 054117 (2021).
800
+ [24] S. Ghosh, T. Chanda, S. Mal, and A. Sen, Fast charging
801
+ of a quantum battery assisted by noise, Phys. Rev. A
802
+ 104, 032207 (2021).
803
+ [25] S.Seah, M. Perarnau-Llobet, G. Haack, N. Brunner, and
804
+ S. Nimmrichter, Quantum Speed-Up in Collisional Bat-
805
+ tery Charging, Phys. Rev. Lett. 127, 100601 (2021).
806
+ [26] James Q. Quach, and William J. Munro, Using Dark
807
+ States to Charge and Stabilize Open Quantum Batteries,
808
+ Phys. Rev. Applied, 14, 024092 (2020).
809
+ [27] JX. Liu, HL. Shi, YH. Shi, XH Wang, and WL. Yang, En-
810
+ tanglement and work extraction in the central-spin quan-
811
+ tum battery, Phys. Rev. B 104, 245418 (2021).
812
+ [28] J. Roßnagel, S. T. Dawkins, K. N. Tolazzi, O. Abah, E.
813
+ Lutz, F. Schmidt-Kaler, K. Singer, A single-atom heat
814
+ engine. Science 352, 325–329 (2016).
815
+ [29] D. von Lindenfels, O. Gr¨ab, C. T. Schmiegelow, V.
816
+ Kaushal, J. Schulz, M. T. Mitchison, J. Goold, F.
817
+ Schmidt-Kaler, and U. G. Poschinger, Spin heat engine
818
+ coupled to a harmonic oscillator flywheel, Phys. Rev.
819
+ Lett. 123, 080602 (2019).
820
+ [30] J. Klatzow, J. N. Becker, P. M. Ledingham, C. Weinzetl,
821
+ K. T. Kaczmarek, D. J. Saunders, J. Nunn, I. A. Walms-
822
+ ley, R. Uzdin, and E. Poem, Experimental demonstration
823
+ of quantum effects in the operation of microscopic heat
824
+ engines, Phys. Rev. Lett. 122, 110601 (2019).
825
+ [31] N. Van Horne, D. Yum, T. Dutta, P. H¨anggi, J. Gong,
826
+ D. Poletti,
827
+ and M. Mukherjee,
828
+ Single-atom energy-
829
+ conversion device with a quantum load, npj Quantum
830
+ Information 6, 37 (2020).
831
+
832
+ 6
833
+ [32] James Q. Quach, Kirsty E. McGhee, Lucia Ganzer, Do-
834
+ minic M. Rouse, Brendon W. Lovett, Erik M. Gauger,
835
+ Jonathan Keeling, Giulio Cerullo, David G. Lidzey, and
836
+ Tersilla Virgili. Superabsorption in an organic micro-
837
+ cavity:
838
+ Toward a quantum battery. Science advances,
839
+ 8(2):eabk3160, (2022).
840
+ [33] Chang-Kang Hu et al, Optimal charging of a super-
841
+ conducting quantum battery, Quantum Sci. Technol. 7
842
+ 045018 (2022)
843
+ [34] J. Goold, M. Huber, A. Riera, L. del Rio and P.
844
+ Skrzypczyk, The role of quantum information in thermo-
845
+ dynamics—a topical review, J. Phys. A: Math. Theor. 49
846
+ 143001 (2016).
847
+ [35] Matteo Lostaglio, An introductory review of the resource
848
+ theory approach to thermodynamics, Rep. Prog. Phys 82
849
+ 114001 (2019).
850
+ [36] V. Narasimhachar, S. Assad, F.C. Binder, J. Thomp-
851
+ son, B. Yadin, and Mile Gu, Thermodynamic resources
852
+ in continuous-variable quantum systems, npj Quantum
853
+ Inf 7, 9 (2021).
854
+ [37] Michael Fleischhauer, Atac Imamoglu, and Jonathan
855
+ P. Marangos, Electromagnetically induced transparency:
856
+ Optics in coherent media, Rev. Mod. Phys. 77, 633
857
+ (2005)
858
+ [38] K. Bergmann, H. Theuer, and B. W. Shore, Coherent
859
+ population transfer among quantum states of atoms and
860
+ molecules, Rev. Mod. Phys. 70, 1003 (1998)
861
+ [39] Petr Kr´al, Ioannis Thanopulos, and Moshe Shapiro, Col-
862
+ loquium: Coherently controlled adiabatic passage, Rev.
863
+ Mod. Phys. 79, 53 (2007)
864
+ [40] Nikolay V. Vitanov, Andon A. Rangelov, Bruce W.
865
+ Shore, and Klaas Bergmann, Stimulated Raman adia-
866
+ batic passage in physics, chemistry, and beyond, Rev.
867
+ Mod. Phys. 89, 015006 (2017)
868
+ [41] Carl E. Wieman, David E. Pritchard, and David J.
869
+ Wineland, Atom cooling, trapping, and quantum manip-
870
+ ulation, Rev. Mod. Phys. 71, S253 (1999)
871
+ [42] D. Leibfried, R. Blatt, C. Monroe, and D. Wineland,
872
+ Quantum dynamics of single trapped ions, Rev. Mod.
873
+ Phys. 75, 281 (2003)
874
+ [43] Klemens Hammerer, Anders S. Sørensen, and Eugene S.
875
+ Polzik, Quantum interface between light and atomic en-
876
+ sembles, Rev. Mod. Phys. 82, 1041 (2010).
877
+ [44] E Brion, L H Pedersen1 and K Mølmer, J. Phys. A: Math.
878
+ Theor. 40, 1033 (2007)
879
+ [45] A. E. Allahverdyan, R. Balian, and Th. M. Nieuwen-
880
+ huizen, Maximal work extraction from finite quantum
881
+ systems, Europhys. Lett. 67, 565 (2004).
882
+ [46] M. F. Santos, E Solano, RL de Matos Filho, Conditional
883
+ large Fock state preparation and field state reconstruc-
884
+ tion in Cavity QED, Phys. Rev. Lett. 87, 93601 (2001).
885
+ [47] MF Santos, Universal and deterministic manipulation of
886
+ the quantum state of harmonic oscillators: a route to
887
+ unitary gates for Fock State qubits. Phys. Rev. Lett. 95,
888
+ 010504 (2005).
889
+ [48] N. Zagury, A. Arag˜ao, J. Casanova, and E. Solano, Uni-
890
+ tary expansion of the time evolution operator, Phys. Rev.
891
+ A, 82, 042110 (2010)
892
+ [49] H. P. Breuer and F. Petruccione, The theory of open
893
+ quantum systems (Oxford University Press, 2002).
894
+
DNFRT4oBgHgl3EQfxDge/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
EdAzT4oBgHgl3EQfG_ss/content/2301.01036v1.pdf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:deaf9bb4f2b6b043ea7ee87421eaf95821498f83d800b3968044290517c5e949
3
+ size 45084191
G9E2T4oBgHgl3EQf-wmW/vector_store/index.faiss ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3f3e7c6f1fbc66f9cded53ddbff69d21ea9b49c15a980ca8ed20617f8d1e6292
3
+ size 3801133
GdE0T4oBgHgl3EQfzgIU/content/tmp_files/2301.02672v1.pdf.txt ADDED
@@ -0,0 +1,1253 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Detecting the heterodyning of gravitational waves
2
+ Jakob Stegmann∗ and Sander M. Vermeulen†
3
+ Gravity Exploration Institute,
4
+ School of Physics and Astronomy, Cardiff University,
5
+ Cardiff, CF24 3AA, United Kingdom
6
+ (Dated: January 10, 2023)
7
+ Gravitational waves modulate the apparent frequencies of other periodic signals. We propose to
8
+ use this effect to detect low-frequency gravitational waves by searching for correlated frequency
9
+ modulations in a large set of well-resolved gravitational wave signals.
10
+ We apply our proposed
11
+ method to the large number of gravitational wave signals from Galactic binary white dwarfs
12
+ that are expected to be detected with the planned space-based gravitational wave detector LISA.
13
+ We show that, given current projections for the number and properties of these sources and the
14
+ sensitivity of the instrument, this method would enable the detection of background gravitational
15
+ wave strain amplitudes of, e.g., A ≃ 10−10 at a frequency F ≃ 10−8 Hz. When using signals from
16
+ binary neutron stars such as those expected to be observed with proposed detectors like DECIGO,
17
+ we expect a sensitivity to gravitational waves competitive with that of current Pulsar Timing
18
+ Arrays.
19
+ This would allow the detection of gravitational waves from, e.g., super-massive black
20
+ hole binaries with chirp masses Mc ≳ 109 M⊙ at a distance D ≃ 10 Mpc. Our results show that
21
+ gravitational-wave detectors could be sensitive at frequencies outside of their designed bandwidth
22
+ using the same infrastructure.
23
+ This has the potential to open up unexplored and otherwise
24
+ inaccessible parts of the gravitational wave spectrum.
25
+ I.
