text
stringlengths 5
1.89M
| meta
dict | domain
stringclasses 1
value |
|---|---|---|
---
abstract: 'Sequences of commuting quantum operators can be parallelized using entanglement. This transformation is behind some optimal quantum metrology protocols and recent results on quantum circuit complexity. We show that dephasing quantum maps in arbitrary dimension can also be parallelized. This implies that for general dephasing noise the protocol with entanglement is not more fragile than the corresponding sequential protocol and, conversely, the sequential protocol is not less effective than the entangled one. We derive this result using tensor networks. Furthermore, we only use transformations strictly valid within string diagrams in dagger compact closed categories. Therefore, they apply verbatim to other theories, such as geometric quantization and topological quantum field theory. This clarifies and characterizes to some extent the role of entanglement in general quantum theories.'
author:
- Sergio Boixo
- Chris Heunen
bibliography:
- 'arxiv-sequentializable.bib'
title: Entangled and sequential quantum protocols with dephasing
---
One of the main goals of quantum computation and quantum information is the understanding of entanglement and its use to surpass the classical bounds for some given task. Quantum metrology is a case in point. It tries to exploit entanglement to measure physical parameters of a system to high precision [@Helstrom1976; @caves_quantum-mechanical_1981; @Holevo1982; @wineland_spin_1992; @braunstein_statistical_1994; @bollinger_optimal_1996; @huelga_improvement_1997; @buzek_optimal_1999; @meyer_experimental_2001; @ulam-orgikh_spin_2001; @fujiwara_estimation_2001; @lee_quantum_2002; @rudolph_quantum_2003; @mitchell_super-resolving_2004; @walther_broglie_2004; @andre_stability_2004; @luis_nonlinear_2004; @bagan_quantum_2004; @dunningham_sub-shot-noise-limited_2004; @chiribella_efficient_2004; @beltran_breaking_2005; @de_burgh_quantum_2005; @boixo_decoherence_2006; @monras_optimal_2006; @hayashi_parallel_2006; @giovannetti_quantum_2006; @fujiwara_strong_2006; @hayashi_quantum_2006; @higgins_entanglement-free_2007; @shaji_qubit_2007; @boixo_generalized_2007; @rey_quantum-limited_2007; @van_dam_optimal_2007; @monras_optimal_2007; @nagata_beating_2007; @boixo_quantum_2008; @choi_bose-einstein_2008; @boixo_quantum-limited_2008; @fujiwara_fibre_2008; @pezze_mach-zehnder_2008; @jones_magnetic_2009; @appel_mesoscopic_2009; @chase_magnetometry_2009; @maldonado-mundo_metrological_2009; @dorner_optimal_2009; @lee_optimization_2009; @aspachs_phase_2009; @maccone_robust_2009; @caves_quantum-circuit_2010; @genoni_optical_2010; @kacprowicz_experimental_2010; @modi_role_2010; @tilma_entanglement_2010; @rivas_precision_2010; @zwierz_general_2010; @kolstrokodynacuteski_phase_2010; @giovannetti_advances_2011; @escher_general_2011; @napolitano_interaction-based_2011; @mullan_improving_2011]. Consider, for example, the phase estimation problem in which one is given a phase gate $e^{-i \phi \sigma_z /2}$ as a black box and one is to find the corresponding phase $\phi$. A standard method to approach this problem is the so-called Ramsey interferometry, where the unitary $e^{-i \phi
\sigma_z /2}$ is applied to each subsystem of an initial quantum state $ \left(\ket 0 + \ket 1\right)^{\otimes n} /\sqrt{2^n}$. We are interested in the scaling with $n$ of the uncertainty $\delta \hat
\phi$ of the estimator $\hat \phi$. In Ramsey interferometry $\delta \hat \phi \propto 1/\sqrt n$, a scaling called the *shot noise* or *standard quantum limit*. In fact, as long as the initial state remains separable and one operator is applied to each subsystem, this bound cannot be surpassed (see supplementary material).
One way to improve this scaling is to introduce entanglement on the probe. The initial state, prior to the evolution, is now transformed into $ \left(\ket{0\cdots 0} +\ket {1 \cdots
1}\right)/ \sqrt 2$. This state can be achieved with an entangling operator (as described in detail below). Applying the phase gate to each subsystem and then a disentangling operator we obtain $ \left(\ket 0
+ e^{-i n \phi} \ket 1\right) \otimes \ket{0 \cdots 0}/\sqrt 2$. The uncertainty scaling is now $\delta \hat \phi \propto 1/n$, which is called the Heisenberg limit [@giovannetti_quantum_2006; @boixo_generalized_2007]. Intuitively, entanglement in the probe improves the measurement sensitivity [@caves_quantum-mechanical_1981]. This principle has been corroborated experimentally on several occasions [@walther_broglie_2004; @mitchell_super-resolving_2004; @nagata_beating_2007].
The scaling $\delta \hat \phi \propto 1/n$ can also be obtained in a different way. The final state of the first qubit in the entangled protocol is $ \left( \ket 0 + e^{-i n \phi} \ket 1\right)/ \sqrt
2$. The same state results when acting with $n$ sequential phase gates $\left(e^{-i \phi \sigma_z/2}\right)^n = e^{-i n \phi \sigma_z /2}$ on the first qubit initialized to $(\ket 0 + \ket 1 )/ \sqrt 2$. Notice that, if with a single application of the phase gate per measurement the uncertainty is constant $\delta \hat \phi \propto 1$, then with $n$ sequential unitaries the uncertainty is $\delta(n \hat \phi) \propto
1$, giving $\delta \hat \phi \propto 1/n$. This sequential version has been explored in frame synchronization [@rudolph_quantum_2003] and clock synchronization [@boixo_decoherence_2006; @higgins_entanglement-free_2007] between two parties.
Given a unitary operator $U$ as a black box and an input state $\ket
\psi = \sum_j c_j \ket{e_j}$, where $\ket{e_j}$ are eigenvectors of $U = \sum e^{i \varphi_j} \ket{e_j} \bra{e_j}$, Kitaev’s phase estimation algorithm [@kitaev_quantum_1995] outputs an approximation of the phase $\varphi_j$ corresponding to $\ket{e_j}$ with probability $|c_j|^2$. This algorithm is known to be optimal [@van_dam_optimal_2007], and it can also be transformed into the sequential protocol [@knill_optimal_2007]. The sequentialization is done by changing the quantum Fourier transform used in Kitaev’s phase estimation into a semi-classical Fourier transform [@griffiths_semiclassical_1996].
The transformation between sequential and entangled protocols has also been used to study the class QNC of quantum circuits with polylogarithmic depth [@moorenilsson:parallel]. This class includes, for instance, standard quantum error-correction encoding and decoding. Interestingly, if we had entangling gates with arbitrary fan-out at our disposal, then certain important functions that require classical circuits of logarithmic depth could be computed by quantum circuits of constant depth [@hoyer_quantum_2003].
In this paper we study the relation between entangled and sequential protocols using *tensor networks* [@schuch_computational_2007; @perez-garcia_matrix_2007; @vidal_entanglement_2007; @verstraete_matrix_2008; @dawson_unifying_2008; @cirac_renormalization_2009; @evenbly_tensor_2011], which encompass similar notations for Liouville space [@mukamel_principles_1999], quantum games [@gutoski_toward_2007] and the so-called quantum combs [@chiribella_quantum_2008; @chiribella_theoretical_2009]. We first introduce this notation for protocols with unitary operators. The diagrams derived are indeed very general, and they hold verbatim for quantum maps. Therefore, we extend the relation between entangled and sequential protocols to the presence of noise.
One reason for the generality of the diagrams used is that we only employ transformations strictly valid within string diagrams in dagger compact closed categories. Consequently, they apply to other quantum theories where the maps are not tensors (for instance, they can be of topological nature). While we refer to the literature for a formal introduction to this subject [@coeckepaquette:practisingphysicist; @selinger:graphicallanguages; @abramskycoecke:categoricalsemantics; @joyalstreet:geometry; @joyalstreet:braided], for the purpose of this paper it suffices to say that dagger compact closed categories capture exactly the necessary mathematically structure that makes the manipulations with string diagrams possible. In particular, composition is denoted by connecting maps, there is an abstract tensor structure between maps indicated by drawing maps in parallel, an abstract dagger action is represented by switching the input and the output of a map, and lines (the input/output of operations) can cross. All these operations are *compatible* between them, so the intuition gained with tensor diagrams still applies. Reciprocally, for tensor networks of quantum maps we will also use some adornments introduced in the context of completely positive categories which keep track (and partly explain) the emergence of transpose operations when carrying out certain manipulations (see Eq. \[eq:quantum\_maps\_adornments\]).
For a given Hilbert space ${\mathcal H}$ and a choice of basis $\{j\}$ the isometry $\delta = \sum_j \ket {jj} \bra {j}$ is an entangling operator. We denote it by $\vcenter{\hbox{\begin{tikzpicture}[scale=0.33]
\begin{pgfonlayer}{nodelayer}
\node [style=none] (0) at (1, .5) {};
\node [style=none] (1) at (-1, 0) {};
\node [style=dot] (2) at (0, 0) {};
\node [style=none] (3) at (1, -.5) {};
\end{pgfonlayer}
\begin{pgfonlayer}{edgelayer}
\draw (1) to (2);
\draw[out=-57,in=183] (2) to (3);
\draw[out=57,in=177] (2) to (0);
\end{pgfonlayer}
\end{tikzpicture}}}$. This is a generalization of the following quantum circuit: $$\begin{aligned}
\label{eq:1}
\vcenter{\vspace*{1.4em}\Qcircuit@C=1em@R=.7em@!R{
& \ctrl{1} &\qw\\
\lstick{\ket{0}} & \targ & \qw
}}
\end{aligned}$$ We now characterize commuting operators in finite-dimensional Hilbert spaces. States are represented by the diagram . For a given state $\ket a$, we can use the entangling operator to obtain a new operator $$\begin{aligned}
\label{eq:abs_phase_point}
\begin{tikzpicture}[scale=0.5, rotate=-90]
\begin{pgfonlayer}{nodelayer}
\node [style=none] (2) at (0, 0) {};
\node [style=box,minimum size = .5cm] (3) at (0, -1) {$\;a\;$};
\node [style=none] (4) at (0, -2) {};
\end{pgfonlayer}
\begin{pgfonlayer}{edgelayer}
\draw (2) to (3);
\draw (3) to (4.center);
\end{pgfonlayer}
\end{tikzpicture}\quad
: = \quad \begin{tikzpicture}[scale=0.33, rotate=-90]
\node [style=circle,draw=black] (0) at (-1, 2) {$a$};
\node [style=none] (1) at (1, 2) {};
\node [style=none] (2) at (-1, 1) {};
\node [style=none] (3) at (1, 1) {};
\node [style=dot] (4) at (0, 0) {};
\node [style=none] (5) at (0, -1) {};
\draw[looseness=1.25, bend left=315] (2.center) to (4);
\draw (4) to (5.center);
\draw (3.center) to (1);
\draw (2.center) to (0);
\draw[looseness=1.25, bend right=45] (4) to (3.center);
\end{tikzpicture}
\end{aligned}$$ which explicitly acts as $$\begin{aligned}
\sum_j \ket x \quad\mapsto\quad \sum_j {\langle a | j \rangle} {\langle j | x \rangle}
\ket{j} = \bra a \left( \sum_j \ket{jj} \bra j \right) \ket x\,.\end{aligned}$$ Commuting operators correspond exactly to different states with the same entangling operator or choice of basis. These operators also commute with the entangling operator (see supplementary material for a diagrammatic proof based on the associative property of $\delta$.) $$\begin{aligned}
\label{eq:associative}
\begin{tikzpicture}[scale=0.4, rotate=-90]
\begin{pgfonlayer}{nodelayer}
\node [style=none] (0) at (-1, 1) {};
\node [style=none] (1) at (1, 1) {};
\node [style=dot] (2) at (0, 0) {};
\node [style=box,minimum size = .5cm] (3) at (0, -1.5) {$\;a\;$};
\node [style=none] (4) at (0, -3) {};
\end{pgfonlayer}
\begin{pgfonlayer}{edgelayer}
\draw (2) to (3);
\draw[bend right=45, looseness=1.5] (2) to (1.center);
\draw (3) to (4.center);
\draw[bend right=45, looseness=1.5] (0.center) to (2);
\end{pgfonlayer}
\end{tikzpicture}
\quad & = \quad
\begin{tikzpicture}[scale=0.4, rotate=-90]
\begin{pgfonlayer}{nodelayer}
\node [style=none] (0) at (-1, 2) {};
\node [style=none] (1) at (1, 2) {};
\node [style=box,minimum size = .5cm] (2) at (-1, 1) {$\;a\;$};
\node [style=none] (3) at (1, 1) {};
\node [style=dot] (4) at (0, 0) {};
\node [style=none] (5) at (0, -1) {};
\end{pgfonlayer}
\begin{pgfonlayer}{edgelayer}
\draw (4) to (5.center);
\draw[bend right=45, looseness=1.50] (2.center) to (4);
\draw[bend right=45, looseness=1.25] (4) to (3.center);
\draw (1.center) to (3.center);
\draw (0.center) to (2);
\end{pgfonlayer}
\end{tikzpicture} \\
\sum_j \ket{jj} \bra{j} a \ket{j} \bra{j}
\quad & = \quad
\sum_j (a \otimes 1) \ket{jj} \bra{j}\end{aligned}$$ The physical interpretation is that there is no difference whatsoever between applying a commuting gate first and then the entangling operator or first entangling and then applying the gate.
Writing for the (unnormalized) equal superposition state $\sum \ket{j}$ for the given choice of basis, and for its adjoint $\sum \bra{j}$, we obtain equivalence between the following two diagrams for any set of commuting operators $\{f_j\}$: $$\begin{aligned}
\label{eq:parallelprotocol}
\vcenter{\hbox{\begin{tikzpicture}[scale=0.4]
\begin{pgfonlayer}{nodelayer}
\node [style=none] (0) at (-1, 2) {};
\node [style=box] (1) at (0, 2) {$\;\;f_n\;\;$};
\node [style=none] (2) at (1, 2) {};
\node [style=none] (3) at (-3, 1) {};
\node [style=dot] (4) at (-2, 1) {};
\node [style=dot] (5) at (2, 1) {};
\node [style=none] (6) at (3, 1) {};
\node [style=none] (7) at (-3.5, 0.5) {${\mathinner{\mskip1mu\raise1pt\vbox{\kern7pt\hbox{.}}
\mskip2mu\raise4pt\hbox{.}\mskip2mu\raise7pt\hbox{.}\mskip1mu}}$};
\node [style=none] (8) at (3.5, 0.8) {$\ddots$};
\node [style=none] (8a) at (0, -1) {$\vdots$};
\node [style=none] (9) at (-4, 0) {};
\node [style=none] (10) at (-1, 0) {};
\node [style=box] (11) at (0, 0) {$f_{n-1}$};
\node [style=none] (12) at (1, 0) {};
\node [style=none] (13) at (4, 0) {};
\node [style=dot] (14) at (-5, -1) {};
\node [style=dot] (15) at (5, -1) {};
\node [style=none] (16) at (-4, -2) {};
\node [style=none] (17) at (-1, -2) {};
\node [style=box] (18) at (0, -2) {$\;\;f_2\;\;$};
\node [style=none] (19) at (1, -2) {};
\node [style=none] (20) at (4, -2) {};
\node [style=dot] (21) at (-7, -2.25) {};
\node [style=dot] (22) at (-6, -2.25) {};
\node [style=dot] (23) at (6, -2.25) {};
\node [style=dot] (24) at (7, -2.25) {};
\node [style=none] (25) at (-5, -3.5) {};
\node [style=none] (26) at (-1, -3.5) {};
\node [style=box] (27) at (0, -3.5) {$\;\;f_1\;\;$};
\node [style=none] (28) at (1, -3.5) {};
\node [style=none] (29) at (5, -3.5) {};
\end{pgfonlayer}
\begin{pgfonlayer}{edgelayer}
\draw (25.center) to (26.center);
\draw[bend left=45] (22) to (14);
\draw (16.center) to (17.center);
\draw[bend left=45, looseness=1.25] (13.center) to (15);
\draw[bend right=45] (29.center) to (23);
\draw (21) to (22);
\draw (19.center) to (20.center);
\draw[bend left=45, looseness=1.25] (2.center) to (5);
\draw[bend left=45, looseness=1.25] (15) to (23);
\draw (26.center) to (28.center);
\draw[bend right=45, looseness=1.25] (20.center) to (15);
\draw[bend left=315] (22) to (25.center);
\draw[bend left=45, looseness=1.25] (14) to (9.center);
\draw[bend right=45, looseness=1.25] (4) to (10.center);
\draw (28.center) to (29.center);
\draw (3.center) to (4);
\draw (10.center) to (12.center);
\draw (23) to (24);
\draw (17.center) to (19.center);
\draw (5) to (6.center);
\draw[bend left=45, looseness=1.25] (4) to (0.center);
\draw (0.center) to (2.center);
\draw[bend right=45, looseness=1.25] (14) to (16.center);
\draw[bend right=45, looseness=1.25] (12.center) to (5);
\end{pgfonlayer}
\end{tikzpicture}}} \end{aligned}$$ $$\begin{aligned}
\label{eq:sequentialprotocol}
\vcenter{\hbox{\begin{tikzpicture}[scale=0.75]
\begin{pgfonlayer}{nodelayer}
\node [style=dot] (0) at (-4.5, 0) {};
\node [style=box] (1) at (-3, 0) {$\;f_{\pi(1)}\;$};
\node [style=box] (2) at (-1.5, 0) {$\;f_{\pi(2)}\;$};
\node [style=none] (3) at (-.5, 0) {};
\node [style=none] (4) at (-0.05, -0.05) {$\cdots$};
\node [style=none] (5) at (.4, 0) {};
\node [style=box] (6) at (1.5, 0) {$f_{\pi(n-1)}$};
\node [style=box] (7) at (3.1, 0) {$\,f_{\pi(n)}\,$};
\node [style=dot] (8) at (4.5, 0) {};
\end{pgfonlayer}
\begin{pgfonlayer}{edgelayer}
\draw (2) to (3.center);
\draw (5.center) to (6);
\draw (7) to (8);
\draw (0) to (1);
\draw (6) to (7);
\draw (1) to (2);
\end{pgfonlayer}
\end{tikzpicture}}}\end{aligned}$$ for any permutation $\pi \in
S(n)$. Diagram corresponds to the entangled protocol in three steps: entangling gate, commuting gate acting on each component, disentangling gate. Diagram clearly corresponds to the sequential protocol. The equivalence follows diagrammatically because every commuting operator $f_j$ can be moved to the beginning of the diagram, and the remaining loops can be cancelled by the isometry property (See supplementary material). Explicitly, the main step is as follows: $$\label{eq:useisometry}
\sum_{k,l} \ket{k} \bra{kk} 1 \otimes f_j \ket{ll} \bra{l}
= f_j
= \big (\sum_{k,l} \ket{k} {\langle kk | ll \rangle} \bra{l} \big ) \circ f_j,$$ where $f_j$ can be represented as in Eq. \[eq:abs\_phase\_point\]. Note that this equivalence also establishes the commutativity of the operators among themselves.
We have shown that the operators defined by the diagram of Eq. commute with the entangling gate. Using the (unnormalized) equal superposition state, we can prove the converse. Given an operator $a$ that commutes with an entangling operator, it is easy to see that the following diagram defines the corresponding state (see supplementary material): $$\begin{aligned}
\label{eq:abs_operator_state}
\begin{tikzpicture}[scale=0.33, rotate=-90]
\begin{pgfonlayer}{nodelayer}
\node [style=circle,draw=black] (3) at (0, 0) {$a$};
\node [style=none] (4) at (0, 2) {};
\end{pgfonlayer}
\begin{pgfonlayer}{edgelayer}
\draw (3) to (4.center);
\end{pgfonlayer}
\end{tikzpicture}
\quad := \quad
\begin{tikzpicture}[scale=0.33, rotate=-90]
\begin{pgfonlayer}{nodelayer}
\node [style=none] (2) at (0, 0) {};
\node [style=box,minimum size = .5cm] (3) at (0, -1.5) {$\;a\;$};
\node [style=dot] (4) at (0, -3) {};
\end{pgfonlayer}
\begin{pgfonlayer}{edgelayer}
\draw (2) to (3);
\draw (3) to (4.center);
\end{pgfonlayer}
\end{tikzpicture}\end{aligned}$$ Explicitly, the state $\ket{a} = \sum_j a \ket{j}$ and the operator $a$ are related by Eq. .