26
+ INTRODUCTION
27
+ The field of gravitational-wave astronomy, as estab-
28
+ lished with the first direct detection of gravitational
29
+ waves (GWs) [1], is still in its infancy. So far, only GWs
30
+ with frequencies between ∼ 10 − 500 Hz produced by
31
+ the coalesecence of black holes and neutron stars with
32
+ masses ∼ 1 − 100 times the mass of our Sun have been
33
+ detected [2]. New detectors and techniques are being de-
34
+ veloped to probe different regions of the GW frequency
35
+ spectrum and to investigate numerous other potential
36
+ GW sources; e.g., rotating neutron stars [3], binary white
37
+ dwarfs (BWDs) [4], intermediate-mass and super-massive
38
+ binary black holes (SMBBHs) [5], a background of pri-
39
+ mordial GWs [6], and dark matter [7, 8].
40
+ The sensitive bandwidth of laser interferometers (the
41
+ only proven type of GW detector), is typically limited
42
+ at low frequencies by spurious accelerations of the test
43
+ masses, and at high frequencies by quantum uncertainty
44
+ in the optical state and an intrinsically decreased re-
45
+ sponse to GWs with wavelengths shorter than the in-
46
+ terferometer’s arms. Laser interferometers can be very
47
+ sensitive at higher frequencies (∼ 1 − 100 MHz), using
48
+ cross-correlation and shorter arms [9, 10]. Increasing the
49
+ sensitivity at lower frequencies is not straightforward,
50
+ and even a space-based instrument such as LISA [11],
51
+ subject to greatly reduced environmental noise, will not
52
+ be sensitive to GWs below ∼ 10−5 Hz. While marginal
53
+ gains have been made in understanding and addressing
54
+ ∗ StegmannJ@cardiff.ac.uk
55
+ † VermeulenSM@cardiff.ac.uk
56
+ the complex amalgam of low-frequency noise contribu-
57
+ tions encountered in laser interferometers (which include
58
+ fundamental quantum limits) [12], it seems unlikely that
59
+ their bandwidth will expand into lower frequencies by
60
+ more than an order of magnitude in the coming decades.
61
+ Other detection techniques to probe new areas of the
62
+ GW spectrum have been proposed and some have been
63
+ tried; none have proven successful in detecting GWs so
64
+ far.
65
+ At high frequencies (kHz – GHz) these include
66
+ techniques that exploit graviton-to-photon conversion
67
+ (known as the inverse Gertsenshtein effect) [13, 14], opti-
68
+ cally levitated sensors, resonant mass detectors [15], and
69
+ more [16]. At low frequencies, currently the only com-
70
+ petitive method to search for GWs is using sets of time-
71
+ resolved observations of pulsars, known as Pulsar Tim-
72
+ ing Arrays (PTAs), which are sensitive in the nHz – µHz
73
+ range [17–26]. GWs incident on the pulsar and/or the
74
+ detector produce deviations of the apparent frequency
75
+ or equivalently the arrival time of the radio pulses that
76
+ are correlated between different pulsars. This detection
77
+ technique thus exploits the interplay of electromagnetic
78
+ pulses with GWs which results in a modulation of the
79
+ pulse frequency. So far, after observing for ∼ 10 yr, PTAs
80
+ have not detected GWs [27–30].
81
+ In this work we propose a new method for detecting
82
+ (low-frequency) GWs using interactions between GWs of
83
+ different frequencies. The basis of the method is the grav-
84
+ itational red- and blueshift induced by one GW onto the
85
+ other. This mechanism can also be viewed as one GW
86
+ perturbing the space-time along the direction of travel of
87
+ the other GW, and thus modulating the arrival times of
88
+ peaks and troughs of the other GW. Mathematically, the
89
+ effect can be described as a multiplication or mixing of
90
+ two GWs. From this description, it can be shown that the
91
+ arXiv:2301.02672v1 [gr-qc] 6 Jan 2023
92
+
93
+ 2
94
+ resulting GW signal contains Fourier components at the
95
+ sum and difference of the frequencies of the two waves,
96
+ with an amplitude proportional to the product of the am-
97
+ plitudes of the individual GWs. This elementary result
98
+ of the mixing of two waves, also known as heterodyn-
99
+ ing, has been used in the processing of electromagnetic
100
+ signals for over a century. Heterodyning effectively pro-
101
+ duces a frequency-shifted copy of one signal (known as a
102
+ sideband) in the frequency range of a readily detectable
103
+ second signal. As we show in this paper, this mechanism
104
+ can be used in GW astronomy, where GW signals de-
105
+ tectable with, e.g., laser interferometers can be used to
106
+ detect low-frequency background GWs. This method of
107
+ searching for low-frequency GWs is conceptually similar
108
+ to the technique used by PTAs, with the crucial difference
109
+ that instead of looking for disturbances in the periodic
110
+ signal of pulsars, we look for disturbances in a periodic
111
+ GW signal. The idea of looking for GW sidebands was
112
+ recently independently proposed by Bustamante-Rosell
113
+ et al. [31], when our paper was in preparation, but their
114
+ analysis and projections differ significantly from ours.
115
+ Our proposed method allows one to expand the sen-
116
+ sitive bandwidth of GW detectors into low-frequency
117
+ regimes using the detectors’ existing infrastructures.
118
+ Moreover, this method could enable a sensitivity to GWs
119
+ in a bandwidth where no other detection methods exist,
120
+ e.g., in the µHz regime where the sensitivity of space-
121
+ based laser interferometers and PTAs leaves a gap.
122
+ Although our method is applicable to general periodic
123
+ GW signals, we focus here on the example of future space-
124
+ based laser-interferometric GW detectors, i.e., LISA [11]
125
+ and DECIGO [32], which are expected to be able to ob-
126
+ serve large numbers of GW signals from BWDs and bi-
127
+ nary neutron stars (BNSs). Using projected parameters
128
+ of the detector and signals for these instruments, we show
129
+ that cross-correlation of many well-resolved GW signals
130
+ can provide sensitivity to secondary low-frequency GWs.
131
+ II.
132
+ THEORY
133
+ We consider a set of N ≫ 1 periodic GW sources
134
+ which could be simultaneously observed for a long time
135
+ (e.g., BWDs in our Galaxy that could be individually
136
+ resolved by LISA [11]). We further assume that these
137
+ sources emit quasi-monochromatic GWs, i.e., that their
138
+ frequency does not significantly change within the obser-
139
+ vation time T (see Sec. V for discussion of the implica-
140
+ tions of relaxing this assumption). In that case we can
141
+ write the GW signal (in units of strain) from the α-th
142
+ periodic source at distance dα as
143
+ hα(t) = aα cos[2πfαt + ϕα],
144
+ (α = 1, 2, . . . , N),
145
+ (1)
146
+ with constant frequency fα, amplitude aα, and initial
147
+ phase ϕα. We refer to these GWs as carrier signals and
148
+ to their sources as carrier sources.
149
+ If there is an incident GW from a secondary, more
150
+ distant source, this GW will perturb the spacetime at
151
+ the location of the carrier sources and at the location of
152
+ the observer. As a consequence, the frequency of the GW
153
+ carrier signals are no longer constant but are modulated
154
+ in time. For a background GW emitted by a distant point
155
+ source in the direction ˆ
156
+ N this frequency modulation of
157
+ the carrier signal is given by [33],
158
+ fα − fα(t)
159
+
160
+ =
161
+ ni
162
+ αnj
163
+ α
164
+ 2(1 + ˆ
165
+ N · ˆnα)
166
+
167
+ hTT
168
+ ij (t) − hTT
169
+ ij (tα)
170
+
171
+ ,
172
+ (2)
173
+ where ˆnα and ni
174
+ α is the unit vector from the observer
175
+ to the α-th carrier source and its components, respec-
176
+ tively, and tα = t−dα(1+ ˆ
177
+ N · ˆnα)/c is the retarded time
178
+ coordinate that accounts for the propagation of the car-
179
+ rier wave. Additionally, hTT
180
+ ij (t) and hTT
181
+ ij (tα) correspond
182
+ to the metric perturbation due to the incident GW at
183
+ the spacetime locations of the carrier source and the ob-
184
+ server, respectively (in the terminology of Pulsar Timing
185
+ Arrays (PTAs) [17, 18], the former is usually referred to
186
+ as the ‘Earth term’ and the latter as the ‘pulsar term’).
187
+ It can be shown that the single-sided frequency spec-
188
+ trum of the modulated signal can then be written as [31]
189
+ ˜hα(f) ≃ aαeiϕαδ(fα − f)
190
+ + 1
191
+ 2aαAIα,Lei(ϕα+ΦL)δ(f − fα + FL)
192
+ + 1
193
+ 2aαAIα,Le−i(ϕα+ΦL)δ(f − fα − FL)
194
+ + 1
195
+ 2aαAIα,Dei(ϕα+Φα,D)δ(f − fα + FD,α)
196
+ + 1
197
+ 2aαAIα,De−i(ϕα+Φα,D)δ(f − fα − FD,α),
198
+ (3)
199
+ where Iα,L,D = (FL,D/fα) K( ˆ
200
+ N, ˆnα, hTT
201
+ ij , dα), and K is
202
+ a purely geometrical factor of order unity that accounts
203
+ for the polarisation, propagation direction, and propaga-
204
+ tion distance of the background and carrier GWs. The
205
+ first term in the spectrum given by Eq. (3) is the Fourier
206
+ component corresponding to the carrier signal at the fre-
207
+ quency f = fα. The modulation due to the background
208
+ GW at the location of the observer manifests as two
209
+ Fourier components with frequencies f = fα±FL (second
210
+ and third term in Eq. 3), which we will refer to as the
211
+ ’local’ sideband terms. Similarly, the modulation of the
212
+ carrier signal due to the background GW at the location
213
+ of the carrier source produces sidebands with frequencies
214
+ f = fα ± FD,α (fourth and fifth term), which we will
215
+ refer to as the ‘distant’ sideband terms. Note that the
216
+ frequency and phase offsets, FL, ΦL, of the ‘local’ terms
217
+ are independent of the carrier (they are equal to the fre-
218
+ quency and phase of the modulating GW at the location
219
+ of the observer), whereas the ’distant’ terms have fre-
220
+ quency and phase offsets FD,α, Φα,D, which depend on
221
+ the location of the carrier source.
222
+ This mechanism, a sort of ‘GW heterodyning’ could
223
+ allow the indirect detection of low-frequency GWs that
224
+ may otherwise be undetectable when a GW detector is
225
+ not sensitive to signals down to a frequency F, but is
226
+
227
+ 3
228
+ sensitive at much higher frequencies fα + F. Using this
229
+ method, the upconverted background signal amplitude
230
+ is Asideband = AaαKF/fα.
231
+ For example, if we take
232
+ the carrier signal to be the GWs emitted by a typical
233
+ BWD (such as the BWDs that LISA aims to detect),
234
+ with frequency fα ∼ 10−2 Hz, and we take the back-
235
+ ground signal to be GWs emitted by a SMBBH with
236
+ amplitude A ∼ 10−12 and frequency FL ∼ 10−8 Hz,
237
+ the background sideband signal appears at an amplitude
238
+ aαIα,L ∼ aα10−6.
239
+ This suppression relative to the carrier would mean the
240
+ background signal amplitude is below the typical noise
241
+ level of the detector. In the following section, we propose
242
+ a method to amplify the signal which utilises the coher-
243
+ ence of the modulation of multiple carrier signals. To this
244
+ end, we construct and add Np = N(N − 1)/2 ≫ 1 dif-
245
+ ferent cross-spectra (one for each pair of carrier sources)
246
+ such that the sideband terms sum up coherently to ex-
247
+ ceed the incoherent random noise.
248
+ III.
249
+ METHODS
250
+ We propose a cross-correlation method for detecting
251
+ a background gravitational wave signal that produces
252
+ phase modulation of carrier GW signals. We will later
253
+ use this method to make quantitative estimates of the
254
+ expected signal-to-noise ratio that can be obtained for
255
+ potential astrophysical GW sources using planned GW
256
+ detectors.
257
+ We consider the time-domain output signal of the GW
258
+ detector s(t) to be given by the sum of N carriers, all
259
+ modulated by a single background GW signal with fre-
260
+ quency F corresponding to either the ‘local’ (F = FL) or
261
+ the ‘distant’ (F = FD) term, and noise n(t) characteristic
262
+ of the detector
263
+ s(t) =
264
+ N
265
+
266
+ α=1
267
+ hα(t) + n(t).
268
+ (4)
269
+ For any carrier, we can apply a demodulation and phase-
270
+ shift to the time-domain detector output and normalise
271
+ it by the modulation index and the carrier amplitude
272
+ sα(t) =
273
+
274
+ 2
275
+ aαIα
276
+ e−i(2πfαt+ϕα) s(t).
277
+ (5)
278
+ This demodulation shifts the frequency of all Fourier
279
+ components in the output by an amount fα, such that
280
+ all sideband (heterodyne) signals are frequency shifted
281
+ to the frequency ±F of the modulating background GW
282
+ that produces them. Moreover, any heterodyne signals
283
+ from background GWs will now appear with a Fourier
284
+ amplitude equal to the background GW strain ampli-
285
+ tude that produces them. In general, the demodulation
286
+ frequency need not be constant in time, but could be ad-
287
+ justed over time to account for time-dependent changes
288
+ in the carrier frequency. Specifically, the demodulation
289
+ frequency and phase could be varied according to a pre-
290
+ determined carrier signal model, or they could be ad-
291
+ justed using feedback control (e.g., through maximising
292
+ the demodulated carrier amplitude) when the frequency
293
+ evolution is unknown a priori. After this frequency and
294
+ phase shift, we can apply an appropriate low-pass filter
295
+ to the data such that other terms, as long as they are
296
+ well-separated from the carrier and modulation sideband,
297
+ need not be considered [34].
298
+ We consider the case where the time-domain detec-
299
+ tor output is discretised with a constant sampling fre-
300
+ quency fs for a total observation time T. Next, we take
301
+ the single-sided discrete Fourier transform of the detec-
302
+ tor output, which yields a discrete complex amplitude
303
+ spectrum Sj
304
+ α for each carrier signal, which will have the
305
+ form
306
+ Sj
307
+ α = AeiΦαδjl(F ) +
308
+
309
+ 2
310
+ aαIα
311
+
312
+ ρj
313
+ α
314
+ T eiηj
315
+ α,
316
+ (6)
317
+ where the index j = 1, 2, . . . , Tfs/2 runs over the fre-
318
+ quency bins, l(F) [35] is the index of the bin that con-
319
+ tains the background signal (δjl is the Kronecker delta),
320
+ ρj
321
+ α is the noise power spectral density of the detector,
322
+ and ηj
323
+ α are the random noise phases (where both noise
324
+ parameters have undergone the frequency and phase shift
325
+ described by Eq. 5).
326
+ The spectrum Sj
327
+ α is unique for
328
+ each carrier signal. As background GWs would modu-
329
+ late all carrier signals coherently (i.e., the sideband phase
330
+ is deterministic), whereas the noise has a random phase,
331
+ cross-correlating different carrier signals is advantageous.
332
+ For each pair of carrier signals (αβ), a cross-spectrum
333
+ Sj
334
+ αβ = Sj
335
+ αSj∗
336
+ β , can be constructed which has the form
337
+ Sj
338
+ αβ = A2ei(Φα−Φβ)δjl(F ) +
339
+ 2
340
+ aαaβIαIβ
341
+
342
+ ρj
343
+ �j
344
+ β
345
+ T
346
+ ei(ηj
347
+ α−ηj
348
+ β),
349
+ (7)
350
+ where Φα − Φβ = Φαβ is the phase difference of the
351
+ modulating signal between the two carrier signals. From
352
+ this expression it can be seen that Φab is deterministic,
353
+ and ηj
354
+ α − ηj
355
+ β = ηj
356
+ αβ is random. Therefore, we can add
357
+ up signal terms from different cross-spectra coherently,
358
+ and the noise will average out. If we have N individ-
359
+ ually resolved carriers at our disposal we can construct
360
+ Np = N(N − 1)/2 different cross spectra and take a co-
361
+ herent weighted average of them
362
+ Sj =
363
+ �Np
364
+ (αβ) λj
365
+ αβSj
366
+ αβ e−iΦαβ
367
+ �Np
368
+ (αβ) λj
369
+ αβ
370
+ ,
371
+ (8)
372
+ where λj
373
+ αβ are the weights of each cross-spectrum. Per-
374
+ forming this coherent summation is possible as long as
375
+ the relative modulation sideband phase Φαβ can be de-
376
+ termined for each carrier pair (αβ). For the modulation
377
+ produced by the background GW at the detector (‘local’
378
+ term), Φαβ = 0 ∀ αβ. For the sideband due to the mod-
379
+ ulation produced at the source of the carrier GW signal
380
+
381
+ 4
382
+ (‘distant’ term), Φαβ is a function of the relative posi-
383
+ tions of the background GW source and the carrier signal
384
+ sources. In this case, Φab can be taken as free parame-
385
+ ters that are fit to the data by maximising the total SNR
386
+ for a particular sideband frequency, which would yield an
387
+ upper estimate of the maximum background GW signal
388
+ power at a certain frequency. Alternatively, a hypothet-
389
+ ical background source position and frequency could be
390
+ assumed, which prescribes a certain set of Φαβ given the
391
+ geometry of the source positions, which would then yield
392
+ an upper limit of the estimated background GW strain
393
+ at that frequency and sky position.
394
+ Note that the coherent average is constructed such
395
+ that
396
+ the
397
+ expected
398
+ real
399
+ part
400
+ of
401
+ the
402
+ signal
403
+ bin
404
+ is
405
+ E
406
+
407
+ Re[Sl(F )]
408
+
409
+ = A2.
410
+ The squared signal-to-noise ratio
411
+ can thus be defined for each bin
412
+ (SNRj)2 =
413
+
414
+ Re[Sj]
415
+ �2
416
+ Var (Re[Sj]).