We have given a complete diagrammatic characterization of all the commuting operators and reviewed the known equivalence of the corresponding entangled and sequential protocols. We now extend this equivalence to quantum maps with general dephasing (a similar question has been addressed before for maps on qubits [@boixo_decoherence_2006]). To distinguish ${\mathcal H}$ from its dual space ${\mathcal H}^*$, we annotate the corresponding wires in diagrams with opposite arrows. Thus operators $c$ from ${\mathcal H}_A$ to ${\mathcal H}_B$ come in four variations: the original, the transpose $c^*$, the adjoint $c^\dagger$, and the conjugate $c_* =
(c^\dagger)^*$. $$\begin{aligned}
& \begin{tikzpicture}[scale=1,rotate=-90]
\begin{pgfonlayer}{nodelayer}
\node (1) at (0,-1) {${\mathcal H}_A$};
\node (2) [style=circle,draw=black] at (0,0) {$c\phantom{^\dag}$};
\node (3) at (0,1) {${\mathcal H}_B$};
\end{pgfonlayer}
\begin{pgfonlayer}{edgelayer}
\draw [arrow] (1) to (2) {};
\draw [arrow] (2) to (3) {};
\end{pgfonlayer}
\end{tikzpicture} &
& \begin{tikzpicture}[scale=1,rotate=-90]
\begin{pgfonlayer}{nodelayer}
\node (1) at (0,-1) {${\mathcal H}_B^*$};
\node (2) [style=circle,draw=black] at (0,0) {$c^*$};
\node (3) at (0,1) {${\mathcal H}_A^*$};
\end{pgfonlayer}
\begin{pgfonlayer}{edgelayer}
\draw [arrow] (3) to (2) {};
\draw [arrow] (2) to (1) {};
\end{pgfonlayer}
\end{tikzpicture} \nonumber \\
& \begin{tikzpicture}[scale=1,rotate=-90]
\begin{pgfonlayer}{nodelayer}
\node (1) at (0,-1) {${\mathcal H}_B$};
\node (2) [style=circle,draw=black] at (0,0) {$c^\dagger$};
\node (3) at (0,1) {${\mathcal H}_A$};
\end{pgfonlayer}
\begin{pgfonlayer}{edgelayer}
\draw [arrow] (1) to (2) {};
\draw [arrow] (2) to (3) {};
\end{pgfonlayer}
\end{tikzpicture} &
& \begin{tikzpicture}[scale=1,rotate=-90]
\begin{pgfonlayer}{nodelayer}
\node (1) at (0,-1) {${\mathcal H}_A^*$};
\node (2) [style=circle,draw=black] at (0,0) {$c_*$};
\node (3) at (0,1) {${\mathcal H}_B^*$};
\end{pgfonlayer}[
\begin{pgfonlayer}{edgelayer}
\draw [arrow] (3) to (2) {};
\draw [arrow] (2) to (1) {};
\end{pgfonlayer}
\end{tikzpicture} \label{eq:quantum_maps_adornments}\end{aligned}$$ We can now use the notation $\vcenter{\smash{\hbox{\begin{tikzpicture}[scale=0.15, rotate=-90]
\node (1) at (1,0) {};
\node (10) at (4, 0) {};
\draw [arrow,looseness=1, markwith={>}, bend left=90] (10) to (1) {};
\end{tikzpicture}}}}$ for the (entangled) state $ \sum_j \ket{\tilde{j} j}\in {\mathcal H}^* \otimes {\mathcal H}$, and $\vcenter{\smash{\hbox{\begin{tikzpicture}[scale=0.15, rotate=90]
\node (1) at (1,0) {};
\node (10) at (4, 0) {};
\draw [arrow,looseness=1, markwith={<}, bend left=90] (10) to (1) {};
\end{tikzpicture}}}}$ for its adjoint. Now, let $\mathcal A(\rho) = \sum_t a_t \cdot \rho \cdot a_t^\dagger$ be a completely positive map with Kraus operators $\{a_t\}$. Define the operator $a = \sum_t \ket t \otimes a_t$ (where the states $\ket
t$ live in a environment Hilbert space ${\mathcal H}_E$). Its diagrammatic representation is $$\begin{aligned}
\begin{tikzpicture}[scale=1, rotate=-90]
\begin{pgfonlayer}{nodelayer}
\node (1) at (0,-1) {};
\node (2) [style=circle,draw=black,thick] at (0,0) {$\mathcal A$};
\node (3) at (0,1) {};
\end{pgfonlayer}
\begin{pgfonlayer}{edgelayer}
\draw [thick] (1) to (2) {};
\draw [thick] (2) to (3) {};
\end{pgfonlayer}
\end{tikzpicture}
=
\begin{tikzpicture}[rotate=-90]
\begin{pgfonlayer}{nodelayer}
\node [style=none] (0) at (-1, 1) {${\mathcal H}_B^*$};
\node [style=none] (1) at (1, 1) {${\mathcal H}_B$};
\node [style=none] (2) at (-1, 0) {};
\node [style=none] (2b) at (-1, -0.3) {};
\node [style=none] (3) at (-0.5, 0) {};
\node [style=none] (4) at (0.5, 0) {};
\node [style=none] (5) at (1, 0) {};
\node [style=none] (5b) at (1, 0.2) {};
\node [style=none] (6) at (-1, -1) {${\mathcal H}_A^*$};
\node [style=none] (7) at (1, -1) {${\mathcal H}_A$};
\node [style=none] at (0, 0.9) {${\mathcal H}_E$};
\node [style=circle,draw=black,fill=white] at (-.75,0) {$a_*$};
\node [style=circle,draw=black,fill=white] at (.75,0) {$\;a\;$};
\end{pgfonlayer}
\begin{pgfonlayer}{edgelayer}
\draw[arrow] (0) to (2);
\draw[arrow, looseness=1.75, bend right=90] (4) to (3);
\draw[arrow] (2b) to (6);
\draw[arrow] (7) to (5);
\draw[arrow] (5b) to (1);
\end{pgfonlayer}
\end{tikzpicture}\end{aligned}$$ The second diagram corresponds to the Stinespring representation of $\mathcal
A$. All completely positive operators can be put in this form. The thick first diagram is defined as a more concise representation. Positive states $B = b^\dagger b$ also have this form, with trivial input space ${\mathcal H}_A = \mathbb{C}$. Every operator between Hilbert spaces can be lifted to a completely positive quantum map. The quantum map obtained by lifting the entangling operator $\delta$ for a given choice of basis has the form $$\begin{aligned}
\label{eq:cp_classical_structure}
\begin{tikzpicture}[scale=0.4, rotate=-90]
\node [style=none] (0) at (-1, 2) {};
\node [style=none] (1) at (1, 2) {};
\node [style=none] (2) at (-1, 1) {};
\node [style=none] (3) at (1, 1) {};
\node [style=dot] (4) at (0, 0) {};
\node [style=none] (5) at (0, -1) {};
\draw[looseness=1.25, bend left=315] (2.center) to (4);
\draw [arrow] (4) to (5.center);
\draw [arrow] (1) to (3.center);
\draw [arrow] (0) to (2.center);
\draw[looseness=1.25, bend right=315] (3.center) to (4);
\node [style=none] (10) at (2, 2) {};
\node [style=none] (11) at (4, 2) {};
\node [style=none] (12) at (2, 1) {};
\node [style=none] (13) at (4, 1) {};
\node [style=dot] (14) at (3, 0) {};
\node [style=none] (15) at (3, -1) {};
\draw[looseness=1.25, bend left=315] (12.center) to (14);
\draw [arrow] (15.center)to (14);
\draw [arrow] (13.center) to (11);
\draw [arrow] (12.center) to (10);
\draw[looseness=1.25, bend right=45] (14) to (13.center);
\end{tikzpicture}
\quad = \quad
\begin{tikzpicture}[scale=0.4, rotate=-90]
\node [style=none] (0) at (-1, 2) {};
\node [style=none] (1) at (1, 2) {};
\node [style=none] (2) at (-1, 1) {};
\node [style=none] (3) at (1, 1) {};
\node [style=dot, thick] (4) at (0, 0) {};
\node [style=none] (5) at (0, -1) {};
\draw[thick,looseness=1.25, bend left=315] (2.center) to (4);
\draw [thick] (4) to (5.center);
\draw [thick] (1) to (3.center);
\draw [thick] (0) to (2.center);
\draw[thick,looseness=1.25, bend right=315] (3.center) to (4);
\end{tikzpicture}
\end{aligned}$$ For any positive state and a choice of basis we can then characterize all commuting quantum maps in that basis. They are obtained exactly as in the diagrams of Eq. , and take the form $$\begin{aligned}
\label{eq:phase_point}
\begin{tikzpicture}[scale=0.4, rotate=-90]
\begin{pgfonlayer}{nodelayer}
\node [style=none] (0) at (-1, 4) {};
\node [style=circle,draw=black] (1) at (1,2) {$b_*$};
\node [style=none] (2) at (-1, 1) {};
\node [style=none] (3) at (1, 1) {};
\node [style=dot] (4) at (0, 0) {};
\node [style=none] (5) at (0, -1) {};
\node [style=circle,draw=black,minimum size = .75cm] (10) at (3, 2) {$\;b\;$};
\node [style=none] (11) at (5, 4) {};
\node [style=none] (12) at (3, 1) {};
\node [style=none] (13) at (5, 1) {};
\node [style=dot] (14) at (4, 0) {};
\node [style=none] (15) at (4, -1) {};
\end{pgfonlayer}
\begin{pgfonlayer}{edgelayer}
\draw[looseness=1.25, bend left=315] (2.center) to (4);
\draw [arrow] (4) to (5.center);
\draw (1) to (3.center);
\draw [arrow] (0) to (2.center);
\draw [arrow, looseness=1.25, bend right=315] (3.center) to (4);
\draw [arrow, markwith={<}, looseness=1.25, bend left=315] (12.center) to (14);
\draw [arrow] (15.center)to (14);
\draw [arrow] (13.center) to (11);
\draw (12.center) to (10);
\draw[looseness=1.25, bend right=45] (14) to (13.center);
\draw [arrow,looseness=1, bend right=90] (10) to (1) {};
\end{pgfonlayer}
\end{tikzpicture}
\quad = \quad
\begin{tikzpicture}[scale=1, rotate=-90]
\begin{pgfonlayer}{nodelayer}
\node (1) at (0,-1) {};
\node (2) [style=box,draw=black,thick] at (0,0) {$\;\;\mathcal B\;\;$};
\node (3) at (0,1) {};
\end{pgfonlayer}
\begin{pgfonlayer}{edgelayer}
\draw [thick] (1) to (2) {};
\draw [thick] (2) to (3) {};
\end{pgfonlayer}
\end{tikzpicture}\end{aligned}$$ Notice that the diagrammatic proofs (as in the supplementary material) stay exactly the same, whereas the direct computations as in Eq. become rather more involved.
All the diagrams introduced for unitary operators that commute with the entangling operator apply verbatim to quantum maps. Indeed, the quantum maps constructed through an entangling map commute and all maps commuting with an entangling map are of this particular form. Further, the equivalence between the entangled and sequential protocols is upheld.
We now give explicit equations for quantum maps that commute with an entangling map. From the defining diagram Eq. we see that the quantum map $\mathcal B$ corresponding to a positive operator $B$ acts as $$\mathcal B(\rho) = B^* \circ \rho\;,$$ where $\circ$ is the Schur product in the basis of the entangling map. All quantum maps of that form are completely positive (already shown diagrammatically), but in addition we want them to be trace preserving, which imposes $$\begin{aligned}
\label{eq:trace_condition}
B_{jj} = 1\;.\end{aligned}$$ The positive operator $B$ has a spectral decomposition $$B = \sum_s \ket{\chi_s} \bra {\chi_s} \;.$$ This gives the Kraus decomposition of $\mathcal B(\rho) = \sum_s b_s \cdot \rho \cdot b_s^\dagger $, with Kraus operators $$b_s = \sum_j {\langle \chi_s | j \rangle} \;\ket j \bra j\;,$$ Notice that in this representation the commuting quantum maps also have commuting Kraus operators.
For unitary phase maps the matrix $B$ must have the form $$(B_{\rm phases})_{jk} = e^{-i (\phi_j - \phi_k) } \ket j \bra k$$ for any choice of phases $\{\phi_j\}$. Maps with *dephasing* with respect to the basis of the entangling map are more interesting. The noise must be dephasing because it does not alter the populations, due to Eq. : $$(\mathcal B(\rho))_{jj} = B_{jj} \rho_{jj} = \rho_{jj}\;.$$ For the qubit case (i.e. when the dimension of the Hilbert space is two), the positive matrix $B$ has the form $$\left(
\begin{array}{cc}
1 & e^{- \gamma - i \phi} \\
e^{-\gamma + i \phi} & 1
\end{array}\right)\;,$$ where $\phi$ is a phase and $\gamma \ge 0$ parametrizes the dephasing. This quantum channel is $$e^{-i \phi \sigma_z /2} \cdot \left( (1-p)\; \rho + p \;\sigma_z \rho \sigma_z \right ) \cdot e^{-i \phi \sigma_z/2}\;,$$ where $p = \frac {1-e^{- \gamma}} 2$ can be interpreted as the probability of a random phase flip.
We can also write explicitly a more generic map that is fully sequentializable. We do that by defining appropriate (unnormalized) vectors $\ket{\chi_s}$ for the positive matrix $B = \sum_s
\ket{\chi_s}\bra{\chi_s}$. These vectors should be orthogonal and for the trace preserving condition should obey $$\begin{aligned}
B_{jj} = \sum_s {\langle j | \chi_s \rangle} {\langle \chi_s | j \rangle} = \sum_s |(\chi_s) _j|^2 = 1\;.\end{aligned}$$ Denote the $n$th roots of unity by $$\begin{aligned}
\omega_j = e^{-i 2 \pi j/n}\;,\end{aligned}$$ where $n$ is the dimension of the underlying Hilbert space. Define $$\begin{aligned}
\ket{\chi_s} = \sqrt{r_s} \sum_j e^{-i \phi_j} \omega^s_j \ket j\end{aligned}$$ for arbitrary phases $\phi_j$ and (dephasing) positive constants $r_j$ such that $\sum_j r_j = 1$. Then $$\begin{aligned}
{\langle \chi_s | \chi_t \rangle} = \sqrt{r_sr_t} \sum_j \omega^{t-s}_j = \sqrt{r_sr_t}n \delta_{s,t}\;,\end{aligned}$$ which makes the vectors orthogonal. Further, $$\begin{aligned}
\sum_s |(\chi_s) _j|^2 = \sum_s r_s = 1\;,\end{aligned}$$ which makes the corresponding completely positive map trace preserving. The Kraus operators of this map are $$\begin{aligned}
b_s = \sqrt{r_s} \sum_j e^{-i (\phi_j + 2 \pi j s /n)} \ket j \bra j\;.\end{aligned}$$ The map without dephasing (a pure rotation) is recovered by choosing $r_s = \delta_{s,0}$.
All diagrams we have used can equally well be interpreted in (dagger compact closed) categories other than tensor networks of Hilbert spaces. In the context of category theory, the morphism corresponding to the entangling gate $\delta$ is called a *classical structure*, because it abstracts the process of copying classical information. It also abstracts a choice of orthonormal basis [@coeckepavlovicvicary:bases; @abramskyheunen:hstar]. The corresponding “equal superposition” state is the unique unit of the classical structure. Sequentialization and parallelization of commuting morphisms are consequences of the *generalized spider theorem* [@coeckeduncan:observables], which is deduced by similar diagrammatic manipulations. Further, a dagger compact closed category with completely positive morphisms can be constructed from any dagger compact closed category [@selinger:completelypositive], giving the corresponding abstract “dephasing morphisms” that commute with the classical structure. See [@heunenboixo:cscp] for more information. Therefore our results have consequences in other theories. For example, conformal or topological quantum field theories can be formulated in terms of dagger compact closed categories of cobordisms [@kock:frobenius; @baez:quandaries; @atiyah:tqft; @segal:cft]. Similarly, geometric quantization can be formulated as concerning classical structures in the dagger compact category of symplectic manifolds and canonical relations [@landsman:quantization]. Our results, giving classes of sequentializable configurations, become interesting in those settings when reading sequentializability as the ability to “trade entanglement for time”.
In summary, we have shown that quantum maps composed of phase gates and general dephasing are exactly the maps that commute with the entangling map. This implies that the sequential and quantum protocols are equivalent. In particular, they are equally sensitive to noise and have exactly the same responsiveness for quantum metrology. The entangled (parallel) protocol is better suited for rapidly changing signals, but is technically challenging. The diagrammatical derivation we presented has advantages over direct computations: it is completely general and, though perhaps unfamiliar, arguably simpler. Also, the abstract point of view sheds some light on the fundamental structures of some quantum protocols and applies to other theories.
#### Acknowledgement. {#acknowledgement. .unnumbered}
Much of this work has been done while both authors were at the Institute for Quantum Information at the California Institute of Technology. We thank Peter Selinger for pointing out [@mccurdyselinger:basicdaggercompactcategories], and David P[é]{}rez Garc[í]{}a, Robert K[ö]{}nig, Peter Love and Spiros Michalakis for discussions. SB acknowledges support from FIS2008-01236 and from Defense Advanced Research Projects Agency award N66001-09-1-2101. CH was supported by the Netherlands Organisation for Scientific Research (NWO).
Bound on the Fisher information scaling for separable states {#sec:fisher_bound}
============================================================
The quantum Cramér-Rao inequality [@braunstein_statistical_1994] states that $$\begin{aligned}
\delta \hat \phi \ge \frac 1 {\sqrt {{\mathcal I}_n}},\end{aligned}$$ where ${\mathcal I}_n$ is the (quantum) Fisher information corresponding to the evolution of the quantum state parametrized by $\phi$. This appendix proves ${\mathcal I}_n \le n {\mathcal I}_{\rm bound}$ for separable states, where ${\mathcal I}_{\rm bound}$ is a bound on the Fisher information for one system. This is a well known fact, which we include here for completeness (see [*e.g.* ]{} [@boixo_quantum-limited_2008] for a different proof). We also note that for subsystems evolving with Hamiltonian $h$ (so the total Hamiltonian is $\sum h$) there exists a bound ${\mathcal I}_{\rm bound} \le \|
h\|$ [@giovannetti_quantum_2006; @boixo_generalized_2007].
Consider a quantum state $\rho$, implicitly parametrized by $\phi$. The quantum Fisher information for the state $\rho$ is defined as $${\mathcal I}_n = {\ensuremath{\mathrm{Tr}}}\rho {\mathcal L}_n^2$$ where the symmetric logarithmic derivative ${\mathcal L}$ is the Hermitian operator implicitly defined by the equation $$\frac 1 2 ( {\mathcal L}_n \rho + \rho {\mathcal L}_n) = \frac {\partial \rho}{\partial \phi}.$$ If follows directly from this definition that $$\begin{aligned}
\label{eq:sld_zero_average}
{\ensuremath{\mathrm{Tr}}}\rho {\mathcal L}= {\ensuremath{\mathrm{Tr}}}\frac 1 2 ( {\mathcal L}_n \rho + \rho {\mathcal L}_n) = {\ensuremath{\mathrm{Tr}}}\frac {\partial \rho}{\partial \phi} = \frac {\partial \; {\ensuremath{\mathrm{Tr}}}\rho}{\partial \phi} = 0.\end{aligned}$$
Now assume that $\rho$ is a product quantum state $\rho_p = \rho^{(1)} \otimes \cdots \otimes \rho^{(n)}$. It is easy to check directly that if ${\mathcal L}^{(j)}$ is the symmetric logarithmic derivative corresponding to $\rho^{(j)}$, then the symmetric logarithmic derivative of the product state $\rho$ is $${\mathcal L}_p = \sum_j 1 \otimes \cdots \otimes 1 \otimes {\mathcal L}^{(j)} \otimes \cdots \otimes 1 = \sum_j {\mathcal L}^{(j)}.$$ The corresponding Fisher information is $$\begin{aligned}
{\mathcal I}_p &= {\ensuremath{\mathrm{Tr}}}\rho_p {\mathcal L}_p^2 = \sum_j {\ensuremath{\mathrm{Tr}}}\rho_p ({\mathcal L}^{(j)})^2 +
\sum_{j \ne k} {\ensuremath{\mathrm{Tr}}}\rho_p {\mathcal L}^{(j)}\otimes {\mathcal L}^{(k)} \nonumber \\ &= \sum_j {\ensuremath{\mathrm{Tr}}}\rho_p ({\mathcal L}^{(j)})^2 \\ &= \sum_j {\mathcal I}_p^{(j)} \le n {\mathcal I}_{\rm bound}.\end{aligned}$$ The fact that ${\ensuremath{\mathrm{Tr}}}\rho_p {\mathcal L}^{(j)}\otimes {\mathcal L}^{(k)} = 0$ for $j \ne
k$ follows almost directly from .
Finally consider an ensemble of states $\rho_e = \sum_j p_j
\rho_j$. To calculate the symmetric logarithmic derivative of an ensemble it is simpler to start with the corresponding quantum superposition $\rho_{ex} = \sum_j p_j \rho_j \otimes \ket j \bra j$. It is easy to check directly that the corresponding symmetric logarithmic derivative for this superposition is ${\mathcal L}_{ex} = \sum_j
{\mathcal L}^{(j)} \otimes \ket j \bra j$. This gives the quantum Fisher information for the superposition $${\mathcal I}_{ex} = {\ensuremath{\mathrm{Tr}}}\rho_{ex} {\mathcal L}_{ex}^2 = \sum_j p_j {\mathcal I}_{ex}^{(j)} \le \max_j \;{\mathcal I}_{ex}^{(j)}.$$ Now, the difference between the quantum ensemble $\rho_e$ and the quantum superposition $\rho_{ex}$ is simply a trace over the auxiliary system that marks the particular state of the ensemble. That is, we forget or lose information. The corresponding Fisher information can only decrease as a result: ${\mathcal I}_{e} \le {\mathcal I}_{ex}$. Formally, this is a consequence of the monotonicity of the Fisher information [@petz_monotone_1996].