417
+ (9)
418
+ It can be shown that an optimal signal-to-noise ratio is
419
+ found by taking the weights [36]
420
+ λj
421
+ αβ =
422
+ Np
423
+
424
+ (γδ)
425
+ ([Cj]−1)αβ,γδ ≃
426
+
427
+ 1
428
+ σj
429
+ ασj
430
+ β
431
+ �2
432
+ = (aαaβIαIβ)2T 2
433
+ 4ρj
434
+ αρj
435
+ β
436
+ ,
437
+ (10)
438
+ where Cj
439
+ αβ,δγ is the pair-wise cross-covariance matrix of
440
+ the cross-spectra Sj
441
+ αβ, Sj
442
+ δγ, and σj
443
+ α,β are the variances of
444
+ frequency bin j in each carrier spectrum (Eq. 6); the
445
+ approximation holds in the weak-signal limit [36]. The
446
+ SNR of a modulating background GW with frequency F
447
+ and amplitude A can now be evaluated
448
+ (SNRl(F ))2 ≃ A4
449
+ 2
450
+ �Np
451
+ (αβ)
452
+
453
+ 1
454
+ σl(F )
455
+ α
456
+ σl(F )
457
+ β
458
+ �2
459
+ .
460
+ (11)
461
+ The GW detector LISA is expected to observe a large
462
+ number of continuous, periodic GW signals from BWDs
463
+ in our Galaxy [4, 11, 37–42]. These BWDs could poten-
464
+ tially serve as carrier sources that allow for the detection
465
+ of low-frequency background GWs as described above.
466
+ The total number and properties of Galactic BWDs
467
+ is subject to large uncertainty. To obtain a quantitative
468
+ projection for the number, frequency, and amplitude of
469
+ BWD GW signals that may be detected with LISA, we
470
+ use an observationally driven parametric model of the
471
+ Galactic white dwarf population, constructed by Korol
472
+ et al. [42][43]. This model builds upon the spectroscopic
473
+ samples of single white dwarfs and BWDs from the Sloan
474
+ Digital Sky Survey (SDSS) and the Supernova Ia Progen-
475
+ itor surveY (SPY) to produce a synthetic population of
476
+ Galactic BWDs which are specified by their component
477
+ masses, orbital frequencies, sky positions, and orienta-
478
+ tions. These source parameters are then used to calculate
479
+ the GW signals of each BWD in the population. Part of
480
+ the BWDs would emit GWs at low frequencies f ≲ 3 mHz
481
+ and are predicted to be so numerous that they are not
482
+ TABLE I. Input parameters used for generating synthetic
483
+ populations of Galactic binary white dwarfs.
484
+ The parame-
485
+ ters ρKorol
486
+ WD,⊙, f Korol
487
+ BWD,4 AU, f Korol
488
+ BWD,amax, and αKorol are used as in-
489
+ put for the algorithm described by Korol et al. [42] to model
490
+ the sets of BWD carrier signals.
491
+ These parameters repre-
492
+ sent the local WD density, the fraction of binaries with semi-
493
+ major axes < 4 AU, the fraction of binaries with semi-major
494
+ axes less than the maximum separation detectable with LISA
495
+ (amax), and a power-law index specifying the BWD semi-
496
+ major axis distribution, respectively (see Korol et al. [42]
497
+ for details). The values of these parameters were chosen to
498
+ correspond to upper (Optimistic), median (Moderate), and
499
+ lower (Pessimistic) observational limits. We chose observa-
500
+ tion times T between 1.0 and 10.0 yr. N indicates the result-
501
+ ing number of BWDs which are individually resolvable with
502
+ LISA.
503
+ Model
504
+ Pessimistic Moderate Optimistic
505
+ ρKorol
506
+ WD,⊙
507
+ [10−3 pc−3]
508
+ 4.11
509
+ 4.49
510
+ 4.87
511
+ f Korol
512
+ BWD,4 AU
513
+ 0.112
514
+ 0.095
515
+ 0.078
516
+ f Korol
517
+ BWD,amax
518
+ 0.008
519
+ 0.009
520
+ 0.010
521
+ αKorol
522
+ −1.18
523
+ −1.30
524
+ −1.45
525
+ T
526
+ [yr]
527
+ 1.0
528
+ 4.0
529
+ 10.0
530
+ N
531
+ 7.0 × 104
532
+ 1.1 × 105
533
+ 1.9 × 105
534
+ individually resolvable but constitute a confusion-limited
535
+ foreground noise [39]. The rest, an estimated number of
536
+ ∼ O(103 – 105) BWDs emit GWs at higher frequencies
537
+ and are expected to be sufficiently loud that they are in-
538
+ dividually resolvable; these are the BWDs which can be
539
+ used as carrier sources in our method.
540
+ We consider three models with different carrier source
541
+ and observation parameters, Pessimistic, Moderate,
542
+ and Optimistic. For these models, we synthesized three
543
+ BWD populations using different input parameters for
544
+ the model of Korol et al. [42]; specifically we vary the
545
+ local WD density ρKorol
546
+ WD,⊙, the WD binary fraction f Korol
547
+ BWD,
548
+ and the power-law index αKorol, which describes the
549
+ BWD semi-major axis distribution (see Korol et al. [42]).
550
+ On the observation side we use three different values for
551
+ the LISA mission lifetime T = 1.0, 4.0, and 10.0 yr, which
552
+ sets the length of observation. To get an upper and lower
553
+ limit for the resulting sensitivity to background GWs, we
554
+ choose the model parameters such that Pessimistic and
555
+ Optimistic models yield the lowest and highest number
556
+ of BWDs within the current observational uncertainty
557
+ while Moderate model corresponds to median values.
558
+ The parameter values of the three different models are
559
+ summarised in Table I.
560
+ In Figure 1, we show the amplitude spectral density
561
+ (ASD) of the BWD carriers for each model together with
562
+ LISA’s projected detector noise amplitude spectral den-
563
+ sity, as in [44], modified to account for the confusion
564
+ noise due to unresolved BWDs derived by Korol et al.
565
+ [42].
566
+ Throughout this work we assume a BWD to be
567
+ individually resolvable if aα
568
+
569
+ T/ρα > 7, although the
570
+ precise threshold does not affect the resulting sensitiv-
571
+
572
+ 5
573
+ FIG. 1.
574
+ Amplitude spectral densities aα
575
+
576
+ T of gravita-
577
+ tional wave signals from individually resolvable binary white
578
+ dwarfs (BWDs) in three different models [42] as a function of
579
+ their frequency f = fα.
580
+ The solid line indicates the root
581
+ of the projected noise power spectral density √ρ of LISA
582
+ [42, 44].
583
+ BWDs are assumed to be individually resolvable
584
+ if aα
585
+
586
+ T/ρα > 7.
587
+ ity due to the dominant contribution of loud sources (see
588
+ Section V).
589
+ IV.
590
+ RESULTS
591
+ We estimate the sensitivity to background gravita-
592
+ tional waves for the three models using our method, as
593
+ in Eq. (11). Figure 2 shows the amplitude A versus fre-
594
+ quency F of a background GW that could be detected
595
+ with SNR = 2, corresponding to a ≃ 95 % detection prob-
596
+ ability.
597
+ The differences between the Pessimistic and
598
+ Optimistic models are less than one order of magnitude
599
+ in A. Our method is sensitive to GWs with frequencies
600
+ as low as F ∼ 10−8 Hz. GWs of these frequencies could
601
+ be present in our Universe, e.g., as part of a (stochas-
602
+ tic) background of GWs emitted by numerous individual
603
+ sources [46]. At a frequency of F ≃ 10−8 Hz our method
604
+ would be sensitive to amplitudes A ≳ 10−10; GWs of
605
+ that amplitude at that frequency could, e.g., be emitted
606
+ by a very massive SMBBH with a chirp mass of several
607
+ ∼ 1010 M⊙ at a distance D = 10 Mpc, which is the scale
608
+ of the Virgo cluster. No other method for detecting GWs
609
+ with frequencies between 10−6 and 10−5 Hz exists.
610
+ We also consider the more general case of a number
611
+ of carrier GW signals observed with any GW detector.
612
+ For this case we assume that all N carrier signals have a
613
+ similar frequency and are detected with the same SNR ∼
614
+
615
+
616
+ T/ρα = const. In Figure 3, we show the correlated
617
+ background GW amplitude that can be detected at an
618
+ SNR of one, as a function of the number and individual
619
+ SNR of the carrier signals.
620
+ We can apply this result to a proposed next-generation
621
+ GW detector such as DECIGO [47, 49, 50], which op-
622
+ erates in the dHz regime and is expected to observe
623
+ GWs from a large number of compact binary stars. As-
624
+ suming DECIGO observes GW signals from a popula-
625
+ tion of N = 105 binary neutron stars (BNSs) each ob-
626
+ served with an SNR of ∼ 104 [47] at a typical frequency
627
+ of fα = 0.1 Hz, it would be possible to detect back-
628
+ ground GWs from SMBBHs with chirp masses of about
629
+ ∼ 109 M⊙ (at a fiducial distance D = 10 Mpc and fre-
630
+ quency F = 10−8 Hz). This would make the sensitivity
631
+ of DECIGO to low-frequency GWs competitive with that
632
+ of current PTAs (cf. Figure 2).