To conclude, we recall that a separable quantum state has the form $\rho_s = \sum_j p_j \rho_1^{(j)} \otimes \cdots \otimes \rho_N^{(j)}$. Putting together the bounds for ensembles and product states we conclude that ${\mathcal I}_s \le n {\mathcal I}_{\rm bound}$, or, in other words, $$\delta \hat \phi \ge \frac 1 {\sqrt{n {\mathcal I}_{\rm bound}}}.$$
Diagrammatic proofs {#sec:diagrams}
===================
Diagrammatically, entangling gates are axiomatically defined by the following properties, called associativity, isometry, commutativity, and Frobenius law [@coeckepaquette:practisingphysicist; @coeckeduncan:observables]. $$\begin{aligned}
\begin{tikzpicture}[scale=0.33, rotate=-90]
\node [style=dot] (1) at (0,-1) {};
\node (5) at (0,1) {};
\node (0) at (0,-2) {};
\node (6) at (1,1) {};
\node (2) at (1,0) {};
\node (4) at (-2,1) {};
\node [style=dot] (3) at (-1,0) {};
\draw (1) to (0);
\draw [bend right=45] (1) to (2);
\draw [bend left=45] (5) to (3);
\draw (6) to (2.west);
\draw [bend right=45] (4) to (3);
\draw [bend right=45] (3) to (1);
\end{tikzpicture}
& =
\begin{tikzpicture}[scale=0.33, rotate=-90]
\node [style=dot] (1) at (0,-1) {};
\node (5) at (0,1) {};
\node (0) at (0,-2) {};
\node (6) at (-1,1) {};
\node (2) at (-1,0) {};
\node (4) at (2,1) {};
\node [style=dot] (3) at (1,0) {};
\draw (1) to (0);
\draw [bend left=45] (1) to (2);
\draw [bend right=45] (5) to (3);
\draw (6) to (2.west);
\draw [bend left=45] (4) to (3);
\draw [bend left=45] (3) to (1);
\end{tikzpicture}
&
\begin{tikzpicture}[scale=0.33, rotate=-90]
\node (3) at (-1,2) {};
\node [style=dot] (1) at (0,1) {};
\node (0) at (0,0) {};
\node (2) at (1,2) {};
\draw [bend right=45] (1) to (2);
\draw [bend right=45] (3) to (1);
\draw (1) to (0);
\end{tikzpicture}
& =
\begin{tikzpicture}[scale=0.33,cross/.style={preaction={draw=white,
-, line width=3pt}}, rotate=-90]
\node [style=dot] (1) at (0,1) {};
\node (0) at (0,0) {};
\node (2) at (1,2) {};
\node (3) at (-1,2) {};
\node (4) at (1,3) {};
\node (5) at (-1,3) {};
\draw (1) to (0);
\draw (1) to [bend right] (2) .. controls (4) and (3) .. (5);
\draw[cross] (1) to [bend left] (3) .. controls (5) and (2) .. (4);
\end{tikzpicture}
\\
\begin{tikzpicture}[scale=0.33, rotate=-90]
\node (0) at (0,-2) {};
\node (3) at (0,2) {};
\node[dot] (2) at (0,1) {};
\node[dot] (1) at (0,-1) {};
\draw [bend right=90] (2) to (1);
\draw (2) to (3);
\draw [bend right=90] (1) to (2);
\draw (0) to (1);
\end{tikzpicture}
& =
\begin{tikzpicture}[scale=0.33, rotate=-90]
\node (0) at (0,-2) {};
\node (3) at (0,1) {};
\draw (0) to (3);
\end{tikzpicture}
&
\begin{tikzpicture}[scale=0.33, rotate=-90]
\node (0) at (-1, 2) {};
\node (1) at (1, 2) {};
\node [style=dot] (2) at (0, 1) {};
\node [style=dot] (3) at (0, 0) {};
\node (4) at (-1, -1) {};
\node (5) at (1, -1) {};
\draw (2) to (3);
\draw[bend left=45] (3) to (5);
\draw[bend left=45] (4) to (3);
\draw[bend right=45] (0) to (2);
\draw[bend right=45] (2) to (1);
\end{tikzpicture}
& =
\begin{tikzpicture}[scale=0.33, rotate=-90]
\node (0) at (-2, 2) {};
\node (1) at (1, 2) {};
\node [style=dot] (2) at (1, 1) {};
\node (3) at (-2, -0) {};
\node (4) at (0, -0) {};
\node (5) at (2, -0) {};
\node [style=dot] (6) at (-1, -1) {};
\node (7) at (-1, -2) {};
\node (8) at (2, -2) {};
\draw (2) to (1);
\draw (6) to (7);
\draw (5) to (8);
\draw[bend left=45] (2) to (5.west);
\draw[bend right=45] (6) to (4.west);
\draw[bend right=45] (3.east) to (6);
\draw[bend left=45] (4.west) to (2);
\draw (0) to (3);
\end{tikzpicture}\end{aligned}$$ It then follows [@abramskyheunen:hstar] from the structure of compact categories that there exists a unique morphism satisfying $$\label{eq:counit}
\begin{tikzpicture}[scale=0.33, rotate=-90]
\node [style=none] (0) at (1, 2) {};
\node [style=dot] (1) at (-1, 1) {};
\node [style=none] (2) at (1, 1) {};
\node [style=dot] (3) at (0, 0) {};
\node [style=none] (4) at (0, -1) {};
\draw[looseness=1.25, bend left=315] (1) to (3);
\draw[looseness=1.25, bend right=45] (3) to (2.center);
\draw (2.center) to (0);
\draw (4) to (3);
\end{tikzpicture}
\; = \;
\begin{tikzpicture}[scale=0.33, rotate=-90]
\node [style=none] (0) at (0, 2) {};
\node [style=none] (1) at (0, -1) {};
\draw (0.center) to (1.center);
\end{tikzpicture}$$
Now Eq. follows immediately from associativity: $$\begin{tikzpicture}[scale=0.33, rotate=-90]
\begin{pgfonlayer}{nodelayer}
\node [style=none] (0) at (-1, 1) {};
\node [style=none] (1) at (1, 1) {};
\node [style=dot] (2) at (0, 0) {};
\node [style=box,minimum size = .5cm] (3) at (0, -1.5) {$\;a\;$};
\node [style=none] (4) at (0, -3) {};
\end{pgfonlayer}
\begin{pgfonlayer}{edgelayer}
\draw (2) to (3);
\draw[bend right=45, looseness=1.5] (2) to (1.center);
\draw (3) to (4.center);
\draw[bend right=45, looseness=1.5] (0.center) to (2);
\end{pgfonlayer}
\end{tikzpicture}
=
\begin{tikzpicture}[scale=0.40, rotate=-90]
\begin{pgfonlayer}{nodelayer}
\node [style=none] (0) at (2, 2) {};
\node [style=none] (1) at (0, 2) {};
\node [style=circle,draw=black] (2) at (-1, 2) {$a$};
\node [style=dot] (3) at (1, 1) {};
\node [style=none] (4) at (-1, 1) {};
\node [style=dot] (5) at (0, 0) {};
\node [style=none] (6) at (0, -1) {};
\end{pgfonlayer}
\begin{pgfonlayer}{edgelayer}
\draw (2) to (4.center); \draw[looseness=1.25, bend left=45] (3)
to (5); \draw[looseness=1.25, bend left=315] (1.center) to (3);
\draw (5) to (6.center); \draw[looseness=1.25, bend right=315]
(0.center) to (3); \draw[looseness=1.25, bend right=45]
(4.center) to (5);
\end{pgfonlayer}
\end{tikzpicture}
=
\begin{tikzpicture}[scale=0.4, rotate=-90]
\begin{pgfonlayer}{nodelayer}
\node [style=circle,draw=black] (0) at (-2, 2.5) {$a$};
\node [style=none] (1) at (0, 2) {};
\node [style=none] (2) at (1, 2) {};
\node [style=dot] (3) at (-1, 1) {};
\node [style=none] (4) at (1, 1) {};
\node [style=dot] (5) at (0, -0) {};
\node [style=none] (6) at (0, -1) {};
\end{pgfonlayer}
\begin{pgfonlayer}{edgelayer}
\draw (2) to (4.center);
\draw[looseness=1.25, bend right=45] (3) to (5);
\draw[looseness=1.25, bend right=315] (1.center) to (3);
\draw (5) to (6.center);
\draw[looseness=1.25, bend left=315] (0) to (3);
\draw[looseness=1.25, bend left=45] (4.center) to (5);
\end{pgfonlayer}
\end{tikzpicture}
=
\begin{tikzpicture}[scale=0.4, rotate=-90]
\begin{pgfonlayer}{nodelayer}
\node [style=none] (0) at (-1, 2) {};
\node [style=none] (1) at (1, 2) {};
\node [style=box,minimum size = .5cm] (2) at (-1, 1) {$\;a\;$};
\node [style=none] (3) at (1, 1) {};
\node [style=dot] (4) at (0, 0) {};
\node [style=none] (5) at (0, -1) {};
\end{pgfonlayer}
\begin{pgfonlayer}{edgelayer}
\draw (4) to (5.center);
\draw[bend right=45, looseness=1.50] (2.center) to (4);
\draw[bend right=45, looseness=1.25] (4) to (3.center);
\draw (1.center) to (3.center);
\draw (0.center) to (2);
\end{pgfonlayer}
\end{tikzpicture}$$
The transformation between the parallel and sequential protocols and , as explicitly given in Eq. , follow directly from isometry: $$\begin{aligned}
&\begin{tikzpicture}[scale=0.4]
\begin{pgfonlayer}{nodelayer}
\node [style=none] (14) at (-2, -1) {};
\node [style=none] (15) at (2, -1) {};
\node [style=dot] (21) at (-4, -2.25) {};
\node [style=dot] (22) at (-3, -2.25) {};
\node [style=dot] (23) at (3, -2.25) {};
\node [style=dot] (24) at (4, -2.25) {};
\node [style=none] (25) at (-2, -3.5) {};
\node [style=none] (26) at (-1, -3.5) {};
\node [style=box] (27) at (0, -3.5) {$\;\;f_1\;\;$};
\node [style=none] (28) at (1, -3.5) {};
\node [style=none] (29) at (2, -3.5) {};
\end{pgfonlayer}
\begin{pgfonlayer}{edgelayer}
\draw (14.center) to (15.center);
\draw (25.center) to (26.center);
\draw[bend left=45] (22) to (14.center);
\draw[bend right=45] (29.center) to (23);
\draw (21) to (22);
\draw[bend left=45, looseness=1.25] (15.center) to (23);
\draw (26.center) to (28.center);
\draw[bend left=315] (22) to (25.center);
\draw (28.center) to (29.center);
\draw (23) to (24);
\end{pgfonlayer}
\end{tikzpicture} \\
\quad &= \quad
\begin{tikzpicture}[scale=0.4]
\begin{pgfonlayer}{nodelayer}
\node [style=none] (14) at (-2, -1) {};
\node [style=none] (15) at (1, -1) {};
\node [style=dot] (21) at (-7, -2.25) {};
\node [style=dot] (22) at (-3, -2.25) {};
\node [style=dot] (23) at (2, -2.25) {};
\node [style=dot] (24) at (3, -2.25) {};
\node [style=none] (25) at (-2, -3.5) {};
\node [style=none] (26) at (-1, -3.5) {};
\node [style=box] (27) at (-5, -2.25) {$\;\;f_1\;\;$};
\node [style=none] (28) at (0, -3.5) {};
\node [style=none] (29) at (1, -3.5) {};
\end{pgfonlayer}
\begin{pgfonlayer}{edgelayer}
\draw (14.center) to (15.center);
\draw (25.center) to (26.center);
\draw[bend left=45] (22) to (14.center);
\draw[bend right=45] (29.center) to (23);
\draw (21) to (22);
\draw[bend left=45, looseness=1.25] (15.center) to (23);
\draw (26.center) to (28.center);
\draw[bend left=315] (22) to (25.center);
\draw (28.center) to (29.center);
\draw (23) to (24);
\end{pgfonlayer}
\end{tikzpicture} \\ & = \quad
\begin{tikzpicture}[scale=0.4]
\begin{pgfonlayer}{nodelayer}
\node [style=dot] (21) at (-7, -2.25) {};
\node [style=dot] (22) at (-3, -2.25) {};
\node [style=box] (27) at (-5, -2.25) {$\;\;f_1\;\;$};
\end{pgfonlayer}
\begin{pgfonlayer}{edgelayer}
\draw (21) to (22);
\end{pgfonlayer}
\end{tikzpicture}\end{aligned}$$
The bijection between the states and operators of Eqs. and is easily seen using the unit property .
Finally, the diagrammatic proof of the commutativity of the linear operators or quantum maps of Eq. can now easily be seen to follow from the associative and commutative properties of the given classical structure. $$\begin{aligned}
& \begin{tikzpicture}[scale=0.5, font=\small, rotate=-90]
\begin{pgfonlayer}{nodelayer}
\node [style=none] (1) at (0,1.5) {};
\node [style=box,minimum size = .5cm] (2) at (0, 0) {$\;b\;$};
\node [style=box,minimum size = .5cm] (3) at (0, -1.5) {$\;a\;$};
\node [style=none] (4) at (0, -3) {};
\end{pgfonlayer}
\begin{pgfonlayer}{edgelayer}
\draw (2) to (1);
\draw (3) to (4.center);
\draw (3) to (2);
\end{pgfonlayer}
\end{tikzpicture}
=
\begin{tikzpicture}[scale=0.5, font=\scriptsize, rotate=-90]
\begin{pgfonlayer}{nodelayer}
\node [style=none] (0) at (2, 2) {};
\node [style=circle,draw=black] (1) at (0, 2) {$b$};
\node [style=circle,draw=black] (2) at (-1, 2) {$a$};
\node [style=dot] (3) at (1, 1) {};
\node [style=none] (4) at (-1, 1) {};
\node [style=dot] (5) at (0, 0) {};
\node [style=none] (6) at (0, -1) {};
\end{pgfonlayer}
\begin{pgfonlayer}{edgelayer}
\draw (2) to (4.center);
\draw[looseness=1.25, bend left=45] (3) to (5);
\draw[looseness=1.25, bend left=325] (1) to (3);
\draw (5) to (6.center);
\draw[looseness=1.25, bend right=315] (0.center) to (3);
\draw[looseness=1.25, bend right=45] (4.center) to (5);
\end{pgfonlayer}
\end{tikzpicture}
=
\begin{tikzpicture}[scale=0.5, font=\small, rotate=-90]
\begin{pgfonlayer}{nodelayer}
\node [style=circle,draw=black] (0) at (-2, 2.5) {$a$};
\node [style=circle,draw=black] (1) at (0, 2.5) {$b$};
\node [style=none] (2) at (1, 2) {};
\node [style=dot] (3) at (-1, 1) {};
\node [style=none] (4) at (1, 1) {};
\node [style=dot] (5) at (0, -0) {};
\node [style=none] (6) at (0, -1) {};
\end{pgfonlayer}
\begin{pgfonlayer}{edgelayer}
\draw (2) to (4.center);
\draw[looseness=1.25, bend right=45] (3) to (5);
\draw[looseness=1.25, bend right=315] (1) to (3);
\draw (5) to (6.center);
\draw[looseness=1.25, bend left=315] (0) to (3);
\draw[looseness=1.25, bend left=45] (4.center) to (5);
\end{pgfonlayer}
\end{tikzpicture}
\\ & =
\begin{tikzpicture}[scale=0.5, font=\small, rotate=-90]
\begin{pgfonlayer}{nodelayer}
\node [style=circle,draw=black] (0) at (-2, 2.5) {$b$};
\node [style=circle,draw=black] (1) at (0, 2.5) {$a$};
\node [style=none] (2) at (1, 2) {};
\node [style=dot] (3) at (-1, 1) {};
\node [style=none] (4) at (1, 1) {};
\node [style=dot] (5) at (0, -0) {};
\node [style=none] (6) at (0, -1) {};
\end{pgfonlayer}
\begin{pgfonlayer}{edgelayer}
\draw (2) to (4.center);
\draw[looseness=1.25, bend right=45] (3) to (5);
\draw[looseness=1.25, bend right=315] (1) to (3);
\draw (5) to (6.center);
\draw[looseness=1.25, bend left=315] (0) to (3);
\draw[looseness=1.25, bend left=45] (4.center) to (5);
\end{pgfonlayer}
\end{tikzpicture}
=
\begin{tikzpicture}[scale=0.5, font=\scriptsize, rotate=-90]
\begin{pgfonlayer}{nodelayer}
\node [style=none] (0) at (2, 2) {};
\node [style=circle,draw=black] (1) at (0, 2) {$a$};
\node [style=circle,draw=black] (2) at (-1, 2) {$b$};
\node [style=dot] (3) at (1, 1) {};
\node [style=none] (4) at (-1, 1) {};
\node [style=dot] (5) at (0, 0) {};
\node [style=none] (6) at (0, -1) {};
\end{pgfonlayer}
\begin{pgfonlayer}{edgelayer}
\draw (2) to (4.center);
\draw[looseness=1.25, bend left=45] (3) to (5);
\draw[looseness=1.25, bend left=325] (1) to (3);
\draw (5) to (6.center);
\draw[looseness=1.25, bend right=315] (0.center) to (3);
\draw[looseness=1.25, bend right=45] (4.center) to (5);
\end{pgfonlayer}
\end{tikzpicture}
=
\begin{tikzpicture}[scale=0.5, font=\small, rotate=-90]
\begin{pgfonlayer}{nodelayer}
\node [style=none] (1) at (0,1.5) {};
\node [style=box,minimum size = .5cm] (2) at (0, 0) {$\;a\;$};
\node [style=box,minimum size = .5cm] (3) at (0, -1.5) {$\;b\;$};
\node [style=none] (4) at (0, -3) {};
\end{pgfonlayer}
\begin{pgfonlayer}{edgelayer}
\draw (2) to (1);
\draw (3) to (4.center);
\draw (3) to (2);
\end{pgfonlayer}
\end{tikzpicture}\end{aligned}$$ These proofs only rely on the four axioms of classical structures, and hence hold in any dagger compact category rather than just for finite-dimensional Hilbert spaces with either linear or quantum maps.
| {
"pile_set_name": "ArXiv"
} | ArXiv |
---
abstract: |
We obtain the mass spectrum and the Higgs self-coupling of the two Higgs doublet model (THDM) in an alternative unification scenario where the parameters of the Higgs potential $\lambda_i$ ($i=1,2,3,4,5$) are determined by imposing their unification with the electroweak gauge couplings. An attractive feature of this scenario is the possibility to determine the Higgs boson masses by evolving the $\lambda_i$,s from the electroweak-Higgs unification scale $M_{GH}$ down to the electroweak scale. The unification condition for the gauge ($g_1,g_2$) and Higgs couplings is written as $g_1=g_2=f(\lambda_i)$, where $g_1=k_Y^{1/2} g_Y$, and $k_Y$ being the normalization constant. Two variants for the unification condition are discussed; Scenario I is defined through the linear relation: $g_1=g_2=k_H(i)\lambda_i(M_{GH})$, while Scenario II assumes a quadratic relation: $g^2_1=g^2_2=k_H(i)\lambda_i(M_{GH})$. In Scenario I, by fixing [*ad hoc*]{} $-k_H(5)=\frac{1}{2}
k_H(4)=\frac{3}{2} k_H(3)=k_H(2)=k_H(1) =1$, taking $\tan\beta=1$ and using the standard normalization ($k_Y=5/3$), we obtain the following spectrum for the Higgs boson masses: $m_{h^0} = 109.1$ GeV, $m_{H^0} = 123.2$ GeV, $m_{A^0} = 115.5$ GeV, and $m_{H^{\pm}}
= 80.3$ GeV, with similar results for other normalizations such as $k_Y=3/2$ and $k_Y=7/4$.
author:
- 'J.L. Díaz-Cruz$^{(a)}$, and A. Rosado$^{(b)}$'
title: 'Electroweak-Higgs Unification in the Two Higgs Doublet Model: Masses and Couplings of the Neutral and Charged Higgs Bosons'
---
Introduction
============
The Standard Model (SM) of the strong and electroweak (EW) interactions has met with extraordinary success; it has been already tested at the level of quantum corrections [@radcorrs1; @radcorrs2]. These corrections give some hints about the nature of the Higgs sector, pointing towards the existence of a relatively light Higgs boson, with a mass of the order of the EW scale, $m_{\phi_{SM}} \simeq v$ [@hixjenser]. However, it is widely believed that the SM cannot be the final theory of particle physics, in particular because the Higgs sector suffers from naturalness problems, and we do not really have a clear understanding of electroweak symmetry breaking (EWSB).
These problems in the Higgs sector can be stated as our present inability to find a satisfactory answer to some questions regarding its structure, which can be stated as follows:
1. What fixes the size (and sign) of the [*dimensionful parameter*]{} $\mu^2_0$ that appears in the Higgs potential?. This parameter determines the scale of EWSB in the SM; in principle it could be as high as the Planck mass, however, it needs to be fixed to much lower values.
2. What is the nature of the quartic Higgs coupling $\lambda$?. This parameter is not associated with a known symmetry, and we expect all interactions in nature to be associated somehow with gauge forces, as these are the ones we understand better [@myghyunif].
An improvement on our understanding of EWSB is provided by the supersymmetric (SUSY) extensions of the SM [@softsusy], where loop corrections to the tree-level parameter $\mu^2_0$ are under control, thus making the Higgs sector more natural. The quartic Higgs couplings is nicely related with gauge couplings through relations of the form: $\lambda=\frac{1}{8} (g^2_2+g^2_Y)$. In the SUSY alternative it is even possible to (indirectly) explain the sign of $\mu^2_0$ as a result loop effects and the breaking of the symmetry between bosons and fermions. Further progress to understand the SM structure is achieved in Grand Unified Theories (GUT), where the strong and electroweak gauge interactions are unified at a high-energy scale ($M_{GUT}$) [@gutrev]. However, certain consequences of the GUT idea seem to indicate that this unification, by itself, may be too drastic (within the minimal SU(5) GUT model one actually gets inexact unification, too large proton decay, doublet-triplet problem, incorrect fermion mass relations, etc.), and some additional theoretical tool is needed to overcome these difficulties. Again, SUSY offers an amelioration of these problems. When SUSY is combined with the GUT program, one gets a more precise gauge coupling unification and some aspects of proton decay and fermion masses are under better control [@myradferm1; @myradferm2].
In order to verify the realization of SUSY-GUT in nature, it will be necessary to observe plenty of new phenomena such as superpartners, proton decay or rare decay modes.
As nice as these ideas may appear, it seems worthwhile to consider other approaches for physics beyond the SM. For instance, it has been shown that additional progress towards understanding the SM origin, can be achieved by postulating the existence of extra dimensions. These theories have received much attention, mainly because of the possibility they offer to address the problems of the SM from a new geometrical perspective. These range from a new approach to the hierarchy problem [@ADD1; @ADD2; @ADD3; @ADD4; @RanSundrum] up to a possible explanation of flavor hierarchies in terms of field localization along the extra dimensions [@nimashmaltz]. Models with extra dimensions have been applied to neutrino physics [@abdeletal1; @abdeletal2; @abdeletal3; @abdeletal4; @Ioannisian:1999sw], Higgs phenomenology [@myhixXD1; @myhixXD2], among many others. In the particular GUT context, it has been shown that it is possible to find viable solutions to the doublet-triplet problem [@hallnomu1; @hallnomu2]. More recently, new methods in strong interactions have also been used as an attempt to revive the old models (TC, ETC, topcolor, etc) [@fathix]. Other ideas have motivated new types of models as well (little Higgs [@littlehix], AdS/CFT composite Higgs models [@hixdual], etc).
In this paper we are interested in exploring further an alternative unification scenario, of weakly-interacting type, that could offer direct understanding of the Higgs sector too and was first discussed in Ref.[@Aranda:2005st]. Namely, we shall explore the consequences of a scenario where the electroweak $SU(2)_L\times
U(1)_Y$ gauge interactions are unified with the Higgs self-interactions at an intermediate scale $M_{GH}$. Here, we explore further this idea within the context of the THDM, which allows us to predict the Higgs spectrum of this model. The dependence of our results on the choice for the normalization for the hypercharge is also discussed, as well as possible test of this EW-Higgs unification idea at future colliders, such as ILC. Besides predicting the Higgs spectrum, namely the masses for the neutral CP-even states ($h^0,H^0$), the neutral CP-odd state ($A^0$) and the Charged Higgs ($H^\pm$), we also discuss the Higgs couplings to gauge bosons and fermions. As we mentioned in our previous paper [@Aranda:2005st], it is relevant to compare our approach with the so called Gauge-Higgs unification program, as they share some similarities. We think that our approach is more model independent, as we first explore the consequences of a parametric unification, without really choosing a definite model at higher energies. In fact, at higher energies both the SUSY models as well as the framework of extra dimensions could work as ultraviolet completion of our approach. The SUSY models could work because they allow to relate the scalar quartic couplings to the gauge couplings, thanks to the D-terms [@myghyunif]. On the other hand, within the extra-dimensions it is also possible to obtain similar relations, when the Higgs fields are identified as the extra-dimensional components of gauge fields [@XDGHix1; @XDGHix2; @ABQuiros1; @ABQuiros2; @ABQuiros3; @Hosotani1; @Hosotani2; @Hosotani3; @Hosotani4; @Hosotani5; @Hosotani6; @Hosotani7; @Hosotani8]. Actually, we feel that the work of Ref.[@gauhixyuku1; @gauhixyuku2] has a similar spirit to ours, in their case they look for gauge unification of the Higgs self-couplings that appear in the superpotential of the NMSSM, and then they justify their work with a concrete model in 7D. However, in the present work, we do not discuss further the unification of the EW-Higgs couplings with the strong constant, which can be realized within the context of extra-dimensional Gauge-Higgs unified theories.
![Prediction for the Higgs boson masses as a function of $k_H(1)$ in the context of the THDM with $k_Y=5/3$, in the frame of Scenario I, taking $\tan\beta=1$ and $m_{top}=170.0$ GeV.[]{data-label="figure1"}](figure1.eps)
![Prediction for the Higgs boson masses as a function of $k_H(1)$ in the context of the THDM with $k_Y=3/2$, in the frame of Scenario I, taking $\tan\beta=1$ and $m_{top}=170.0$ GeV.[]{data-label="figure2"}](figure2.eps)
![Prediction for the Higgs boson masses as a function of $k_H(1)$ in the context of the THDM with $k_Y=7/4$, in the frame of Scenario I, taking $\tan\beta=1$ and $m_{top}=170.0$ GeV.[]{data-label="figure3"}](figure3.eps)
![Prediction for the $sin^2(\beta-\alpha)$ as a function of $k_H(1)$ in the context of the THDM for $k_Y=5/3,3/2,7/4$, in the frame of Scenario I, taking $\tan\beta=1$ and $m_{top}=170$ GeV.[]{data-label="figure4"}](figure4.eps)
![Prediction for the Higgs boson masses as a function of $k_H(1)$ in the context of the THDM with $k_Y=5/3$, in the frame of Scenario II, taking $\tan\beta=1$ and $m_{top}=170.0$ GeV.[]{data-label="figure5"}](figure5.eps)
![Prediction for the Higgs boson masses as a function of $k_H(1)$ in the context of the THDM with $k_Y=3/2$, in the frame of Scenario II, taking $\tan\beta=1$ and $m_{top}=170.0$ GeV.[]{data-label="figure6"}](figure6.eps)
![Prediction for the Higgs boson masses as a function of $k_H(1)$ in the context of the THDM with $k_Y=7/4$, in the frame of Scenario II, taking $\tan\beta=1$ and $m_{top}=170.0$ GeV.[]{data-label="figure7"}](figure7.eps)
![Prediction for the $sin^2(\beta-\alpha)$ as a function of $k_H(1)$ in the context of the THDM for $k_Y=5/3,3/2,7/4$, in the frame of Scenario II, taking $\tan\beta=1$ and $m_{top}=170$ GeV.[]{data-label="figure8"}](figure8.eps)
Gauge-Higgs unification in the SM: Review.
==========================================
In the EW-Higgs unified scenario, one assumes that there exists a scale where the gauge couplings constants $g_1,g_2$, associated with the gauge symmetry $SU(2)_L\times U(1)_Y$, are unified, and that at this scale they also get unified with the SM Higgs self coupling $\lambda$, i.e. $g_1=g_2=f(\lambda)$ at $M_{GH}$. The precise relation between $g_1$ and $g_Y$ (the SM hypercharge coupling) involves a normalization factor $k_Y$, i.e. $g_1=k_Y^{1/2}g_Y $, which depends on the unification model. The standard normalization gives $k_Y=5/3$, which is associated with minimal models such as $SU(5), SO(10), E_6$. However, in the context of string theory it is possible to have such standard normalization without even having a unification group. For other unification groups that involve additional $U(1)$ factors, one would have exotic normalizations too, and similarly for the case of GUT models in extra-dimensions. In what follows we shall present results for the cases: $k_Y=5/3,3/2$ and $7/4$, which indeed arise in string-inspired models [@Dienes:1996yh]. Note that these values fall in the range $3/2
< 5/3 < 7/4$ and so they can illustrate what happens when one chooses a value below or above the standard normalization. The form of the unification condition will depend on the particular realization of this scenario, which could be as generic as possible. However, in order to be able to make predictions for the Higgs boson mass, we shall consider two specific realizations. Scenario I will be based on the linear relation: $g_2=g_1=k_H
\lambda(M_{GH})$, where the factor $k_H$ is included in order to retain some generality, for instance to take into account possible unknown group theoretical or normalization factors. Motivated by specific models, such as SUSY itself, as well as an argument based on the power counting of the beta coefficients in the RGE for scalar couplings, [*i.e.*]{}, the fact that $\beta_\lambda$ goes as $O(g^4)$, we shall also define Scenario II, through the quadratic unification condition: $g^2_1=g^2_2=k_H\lambda (M_{GH})$. The expressions for the SM renormalization group equations at the two-loop level can be found in Ref. [@langa1].
In practice, one determines first the scale $M_{GH}$ at which $g_2$ and $g_1$ are unified, then one fixes the quartic Higgs coupling $\lambda$ by imposing the unification condition and finally, by evolving the quartic Higgs coupling down to the EW scale, we are able to predict the Higgs boson mass. For the numerical calculations, discussed in Ref.[@Aranda:2005st], we employed the full two-loop SM renormalization group equations involving the gauge coupling constants $g_{1,2,3}$, the Higgs self-coupling $\lambda$, the top-quark Yukawa coupling $g_t$, and the parameter $k_Y$ [@rgebsm; @langa1]. We also take the values for the coupling constants as reported in the Review of Particle Properties [@revPP], while for the top quark mass we take the value recently reported in [@lasttopm1; @lasttopm2].
Now, let us summarize our previous results with the full numerical analysis. For $k_Y=5/3$ we find that $M_{GH} \cong 1.0 \times
10^{13}$ GeV and by taking $\tan\beta=1$, results for the Higgs boson mass are given as a function of the parameter $k_H$ over a range $10^{-1} < k_H < 10^2$, which covers three orders of magnitude (We stress here that the expected natural value for $k_H$ is 1). For such a range of $k_H$, the Higgs boson mass takes the values: $176 <
m_H < 275$ GeV for Scenario I, while for $k_H=1$ we obtain a prediction for the Higgs boson mass: $m_H=229,\,234,\,241$ GeV, for a top quark mass of $m_{top}=165,\,170,\,175$ GeV [@lasttopm1; @lasttopm2], respectively. On the other hand, for Scenario II, we find that the Higgs boson mass can take the values: $175 < m_H < 269$ GeV, while for $k_H=1$ we obtain: $m_H=214,\,222,\,230$ GeV.
Then, when we compare our results with the Higgs boson mass obtained from EW precision measurements, which imply $m_H \lsim 190$ GeV, we notice that in order to get compatibility with such value, our model seems to prefer high values of $k_H$. For instance, by taking the lowest value that we consider here for the top mass, $m_t=165$ GeV, and fixing $k_H=10^2$, we obtain the minimum value for the Higgs boson mass equal to $m_H=176$ GeV in Scenario I, while Scenario II implies a minimal value that is slightly lower, $m_H=175$ GeV.
For $k_Y=3/2$ we find that $M_{GH}= 4.9 \times 10^{14}$ GeV, higher than in the previous case, but for which one still gets a mass gap between $M_{GH}$ and a possible $M_{GUT}$. In this case and by taking $\tan\beta=1$ we find values a little higher for the Higgs boson mass, for instance, for $k_H=1$, one gets $m_H=225,\,232,\,238$ ($m_H=212,\,220,\,218$) GeV for scenario I (II).
On the other hand, for $k_Y=7/4$ we find that $M_{GH}= 1.8 \times
10^{12}$ GeV, which is lower than that of the previous cases, and has an even larger mass gap between $M_{GH}$ and a possible $M_{GUT}$. In this case and by taking $\tan\beta=1$ we also find values slightly higher for the Higgs boson mass, for instance, for $k_H=1$, one gets $m_H=230,\,236,\,243$ ($m_H=215,\,223,\,231$) GeV for scenario I (II).
At this point, rather than continuing discussions on the precise Higgs boson mass, we would like to emphasize that our approach based on the EW-Higgs unification idea is very successful in giving a Higgs boson mass that has indeed the correct order of magnitude, and that once measured at the LHC we will be able to fix the parameter $k_H$ and find connections with other approaches for physics beyond the SM, such as the one to be discussed next.
In fact, for the Higgs boson mass range that is predicted in our approach, it turns out that the Higgs will decay predominantly into the mode $h\to ZZ$, which may provide us with good chances to measure the Higgs boson mass within a precision of $5 \%$ [@Duhrssen:2004cv1; @Duhrssen:2004cv2], thus making it possible to bound $k_H$ to within a few percent level. Further tests of our EW-Higgs unification hypothesis would involve testing more implications of the quartic Higgs coupling. For instance one could use the production of Higgs pairs ($e^+ e^- \to \nu \bar{\nu} h h$) at a future linear collider, such as the ILC. This is just another example of the complementarity of future studies at LHC and ILC.
![Prediction for the Higgs boson masses as a function of $\tan\beta$ in the context of the THDM with $k_Y=5/3$, in the frame of Scenario I, taking $k_H(1)=1$ and $m_{top}=170.0$ GeV.[]{data-label="figure9"}](figure9.eps)
![Prediction for the Higgs boson masses as a function of $\tan\beta$ in the context of the THDM with $k_Y=3/2$, in the frame of Scenario I, taking $k_H(1)=1$ and $m_{top}=170.0$ GeV.[]{data-label="figure10"}](figure10.eps)
![Prediction for the Higgs boson masses as a function of $\tan\beta$ in the context of the THDM with $k_Y=7/4$, in the frame of Scenario I, taking $k_H(1)=1$ and $m_{top}=170.0$ GeV.[]{data-label="figure11"}](figure11.eps)
![Prediction for the Higgs-fermion couplings as a function of $\tan\beta$ in the context of the THDM with $k_Y=5/3$, in the frame of Scenario I, taking $k_H(1)=1$ and $m_{top}=170.0$ GeV. The curves correspond to: 1) $h^0 t \bar{t}$, 2) $H^0 t \bar{t}$, 3) $A^0 t
\bar{t}$, 4) $-h^0 b \bar{b}$, 5) $H^0 b \bar{b}$ and 6) $A^0 b
\bar{b}$.[]{data-label="figure12"}](figure12.eps)
![Prediction for the Higgs-fermion couplings as a function of $\tan\beta$ in the context of the THDM with $k_Y=3/2$, in the frame of Scenario I, taking $k_H(1)=1$ and $m_{top}=170.0$ GeV. The curves correspond to: 1) $h^0 t \bar{t}$, 2) $H^0 t \bar{t}$, 3) $A^0 t
\bar{t}$, 4) $-h^0 b \bar{b}$, 5) $H^0 b \bar{b}$ and 6) $A^0 b
\bar{b}$.[]{data-label="figure13"}](figure13.eps)
![Prediction for the Higgs-fermion couplings as a function of $\tan\beta$ in the context of the THDM with $k_Y=7/4$, in the frame of Scenario I, taking $k_H(1)=1$ and $m_{top}=170.0$ GeV. The curves correspond to: 1) $h^0 t \bar{t}$, 2) $H^0 t \bar{t}$, 3) $A^0 t
\bar{t}$, 4) $-h^0 b \bar{b}$, 5) $H^0 b \bar{b}$ and 6) $A^0 b
\bar{b}$.[]{data-label="figure14"}](figure14.eps)
![Prediction for the Higgs-boson couplings as a function of $\tan\beta$ in the context of the THDM with $k_Y=5/3$, in the frame of Scenario I, taking $k_H(1)=1$ and $m_{top}=170.0$ GeV. The curves correspond to: 1) $h^0 V V=g_{h^0 V V}/g_{h_{sm}^0 V V}$, 2) $H^0 V
V=g_{H^0 V V}/g_{h_{sm}^0 V V}$, 3) $|h^0 V V|^2=|g_{h^0 V
V}/g_{h_{sm}^0 V V}|^2$ and 4) $|H^0 V V|^2=|g_{H^0 V V}/g_{h_{sm}^0
V V}|^2$, where $V=W$ or $Z$.[]{data-label="figure15"}](figure15.eps)
![Prediction for the Higgs-boson couplings as a function of $\tan\beta$ in the context of the THDM with $k_Y=3/2$, in the frame of Scenario I, taking $k_H(1)=1$ and $m_{top}=170.0$ GeV. The curves correspond to: 1) $h^0 V V=g_{h^0 V V}/g_{h_{sm}^0 V V}$, 2) $H^0 V
V=g_{H^0 V V}/g_{h_{sm}^0 V V}$, 3) $|h^0 V V|^2=|g_{h^0 V
V}/g_{h_{sm}^0 V V}|^2$ and 4) $|H^0 V V|^2=|g_{H^0 V V}/g_{h_{sm}^0
V V}|^2$, where $V=W$ or $Z$.[]{data-label="figure16"}](figure16.eps)
![Prediction for the Higgs-boson couplings as a function of $\tan\beta$ in the context of the THDM with $k_Y=7/4$, in the frame of Scenario I, taking $k_H(1)=1$ and $m_{top}=170.0$ GeV. The curves correspond to: 1) $h^0 V V=g_{h^0 V V}/g_{h_{sm}^0 V V}$, 2) $H^0 V
V=g_{H^0 V V}/g_{h_{sm}^0 V V}$, 3) $|h^0 V V|^2=|g_{h^0 V
V}/g_{h_{sm}^0 V V}|^2$ and 4) $|H^0 V V|^2=|g_{H^0 V V}/g_{h_{sm}^0
V V}|^2$, where $V=W$ or $Z$.[]{data-label="figure17"}](figure17.eps)
EW-Higgs unification in the Two-Higgs doublet model.
====================================================
Let us now discuss the implications of EW-Higgs unification for the two-Higgs doublet model (THDM). This model includes two scalar doublets ($\Phi_1$, $\Phi_2$), and the Higgs potential can be written as follows [@thdmky]: $$\begin{aligned}
V(\Phi_1,\Phi_2) &=& \mu^2_1 \Phi_1^\dagger \Phi_1
+ \mu^2_2 \Phi_2^\dagger\Phi_2
+ \lambda_1 (\Phi_1^\dagger \Phi_1)^2
+ \lambda_2 (\Phi_2^\dagger \Phi_2)^2
+ \lambda_3 (\Phi_1^\dagger \Phi_1) (\Phi_2^\dagger \Phi_2) \nonumber\\
& & + \lambda_4 (\Phi_1^\dagger \Phi_2) (\Phi_2^\dagger \Phi_1)
+ \frac{1}{2} \lambda_5 [(\Phi_1^\dagger \Phi_2)^2 +
(\Phi_2^\dagger \Phi_1)^2] \ .\end{aligned}$$
It is clear that by absorbing a phase in the definition of $\Phi_2$, one can make $\lambda_5$ real and negative, which pushes all potential CP violating effects into the Yukawa sector:
$$\lambda_5 \leq 0.$$
In order to avoid spontaneous breakdown of the electromagnetic $U(1)$ [@loren1], the vacuum expectation values must have the following form:
$$\langle \Phi_1 \rangle = \left(
\begin{array}{c}
0 \\
v_1
\end{array}
\right) \,\,\, , \,\,\, \langle \Phi_2 \rangle = \left(
\begin{array}{c}
0 \\
v_2
\end{array}
\right) \,\,\, ,$$
$v_1^2 + v_2^2 \equiv v^2 = (246 GeV)^2$. This configuration is indeed a minimum of the tree level potential if the following conditions are satisfied. $$\begin{aligned}
\lambda_1 &\geq& 0 \ , \nonumber \\
\lambda_2 &\geq& 0 \ , \nonumber \\
\lambda_4 + \lambda_5 &\leq& 0 \ , \nonumber \\
4\lambda_1 \lambda_2 &\geq& (\lambda_3 + \lambda_4 + \lambda_5)^2 \ .\end{aligned}$$
The scalar spectrum in this model includes two CP-even states ($h^0,H^0$), one CP-odd ($A^0$) and two charged Higgs bosons ($H^{\pm}$). The tree level expressions for the masses and mixing angles are given as follows:
$$\begin{aligned}
\tan\beta &=& \frac{v_2}{v_1} \ , \\
\sin\alpha &=& -(\mbox{sgn} \, C) \left[\frac{1}{2}
\frac{\sqrt{(A-B)^2+4C^2}-(B-A)}{\sqrt{(A-B)^2+4C^2}} \right]^{1/2} \ , \\
\cos\alpha &=& \left[\frac{1}{2}
\frac{\sqrt{(A-B)^2+4C^2}+(B-A)}{\sqrt{(A-B)^2+4C^2}} \right]^{1/2} \ , \\
M^2_{H^{\pm}} &=& - \frac{1}{2} (\lambda_4+\lambda_5) v^2 \ , \\
M^2_{A^0} &=& - \lambda_5 v^2 \ , \\
M^2_{H^0,h^0} &=& \frac{1}{2} \left[A+B {\pm} \sqrt{(A-B)^2+4C^2} \right] \ ,\end{aligned}$$
where $A=2 \lambda_1 v^2_1$, $B=2 \lambda_2 v^2_2$, $C=(\lambda_3 + \lambda_4 + \lambda_5) v_1 v_2$.
The two Higgs doublet models are described by 7 independent parameters which can be taken to be $\alpha$, $\beta$, $m_{H^{\pm}}$, $m_{H^{0}}$, $m_{h^{0}}$, $m_{A^{0}}$, while the top quark mass is given as:
$$m_t = g_t v \sin\beta \ .$$
Now, we write the THDM renormalization group equations at the one loop level involving the gauge coupling constants $g_{1,2,3}$, the Higgs self-couplings $\lambda_{1,2,3,4,5}$, the top-quark Yukawa coupling $g_t$, and the parameter $k_Y$, as follows [@thdmky; @langa1]:
$$\begin{aligned}
\frac{dg_i}{dt} &=& \frac{b_i^{thdm}}{16 \pi^2 } g^3_i \ , \\
\frac{dg_t}{dt} &=& \frac{g_t}{16 \pi^2 } \left[ \frac{9}{2} g^2_t -
\left(\frac{17}{12k_{Y}}g^2_1+ \frac{9}{4}g^2_2+8g^2_3\right)\right] \ , \\
\frac{d\lambda_{1}}{dt} &=& \frac{1}{16 \pi^2 }\left[24\lambda_{1}^2 +
2\lambda_{3}^2 + 2\lambda_{3}\lambda_{4} + \lambda_{5}^2 +
\lambda_{4}^2 - 3\lambda_{1}(3g^2_2+\frac{1}{k_{Y}}g^2_1) +
12\lambda_{1}g^2_t \right. \nonumber \\
& & \left. + \frac{9}{8}g^4_2 + \frac{3}{4k_{Y}}g^2_1g^2_2 +
\frac{3}{8k^2_{Y}}g^4_1 - 6g^4_t \right] \ , \\
\frac{d\lambda_{2}}{dt} &=& \frac{1}{16 \pi^2 }\left[24\lambda_{2}^2 +
2\lambda_{3}^2 + 2\lambda_{3}\lambda_{4} + \lambda_{5}^2 +
\lambda_{4}^2 - 3\lambda_{2}(3g^2_2+\frac{1}{k_{Y}}g^2_1)
\right. \nonumber \\
& & \left. + \frac{9}{8}g^4_2 + \frac{3}{4k_{Y}}g^2_1g^2_2 +
\frac{3}{8k^2_{Y}}g^4_1 \right] \ , \\
\frac{d\lambda_{3}}{dt} &=& \frac{1}{16 \pi^2 }\left[
4(\lambda_{1} + \lambda_{2})(3\lambda_{3} + \lambda_{4})
+ 4\lambda^2_{3}+2\lambda^2_{4} + 2\lambda_{5}^2
- 3\lambda_{3}(3g^2_2+\frac{1}{k_{Y}}g^2_1) \right. \nonumber \\
& & \left. + 6\lambda_{3}g^2_t+ \frac{9}{4}g^4_2 - \frac{3}{2k_{Y}}g^2_1g^2_2
+ \frac{3}{4k^2_{Y}}g^4_1 \right] \ , \\
\frac{d\lambda_{4}}{dt} &=& \frac{1}{16 \pi^2 }\left[
4\lambda_{4}(\lambda_{1} + \lambda_{2} + 2\lambda_{3} + \lambda_{4})
+ 8\lambda_{5}^2
- 3\lambda_{4}(3g^2_2+\frac{1}{k_{Y}}g^2_1) \right. \nonumber \\
& & \left. + 6\lambda_{4}g^2_t+ \frac{3}{k_{Y}}g^2_1g^2_2
\right] \ , \\
\frac{d\lambda_{5}}{dt} &=& \frac{1}{16 \pi^2 }\left[\lambda_{5}
\left((4\lambda_{1} + 4\lambda_{2} + 8\lambda_{3} + 12\lambda_{4} -
3(3g^2_2+\frac{1}{k_{Y}}g^2_1) + 6g^2_t \right) \right] \ ,\end{aligned}$$
where $(b_1^{thdm},b_2^{thdm},b_3^{thdm})=(7/k_{Y},-3,-7)$; $\mu$ denotes the scale at which the coupling constants are defined, and $t=\log(\mu/\mu_0)$.