633
+ For reference, we also indicate in Figure 3 the sensi-
634
+ tivity that could be obtained using ∼ 105 carrier signals
635
+ with an SNR ∼ 102 from compact binary coalescences,
636
+ as expected to be detected using both Einstein Telescope
637
+ (ET) and Cosmic Explorer (CE) [48]. These carrier sig-
638
+ nals would have frequencies between 10 and 103 Hz and
639
+ could be observed for a duration T ≲ 103 s, which means
640
+ the minimum detectable background GW frequency us-
641
+ ing our method is F ∼ 10−3 Hz. Coherent background
642
+ GW signals may be searched for using non-coincident
643
+ carrier signals with a slight modification of the method
644
+ described in Sec. III; a frequency-dependent phase correc-
645
+ tion (φcorr = 2πTdiffF) must be applied to each carrier’s
646
+ demodulated spectrum (Eq. 6), for a time difference be-
647
+ tween the signals Tdiff. In case the background GW signal
648
+ has a coherence time much shorter than the total obser-
649
+ vation time for all signals (i.e., the detector’s lifetime),
650
+ only coincident carrier signals can be cross-correlated to
651
+ gain sensitvity.
652
+ The sensitivity of our method is fundamentally lim-
653
+ ited to frequencies F ≳ 1/T, as for lower frequencies
654
+ the background signal cannot be distinguished from the
655
+ carrier [31]. The same low-frequency limit due to obser-
656
+ vation time exists for PTAs. The high-frequency limit of
657
+ our method is set by the Nyquist frequency of the detec-
658
+ tor output sampling, fs/2, where for LISA fs ∼ 1 Hz
659
+ [31]. PTAs have a much smaller sensitive bandwidth due
660
+ to the low observation cadence of radio telescopes (once
661
+ every several days or less).
662
+ V.
663
+ DISCUSSION
664
+ There are several effects that could in practice degrade
665
+ the sensitivity that would be obtained using our method.
666
+ Of particular concern is phase noise imparted by the
667
+ data acquisition system of the gravitational-wave detec-
668
+ tor. As this noise would appear as modulations of the car-
669
+ rier signal, it would obfuscate any background GWs that
670
+ produce the same effect. Phase noise in the data acqui-
671
+ sition system, due to, e.g., timing jitter of the sampling
672
+
673
+ Noise ASD
674
+ Moderate BWD ASD
675
+ Optimistic BWDASD
676
+ Pessimistic BWD ASD
677
+ 10-16
678
+ -18
679
+ 10
680
+ 10
681
+ 20
682
+ 10
683
+ 10
684
+ 10-3
685
+ 10-2
686
+ 10-1
687
+ Frequency f [Hz]6
688
+ FIG. 2. Sensitivity to low-frequency gravitational waves (GWs) that can be obtained by searching for correlated modulations
689
+ in a set of well-resolved GW signals from binary white dwarfs (BWDs), as expected to be detected with LISA. For reference,
690
+ we show the expected GW amplitudes of super-massive binary black holes with chirp masses ranging from 108 to 1011 M⊙ at a
691
+ fiducial distance D = 10 Mpc. We also show sensitivity curves from Pulsar Timing Arrays (PPTA [27]; EPTA [29]; NANOGrav
692
+ [45]). The detection threshold (SNR = 2) is chosen to allow a consistent comparison to reported PTA sensitivities. In practice,
693
+ we expect our method to show a reduction in sensitivity around F ≃ 1/yr ≃ 32 nHz as seen for PTAs, where it would be difficult
694
+ to distinguish a background GW from the Doppler modulation due the annual motion of LISA around the sun. The sensitivity
695
+ of our method is limited to frequencies F ≳ 1/T (e.g., 32 nHz in the Pessimistic model), below which the sensitivity is limited
696
+ by the finite width of the frequency bins.
697
+ clocks, would produce irreducible correlated noise in the
698
+ demodulated cross-spectra of different carriers. This ef-
699
+ fect might only be reduced by cross-correlating data ob-
700
+ tained with different uncorrelated oscillators. Similarly,
701
+ stochastic phase noise intrinsic to the carrier GW sig-
702
+ nal would reduce sensitivity to background GWs. In this
703
+ case the effect on the sensitivity is limited as this noise
704
+ will be uncorrelated between carriers and will be reduced
705
+ in the average cross-spectrum (Eq. 8).
706
+ In addition to these effective stochastic fluctuations
707
+ of the carrier signal, there could be deterministic fre-
708
+ quency changes of the carrier and background GWs. If
709
+ the frequency of the background GWs changes signifi-
710
+ cantly over the measurement time, i.e., if the GW back-
711
+ ground power spectral density is non-stationary, the co-
712
+ herent signal power would be spread over multiple fre-
713
+ quency bins, leading to a lower SNR in each bin.
714
+ An
715
+ SMBBH background source might undergo a significant
716
+ frequency evolution as its orbital period decays due to
717
+ energy loss by GW emission. Figure 4 shows that this
718
+ frequency change ˙F (‘chirp’) would not not be significant
719
+ for SMBBHs (Mc ≳ 109) over the duration of observa-
720
+ tion T ≃ 1 – 10 yr. Figure 4 also shows the expected fre-
721
+ quency changes of the LISA and DECIGO carrier signals.
722
+ In particular, it shows that most DECIGO BNSs undergo
723
+ significant frequency evolution over the duration of the
724
+ detected signal. As discussed in Sec. III, these frequency
725
+ changes could be compensated for at the demodulation
726
+ stage.
727
+ Non-stationarity of the background GW PSD has an-
728
+ other effect; the frequency change over a time equal
729
+ to the typical light travel time between the carrier
730
+ source and observer determines the frequency-space sep-
731
+ aration of the ‘local’ and ‘distant’ sideband terms, i.e.,
732
+ |FL − FD| ∝ dα ˙F/c. If these terms are not separated in
733
+ the spectrum, i.e., when |FL −FD| ≲ 1/T, coherent sum-
734
+ mation of the ‘local’ terms of different cross-spectra is
735
+ still possible but the ‘distant’ terms would add a small in-
736
+ coherent noise-like contribution to any signal bin. The in-
737
+ set of Figure 4 shows that given typical light travel times
738
+ between BWDs and the LISA detector of dα/c ≃ 10−1 –
739
+ 101 kpc/c [41], both separated and non-separated side-
740
+ bands could be observed for background SMBBH GW
741
+ sources. On the other hand, DECIGO will observe car-
742
+ rier signals from BNSs at much larger distances, e.g.,
743
+ dα ≃ 104 kpc for a GW170817-like event [52], and there-
744
+ fore ‘local’ and ’distant’ sidebands produced by a back-
745
+ ground SMBBH source (Mc ≳ 109 M⊙) would be well-
746
+ separated in DECIGO data.
747
+ We note that for the sensitivity projections for LISA,
748
+
749
+ 10-8
750
+ 10l1 Mo
751
+ Strain amplitude A
752
+ 1010Mo
753
+ Pessimistic
754
+ 12
755
+ 10
756
+ 109 Mo
757
+ Moderate
758
+ M
759
+ Optimistic.
760
+ NANOGrav
761
+ 08 M
762
+ 10-14
763
+ EPTA
764
+ PPTA
765
+ 10-8
766
+ 10-7
767
+ 10-6
768
+ 10-5
769
+ Frequency F [Hz]7
770
+ FIG. 3.
771
+ Order-of-magnitude estimate for the sensitivity to
772
+ background gravitational waves (GWs) by cross-correlating a
773
+ generic set of a number of GW signals N that are each de-
774
+ tected with a certain SNR (‘Carrier SNR’). The sensitivity
775
+ (given by the colour scale) is expressed as the product of the
776
+ background amplitude A times the typical frequency ratio of
777
+ the background and carrier signals fα/F, where the detection
778
+ threshold corresponds to an SNR equal to one. Furthermore,
779
+ we indicate the sensitivity that could be obtained using a set
780
+ of GW signals in the dHz regime from binary neutron stars as
781
+ carriers, which could be done using data from DECIGO [47],
782
+ and similarly the sensitivity using carrier signals detected us-
783
+ ing ET and CE [48]. We also show the sensitivity that could
784
+ be obtained using the average SNR of binary white dwarf
785
+ signals detected by LISA (in the Moderate model), as expli-
786
+ cated in Fig. 2. For these detectors we assume typical carrier
787
+ frequencies of fα ≃ 0.1 Hz (DECIGO), 10 Hz (ET/CE), and
788
+ 10−3 Hz (LISA). For reference, we show contour lines that cor-
789
+ respond to GW amplitudes from super-massive binary black
790
+ holes with chirp masses ranging from 109 to 1011 M⊙ at a
791
+ fiducial distance D = 10 Mpc, with a background frequency
792
+ F = 10−8 Hz, and a carrier frequency fα = 0.1 Hz.