The form of the unification condition will depend on the particular realization of this scenario, which could be as generic as possible. However, in order to make predictions for the Higgs mass, we shall consider again two specific realizations. Scenario I will be based on the linear relation: $$g_1=g_2=k_H(i) \, \lambda_i(M_{GH}) \hspace{2.5cm} (i=1,2,3,4,5) \, ,$$ where the factors $k_H(i)$ are included in order to take into account possible unknown group theoretical or normalization factors. We shall also define Scenario II, which uses quadratic unification conditions, as follows: $$g^2_1=g^2_2=k_H(i) \, \lambda_i(M_{GH}) \hspace{2.5cm} (i=1,2,3,4,5) \, .$$
Now, we present first the results of the numerical analysis for the Higgs bosons masses in the context of the two Higgs-doublet model for $\tan\beta=1$ and taking $m_{top}=170.0$ GeV. In order to get an idea of the behavior of the masses of the Higgs bosons ($h^0$, $H^0$, $A^0$, $H^{\pm}$) we make the following [*ad hoc*]{} choice: $$-k_H(5)=\frac{1}{2} k_H(4)=\frac{3}{2} k_H(3)=k_H(2)=k_H(1) \, ,$$ for both Scenarios I and II.
It will be also presented in this section a complete discussion on the resulting couplings of the neutral CP even Higgs bosons with gauge vector boson pairs in the THDM, which are related to the corresponding SM couplings as follows [@HHG]: $$\label{bosoncoup} \frac{g_{h^0 V
V}}{g_{h_{sm}^0 V V}}=\sin(\beta-\alpha)\, , \mbox{\hspace{1cm}}
\frac{g_{H^0 V V}}{g_{h_{sm}^0 V V}}=\cos(\beta-\alpha)\, ,$$ where $V=W$ or $Z$. For the moment it suffices to stress that the factor $\sin^2(\beta-\alpha)$ fixes the coupling of the lightest CP even Higgs boson with ZZ pairs, relative to the SM value, and therefore scales the result for the cross-section of the reaction $e^+ \, e^- \to h^0 + Z$, which in turn allow us to determine the Higgs masses within LEP bounds. Hence, results for the Higgs bosons masses and $\sin^2(\beta-\alpha)$ are given as a function of the parameter $k_H(1)$, looking for regions which are acceptable according to the available experimental data. In fact, first we will make use of the experimental results reported in the Table 14 of Ref.[@unknown:2006cr] which allow, assuming SM decay rates, a simple and direct check of our results for $m_{h^0}$ and $\sin^2(\beta-\alpha)$. We would like to emphasize the following: Even though the analysis of the EW-Higgs Unification within the THDM implies that the lightest neutral CP-even Higgs boson has a mass ($\sim 100$ GeV) that is somewhat below the LEP bounds, 114.4 GeV [@unknown:2006cr; @Barate:2003sz], it should be mentioned that this bound refers to the SM Higgs boson. The bound on the lightest Higgs boson of the THDM depends on the factor $\sin^2(\beta-\alpha)$, which could be less than 1, thus resulting in weaker Higgs boson mass bounds. Secondly, we will use the experimental bound reported for $m_{H^{\pm}}$ in the literature [@Heister:2002ev]: $$\label{bound1}
m_{H^{\pm}} > 79.3 \, \mbox{GeV \hspace{1.0cm} (95\% C.L.)} \, ,\\$$ Even though these two comparisons lead to a parameter space drastically reduced, from Figs.1-8 we observe that there is still an allowed region for Scenarios I and II, [*viz*]{}, $0.4 \lsim k_H(1)
\lsim 1.1$ for Scenario I and $0.15 \lsim k_H(1) \lsim 0.55$ for Scenario II.
From now on, we will restrict ourselves to continue our numerical analysis only in Scenario I, assuming $k_H(1)=1$
Now we present results in terms of the parameter $\tan\beta$ for the lightest neutral and the charged Higgs boson masses ($m_{h^0}$ and $m_{H^{\pm}}$) and the coupling of the lightest neutral CP-even Higgs boson with $ZZ$ pairs, relative to the corresponding SM value ($|g_{h^0 Z Z}/g_{h_{sm}^0 Z Z}|^2=\sin^2(\beta-\alpha)$), looking again for regions which are acceptable according to the currently available experimental data, for $k_Y=5/3$ (Table \[tab:t1\]), $k_Y=3/2$ (Table \[tab:t2\]) and $k_Y=7/4$ (Table \[tab:t3\]). As can be seen from Tables \[tab:t1\]-\[tab:t3\] there are values of $\tan\beta$ where the ratio $|g_{h^0 Z Z}/g_{h_{sm}^0 Z
Z}|^2$ is substantially reduced, which therefore will allow to overcome the constraints imposed by the LEP search for neutral Higgs bosons. Lastly, taking into account the bound on $m_{H^{\pm}}$ given in expression (\[bound1\]), we conclude that the following regions for $\tan\beta$ are experimentally allowed: $$\label{range1}
0.975 \leq \tan\beta \leq 1.15 \, \mbox{\hspace{1.0cm} for } k_Y=5/3 \, ,\\$$ $$\label{range2}
0.975 \leq \tan\beta \leq 1.20 \, \mbox{\hspace{1.0cm} for } k_Y=3/2 \, ,\\$$ $$\label{range3}
0.95 \leq \tan\beta \leq 1.125 \, \mbox{\hspace{1.0cm} for } k_Y=7/4 \, .\\$$
Now, using the ranges given in (\[range1\]),(\[range2\]), and (\[range3\]) we plot in Figs. 9, 10, and 11, the results for the Higgs boson masses as a function of $\tan\beta$ for $k_Y=5/3$, $k_Y=3/2$ and $k_Y=7/4$, respectively. We present also the same results in Tables \[tab:t4\], \[tab:t5\] and \[tab:t6\].
Let us discuss briefly the results of the numerical analysis of the Higgs mass spectrum. For $k_Y=5/3$, we find that $M_{GH}= 1.3 \times
10^{13}$ GeV, and by taking $\tan\beta=1$ we obtain the following Higgs mass spectrum $m_{h^0} = 109.1$ GeV, $m_{H^0} = 123.2$ GeV, $m_{A^0} = 115.5$ GeV, and $m_{H^{\pm}} = 80.3$ GeV. In turn, for $k_Y=3/2$ we find that $M_{GH}= 5.9 \times 10^{14}$ GeV, somewhat higher than in the previous case, but for which one still gets a mass gap between $M_{GH}$ and a possible $M_{GUT}$. One finds similar values for $m_{H^0}$, $m_{A^0}$, and $m_{H^{\pm}}$ and a little lower values for $m_{h^0}$, for instance for $\tan\beta=1$, we get $m_{h^0} = 102.3$ GeV, $m_{H^0} = 122.4$ GeV, $m_{A^0} =
112.8$ GeV, and $m_{H^{\pm}} = 80.3$ GeV. On the other hand, for $k_Y=7/4$ we find that $M_{GH}= 2.2 \times 10^{12}$ GeV, which is lower than that of the previous cases, and has an even larger mass gap between $M_{GH}$ and a possible $M_{GUT}$. We obtain similar values for $m_{H^0}$, $m_{A^0}$, and $m_{H^{\pm}}$ and a little higher values for $m_{h^0}$. For instance, for $\tan\beta=1$, one gets $m_{h^0} = 112.3$ GeV, $m_{H^0} = 123.0$ GeV, $m_{A^0} = 117.4$ GeV, and $m_{H^{\pm}} = 80.6$ GeV.
The numerical results presented in Figs. 9-11 (Tables \[tab:t4\]-\[tab:t6\]) lead us to conclude that the Higgs mass spectrum is almost independent of the value of $k_Y$. However, the unification scale $M_{GH}$ depends strongly on the value of $k_Y$, going from $2.2 \times 10^{12}$ GeV up to $5.9 \times 10^{14}$ GeV for $7/4 > k_Y > 3/2$ (for $k_Y=5/3$, it is obtained $M_{GH}=1.3
\times 10^{13}$ GeV).
At this point, we want to recall the relation between the Higgs-fermion couplings, which can be expressed relative to the SM value and is given by [@HHG]: $$\begin{aligned}
\label{fermioncoup}
H^0 t \bar{t} : \frac{\sin\alpha}{\sin\beta} \, ,& \mbox{\hspace{2cm}} &H^0 b \bar{b}:\frac{\cos\alpha}{\cos\beta} \, , \nonumber\\
h^0 t \bar{t} : \frac{\cos\alpha}{\sin\beta} \, ,& &h^0 b \bar{b}:\frac{-\sin\alpha}{\cos\beta} \, , \\
A^0 t \bar{t} : \cot\beta \, ,& &A^0 b \bar{b}:\tan\beta \, .
\nonumber\end{aligned}$$
Now, we use the ranges given in (\[range1\]),(\[range2\]), and (\[range3\]), and we present in Figs. 12, 13, and 14, the results for the fermion couplings a a function of $\tan\beta$ for $k_Y=5/3$, $k_Y=3/2$ and $k_Y=7/4$, respectively.
Finally, making use of the ranges given in (\[range1\]),(\[range2\]), and (\[range3\]), we present in Figs. 15, 16, and 17, the results for the Higgs-boson couplings as a function of $\tan\beta$ for $k_Y=5/3$, $k_Y=3/2$ and $k_Y=7/4$, respectively.
From our results shown in Figs. 9-17 and Tables \[tab:t4\]-\[tab:t6\], we also conclude that the Higgs mass spectrum does not depend strongly on the value of $\tan\beta$. On the other hand, the fermion couplings and the boson couplings depend strongly on the value of $\tan\beta$.
We find that for $\tan\beta=1$, the coupling of $h^0$ to up-type (d-type) quarks is suppressed (enhanced), which will have important phenomenological consequences [@Balazs:1998nt; @Diaz-Cruz:1998qc]: For instance it will suppress the production of $h^0$ at hadron colliders through gluon fusion, while the associated production with $b\bar{b}$ quarks will be enhanced. The couplings of $H^0$ show the opposite behavior, namely the couplings with d-type (up-type) quarks are suppressed (enhanced). This behavior changes as $\tan\beta$ takes higher values, and is reversed already for $\tan\beta=1.1$. Similar results are obtained for other normalizations. We end this section by saying that similar results are obtained in the experimentally allowed regions for Scenario II.
$\;\tan\beta\;$ $\;m_{h^0}\;$ (GeV) $\;\sin^{2}(\beta-\alpha)\;$ $\;m_{H^{\pm}}$ (GeV)$\;$
----------------- --------------------- ------------------------------ ---------------------------
$ \;0.900 \;$ $\; 105.3 \;$ $\; 0.3674 \;$ $\; 77.26\;$
$ \;0.925 \;$ $\; 106.5 \;$ $\; 0.3155 \;$ $\; 78.13\;$
$ \;0.950 \;$ $\; 107.6 \;$ $\; 0.2540 \;$ $\; 78.92\;$
$ \;0.975 \;$ $\; 108.5 \;$ $\; 0.1816 \;$ $\; 79.63\;$
$ \;1.000 \;$ $\; 109.1 \;$ $\; 0.1020 \;$ $\; 80.28\;$
$ \;1.025 \;$ $\; 109.3 \;$ $\; 0.0315 \;$ $\; 80.87\;$
$ \;1.050 \;$ $\; 109.0 \;$ $\; 0.0001 \;$ $\; 81.41\;$
$ \;1.075 \;$ $\; 108.2 \;$ $\; 0.0238 \;$ $\; 81.90\;$
$ \;1.100 \;$ $\; 107.0 \;$ $\; 0.0828 \;$ $\; 82.36\;$
$ \;1.125 \;$ $\; 105.6 \;$ $\; 0.1497 \;$ $\; 82.78\;$
$ \;1.150 \;$ $\; 104.0 \;$ $\; 0.2107 \;$ $\; 83.17\;$
$ \;1.175 \;$ $\; 102.4 \;$ $\; 0.2629 \;$ $\; 83.53\;$
$ \;1.200 \;$ $\; 100.8 \;$ $\; 0.3068 \;$ $\; 83.86\;$
$ \;1.225 \;$ $\; 99.24 \;$ $\; 0.3442 \;$ $\; 84.17\;$
$ \;1.250 \;$ $\; 97.70 \;$ $\; 0.3763 \;$ $\; 84.46\;$
$ \;1.275 \;$ $\; 96.21 \;$ $\; 0.4044 \;$ $\; 84.73\;$
$ \;1.300 \;$ $\; 94.75 \;$ $\; 0.4293 \;$ $\; 84.99\;$
: Prediction for the lightest neutral Higgs boson $h^0$ mass, $\sin^{2}(\beta-\alpha)$ and the charged Higgs boson $H^{\pm}$ mass as a function of $\tan\beta$ in the context of the THDM with $k_Y=5/3$, in the frame of Scenario I with $k_H(1)=1$, taking $m_{top}=170.0$ GeV.[]{data-label="tab:t1"}
$\;\tan\beta\;$ $\;m_{h^0}\;$ (GeV) $\;\sin^{2}(\beta-\alpha)\;$ $\;m_{H^{\pm}}$ (GeV)$\;$
----------------- --------------------- ------------------------------ ---------------------------
$ \;0.900 \;$ $\; 99.93 \;$ $\; 0.3063 \;$ $\; 76.91\;$
$ \;0.925 \;$ $\; 100.8 \;$ $\; 0.2503 \;$ $\; 77.89\;$
$ \;0.950 \;$ $\; 101.6 \;$ $\; 0.1898 \;$ $\; 78.78\;$
$ \;0.975 \;$ $\; 102.1 \;$ $\; 0.1277 \;$ $\; 79.57\;$
$ \;1.000 \;$ $\; 102.3 \;$ $\; 0.0697 \;$ $\; 80.30\;$
$ \;1.025 \;$ $\; 102.3 \;$ $\; 0.0249 \;$ $\; 80.95\;$
$ \;1.050 \;$ $\; 101.9 \;$ $\; 0.0020 \;$ $\; 81.55\;$
$ \;1.075 \;$ $\; 101.2 \;$ $\; 0.0044 \;$ $\; 82.10\;$
$ \;1.100 \;$ $\; 100.2 \;$ $\; 0.0279 \;$ $\; 82.61\;$
$ \;1.125 \;$ $\; 99.06 \;$ $\; 0.0643 \;$ $\; 83.07\;$
$ \;1.150 \;$ $\; 97.78 \;$ $\; 0.1060 \;$ $\; 83.50\;$
$ \;1.175 \;$ $\; 96.43 \;$ $\; 0.1483 \;$ $\; 83.90\;$
$ \;1.200 \;$ $\; 95.04 \;$ $\; 0.1886 \;$ $\; 84.27\;$
$ \;1.225 \;$ $\; 93.65 \;$ $\; 0.2259 \;$ $\; 84.61\;$
$ \;1.250 \;$ $\; 92.27 \;$ $\; 0.2601 \;$ $\; 84.93\;$
$ \;1.275 \;$ $\; 90.90 \;$ $\; 0.2913 \;$ $\; 85.23\;$
$ \;1.300 \;$ $\; 89.56 \;$ $\; 0.3198 \;$ $\; 85.51\;$
: Prediction for the lightest neutral Higgs boson $h^0$ mass, $\sin^{2}(\beta-\alpha)$ and the charged Higgs boson $H^{\pm}$ mass as a function of $\tan\beta$ in the context of the THDM with $k_Y=3/2$, in the frame of Scenario I with $k_H(1)=1$, taking $m_{top}=170.0$ GeV.[]{data-label="tab:t2"}
$\;\tan\beta\;$ $\;m_{h^0}\;$ (GeV) $\;\sin^{2}(\beta-\alpha)\;$ $\;m_{H^{\pm}}$ (GeV)$\;$
----------------- --------------------- ------------------------------ ---------------------------
$ \; 0.900 \;$ $\; 107.9 \;$ $\; 0.4004 \;$ $\; 77.79\;$
$ \; 0.925 \;$ $\; 109.3 \;$ $\; 0.3518 \;$ $\; 78.61\;$
$ \; 0.950 \;$ $\; 110.5 \;$ $\; 0.2908 \;$ $\; 79.35\;$
$ \; 0.975 \;$ $\; 111.6 \;$ $\; 0.2121 \;$ $\; 80.02\;$
$ \; 1.000 \;$ $\; 112.3 \;$ $\; 0.1144 \;$ $\; 80.63\;$
$ \; 1.025 \;$ $\; 112.6 \;$ $\; 0.0229 \;$ $\; 81.18\;$
$ \; 1.050 \;$ $\; 112.2 \;$ $\; 0.0064 \;$ $\; 81.69\;$
$ \; 1.075 \;$ $\; 111.1 \;$ $\; 0.0777 \;$ $\; 82.16\;$
$ \; 1.100 \;$ $\; 109.5 \;$ $\; 0.1693 \;$ $\; 82.58\;$
$ \; 1.125 \;$ $\; 107.9 \;$ $\; 0.2460 \;$ $\; 82.98\;$
$ \; 1.150 \;$ $\; 106.1 \;$ $\; 0.3054 \;$ $\; 83.34\;$
$ \; 1.175 \;$ $\; 104.4 \;$ $\; 0.3519 \;$ $\; 83.68\;$
$ \; 1.200 \;$ $\; 102.7 \;$ $\; 0.3893 \;$ $\; 84.00\;$
$ \; 1.225 \;$ $\; 101.1 \;$ $\; 0.4204 \;$ $\; 84.29\;$
$ \; 1.250 \;$ $\; 99.50 \;$ $\; 0.4470 \;$ $\; 84.56\;$
$ \; 1.275 \;$ $\; 97.97 \;$ $\; 0.4701 \;$ $\; 84.82\;$
$ \; 1.300 \;$ $\; 96.48 \;$ $\; 0.4907 \;$ $\; 85.06\;$
: Prediction for the lightest neutral Higgs boson $h^0$ mass, $\sin^{2}(\beta-\alpha)$ and the charged Higgs boson $H^{\pm}$ mass as a function of $\tan\beta$ in the context of the THDM with $k_Y=7/4$, in the frame of Scenario I with $k_H(1)=1$, taking $m_{top}=170.0$ GeV.[]{data-label="tab:t3"}
$\;\tan\beta\;$ $\;m_{h^0}\;$ (GeV) $\;m_{H^0}\;$ (GeV) $\;m_{A^0}\;$ (GeV) $\;m_{H^{\pm}}$ (GeV)$\;$
----------------- --------------------- --------------------- --------------------- ---------------------------
$\;0.975\;$ $\;108.5\;$ $\;125.4\;$ $\;114.7\;$ $\;79.63\;$
$\;1.000\;$ $\;109.1\;$ $\;123.2\;$ $\;115.5\;$ $\;80.28\;$
$\;1.025\;$ $\;109.3\;$ $\;121.6\;$ $\;116.2\;$ $\;80.87\;$
$\;1.050\;$ $\;109.0\;$ $\;120.6\;$ $\;116.9\;$ $\;81.41\;$
$\;1.075\;$ $\;108.2\;$ $\;120.2\;$ $\;117.6\;$ $\;81.90\;$
$\;1.100\;$ $\;107.0\;$ $\;120.3\;$ $\;118.2\;$ $\;82.36\;$
$\;1.125\;$ $\;105.6\;$ $\;120.7\;$ $\;118.7\;$ $\;82.78\;$
$\;1.150\;$ $\;104.0\;$ $\;121.3\;$ $\;119.2\;$ $\;83.17\;$
: Predicted Higgs mass spectrum as a function of $\tan\beta$ in the context of the THDM with $k_Y=5/3$, in the frame of Scenario I with $k_H(1)=1$, taking $m_{top}=170.0$ GeV.[]{data-label="tab:t4"}
$\;\tan\beta\;$ $\;m_{h^0}\;$ (GeV) $\;m_{H^0}\;$ (GeV) $\;m_{A^0}\;$ (GeV) $\;m_{H^{\pm}}$ (GeV)$\;$
----------------- --------------------- --------------------- --------------------- ---------------------------
$\;0.975\;$ $\;102.1\;$ $\;124.4\;$ $\;111.8\;$ $\;79.57\;$
$\;1.000\;$ $\;102.3\;$ $\;122.4\;$ $\;112.8\;$ $\;80.30\;$
$\;1.025\;$ $\;102.3\;$ $\;120.9\;$ $\;113.6\;$ $\;80.95\;$
$\;1.050\;$ $\;101.9\;$ $\;119.9\;$ $\;114.3\;$ $\;81.55\;$
$\;1.075\;$ $\;101.2\;$ $\;119.3\;$ $\;115.0\;$ $\;82.10\;$
$\;1.100\;$ $\;100.2\;$ $\;119.1\;$ $\;115.6\;$ $\;82.61\;$
$\;1.125\;$ $\;99.06\;$ $\;119.1\;$ $\;116.2\;$ $\;83.07\;$
$\;1.150\;$ $\;97.78\;$ $\;119.4\;$ $\;116.8\;$ $\;83.50\;$
$\;1.175\;$ $\;96.43\;$ $\;119.7\;$ $\;117.3\;$ $\;83.90\;$
$\;1.200\;$ $\;95.04\;$ $\;120.2\;$ $\;117.7\;$ $\;84.27\;$
: Predicted Higgs mass spectrum as a function of $\tan\beta$ in the context of the THDM with $k_Y=3/2$, in the frame of Scenario I with $k_H(1)=1$, taking $m_{top}=170.0$ GeV.[]{data-label="tab:t5"}
$\;\tan\beta\;$ $\;m_{h^0}\;$ (GeV) $\;m_{H^0}\;$ (GeV) $\;m_{A^0}\;$ (GeV) $\;m_{H^{\pm}}$ (GeV)$\;$
----------------- --------------------- --------------------- --------------------- ---------------------------
$\;0.950\;$ $\;110.5\;$ $\;127.9\;$ $\;115.7\;$ $\;79.35\;$
$\;0.975\;$ $\;111.6\;$ $\;125.2\;$ $\;116.6\;$ $\;80.02\;$
$\;1.000\;$ $\;112.3\;$ $\;123.0\;$ $\;117.4\;$ $\;80.63\;$
$\;1.025\;$ $\;112.6\;$ $\;121.3\;$ $\;118.1\;$ $\;81.18\;$
$\;1.050\;$ $\;112.2\;$ $\;120.6\;$ $\;118.8\;$ $\;81.69\;$
$\;1.075\;$ $\;111.1\;$ $\;120.6\;$ $\;119.4\;$ $\;82.16\;$
$\;1.100\;$ $\;109.5\;$ $\;121.1\;$ $\;119.9\;$ $\;82.58\;$
$\;1.125\;$ $\;107.9\;$ $\;121.8\;$ $\;120.5\;$ $\;82.98\;$
: Predicted Higgs mass spectrum as a function of $\tan\beta$ in the context of the THDM with $k_Y=7/4$, in the frame of Scenario I with $k_H(1)=1$, taking $m_{top}=170.0$ GeV.[]{data-label="tab:t6"}
Comments and conclusions
========================
In this paper we have obtained the Higgs mass spectrum, the Higgs-fermion couplings and the Higgs-boson couplings of the THDM in a framework where it is possible to unify the Higgs self-coupling with the gauge interactions.