793
+ the number N of individually resolvable BWDs in our
794
+ models (see Table I) is larger by a factor up to ∼ 10 com-
795
+ pared to previous estimates from Galaxy models com-
796
+ bined with a binary population model [4, 39, 41, 53, 54]
797
+ which reflects the large uncertainty of current predictions
798
+ about the detectable BWD population. However, the ex-
799
+ act total number of BWDs does not significantly affect
800
+ the estimated sensitivity because the ∼ O(103) loudest
801
+ BWDs signals provide the dominant contribution to the
802
+ sensitivity. This is shown in Figure 5; where we plot the
803
+ normalised cumulative contribution of BWDs to the total
804
+ SNR. It can be seen that several 102 to 103 BWDs are
805
+ enough to achieve similar sensitivities to the total BWD
806
+ population.
807
+ FIG. 4.
808
+ Timescale f/ ˙f = (5/96)(c3/GMc)5/3(πf)−8/3 at
809
+ which the frequency f of a compact binary with chirp
810
+ mass Mc significantly increases due to energy loss through
811
+ gravitational-wave emission. Coloured boxes indicate the pa-
812
+ rameter regions of background super-massive binary black
813
+ holes (SMBBHs), LISA binary white dwarfs (BWDs), and
814
+ DECIGO binary neutron stars (BNSs). This shows that LISA
815
+ BWDs and most of the SMBBHs would not undergo signifi-
816
+ cant frequency changes within the observation time T ≃ 1 –
817
+ 10 yr, whereas most DECIGO BNSs would. The inset shows
818
+ whether the SMBBHs would exhibit significant frequency
819
+ changes within typical light travel times between a carrier
820
+ source and the observer, i.e., whether ‘local’ and ‘distant’
821
+ sidebands overlap or not. For this figure we take the max-
822
+ imum GW frequency emitted by SMBBHs to correspond to
823
+ the Innermost Stable Circular Orbit f ≲ 1 kHz (M⊙/Mc) eval-
824
+ uated for equal-mass binaries [51], which causes the diagonal
825
+ cut-off.
826
+ VI.
827
+ CONCLUSION
828
+ In this work, we have outlined a method to use a set
829
+ of carrier gravitational wave sources to search for cor-
830
+ related frequency modulations caused by low-frequency
831
+ background GWs.
832
+ In this method demodulated cross-
833
+ spectra of carrier sources are added coherently and with
834
+ optimal weights such that any modulation common to
835
+ the carrier sources is amplified with respect to random
836
+ detector noise.
837
+ We considered the case of using our method to search
838
+ for low-frequency GWs in data from LISA, which is ex-
839
+ pected to detect GWs from a large number of Galac-
840
+ tic binary white dwarfs. The projected sensitivity that
841
+ could thus be obtained (Figure 2) ranges from strain am-
842
+ plitudes of A ∼ 10−10 at F ∼ 10−8 Hz to ∼ 10−7 at
843
+ ∼ 10−5 Hz, and would cover a part of the GW spectrum
844
+ where no other detection methods are currently available.
845
+
846
+ 105
847
+ 100
848
+ 10-1
849
+ 109 Mo×
850
+ 104
851
+ 10-2
852
+ DECIGO
853
+ 10-
854
+ 3
855
+ Carrier SNR
856
+ 103
857
+ 10-4
858
+ ET&CE
859
+ 10-5
860
+ 102
861
+ X
862
+ 1011 Ma
863
+ 10-6
864
+ LISA
865
+ 10-7
866
+ 101
867
+ 10-8
868
+ 100
869
+ 10-9
870
+ 101
871
+ 103
872
+ 105
873
+ 107
874
+ Number of carriers N1011
875
+ 109
876
+ Mc [Mo]
877
+ SMBBHS
878
+ 107
879
+ 100
880
+ 106
881
+ kpc
882
+ SMBBHS
883
+ kpc
884
+ kpc
885
+ kpc
886
+ c
887
+ c
888
+ 105
889
+ 10-9
890
+ 10-7
891
+ 10-5
892
+ f [Hz]
893
+ 12
894
+ Insignificant chirp
895
+ Significant chirp
896
+ 10yr
897
+ f/f>T
898
+ f/f<T
899
+ DECIGO BNSS
900
+ LISA BWDS8
901
+ FIG. 5. Cumulative SNR of a background gravitational wave
902
+ signal as a function of the n loudest binary white dwarfs
903
+ (BWDs) in the set of carrier signals. Stars at the end of each
904
+ line indicate the total number N of binaries in each model. In
905
+ any model several 102 to 103 of the loudest BWDs are enough
906
+ to achieve sensitivities similar to the entire sample.
907
+ This sensitivity could potentially enable the detection
908
+ of very massive SMBBHs with a chirp mass of several
909
+ 1010 M⊙ at a distance of D = 10 Mpc, if such systems
910
+ exist.
911
+ Single super-massive BHs of several ∼ 1010 M⊙
912
+ would be close to theoretical mass upper limits above
913
+ which they cannot grow through luminous gas accretion
914
+ [55], and so far candidates have only been observed at
915
+ distances of more than several ∼ 100 Mpc [e.g., 56, 57].
916
+ Our results show that an even better sensitivity could
917
+ be achieved using GW signals from compact binary stars
918
+ detectable with next-generation GW detectors that op-
919
+ erate in the dHz regime. In particular, using signals of
920
+ binary neutron stars expected to be detected with DE-
921
+ CIGO would yield a sensitivity competitive with that of
922
+ current pulsar timing arrays.
923
+ Our results show that future detectors designed to
924
+ detect GW signals in a higher frequency range could be
925
+ used to indirectly probe GWs down to the frequencies
926
+ given by the inverse instrument lifetime. Conveniently,
927
+ this could be achieved without modification of the
928
+ detector designs and with the same data outputs. This
929
+ method could therefore prove a valuable tool in the
930
+ exploration of the gravitational-wave spectrum and the
931
+ development of gravitational-wave astronomy in general.
932
+ ACKNOWLEDGEMENTS
933
+ We thank Vivien Raymond, Bangalore Sathyaprakash,
934
+ Fabio Antonini, Hartmut Grote, Guido M¨uller, Antoine
935
+ Petiteau, Martin Hewitson, Lucio Mayer, and Valeriya
936
+ Korol for helpful input and discussions.
937
+ DATA AVAILABILITY
938
+ The data underlying this article will be shared on rea-
939
+ sonable request to the authors.
940
+ [1] Abbott, B. P. et al., Observation of Gravitational Waves
941
+ from a Binary Black Hole Merger, Phys. Rev. Lett. 116,
942
+ 061102 (2016), arXiv:1602.03837 [gr-qc].
943
+ [2] The LIGO Scientific Collaboration, The population
944
+ of
945
+ merging
946
+ compact
947
+ binaries
948
+ inferred
949
+ using
950
+ grav-
951
+ itational
952
+ waves
953
+ through
954
+ GWTC-3,
955
+ arXiv
956
+ e-prints
957
+ ,
958
+ arXiv:2111.03634
959
+ (2021),
960
+ arXiv:2111.03634
961
+ [astro-
962
+ ph.HE].
963
+ [3] B. J. Owen, L. Lindblom, C. Cutler, B. F. Schutz, A. Vec-
964
+ chio, and N. Andersson, Gravitational waves from hot
965
+ young rapidly rotating neutron stars, Physical Review D
966
+ 58, 084020 (1998), arXiv:gr-qc/9804044 [gr-qc].
967
+ [4] G. Nelemans, L. R. Yungelson, and S. F. Portegies Zwart,
968
+ The gravitational wave signal from the Galactic disk pop-
969
+ ulation of binaries containing two compact objects, As-
970
+ tronomy & Astrophysics 375, 890 (2001), arXiv:astro-
971
+ ph/0105221 [astro-ph].
972
+ [5] A. Klein, E. Barausse, A. Sesana, A. Petiteau, E. Berti,
973
+ S. Babak, J. Gair, S. Aoudia, I. Hinder, F. Ohme, and
974
+ B. Wardell, Science with the space-based interferometer
975
+ eLISA: Supermassive black hole binaries, Physical Re-
976
+ view D 93, 024003 (2016), arXiv:1511.05581 [gr-qc].
977
+ [6] V. Mandic and A. Buonanno, Accessibility of the pre-
978
+ big-bang models to LIGO, Physical Review D 73, 063008
979
+ (2006), arXiv:astro-ph/0510341 [astro-ph].
980
+ [7] B. Carr, F. K¨uhnel, and M. Sandstad, Primordial black
981
+ holes as dark matter, Physical Review D 94, 083504
982
+ (2016), arXiv:1607.06077 [astro-ph.CO].
983
+ [8] R. Brito, S. Ghosh, E. Barausse, E. Berti, V. Cardoso,
984
+ I. Dvorkin, A. Klein, and P. Pani, Gravitational wave
985
+ searches for ultralight bosons with LIGO and LISA,
986
+ Physical Review D 96, 10.1103/PhysRevD.96.064050
987
+ (2017), arXiv: 1706.06311.