The hypercharge normalization plays an important role to identify the EW-Higgs unification scale. For the canonical value $k_Y=5/3$ we get $M_{GH} = 1.3 \times 10^{13}$ GeV. For lower values, such as $k_Y=3/2$ the scale is $M_{GH} = 5.9 \times 10^{14}$ GeV, which is closer to the GUT scale ($\approx 10^{16}$ GeV) but for higher values, such as $k_Y=7/4$, which gives $M_{GH} = 2.2 \times 10^{12}$ GeV, the EW-Higgs unification becomes clearly distinctive.
The present approach still lacks a solution to the hierarchy problem; at the moment we have to affiliate to argument that fundamental physics could accept some fine-tuning [@splitsusy]. Another option would be to consider one of the simplest early attempts to solve the problem of quadratic divergences in the SM, namely through an accidental cancellation [@veltmanqd]. In fact, such kind of cancellation implies a relationship between the quartic Higgs coupling and the Yukawa and gauge constants, which has the form: $\lambda = y^2_t -\frac{1}{8} [ 3g^2 +g'^2]$. Unfortunately, this relation implies a Higgs mass $m_\phi=316$ GeV, and that seems already excluded. Nevertheless, this relation could work if one takes into account the running of the coupling and Yukawa constants. This particular point will be the subject of future investigations.
[99]{}
For a review of radiative corrections see: \[LEP Collaborations\], arXiv:hep-ex/0412015. See also: P. Langacker, arXiv:hep-ph/0211065. U. Baur [*et al.*]{} \[The Snowmass Working Group on Precision Electroweak Measurements Collaboration\], in [*Proc. of the APS/DPF/DPB Summer Study on the Future of Particle Physics (Snowmass 2001)* ]{} ed. N. Graf, eConf [**C010630**]{}, P1WG1 (2001) \[arXiv:hep-ph/0202001\]. J. L. Diaz-Cruz, Mod. Phys. Lett. A [**20**]{}, 2397 (2005) \[arXiv:hep-ph/0409216\]. For a review see: D. J. H. Chung, L. L. Everett, G. L. Kane, S. F. King, J. Lykken and L. T. Wang, arXiv:hep-ph/0312378. For a recent review of GUTs see: R.N. Mohapatra, hep-ph/0412050.
J. L. Diaz-Cruz, H. Murayama and A. Pierce, Phys. Rev. D[**65**]{}, 075011 (2002) \[arXiv:hep-ph/0012275\]. J. L. Diaz-Cruz and J. Ferrandis, arXiv:hep-ph/0504094. N. Arkani-Hamed, S. Dimopoulos and G. R. Dvali, Phys. Lett. B[**429**]{}, 263 (1998). I. Antoniadis, N. Arkani-Hamed, S. Dimopoulos and G. R. Dvali, Phys. Lett. B[**436**]{}, 257 (1998).
D. Cremades, L.E. Ibañez and F. Marchesasno, Nucl. Phys. B[**643**]{}, 93 (2002).
C. Kokorelis, Nucl. Phys. B[**677**]{}, 115 (2004).
L. Randall and R. Sundrum, Phys. Rev. Lett. [**83**]{}, 3370 (1999). N. Arkani-Hamed and M. Schmaltz, Phys. Rev. D[**61**]{}, 033005 (2000) \[arXiv:hep-ph/9903417\]. R. Barbieri, P. Creminelli and A. Strumia, Nucl. Phys. B[**585**]{}, 28 (2000) \[arXiv:hep-ph/0002199\]. N. Arkani-Hamed, S. Dimopoulos, G. R. Dvali and J. March-Russell, Phys. Rev. D[**65**]{}, 024032 (2002) \[arXiv:hep-ph/9811448\]. R. N. Mohapatra and A. Perez-Lorenzana, Phys. Rev. D[**66**]{}, 035005 (2002) \[arXiv:hep-ph/0205347\]. N. Cosme, J. M. Frere, Y. Gouverneur, F. S. Ling, D. Monderen and V. Van Elewyck, Phys. Rev. D[**63**]{}, 113018 (2001) \[arXiv:hep-ph/0010192\]. A. Ioannisian and J. W. F. Valle, Phys. Rev. D [**63**]{}, 073002 (2001) \[arXiv:hep-ph/9911349\]. A. Aranda, C. Balazs and J. L. Diaz-Cruz, Nucl. Phys. B[**670**]{}, 90 (2003) \[arXiv:hep-ph/0212133\]. A. Aranda and J. L. Diaz-Cruz, Mod. Phys. Lett. A[**20**]{}, 203 (2005) \[arXiv:hep-ph/0207059\]. K. R. Dienes, E. Dudas and T. Gherghetta, Phys. Lett. B[**436**]{}, 55 (1998). L. J. Hall, Y. Nomura and D. R. Smith, Nucl. Phys. B[**639**]{}, 307 (2002) \[arXiv:hep-ph/0107331\]. See for instance the so-called fat Higgs models: R. Harnik et al., Phys. Rev. D[**70**]{} (2004) 015002.
For a review of little Higgs models see: Jay G. Wacker, arXive: hep-ph/0208235.
For models with composite Higgs using AdS-CFT duality see: K. Agashe, R. Contino and A. Pomarol, Nucl. Phys. B[**719**]{} (2005) 165.
A. Aranda, J. L. Diaz-Cruz and A. Rosado, “Electroweak - Higgs unification and the Higgs boson mass,” arXiv:hep-ph/0507230 (to be published in Int. Jour. Mod. Phys. A). N. S. Manton, Nucl. Phys. B[**158**]{}, 141 (1979). D. B. Fairlie, Phys. Lett. B [**82**]{}, 97 (1979). I. Antoniadis, K. Benakli and M. Quiros, New J. Phys. [**3**]{}, 20 (2001) \[arXiv:hep-th/0108005\]. Y. Hosotani, S. Noda and K. Takenaga, Phys. Lett. B[**607**]{}, 276 (2005) \[arXiv:hep-ph/0410193\]. C. Csaki, C. Grojean and H. Murayama, Phys. Rev. D[**67**]{}, 085012 (2003) \[arXiv:hep-ph/0210133\]. Y. Hosotani, S. Noda and K. Takenaga, Phys. Lett. B[**607**]{}, 276 (2005) \[arXiv:hep-ph/0410193\]. C. A. Scrucca, M. Serone, L. Silvestrini and A. Wulzer, JHEP [**0402**]{}, 049 (2004) \[arXiv:hep-th/0312267\]. C. A. Scrucca, M. Serone and L. Silvestrini, Nucl. Phys. B [**669**]{}, 128 (2003) \[arXiv:hep-ph/0304220\]. G. Burdman and Y. Nomura, Nucl. Phys. B [**656**]{}, 3 (2003) \[arXiv:hep-ph/0210257\]. N. Haba and Y. Shimizu, Phys. Rev. D [**67**]{}, 095001 (2003) \[Erratum-ibid. D [**69**]{}, 059902 (2004)\] \[arXiv:hep-ph/0212166\]. I. Gogoladze, Y. Mimura and S. Nandi, Phys. Lett. B [**560**]{}, 204 (2003) \[arXiv:hep-ph/0301014\]. I. Gogoladze, Y. Mimura and S. Nandi, Phys. Lett. B [**562**]{}, 307 (2003) \[arXiv:hep-ph/0302176\]. I. Gogoladze, Y. Mimura and S. Nandi, Phys. Rev. D [**69**]{}, 075006 (2004) \[arXiv:hep-ph/0311127\]. I. Gogoladze, Y. Mimura and S. Nandi, Phys. Lett. B[**560**]{}, 204 (2003) \[arXiv:hep-ph/0301014\]. I. Gogoladze, T. Li, Y. Mimura and S. Nandi, Phys. Rev. D [**72**]{}, 055006 (2005) \[arXiv:hep-ph/0504082\]. K. R. Dienes and J. March-Russell, Nucl. Phys. B [**479**]{}, 113 (1996) \[arXiv:hep-th/9604112\]. V. Barger, J. Jiang, P. Langacker and Y. Li, Nucl. Phys. B[**726**]{} (2005) 149.
V. Barger, M.S. Berger and M.S. Ohman, Phys. Rev. D[**47**]{} (1993) 1093.
Review of Particle Physics. Particle data group. Phys. Lett. B[**592**]{} (2004) 1.
The values for the top mass used here are based in the data recently reported in: A. Abulencia [*et al.*]{} \[CDF Collaboration\], “Measurement of the top quark mass using template methods on dilepton events in proton antiproton collisions at s\*\*(1/2) = 1.96-TeV,” arXiv:hep-ex/0602008. See also: A. Abulencia [*et al.*]{} \[CDF Collaboration\], “Top quark mass measurement from dilepton events at CDF II with the matrix-element method,” arXiv:hep-ex/0605118. M. Duhrssen, S. Heinemeyer, H. Logan, D. Rainwater, G. Weiglein and D. Zeppenfeld, Phys. Rev. D [**70**]{}, 113009 (2004) \[arXiv:hep-ph/0406323\]. J. L. Diaz-Cruz and D. A. Lopez-Falcon, Phys. Lett. B [**568**]{}, 245 (2003) \[arXiv:hep-ph/0304212\]. D. Kominis and R. Sekhar, Phys. Lett. B[**304**]{} (1993) 152.
J.L. Diaz-Cruz and A. Mendez, Nucl. Phys. B[**380**]{}, 39 (1992).
J. Gunion, H. Haber, G. Kane and S. Dawson, [*The Higgs Hunter’s Guide*]{}, Addison-Wesley Publishig Company, Reading, MA, 1990.
\[ALEPH Collaboration\], arXiv:hep-ex/0602042. R. Barate [*et al.*]{} \[LEP Working Group for Higgs boson searches\], Phys. Lett. B [**565**]{}, 61 (2003) \[arXiv:hep-ex/0306033\]. A. Heister [*et al.*]{} \[ALEPH Collaboration\], Phys. Lett. B [**543**]{}, 1 (2002) \[arXiv:hep-ex/0207054\]. C. Balazs, J. L. Diaz-Cruz, H. J. He, T. Tait and C. P. Yuan, Phys. Rev. D [**59**]{}, 055016 (1999) \[arXiv:hep-ph/9807349\].
J. L. Diaz-Cruz, H. J. He, T. Tait and C. P. Yuan, Phys. Rev. Lett. [**80**]{}, 4641 (1998) \[arXiv:hep-ph/9802294\]. N. Arkani-Hamed and S. Dimopoulos, arXiv:hep-th/0405159. M. J. G. Veltman, Acta Phys. Polon. B[**12**]{}, 437 (1981).
| {
"pile_set_name": "ArXiv"
} | ArXiv |
ITP-UH-17/06\
hep-th/0608048
[**Vector-multiplet effective action in the\
non-anticommutative charged hypermultiplet model**]{}\
[ **I.L. Buchbinder $^+$[^1], O. Lechtenfeld $^\ddag$[^2], I.B. Samsonov $^{*}$[^3]**]{}\
[*$^+$ Department of Theoretical Physics, Tomsk State Pedagogical University,\
Tomsk 634041, Russia\
$^\ddag$ Institut f" ur Theoretische Physik, Leibniz Universit" at Hannover,\
Appelstraße 2, D-30167 Hannover, Germany\
$^*$ Laboratory of Mathematical Physics, Tomsk Polytechnic University,\
30 Lenin Ave, Tomsk 634050, Russia* ]{}\
**Abstract**
We investigate the quantum aspects of a charged hypermultiplet in deformed $\cN{=}(1,1)$ superspace with singlet non-anticommutative deformation of supersymmetry. This model is a “star” modification of the hypermultiplet interacting with a background Abelian vector superfield. We prove that this model is renormalizable in the sense that the divergent part of the effective action is proportional to the $\cN{=}(1,0)$ non-anticommutative super Yang-Mills action. We also calculate the finite part of the low-energy effective action depending on the vector multiplet, which corresponds to the (anti)holomorphic potential. The holomorphic piece is just deformed to the star-generalization of the standard holomorphic potential, while the antiholomorphic piece is not. We also reveal the component structure and find the deformation of the mass and the central charge.
Introduction and Summary
========================
In this paper we study the quantum aspects of $\cN{=}(1,0)$ non-anticommutative theories with singlet deformation of supersymmetry. To provide the motivation of this work let us briefly summarize the most important achievements and problems concerning non-anticommutative theories with $\cN{=}(1/2,0)$ and $\cN{=}(1,0)$ supersymmetry.
The interest in $\cN{=}(1/2,0)$ non-anticommutative deformations of supersymmetry originated with the papers [@Strings; @Strings1; @Strings2], where these deformations were derived from superstring theory on a constant graviphoton background. As a rule, such deformations break supersymmetry only in the chiral sector of superspace, which is possible in Euclidean superspace. The key feature of non-anticommutative deformations on the quantum level is the preservation of renormalizability, which was established for $\cN{=}(1/2,0)$ Wess-Zumino [@WZ] and super Yang-Mills (SYM) [@SYM12; @Jack; @Penati] models. This result is very non-trivial since the non-anticommutative deformations involve a parameter with negative mass dimension which plays the role of a new coupling constant. Since such theories appear to be renormalizable, their quantum dynamics should be explored. Indeed, the low-energy effective action of such models was considered in [@B], where the corrections due to the non-anticommutative deformation were calculated. These results provide a promising new method of partial supersymmetry breaking which preserves renormalizability. New remarkable quantum features suggest interesting physical applications (see, e.g., [@Strings1] for modifications of the glueball superpotential and the expectation value of the glueball field in $\cN{=}(1/2,0)$ supersymmetric field theories).
These surprises of $\cN{=}(1/2,0)$ theories motivated analogous investigations of deformed [*extended*]{} supersymmetric theories. In this case we distinguish several types of non-anticommutative deformations. The simplest one depends on a single scalar parameter $I$ which appears in the anticommutator of the chiral $\cN{=(}1,1)$ Grassmann coordinates, {\^\_i,\^\_j }\_=2 I\^ \_[ij]{}. \[eq1\] Such a deformation was introduced in [@singlet] and was named a chiral singlet deformation; the corresponding field theories are referred to as $\cN{=}(1,0)$ non-anticommutative. Although more general types of deformations of extended supersymmetry have been considered in [@CILQ; @Araki1], here we shall restrict ourselves to the singlet deformation (\[eq1\]), since this case is most elaborated now on the classical level [@FILSZ]-[@Araki] and its stringy origins have been established [@FILSZ].
The quantum aspects of $\cN{=}(1,0)$ non-anticommutative theories are more involved. In [@FILSZ; @ILZ; @IZ] it was shown that the $\cN{=}(1,0)$ SYM and hypermultiplet models acquire a number of new classical interaction terms even in the Abelian case. In principle, such terms can spoil renormalizability. In our recent paper [@my] we addressed this problem in two such $\cN{=}(1,0)$ theories, namely the Abelian SYM model and the [*neutral*]{} hypermultiplet interacting with an Abelian gauge superfield. By computing all divergent contributions to the effective action we proved that both these models remain renormalizable. Note that these theories are deformations of free ones, and thus all interaction terms vanish in the undeformed limit $I\to0$. A physically more important example is the [*charged*]{} hypermultiplet model, which – already prior to the deformation – features the interaction of a hypermultiplet with a background Abelian vector superfield. In particular, the low-energy effective action of this theory is goverend by the so-called holomorphic potential, which plays a significant role in the Seiberg-Witten theory [@Seiberg]. Therefore, in the present work we study the low-energy effective action and renormalizability for the $\cN{=}(1,0)$ non-anticommutative charged hypermultiplet model.
Theories with extended supersymmetry are most naturally described within the harmonic superspace approach [@HSS; @Book]. Hence, we will consider the $\cN{=}(1,0)$ non-anticommutative charged hypermultiplet in the harmonic superspace that was studied on the classical level in [@ILZ; @IZ]. We are interested in the non-anticommutative deformation of the holomorphic effective action that was discussed in [@Buch; @Eremin] using the harmonic superspace approach. Here we generalize the results of these works to the non-anticommutatively deformed hypermultiplet theory and compute the leading contributions to the effective action.
One of our main results is the proof of one-loop renormalizability of the deformed charged hypermultiplet system. It supports the idea that the non-anticommutative deformations in general do not spoil the renormalizability of supersymmetric theories. Next, we find the leading contributions to the (anti)holomorphic effective action including non-anticommutative corrections. We observe that the holomorphic and antiholomorphic pieces are deformed differently: the holomorphic piece is nothing but the star-generalization of the standard holomorphic potential while the antiholomorphic piece is not. We study also the component structure of the deformed effective action and derive the corrections to the standard terms in the (anti)holomorphic potential for the bosonic component fields. The deformation of the mass and the central charge due to non-anticommutativity are found as well.
The paper is organized as follows. In Sect. 2 we review the basic aspects of the undeformed charged hypermultiplet theory for later comparison with the deformed ones. Sect. 3 summarizes the most essential points about the $\cN{=}(1,0)$ non-anticommutative charged hypermultiplet on the classical level. In Sect. 4 the leading contributions to the low-energy effective action are computed, culminating in the renormalizability proof. In Sect. 5 we derive the component structure of the (anti)holomorphic effective action found in the previous section and analyze the new terms which appear due to the deformation. Sect. 6 contains our comments on the non-anticommutative deformation of mass and central charge in the charged hypermultiplet model. In the Conclusions we discuss the obtained results and point out some tempting unsolved problems. An Appendix collects some properties of a special function which encodes the deformation of the holomorphic potential.
0
The undeformed theory
=====================
In this section we review briefly the known results concerning the model of charged hypermultiplet and its effective action.
The $\cN=2$ supersymmetric models are most naturally described within the $\cN=2$ harmonic superspace approach [@HSS; @Book]. In particular, the classical action of charged hypermultiplet interacting with the Abelian vector superfield is given by S=d\^[(-4)]{}du q\^+(D\^[++]{}+V\^[++]{})q\^+. \[e1\] Here $q^+$ and its conjugate $\breve q^+$ are complex analytic superfields which describe the hypermultiplet, while $V^{++}$ is a real analytic superfield which corresponds to the vector multiplet. The integration in (\[e1\]) is performed over $\cN=2$ analytic superspace with the measure $d\zeta^{(-4)}du$. The action (\[e1\]) is invariant under the following (Abelian) gauge transformations of superfields q\^+=q\^+,q\^+=-q\^+,V\^[++]{}=-D\^[++]{}, \[e2\] where $\lambda$ is a real analytic superfield.
The general structure of the low-energy effective action in the model (\[e1\]) is given by (see, e.g., [@Buch; @Eremin]) =d\^4x d\^4 [F]{}(W)+d\^4x d\^4| |[F]{}(|W)+ d\^4x d\^8 [H]{}(W,|W), \[e3\] where ${\cal F}$ is holomorphic potential, $\bar{\cal F}$ is antiholomorphic potential, ${\cal H}$ is non-holomorphic potential. The strength superfields $W$, $\bar W$ are gauge invariant objects which are expressed through the gauge prepotential as follows |W=-14 D\^[+]{}D\^+\_V\^[–]{},W=-14 |D\^+\_|D\^[+]{}V\^[–]{}. \[e4\] where V\^[–]{}(z,u)=du\_1 \[e6\] is a solution of the zero-curvature equation D\^[++]{}V\^[–]{}-D\^[–]{}V\^[++]{}=0. \[e5\]
Note that in the Abelian case the superfields $W$ and $\bar W$ are chiral and antichiral respectively D\^\_|W=0,|D\^\_W=0. \[e7\] Therefore the functions ${\cal F}(W)$ and $\bar{\cal F}(\bar W)$, which are related to each other by complex conjugation, are integrated over the chiral and antichiral superspaces respectively.