988
+ [9] S. M. Vermeulen, L. Aiello, A. Ejlli, W. L. Griffiths, A. L.
989
+ James, K. L. Dooley, and H. Grote, An experiment for
990
+ observing quantum gravity phenomena using twin table-
991
+ top 3D interferometers, Classical and Quantum Gravity
992
+ 38, 085008 (2021).
993
+ [10] A. Chou, H. Glass, H. R. Gustafson, C. Hogan, B. L.
994
+ Kamai, O. Kwon, R. Lanza, L. McCuller, S. S. Meyer,
995
+ J. Richardson, C. Stoughton, R. Tomlin, and R. Weiss,
996
+ The Holometer:
997
+ An Instrument to Probe Planckian
998
+ Quantum Geometry, Classical and Quantum Gravity 34,
999
+ 065005 (2017), arXiv: 1611.08265.
1000
+
1001
+ 100
1002
+ SNR(n ≤ N)/SNR(n = N)
1003
+ 10
1004
+ 10-2
1005
+ Pessimistic
1006
+ Moderate
1007
+ Optimistic
1008
+ 100
1009
+ 102
1010
+ 104
1011
+ n loudest BWDs9
1012
+ [11] P. Amaro-Seoane et al., Laser Interferometer Space
1013
+ Antenna,
1014
+ arXiv
1015
+ e-prints
1016
+ ,
1017
+ arXiv:1702.00786
1018
+ (2017),
1019
+ arXiv:1702.00786 [astro-ph.IM].
1020
+ [12] A. Buikema et al., Sensitivity and performance of the ad-
1021
+ vanced ligo detectors in the third observing run, Physical
1022
+ Review D 102, 062003 (2020).
1023
+ [13] M. E. Gertsenshtein, Wave resonance of light and gravi-
1024
+ tional waves, Sov Phys JETP 14, 84 (1962).
1025
+ [14] A. Ejlli, D. Ejlli, A. M. Cruise, G. Pisano, and H. Grote,
1026
+ Upper limits on the amplitude of ultra-high-frequency
1027
+ gravitational waves from graviton to photon conversion,
1028
+ Eur. Phys. J. C 79, 1032 (2019), eprint: 1908.00232.
1029
+ [15] M. Goryachev and M. E. Tobar, Gravitational wave de-
1030
+ tection with high frequency phonon trapping acoustic
1031
+ cavities, Physical Review D 90, 102005 (2014), publisher:
1032
+ American Physical Society.
1033
+ [16] N. Aggarwal, O. D. Aguiar, A. Bauswein, G. Cella,
1034
+ S. Clesse, A. M. Cruise, V. Domcke, D. G. Figueroa,
1035
+ A.
1036
+ Geraci,
1037
+ M.
1038
+ Goryachev,
1039
+ H.
1040
+ Grote,
1041
+ M.
1042
+ Hind-
1043
+ marsh, F. Muia, N. Mukund, D. Ottaway, M. Peloso,
1044
+ F. Quevedo, A. Ricciardone, J. Steinlechner, S. Stein-
1045
+ lechner, S. Sun, M. E. Tobar, F. Torrenti, C. ��Unal, and
1046
+ G. White, Challenges and opportunities of gravitational-
1047
+ wave searches at MHz to GHz frequencies, Living Re-
1048
+ views in Relativity 24, 4 (2021).
1049
+ [17] S. Detweiler, Pulsar timing measurements and the search
1050
+ for gravitational waves, Astrophys. J. 234, 1100 (1979).
1051
+ [18] M. V. Sazhin, Opportunities for detecting ultralong grav-
1052
+ itational waves, Sov. Astron. 22, 36 (1978).
1053
+ [19] B. Mashhoon, On the contribution of a stochastic back-
1054
+ ground of gravitationnal radiation to the timing noise
1055
+ of pulsars., Monthly Notices of the Royal Astronomical
1056
+ Society 199, 659 (1982).
1057
+ [20] B. Bertotti, B. J. Carr, and M. J. Rees, Limits from the
1058
+ timing of pulsars on the cosmic gravitational wave back-
1059
+ ground., Monthly Notices of the Royal Astronomical So-
1060
+ ciety 203, 945 (1983).
1061
+ [21] R. W. Hellings and G. S. Downs, Upper limits on the
1062
+ isotropic gravitational radiation background from pulsar
1063
+ timing analysis., The Astrophysical Journal Letters 265,
1064
+ L39 (1983).
1065
+ [22] R. S. Foster and D. C. Backer, Constructing a Pulsar
1066
+ Timing Array, Astrophys. J. 361, 300 (1990).
1067
+ [23] V. M. Kaspi, J. H. Taylor, and M. F. Ryba, High-
1068
+ Precision Timing of Millisecond Pulsars. III. Long-Term
1069
+ Monitoring of PSRs B1855+09 and B1937+21, Astro-
1070
+ phys. J. 428, 713 (1994).
1071
+ [24] F. A. Jenet, G. B. Hobbs, K. J. Lee, and R. N. Manch-
1072
+ ester, Detecting the Stochastic Gravitational Wave Back-
1073
+ ground Using Pulsar Timing, The Astrophysical Journal
1074
+ Letters 625, L123 (2005), arXiv:astro-ph/0504458 [astro-
1075
+ ph].
1076
+ [25] F. A. Jenet, G. B. Hobbs, W. van Straten, R. N. Manch-
1077
+ ester, M. Bailes, J. P. W. Verbiest, R. T. Edwards, A. W.
1078
+ Hotan, J. M. Sarkissian, and S. M. Ord, Upper Bounds on
1079
+ the Low-Frequency Stochastic Gravitational Wave Back-
1080
+ ground from Pulsar Timing Observations: Current Lim-
1081
+ its and Future Prospects, Astrophys. J. 653, 1571 (2006),
1082
+ arXiv:astro-ph/0609013 [astro-ph].
1083
+ [26] G. Hobbs, F. Jenet, K. J. Lee, J. P. W. Verbiest, D. Yard-
1084
+ ley, R. Manchester, A. Lommen, W. Coles, R. Edwards,
1085
+ and C. Shettigara, TEMPO2: a new pulsar timing pack-
1086
+ age - III. Gravitational wave simulation, Monthly No-
1087
+ tices of the Royal Astronomical Society 394, 1945 (2009),
1088
+ arXiv:0901.0592 [astro-ph.SR].
1089
+ [27] D. R. B. Yardley, G. B. Hobbs, F. A. Jenet, J. P. W. Ver-
1090
+ biest, Z. L. Wen, R. N. Manchester, W. A. Coles, W. van
1091
+ Straten, M. Bailes, N. D. R. Bhat, S. Burke-Spolaor,
1092
+ D. J. Champion, A. W. Hotan, and J. M. Sarkissian,
1093
+ The sensitivity of the Parkes Pulsar Timing Array to
1094
+ individual sources of gravitational waves, Monthly No-
1095
+ tices of the Royal Astronomical Society 407, 669 (2010),
1096
+ arXiv:1005.1667 [astro-ph.GA].
1097
+ [28] J. P. W. Verbiest et al., The International Pulsar
1098
+ Timing Array:
1099
+ First data release, Monthly Notices
1100
+ of the Royal Astronomical Society 458, 1267 (2016),
1101
+ arXiv:1602.03640 [astro-ph.IM].
1102
+ [29] S. Babak et al., European Pulsar Timing Array lim-
1103
+ its on continuous gravitational waves from individ-
1104
+ ual supermassive black hole binaries, Monthly Notices
1105
+ of the Royal Astronomical Society 455, 1665 (2016),
1106
+ arXiv:1509.02165 [astro-ph.CO].
1107
+ [30] Z. Arzoumanian et al., The NANOGrav 12.5 yr Data Set:
1108
+ Search for an Isotropic Stochastic Gravitational-wave
1109
+ Background, The Astrophysical Journal Letters 905, L34
1110
+ (2020), arXiv:2009.04496 [astro-ph.HE].
1111
+ [31] M.
1112
+ J.
1113
+ Bustamante-Rosell,
1114
+ J.
1115
+ Meyers,
1116
+ N.
1117
+ Pearson,
1118
+ C. Trendafilova, and A. Zimmerman, Gravitational wave
1119
+ timing array, Physical Review D 105, 044005 (2022),
1120
+ arXiv:2107.02788 [gr-qc].
1121
+ [32] S. Kawamura et al., The Japanese space gravitational
1122
+ wave antenna: DECIGO, Classical and Quantum Grav-
1123
+ ity 28, 094011 (2011).
1124
+ [33] M. Maggiore, Gravitational Waves, Vol. 2 (Oxford Uni-
1125
+ versity Press, Oxford, 2018).
1126
+ [34] The sidebands due to the local modulation can be con-
1127
+ sidered well-separated in the frequency domain from
1128
+ the sidebands due to the distant modulation (pul-
1129
+ sar/distant term) when |FL − FD| ≫ 1/T. We also as-
1130
+ sume all carrier signals are well-separated from each other
1131
+ (|fα − fβ| ≫ 1/T ∀ α, β).
1132
+ [35] l(F) = ⌈FT + 1
1133
+ 2⌉.
1134
+ [36] J.
1135
+ D.
1136
+ Romano,
1137
+ J.
1138
+ S.
1139
+ Hazboun,
1140
+ X.
1141
+ Siemens,
1142
+ and
1143
+ A. M. Archibald, Common-spectrum process versus
1144
+ cross-correlation for gravitational-wave searches using
1145
+ pulsar timing arrays, Phys. Rev. D 103, 063027 (2021).
1146
+ [37] D. Hils, P. L. Bender, and R. F. Webbink, Gravitational
1147
+ Radiation from the Galaxy, Astrophys. J. 360, 75 (1990).
1148
+ [38] A. J. Ruiter,
1149
+ K. Belczynski,
1150
+ M. Benacquista, and
1151
+ K. Holley-Bockelmann, The Contribution of Halo White
1152
+ Dwarf Binaries to the Laser Interferometer Space An-
1153
+ tenna Signal, Astrophys. J. 693, 383 (2009).
1154
+ [39] A. J. Ruiter, K. Belczynski, M. Benacquista, S. L. Lar-
1155
+ son, and G. Williams, The LISA Gravitational Wave
1156
+ Foreground: A Study of Double White Dwarfs, Astro-
1157
+ phys. J. 717, 1006 (2010), arXiv:0705.3272 [astro-ph].
1158
+ [40] S. Toonen, G. Nelemans, and S. Portegies Zwart, Su-
1159
+ pernova Type Ia progenitors from merging double white
1160
+ dwarfs. Using a new population synthesis model, Astron-
1161
+ omy & Astrophysics 546, A70 (2012), arXiv:1208.6446
1162
+ [astro-ph.HE].
1163
+ [41] A. Lamberts, S. Blunt, T. B. Littenberg, S. Garrison-
1164
+ Kimmel, T. Kupfer, and R. E. Sanderson, Predicting
1165
+ the LISA white dwarf binary population in the Milky
1166
+ Way with cosmological simulations, Monthly Notices
1167
+ of the Royal Astronomical Society 490, 5888 (2019),
1168
+ arXiv:1907.00014 [astro-ph.HE].
1169
+
1170
+ 10
1171
+ [42] V. Korol, N. Hallakoun, S. Toonen, and N. Karnesis, Ob-
1172
+ servationally driven Galactic double white dwarf popula-
1173
+ tion for LISA, Monthly Notices of the Royal Astronom-
1174
+ ical Society 511, 5936 (2022), arXiv:2109.10972 [astro-
1175
+ ph.HE].
1176
+ [43] https://gitlab.in2p3.fr/korol/
1177
+ observationally-driven-population-of-galactic-binaries,
1178
+ accessed January 6, 2023.
1179
+ [44] T. Robson, N. J. Cornish, and C. Liu, The construction
1180
+ and use of LISA sensitivity curves, Classical and Quan-
1181
+ tum Gravity 36, 105011 (2019), arXiv:1803.01944 [astro-
1182
+ ph.HE].
1183
+ [45] K. Aggarwal et al., The NANOGrav 11 yr Data Set: Lim-
1184
+ its on Gravitational Waves from Individual Supermas-
1185
+ sive Black Hole Binaries, Astrophys. J. 880, 116 (2019),
1186
+ arXiv:1812.11585 [astro-ph.GA].
1187
+ [46] A. Sesana, A. Vecchio, and C. N. Colacino, The stochastic
1188
+ gravitational-wave background from massive black hole
1189
+ binary systems: implications for observations with Pulsar
1190
+ Timing Arrays, Monthly Notices of the Royal Astronom-
1191
+ ical Society 390, 192 (2008), arXiv:0804.4476 [astro-ph].
1192
+ [47] N. Seto, S. Kawamura, and T. Nakamura, Possibility
1193
+ of Direct Measurement of the Acceleration of the Uni-
1194
+ verse Using 0.1 Hz Band Laser Interferometer Gravita-
1195
+ tional Wave Antenna in Space, PRL 87, 221103 (2001),
1196
+ arXiv:astro-ph/0108011 [astro-ph].
1197
+ [48] M. Maggiore, C. Van Den Broeck, N. Bartolo, E. Bel-
1198
+ gacem,
1199
+ D. Bertacca,
1200
+ M. A. Bizouard,
1201
+ M. Branch-
1202
+ esi, S. Clesse, S. Foffa, J. Garc´ıa-Bellido, S. Grimm,
1203
+ J. Harms, T. Hinderer, S. Matarrese, C. Palomba,
1204
+ M. Peloso, A. Ricciardone, and M. Sakellariadou, Science
1205
+ case for the Einstein telescope, Journal of Cosmology and
1206
+ Astroparticle Physics 2020 (3), 050, arXiv:1912.02622
1207
+ [astro-ph.CO].
1208
+ [49] K. Yagi and N. Seto, Detector configuration of DE-
1209
+ CIGO/BBO and identification of cosmological neutron-
1210
+ star binaries, Physical Review D 83, 044011 (2011),
1211
+ arXiv:1101.3940 [astro-ph.CO].
1212
+ [50] S. Isoyama, H. Nakano, and T. Nakamura, Multiband
1213
+ gravitational-wave astronomy: Observing binary inspi-
1214
+ rals with a decihertz detector, B-DECIGO, Progress
1215
+ of Theoretical and Experimental Physics 2018, 073E01
1216
+ (2018), arXiv:1802.06977 [gr-qc].
1217
+ [51] M. Maggiore, Gravitational Waves, Vol. 1 (Oxford Uni-
1218
+ versity Press, Oxford, 2008).
1219
+ [52] Abbott, B. P. et al., Multi-messenger Observations of a
1220
+ Binary Neutron Star Merger, The Astrophysical Journal
1221
+ Letters 848, L12 (2017), arXiv:1710.05833 [astro-ph.HE].
1222
+ [53] G. Nelemans, S. F. Portegies Zwart, F. Verbunt, and
1223
+ L. R. Yungelson, Population synthesis for double white
1224
+ dwarfs. II. Semi-detached systems: AM CVn stars, As-
1225
+ tronomy & Astrophysics 368, 939 (2001), arXiv:astro-
1226
+ ph/0101123 [astro-ph].
1227
+ [54] S. Nissanke, M. Vallisneri, G. Nelemans, and T. A.
1228
+ Prince,
1229
+ Gravitational-wave
1230
+ Emission
1231
+ from
1232
+ Compact
1233
+ Galactic
1234
+ Binaries,
1235
+ Astrophys.
1236
+ J.
1237
+ 758,
1238
+ 131
1239
+ (2012),
1240
+ arXiv:1201.4613 [astro-ph.GA].
1241
+ [55] A. King, How big can a black hole grow?, Monthly No-
1242
+ tices of the Royal Astronomical Society 456, L109 (2016),
1243
+ arXiv:1511.08502 [astro-ph.GA].
1244
+ [56] B. T. Dullo, A. W. Graham, and J. H. Knapen, A re-
1245
+ markably large depleted core in the Abell 2029 BCG IC
1246
+ 1101, Monthly Notices of the Royal Astronomical Society
1247
+ 471, 2321 (2017), arXiv:1707.02277 [astro-ph.GA].
1248
+ [57] B. T. Dullo, The Most Massive Galaxies with Large
1249
+ Depleted Cores:
1250
+ Structural Parameter Relations and
1251
+ Black Hole Masses, Astrophys. J. 886, 80 (2019),
1252
+ arXiv:1910.10240 [astro-ph.GA].
1253
+
GdE0T4oBgHgl3EQfzgIU/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
J9FOT4oBgHgl3EQfyjQh/content/2301.12928v1.pdf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dea0d376de2caa0e0bf669e8876cc44f538d311699b413117c65c2bca84508cc
3
+ size 218222
J9FOT4oBgHgl3EQfyjQh/vector_store/index.faiss ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:24505ee560ff77e6269e4567c5202375f0b20a2f8f9ab598e748994db9be7ea3
3
+ size 3538989
K9AyT4oBgHgl3EQfTveL/content/2301.00112v1.pdf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:617ae15b7a91bd757353d8f940c4005a6a8bae6db6630fcea2e8480f69de016c
3
+ size 350924
K9AyT4oBgHgl3EQfTveL/vector_store/index.faiss ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:96d36d5e35b8e10305b9ff0f17b083005ce031b8916ced0afaf45b738f427ed4
3
+ size 2621485
K9AyT4oBgHgl3EQfTveL/vector_store/index.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9423b6427a4a0e86d4092edf31e2f1b4aef135bf6602e7c4c6f68eb3feb672a0
3
+ size 92163
K9AzT4oBgHgl3EQfVvwt/vector_store/index.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0422682a5a14396bef025741b44d2fed0b28eb07cf7f91d234e8b44a0f44f382
3
+ size 246628
KE Research Center Intro/content/知识工程研究中心汇报20230710.pdf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2e05828542d76d25e0c7596f49c743cc54176accb6826681c453bc8675f82801
3
+ size 1432699