The perturbative low-energy effective action in the model (\[e1\]) is studied now in details, (see, e.g., [@Buch]-[@Dragon]) where both holomorphic and non-holomorphic contributions have been found. In particular, the holomorphic part of the effective action which is leading in the low-energy approximation is given by (W)=-W\^2W, \[e8\] where $\Lambda$ is some scale.
The strength superfields $W$, $\bar W$ have the following component structure in the bosonic sector W=+(\^+\_[mn]{}\^-)F\_[mn]{}+…, |W= |+(|\^+\_[mn]{}|\^-)F\_[mn]{}+…, \[e9\] where $\phi,\ \bar\phi$ are scalar fields, $F_{mn}=\partial_m A_n-\partial_n A_m$ is the Maxwell strength and dots stand for the terms with spinors $\Psi^i_\alpha,\ \bar\Psi_{i\dot\alpha}$ and auxiliary fields ${\cal D}^{kl}$ as well as the terms with spatial derivatives. The component structure of holomorphic effective action can be most easily derived in the following approximation
[c]{} =const,|=const,F\_[mn]{}=const,\
\^i\_=|\_[i]{}=[D]{}\^[kl]{}=0.
\[e10\] Substituting the strength superfields (\[e9\]) into (\[e3\]), one gets the component structure of the holomorphic part of the effective action \_[hol]{}=d\^4x d\^4 [F]{}(W)= -1[32\^2]{}d\^4x (F\_[mn]{}F\_[mn]{}+F\_[mn]{}F\_[mn]{})( +32 )+…, \[e11\] where $\tilde F_{mn}=\frac12\varepsilon_{mnrs}F_{rs} $ and dots stand for the higher terms which are not essential in the approximation (\[e10\]). The constant 3/2 in (\[e11\]) can be removed by the shift of the parameter $\Lambda$, however it will be important when we will consider the deformation of (anti)holomorphic effective action due to non-anticommutativity. The antiholomorphic part of effective action is given by the complex conjugation of the action (\[e11\]).
0
Non-anticommutative charged hypermultiplet model
================================================
The chiral singlet deformation of $\cN=(1,1)$ superspace was introduced in [@singlet] and the corresponding field models were studied in [@FILSZ]-[@my]. Such a deformation is effectively taken into account by the star product operator =, \[e12\] which should be placed everywhere instead of usual product of superfields in the classical actions. The constant $I$ here is a parameter of non-anticommutativity, $Q^i_\alpha$ are the supercharges. In particular, the non-anticommutative generalization of the action (\[e1\]) is given by [@ILZ] S=d\^[(-4)]{} du q\^+\^[++]{}q\^+, \[e13\] where we use the notations \^[++]{}=D\^[++]{}+V\^[++]{},\^[–]{}=D\^[–]{}+V\^[–]{}. \[13.1\] This action is invariant under the following gauge transformations q\^+=-q\^+,q\^+=q\^+,V\^[++]{}=-D\^[++]{}-\[V\^[++]{},\]\_. \[e14\] which are the non-anticommutative generalizations of the usual ones (\[e2\]).
In the deformed case the strength superfields $W$, $\bar W$ are defined by the standard equations (\[e4\]), but the superfield $V^{--}$ is now given by a series V\^[–]{}(z,u)=\_[n=1]{}\^(-1)\^n du\_1…du\_n , \[e15\] which solves the star-deformed zero-curvature equation [@FILSZ] D\^[++]{}V\^[–]{}-D\^[–]{}V\^[++]{}+\[V\^[++]{},V\^[–]{}\]\_=0. \[e16\]
Let us introduce the “bridge” superfield $\Omega(z,u)$ as a general $\cN=(1,1)$ superfield which relates the covariant harmonic derivatives $\nabla^{\pm\pm}$ with the plain ones $D^{\pm\pm}$: \^[++]{}=e\^\_D\^[++]{}e\^[-]{}\_,\^[–]{}=e\^\_D\^[–]{}e\^[-]{}\_. \[e17\] The bridge superfield was originally introduced in [@HSS] for the undeformed $\cN=2$ SYM theory as an operator relating the $\cN=2$ superfields in the $\tau$- and $\lambda$-frames. Using the bridge superfield $\Omega$ one can alternatively rewrite the equation (\[e15\]) in the following two equivalent forms V\^[–]{}(z,u)=du’ =du’. \[e18\] The expression (\[e18\]) can be checked directly to satisfy the zero-curvature condition (\[e16\]).
It is well known [@Book] that the [*free*]{} propagator in the hypermultiplet model (\[e1\]) is given by the following expression [^4] G\_0\^[(1,1)]{}(1|2)=-1(D\^+\_1)\^4(D\^+\_2)\^4 \[e18.1\] which solves the equation $D^{++}G_0^{(1,1)}(1|2)=\delta_A^{(3,1)}(1|2)$, where $\delta_A^{(3,1)}(1|2)$ is the analytic delta-function. Let us define now the [*full*]{} propagator in the model (\[e13\]) as a distribution satisfying the equation \^[++]{}G\^[(1,1)]{}(1|2)=\_A\^[(3,1)]{}(1|2). \[e19\] The solution of (\[e19\]) can formally be written as G\^[(1,1)]{}(1|2)=-1[\_]{}(D\^+\_1)\^4(D\^+\_2)\^4{ e\_\^[(1)]{}e\_\^[-(2)]{} }, \[e20\] where $\hat\square_\star$ is a covariant box operator \_=-12(D\^+)\^4\^[–]{}\^[–]{}. \[e21\] Clearly, it moves an analytic superfield to another analytic one. The operator (\[e21\]), acting on the analytic superfield, can be represented in the form $$\begin{aligned}
\hat\square_\star&=&\nabla^m\star\nabla_m
-\frac 12(\nabla^{+\alpha}\star W)\star\nabla^-_\alpha
-\frac 12(\bar\nabla^+_{\dot\alpha}\star \bar W)\star\bar\nabla^{-\dot\alpha}
+\frac 14(\nabla^{+\alpha}\star\nabla^+_\alpha\star W)\star\nabla^{--}
\nonumber\\&&
-\frac 18[\nabla^{+\alpha},\nabla^-_\alpha]_\star\star W
-\frac12\{W,\bar W \}_\star.
\label{e22}\end{aligned}$$ Here $\nabla^\pm_\alpha=D^\pm_\alpha+V^\pm_\alpha$, $\bar\nabla^\pm_{\dot\alpha}=\bar D^\pm_{\dot\alpha}+\bar V^\pm_{\dot\alpha}$ are covariant spinor derivatives. Note that the expression (\[e22\]) has a similar form as in the undeformed theory [@backgr] with the simple star-modification of multiplication of superfields. This result is not surprising since eq. (\[e22\]) is derived from (\[e21\]) only by using the (anti)commutation relations between spinor derivatives which have the same form as in the undeformed theory.
0
Computation of low-energy effective action
==========================================
General structure of low-energy effective action
------------------------------------------------
The model (\[e13\]) is non-anticommutative deformation of the model (\[e1\]). One can think naively that the effective potentials ${\cal F}$, $\bar{\cal F}$, ${\cal H}$ in the effective action (\[e3\]) will get the analogous deformation (W)\_(W),|[F]{}(|W)|[F]{}\_(|W),(W,|W)\_(W,|W). \[e23\] However, we will show that this assertion is not true for the antiholomorphic potential in the sense that no any action can be constructed with a function $\bar {\cal F}_\star(\bar W)$ integrated over the antichiral superspace. Indeed, the strength superfields $W$, $\bar W$ are not (anti)chiral, but [*covariantly*]{} (anti)chiral D\^+\_|W=\^-\_|W=0,|D\^+\_W=|\^-\_W=0. \[e24\] Therefore the expression $\int d^4x d^4\bar\theta \bar{\cal F}_\star (\bar
W)$ depends on $\theta$ variables D\^-\_d\^4x d\^4| |[F]{}\_(|W) = d\^4x d\^4| \[|[F]{}\_(|W),V\^-\_\]\_0. \[e25\] The rhs of (\[e25\]) does not vanish since the star-product is not cyclic under $d^4xd^4\bar\theta$ integration. Moreover, the expression $\int d^4x d^4\bar\theta \bar{\cal F}_\star(\bar W)$ violates the gauge invariance. Indeed, the strength superfields transform covariantly under the gauge transformations (\[e14\]) of gauge superfield W=\[, W\]\_,|W= \[, |W\]\_. \[e26\] Since the superfields $\lambda$ and $\bar W$ are not antichiral, we have d\^4x d\^4| |[F]{}\_(|W)= d\^4x d\^4| \[, |[F]{}\_(|W)\]\_0. \[e27\] The similar remarks on the gauge invariance in holomorphic and antiholomorphic parts of classical action in $\cN=(1/2,0)$ gauge theory are given in [@Penati]. Note also that there is no such a problem with the holomorphic potential ${\cal
F}_\star(W)$, as it is pointed out in [@FILSZ] for the case of classical $\cN=(1,0)$ SYM action.
Taking into account these remarks we propose that the general form of the low-energy effective action in the hypermultiplet model (\[e13\]) is given by =d\^4x d\^4 [F]{}\_(W)+ d\^4xd\^8 [H]{}\_(V\^[++]{},V\^[–]{}, W,|W). \[e28\] Here we assume that the possible terms in the low-energy effective action which correspond to the antiholomorphic potential in the limit $I\to 0$ is included into the function ${\cal
H}_\star(V^{++},V^{--}, W,\bar W)$ integrated in full superspace. The direct computations will specify these functions ${\cal
F}_\star$ and ${\cal H}_\star$.
For the further considerations it will be more convenient to study the variation of effective action $\delta\Gamma$ rather then $\Gamma$ itself. In particular, given the holomorphic part of the action (\[e28\]) \_[hol]{}=d\^4x d\^4 [F]{}\_(W), \[e29\] using the same steps as in the undeformed non-Abelian $\cN=2$ supergauge theory [@Dragon], one can write its variation either in the analytic superspace \_[hol]{}=d\^[(-4)]{} du V\^[++]{}, \[e30\] or in the full superspace \_[hol]{}=d\^[12]{}z du V\^[++]{}V\^[–]{}1W’\_(W). \[e31\]
One-loop effective action
-------------------------
The one-loop effective action in the model (\[e13\]) is defined by the following formal expression [^5] = =(\^[++]{})=-G\^[(1,1)]{}(1|2), \[e32\] where $G^{(1,1)}(1|2)$ is given by eq. (\[e20\]). It is easy to find the variation of (\[e32\]) ==d\^[(-4)]{} du V\^[++]{}(1)G\^[(1,1)]{}(1|2)|\_[(1)=(2)]{}. \[e33\] There is an important relation between full and free hypermultiplet propagators G\^[(1,1)]{}(1|3)=G\_0\^[(1,1)]{}(1|3)-d\_2\^[(-4)]{} du\_2 G\_0\^[(1,1)]{}(1|2)V\^[++]{}(2)G\^[(1,1)]{}(2|3) \[e34\] which can be checked directly to satisfy (\[e19\]). Substituting (\[e34\]) into (\[e33\]), we find =-d\_1\^[(-4)]{} du\_1 d\_2\^[(-4)]{} du\_2 V\^[++]{}(1)G\_0\^[(1,1)]{}(1|2)V\^[++]{}(2)G\^[(1,1)]{}(2|1). \[e35\] Taking into account the exact form of the propagators (\[e18.1\],\[e20\]), we rewrite eq. (\[e35\]) as follows $$\begin{aligned}
\delta\Gamma&=&-\int d\zeta_1^{(-4)} d\zeta_2^{(-4)} du_1 du_2\,
\delta V^{++}(1)\star\frac1\square(D^+_1)^4(D^+_2)^4
\frac{\delta^{12}(z_1-z_2)}{(u^+_1u^+_2)^3}
\nonumber\\&&\times
V^{++}(2)\star\frac1{\hat\square_{\star(2)}}\star
(D^+_1)^4(D^+_2)^4\left\{
e^{\Omega(2)}_\star\star e^{-\Omega(1)}_\star
\frac{\delta^{12}(z_2-z_1)}{(u^+_2u^+_1)^3}
\right\}.
\label{e36}\end{aligned}$$ Now we take off the derivatives $(D^+_1)^4(D^+_2)^4$ from the first delta-function to restore the full superspace measure $$\begin{aligned}
\delta\Gamma&=&-\int d^{12}z_1 d^{12}z_2 du_1 du_2\,
\delta V^{++}(1)\star\frac1\square
\frac{\delta^{12}(z_1-z_2)}{(u^+_1u^+_2)^3}
\nonumber\\&&\times
V^{++}(2)\star\frac1{\hat\square_{\star(2)}}\star
(D^+_1)^4(D^+_2)^4\left\{
e^{\Omega(2)}_\star\star e^{-\Omega(1)}_\star\star
\frac{\delta^{12}(z_2-z_1)}{(u^+_2u^+_1)^3}
\right\}.
\label{e37}\end{aligned}$$ The equation (\[e37\]) is a starting point for further calculations of different contributions to the effective action.
Divergent part of effective action
----------------------------------
To derive the divergent part of the effective action it is sufficient to consider the approximation 1[\_]{}1\[eq77\] since all other terms in the operator $\hat\square_\star$ result to higher powers of momenta in the denominator. Upon the condition (\[eq77\]), the variation of effective action (\[e37\]) simplifies essentially $$\begin{aligned}
\delta\Gamma_{div}&=&\int d^{12}z_1 d^{12}z_2 \frac{du_1 du_2}{(u^+_1u^+_2)^6}
\delta V^{++}(1)\star\frac1\square
\delta^{12}(z_1-z_2)
\nonumber\\&&\times
V^{++}(2)\star\frac1\square
(D^+_1)^4(D^+_2)^4\left\{
e^{\Omega(2)}_\star\star e^{-\Omega(1)}_\star\star
\delta^{12}(z_2-z_1)\right\}.
\label{eq78}\end{aligned}$$ We have to apply the identity \^8(\_1-\_2)(D\_1\^+)\^4(D\^+\_2)\^4\^[12]{}(z\_1-z\_2)= (u\^+\_1u\^+\_2)\^4\^[12]{}(z\_1-z\_2) \[eq79\] to shrink the integration over the Grassmann variables to a point. Note that all the derivatives $D^+$ in eq. (\[eq78\]) must hit the delta function, otherwise the result is zero since there are exactly eight such derivatives to apply (\[eq79\]). Moreover, the presence of star-product can not modify the relation (\[eq79\]). Calculating the divergent momentum integral \_[div]{} = ,(0) \[eq21\] and applying the equation (\[eq79\]), the integration over $d^{12}z_2$ in (\[eq78\]) can be performed resulting to \_[div]{}=1[16\^2]{}d\^[12]{}z du\_1 V\^[++]{}(z,u\_1)du\_2. \[eq80\] Using the relation (\[e18\]) we obtain finally \_[div]{}=1[16\^2]{} d\^[12]{}z du V\^[++]{}V\^[–]{}. \[eq81\] The variation (\[eq81\]) can be easily integrated with the help of eq. (\[e31\]) \_[div]{}=1[32\^2]{}d\^4x d\^4 W\^2. \[eq81\_\] We see that the divergent part of effective action is proportional to the classical action in $\cN=(1,0)$ SYM model. In this sense the non-anticommutative charged hypermultiplet model (\[e13\]) is renormalizable.
Finite part of the effective action
-----------------------------------
Now we will derive the finite part of the effective action of deformed hypermultiplet model. We start with the expression (\[e37\]) applying the following approximation 1[\_]{}1[-12{W,|W }\_]{}. \[eq82\] This means that we neglect all spatial and spinor covariant derivatives of strength superfields in the decomposition (\[e22\]) \_m W=\_m|W=0,\^[+]{}W=0,|\^+\_|W=0. \[eq83\] Exactly such an approximation (\[eq83\]) is sufficient for deriving the (anti)holomorphic contributions. Therefore the effective action (\[e37\]) can be rewritten as follows $$\begin{aligned}
\delta\Gamma&=&\int d^{12}z_1 d^{12}z_2 \frac{du_1 du_2}{(u^+_1u^+_2)^6}
\delta V^{++}(1)\star\frac1\square
\delta^{12}(z_1-z_2)
\nonumber\\&&\times
V^{++}(2)\star\frac1{\square-\frac12\{W,\bar W \}_\star}
(D^+_1)^4(D^+_2)^4\left\{
e^{\Omega(2)}_\star\star e^{-\Omega(1)}_\star\star
\delta^{12}(z_2-z_1)\right\}.
\label{eq84}\end{aligned}$$ Once again, all the derivatives $D^+$ in the second line of (\[eq84\]) must hit the delta-function. Applying the identity (\[eq79\]) the integration over $d^8\theta_2$ is performed $$\begin{aligned}
\delta\Gamma&=&\int d^{12}z_1 d^4x_2 \frac{du_1 du_2}{(u^+_1u^+_2)^2}
\delta V^{++}(x_1,\theta,u_1)\star V^{++}(x_2,\theta,u_2)
\star e^{\Omega(2)}_\star\star e^{-\Omega(1)}_\star
\nonumber\\&&
\star\frac1\square
\delta^4(x_1-x_2)\frac1{\square-\frac12\{W,\bar W \}_\star}
\delta^4(x_2-x_1).
\label{eq85}\end{aligned}$$ In the momentum space the second line of (\[eq85\]) reads 1[k\^2]{}1[k\^2+12{W,|W}\_]{} =-\_+([divergent term]{}), \[eq86\] where $\Lambda$ is an arbitrary constant of dimension $+1$. Note that the integral (\[eq86\]) has the logarithmic divergence. We consider here only its finite part since the divergent contribution have been calculated above. As a result, the finite part of (\[eq85\]) is given by $$\begin{aligned}
\delta\Gamma&=&-\frac1{16\pi^2}\int d^{12}z du_1\,
\delta V^{++}(x,\theta,u_1)\star \int du_2
\frac{V^{++}(x,\theta,u_2)\star
e^{\Omega(x,\theta,u_2)}_\star\star
e^{-\Omega(x,\theta,u_1)}_\star}{(u^+_1u^+_2)^2}
\nonumber\\&&
\star\ln_\star\frac{\{W,\bar W\}_\star}{2\Lambda^2}.
\label{eq87}\end{aligned}$$ Applying the identity (\[e18\]), we conclude =-1[16\^2]{}d\^[12]{}z du V\^[++]{}V\^[–]{}\_. \[eq88\]
The variation of the effective action (\[eq88\]) is one of the main results of the present work. It gives us the part of the low-energy effective action depending on the strength superfields without derivatives.
If the parameter of non-anticommutativity tends to zero, $I\to0$, the equation (\[eq88\]) reduces to the holomorphic and antiholomorphic parts of the effective action \_[(I=0)]{}=-1[16\^2]{}d\^[12]{}z du V\^[++]{} V\^[–]{} (W+). \[eq89\] The variation (\[eq89\]) exactly corresponds to the holomorphic potential (\[e8\]) and its conjugate.
Note that when $I\ne0$, the logarithm in (\[eq88\]) can not be represented in a form of a sum of holomorphic and antiholomorphic parts. Therefore, the variation of the effective action (\[eq88\]) is responsible for both holomorphic, antiholomorphic and non-holomorphic contributions to the effective action.
Holomorphic contribution
------------------------
Let us single out purely holomorphic part from the expression (\[eq88\]). For this purpose we restrict the background strength superfield $\bar W$ to be constant |W=|[**W**]{}=const, \[eq90\] and the log function in (\[eq88\]) simplifies to \_= \_+. \[eq91\] The holomorphic part is now given by \_[hol]{}=-1[16\^2]{}d\^[12]{}z du V\^[++]{}V\^[–]{}\_W. \[eq93.1\] According to the equation (\[e31\]), the variation (\[eq93.1\]) can easily be integrated: \_[hol]{}=-1[32\^2]{}d\^4x d\^4 WW\_. \[eq93.2\] As a result, we proved that the holomorphic part of effective action in the hypermultiplet model is nothing but a star-product generalization of a standard holomorphic potential (\[e8\]).
Antiholomorphic contribution
----------------------------
Similarly, the antiholomorphic part of the effective action can be found from (\[eq88\]) when we restrict the strength $W$ to be constant W=[**W**]{}=const. \[e93.2.1\] The antiholomorphic part now reads \_[antihol]{}=1[16\^2]{}d\^[12]{}z du V\^[++]{}V\^[–]{}\_. \[eq93.3\] In contrast to the variation (\[eq93.1\]), the expression (\[eq93.3\]) can not be so easily integrated since there is no antiholomorphic potential written in the antichiral superspace. However, one can readily find the (part of) effective equation of motion corresponding to the variation (\[eq93.3\]) =1[16\^2]{}(|D\^+)\^2. \[eq93.4\] The variation (\[eq93.3\]) will be integrated in the next section only for some particular choice of background gauge superfields.
0
Component structure of low-energy effective action
==================================================
We are interested in the leading component terms of the (anti)holomorphic effective actions (\[eq93.2\]) and (\[eq93.3\]) in the bosonic sector. Clearly, these effective actions should give the corrections to the corresponding expression (\[e11\]) in the undeformed theory. Therefore we will use the same ansatz (\[e10\]) for the component fields. Here we will follow the works [@FILSZ; @ILZ], using the same conventions and notations for the component fields and sigma-matrices.
The scalar and vector fields enter the prepotential $V^{++}$ in the Wess-Zumino gauge as follows V\^[++]{}\_[WZ]{}=(\^+)\^2|+(|\^+)+(\^+\_m|\^+)A\_m -2i(|\^+)\^2(\^+\^-)\_m A\_m -(|\^+)\^2(\^-\_[mn]{}\^+)F\_[mn]{}. \[eqq100\] The prepotential $V^{--}$ is defined as a solution of zero-curvature equation (\[e16\]). Unfolding the star-product in (\[e16\]), we have $$\begin{aligned}
&&D^{++}V^{--}-D^{--}V^{++}_{WZ}+2I[\partial_+^\alpha V^{++}_{WZ}\partial_{-\alpha}V^{--}
-\partial_-^\alpha V^{++}_{WZ}\partial_{+\alpha}V^{--}]
\nonumber\\
&&+\frac{I^3}2[\partial_-^\alpha(\partial_+)^2V^{++}_{WZ}
\partial_{+\alpha}(\partial_-)^2V^{--}
-\partial_+^\alpha(\partial_-)^2V^{++}_{WZ}\partial_{-\alpha}(\partial_+)^2V^{--}]=0,
\label{eqq102}\end{aligned}$$ where \_[+]{}=,\_[-]{}=. \[eqq103\] One can look for the prepotential $V^{--}$ in the following form $$\begin{aligned}
V^{--}&=&v^{--}+\bar\theta^-_{\dot\alpha}v^{-\dot\alpha}+
(\bar\theta^-)^2 A
+(\bar\theta^+\bar\theta^-)\varphi^{--}\nonumber\\
&&+(\bar\theta^+\tilde\sigma^{mn}\bar\theta^-)\varphi^{--}_{mn}
+(\bar\theta^-)^2\bar\theta^+_{\dot\alpha}\tau^{-\dot\alpha}
+(\bar\theta^+)^2(\bar\theta^-)^2\tau^{--},
\label{eqq107}\end{aligned}$$ where all fields in the rhs of eq. (\[eqq107\]) depend only on $\theta^+_\alpha,\ \theta^-_\alpha$ variables. The superfields $v^{--}$, $v^{-\dot\alpha}$, $\varphi^{--}$, $A$, $\tau^{-\dot\alpha}$, $\tau^{--}$ should be found from the eq. (\[eqq102\]). The iterative procedure of solving eq. (\[eqq102\]) is given in [@FILSZ]. Following the same steps we find $$\begin{aligned}
v^{--}&=&(\theta^-)^2\frac{\bar\phi}{1+4I\bar\phi},
\label{eqq108}\\
v^{-\dot\alpha}&=&\frac{(\theta^-\sigma_m)^{\dot\alpha}A_m}{1+4I\bar\phi},
\label{eqq109}\\
\varphi^{--}&=&-\frac{2i(\theta^-)^2\partial_mA_m}{1+4I\bar\phi},
\label{eqq110}\\
A&=&\phi+\frac{4IA_mA_m}{1+4I\bar\phi}
+(\theta^+\sigma_{mn}\theta^-)F_{mn},
\label{eqq111}\\
\tau^{-\dot\alpha}&=&\frac{4I(\theta^-\sigma_{mn})^\alpha F_{mn}\sigma_{
r\alpha}{}^{\dot\alpha}A_r}{1+4I\bar\phi},
\label{eqq112}\\
\tau^{--}&=&\frac{4I(\theta^-)^2(F_{mn}F_{mn}+F_{mn}\tilde
F_{mn})}{1+4I\bar\phi}.
\label{eqq113}\end{aligned}$$
Now we obtain the component structure of the strength superfields $$\begin{aligned}
W&=&-\frac14(\bar D^+)^2V^{--}=\phi+\frac{4IA_m
A_m}{1+4I\bar\phi}+(\theta^+\sigma_{mn}\theta^-)F_{mn},
\label{eqq114}\\
\bar W&=&-\frac14( D^+)^2V^{--}=\frac{\bar\phi}{1+4I\bar\phi}+
(\bar\theta^+\tilde\sigma_{mn}\bar\theta^-)\frac{F_{mn}}{1+4I\bar\phi}.
\label{eqq115}\end{aligned}$$ Note that the strength superfields (\[eqq114\]) and (\[eqq115\]) are deformed differently. It is clear that in the limit $I\to0$ these expressions coincide with the undeformed ones (\[e9\]). Introducing the notations =+,|=,\_[mn]{}=, \[eqq116\] the eqs. (\[eqq114\],\[eqq115\]) can be written in a form similar to eq. (\[e9\]) W=+(\^+\_[mn]{}\^-)F\_[mn]{},|W=|+(|\^+\_[mn]{}|\^-)[**F**]{}\_[mn]{}. \[eqq117\]
To find the component structure of the holomorphic potential (\[eq93.2\],\[eq93.3\]) we have to compute the following quontities $$\begin{aligned}
W\star W&=&\Phi^2+4I^2(F^2+F\tilde F)+
2(\theta^+\sigma_{mn}\theta^-)F_{mn}\Phi
+(\theta^+)^2(\theta^-)^2(F^2+F\tilde F),
\label{eqq121}\\
\ln_\star\frac W\Lambda&=&\ln\frac\Phi\Lambda+\frac14\ln\left[
1-\frac{8I^2(F^2+F\tilde F)}{\Phi^2}\right]
+(\theta^+\sigma_{mn}\theta^-)\frac{F_{mn}}{\Phi}
\frac{{\rm arcth}\sqrt{\frac{8I^2(F^2+F\tilde F)}{\Phi^2}}}{
\sqrt{\frac{8I^2(F^2+F\tilde F)}{\Phi^2}}}
\nonumber\\&&
+\frac1{16I^2}(\theta^+)^2(\theta^-)^2\ln\left[
1-\frac{8I^2(F^2+F\tilde F)}{\Phi^2}
\right],
\label{eqq122}\\
\bar W\star\bar W&=&\bar W\bar W=
\bar\Phi^2+
2(\bar\theta^+\tilde\sigma_{mn}\bar\theta^-)F_{mn}\bar\Phi
+(\bar\theta^+)^2(\bar\theta^-)^2(F^2+F\tilde F),
\label{eqq122.1}\\
\ln_\star\frac{\bar W}\Lambda&=&
\ln\frac{\bar W}{\Lambda}=
\ln\frac{\bar\Phi}\Lambda
+(\bar\theta^+\tilde\sigma_{mn}\bar\theta^-)
\frac{{\bf F}_{mn}}{\bar\Phi}
-\frac12(\bar\theta^+)^2(\bar\theta^-)^2\frac{{\bf F}^2+{\bf F}\tilde {\bf
F}}{\bar\Phi^2}.
\label{eqq122.2}\end{aligned}$$ As a result, substituting the expressions (\[eqq121\],\[eqq122\]) into eq. (\[eq93.2\]), we find the component structure of the holomorphic effective action \_[hol]{}=-1[32\^2]{}d\^4x (F\^2+FF) , \[eqq124\] where $$\begin{aligned}
\Delta(X)&=&\frac12(1-X)^2
\ln(X-1)+\frac12(1+ X)^2\ln(1+ X)-(1+X^2)\ln X,
\label{eqq126}\\
X(\Phi,F_{mn})&=&\frac\Phi{2I\sqrt{2(F^2+F\tilde
F)}}.
\label{eqq124.1}\end{aligned}$$ The equation (\[eqq124\]) shows that the function $\Delta(X)$ is a correction due to non-anticommutativity to the standard holomorphic effective action (\[e11\]). This function has the smooth limit at $I\to0$ \_[I0]{}(X)=32. \[eqq127\] Note that exactly the constant 3/2 stands in the rhs in eq. (\[e11\]) which was not essential in the undeformed theory, but now this constant is replaced by the function $\Delta(X)$. More detailed studies of the function $\Delta(X)$ are given in the Appendix.
Let us consider the antiholomorphic potential when the strength superfield $\bar W$ is defined by the component expression (\[eqq115\]) with the fields $F_{mn}$ and $\bar\phi$ being constant. The equations (\[eqq122.1\],\[eqq122.2\]) show that the star-product can be omitted for such an approximation. Therefore the variation (\[eq93.3\]) can be integrated is the same way as in the undeformed theory $$\begin{aligned}
\Gamma_{antihol}&=&-\frac1{32\pi^2}\int d^4x d^4\bar\theta
\,\bar W^2\ln\frac{\bar W}{\Lambda}
=-\frac1{32\pi^2}\int d^4x ({\bf F}^2+{\bf F}\tilde {\bf F})
\left(\ln\frac{\bar\Phi}{\Lambda}+\frac32\right)
\nonumber\\&&
=-\frac1{32\pi^2}\int d^4x \frac{(F^2+F\tilde F)}{
(1+4I\bar\phi)^2}\left(
\ln\frac{\bar\phi}{\Lambda(1+4I\bar\phi)}+\frac32\right).
\label{eqq135}\end{aligned}$$ Here the equations (\[eqq122.1\],\[eqq122.2\],\[eqq116\]) have been used. We see that the non-anticommutativity manifests itself here by a simple rescaling of fields by the factor $1/(1+4I\bar\phi)$. In the limit $I\to0$ the expression (\[eqq135\]) reduces to the standard one for the antiholomorphic potential.
Deformation of the central charge and the mass
==============================================
It is well known [@Buch] that the model of hypermultiplet interacting with the external vector superfield possesses non-trivial central charge which is related to the mass of the hypermultiplet via BPS relation. Let us find the deformation of central charge and the mass in the case of non-anticommutative singlet deformation under considerations.
The central charges in the $\cN=2$ theories arise effectively from non-vanishing vacuum expectation values of scalar fields [@Buch] =a,|=|a. \[eq94\_\] The constants $a$, $\bar a$ enter the prepotentials (\[eqq100\],\[eqq107\]) as follows $$\begin{aligned}
{\bf V}^{++}&=&a(\bar\theta^+)^2+\bar a(\theta^+)^2,
\label{eq94}\\
{\bf V}^{--}&=&a(\bar\theta^-)^2+\frac{\bar a}{1+4I\bar a}(\theta^-)^2.
\label{eq95}\end{aligned}$$ The corresponding strength superfields (\[eqq114\],\[eqq115\]) read now =a,|[**W**]{}=. \[eq96\] We see that the strength $\bar {\bf W}$ is deformed by the non-anticommutativity while ${\bf W}$ is not.
Now, using the standard relations $V^-_{\alpha}=-D^+_\alpha V^{--}$, $\bar V^-_{\dot\alpha}=-\bar D^+_{\dot\alpha} V^{--}$, we derive the covariant spinor derivatives, corresponding to the prepotentials (\[eq94\],\[eq95\])
[ll]{} |[**D**]{}\^+\_=|D\^+\_, & [**D**]{}\^+\_=D\^+\_,\
|[**D**]{}\^-\_=|D\^-\_+2a|\^-\_, & [**D**]{}\^-\_=D\^-\_-\^-\_.
\[eq97\] The supercharges anticommuting with the derivatives (\[eq97\]) are
[ll]{} |[**Q**]{}\^+\_=|Q\^+\_-2a|\^+\_, & [**Q**]{}\^+\_=Q\^+\_+\^+\_,\
|[**Q**]{}\^-\_=|Q\^-\_,& [**Q**]{}\^-\_=Q\^-\_.
\[eq98\] Note that only the supercharge ${\bf Q}^+_\alpha$ is deformed by the non-anticommutativity. It is easy to find now the anticommutation relation between the supercharges (\[eq98\]) {|[**Q**]{}\^+\_,|[**Q**]{}\^-\_ }=2|Z\_,{[**Q**]{}\^+\_,[**Q**]{}\^-\_}=-2Z\_[ ]{}, \[eq99\] where the central charges $Z$, $\bar Z$ are given by |Z=a,Z=. \[eq100\] As a result, only the central charge $Z$ is deformed by the non-anticommutativity while $\bar Z$ is not.
It is well known that in the presence of central charge the hypermultiplet acquires the BPS mass. To find the mass, let us consider the operator =-12(D\^+)\^4(\^[–]{})\^2=-m\^2, \[eq101\] where $\nnabla^{--}=D^{--}+{\bf V}^{--}$ and m\^2=Z|Z=. \[eq102\] It is easy to see that $m^2$, given by eq. (\[eq102\]), is a mass squared of the hypermultiplet. Indeed, let us consider the hypermultiplet model interacting with the background vector superfield (\[eq94\]) =d\^[(-4)]{} duq\^+(D\^[++]{}+[**V**]{}\^[++]{})q\^+. \[eq103\] The model (\[eq103\]) is effectively described by the massive propagator \^[(1,1)]{}(1|2)=-1[-m\^2]{}(D\^+\_1)\^4(D\^+\_2)\^4 {e\^[[****]{}(1)]{}\_e\^[-[****]{}(2)]{}\_ }, \[eq104\] which can be derived in the same way as the general one (\[e20\]). Here $\bf\Omega$ is a bridge superfield corresponding to the prepotential ${\bf V}^{++}$ (\[eq94\]).
Note that the mass squared (\[eq102\]) and the central charges (\[eq100\]) are deformed in such a way that the BPS relation $m^2=Z\bar
Z$ is conserved.
Conclusions
===========
In this paper we studied the low-energy effective action and renormalizability of $\cN{=}(1,0)$ non-anticommutative charged hypermultiplet theory. This model describes the interaction of a hypermultiplet with an Abelian vector superfield under the singlet chiral deformation of supersymmetry. Let us summarize the basic results obtained in the present work.
1. The procedure of perturbative quantum computation of the low-energy effective action for the deformed charged hypermultiplet model is developed within the harmonic superspace approach.
2. Using this procedure, the divergent part of the effective action is calculated and is shown to be proportional to the classical action of $\cN{=}(1,0)$ non-anticommutative SYM theory. In this sense the present model is renormalizable.
3. The general structure of the low-energy effective action of this theory is revealed. Away from the undeformed limit, the antiholomorphic piece no longer exists by itself but is incorporated in a [*full*]{} $\cN{=}(1,1)$ superspace integral.
4. The holomorphic effective action is calculated and remains a [*chiral*]{} superspace integral. It is shown to be given by the holomorphic potential which is a star-generalization of the undeformed one. The contribution to (the variation of) the effective action that corresponds to the antiholomorphic potential is also found as an expression written in full $\cN{=}(1,1)$ superspace.
5. The component structure of the (anti)holomorphic effective action is studied in the bosonic sector in the constant-fields approximation. It is shown that the holomorphic and antiholomorphic potentials still get deformed differently. In this approximation the deformed holomorphic potential (\[eqq124\]) acquires the extra terms given by the function (\[eqq126\]). For the antiholomorphic piece, it is shown that the deformation merely effects a rescaling of component fields by a factor of $(1+4I\bar\phi)^{-1}$, where $\bar\phi$ is one of the two scalar fields of the vector multiplet.
6. The deformation of the mass and the central charge is found. It is shown that the mass-squared as well as the central charge are rescaled by the same factor $(1+4I\bar\phi)^{-1}$, preserving the relation between them.
In the light of the present results, it would be rewarding to solve the following problems concerning the quantum aspects of non-anticommutative theories with extended supersymmetry. First, it is tempting to determine for the hypermultiplet model the deformation of the next-to-leading terms in the effective action, which are necessarily non-holomorphic. Also, one should develop the non-Abelian generalization. Next, it is important to perform an analogous investigation for the pure SYM theory, since in the undeformed case this model (the Seiberg-Witten theory) plays an important role in modern theoretical physics. Finally, it would be interesting to extend the quantum studies of non-anticommutative theories to the case of non-singlet (i.e. more general) deformations of supersymmetry, as considered particularly in [@CILQ] on the classical level.
Acknowledgements {#acknowledgements .unnumbered}
================
The present work is supported particularly by the DFG grant, project No 436 RUS 113/669/0-3 and by INTAS grant. The work of I.L.B and I.B.S. was supported by RFBR grant, project No 06-02-16346, joint RFBR-DFG grant, project No 06-02-04012 and LRSS grant, project No 4489.2006.2. IBS acknowledges the support from the grant of the President of Russian Federation, project No 7110.2006.2. O.L. acknowledges support from the DFG grant Le-838/7.
0
Appendix
========
Let us give some more comments about the function $\Delta(X)$ given by eq. (\[eqq126\]).
[**i.**]{} Due to the presence of log function, the expression (\[eqq126\]) is well-defined if $X>1$, or >2I. \[eqq128\] The equation (\[eqq128\]) means that the vacuum values for the scalar field $\phi$ are bounded below if $I\ne0$.
[**ii.**]{} In the region (\[eqq128\]) the function $3/2-\Delta(X)$ is monotone decreasing. It is plotted in the Fig. 1a.
[**iii.**]{} If one relaxes the condition (\[eqq127\]), the function $\Delta(X)$, as well as the holomorphic effective action, acquires an imaginary part. The expressions ${\rm Re}(\ln
X+\Delta(X))$, ${\rm Im}(\ln X+\Delta(X))$ are plotted in Fig. 1b in comparison with the function $\ln X+3/2$ which is responsible for the bosonic part of the holomorphic effective action (\[e11\]) in the undeformed case. It is interesting to note that the function $\ln X+\Delta(X)$ has no logarithmic singularity at the origin $X=0$ when $I\ne0$.
[99]{} J. de Boer, P.A. Grassi, P. van Nieuwenhuizen, [*Non-commutative superspace from string theory*]{}, Phys. Lett. [**B574**]{} (2003) 98, [hep-th/0302078]{}. H. Ooguri, C. Vafa, [*The C-deformation of gluino and non-planar diagrams*]{}, Adv. Theor. Math. Phys. [**7**]{} (2003) 53, [hep-th/0302109]{}; [*Gravity induced C-deformation*]{}, Adv. Theor. Math. Phys. [**7**]{} (2004) 405, [hep-th/0303063]{}. N. Berkovits, N. Seiberg, [*Superstrings in graviphoton background and $N=1/2+3/2$ supersymmetry*]{}, JHEP [**0307**]{} (2003) 010, [hep-th/0306226]{}; N. Seiberg, [*Noncommutative superspace, $N=1/2$ supersymmetry, field theory and string theory*]{}, JHEP [**0306**]{} (2003) 010, [hep-th/0305248]{}. R. Britto, B. Feng, [*N=1/2 Wess-Zumino model is renormalizable*]{}, Phys. Rev. Lett. [**91**]{} (2003) 201601, [hep-th/0307165]{}; R. Britto, B. Feng, S.-J. Rey, [*Deformed superspace, N=1/2 supersymmetry and nonrenormalization theorems*]{}, JHEP [**0307**]{} (2003) 067, [hep-th/0306215]{}; S. Terashima, J.-T. Yee, [*Comments on noncommutative superspace*]{}, JHEP [**0312**]{} (2003) 053, [hep-th/0306237]{}; A. Romagnoni, [*Renormalizability of N=1/2 Wess-Zumino model in superspace*]{}, JHEP [**0310**]{} (2003) 016, [hep-th/0307209]{}; M.T. Grisaru, S. Penati, A. Romagnoni, [*Nonanticommutative superspace and N=1/2 WZ model*]{}, Class.Quant.Grav. [**21**]{} (2004) S1391, [hep-th/0401174]{}; [*Two-loop renormalization for nonanticommutative N=1/2 supersymmetric WZ Model*]{}, JHEP [**0308**]{} (2003) 003, [hep-th/0307099]{}; D. Berenstein, S.-J. Rey, [*Wilsonian proof for renormalizability of N=1/2 supersymmetric field theories*]{}, Phys. Rev. [**D68**]{} (2003) 121701, [hep-th/0308049]{}; O. Lunin, S.-J. Rey, [*Renormalizability of non(anti)commutative gauge theories with N=1/2 supersymmetry*]{}, JHEP [**0309**]{} (2003) 045, [hep-th/0307275]{}; R. Britto, B. Feng, S.-J. Rey, [*Non(anti)commutative superspace, UV/IR mixing and open Wilson lines*]{}, JHEP [**0308**]{} (2003) 001, [hep-th/0307091]{}; M. Alishahiha, A. Ghodsi, N. Sadooghi, [*One-loop perturbative corrections to non(anti)commutativity parameter of N=1/2 supersymmetric U(N) gauge theory*]{}, Nucl. Phys. [**B691**]{} (2004) 111, [hep-th/0309037]{}. I. Jack, D.R.T. Jones, L.A. Worthy, [*One-loop renormalisation of N=1/2 supersymmetric gauge theory*]{}, Phys. Lett. [**B611**]{} (2005) 199, [hep-th/0412009]{}; [*One-loop renormalisation of general N=1/2 supersymmetric gauge theory*]{}, Phys. Rev. [**D72**]{} (2005) 065002, [hep-th/0505248]{}; [*One-loop renormalisation of massive N=1/2 supersymmetric gauge theory*]{}, [hep-th/0607194]{}; [*One-loop renormalisation of N=1/2 supersymmetric gauge theory in the adjoint representation*]{}, [hep-th/0607195]{}. S. Penati, A. Romagnoni, [*Covariant quantization of N=1/2 SYM theories and supergauge invariance*]{}, JHEP [**0502**]{} (2005) 064, [hep-th/0412041]{}; M.T. Grisaru, S. Penati, A. Romagnoni, [*Non(anti)commutative SYM theory: renormalization in superspace*]{}, JHEP [**0602**]{} (2006) 043, [hep-th/0510175]{}; A.T. Banin, I.L. Buchbinder, N.G. Pletnev, [*Chiral effective potential in N=1/2 non-commutative Wess-Zumino model*]{}, JHEP [**0407**]{} (2004) 011, [hep-th/0405063]{}; O.D. Azorkina, A.T. Banin, I.L. Buchbinder, N.G. Pletnev, [*One-loop effective potential in N=1/2 generic chiral superfield model*]{}, Phys. Lett. [**B635**]{} (2006) 50, [hep-th/0601045]{}; [*Construction of effective action in nonanticommutative supersymmetric field theories*]{}, Phys. Lett. [**B633**]{} (2006) 389, [hep-th/0509193]{}. E. Ivanov, O. Lechtenfeld, B. Zupnik, [*Nilpotent deformations of $N=2$ superspace*]{}, JHEP [**0402**]{} (2004) 012, [hep-th/0308012]{}; S. Ferrara, E. Sokatchev, [*Non-anticommutative N=2 super-Yang-Mills theory with singlet deformation*]{}, Phys. Lett. [**B579**]{} (2004) 226, [hep-th/0308021]{}. A. De Castro, E. Ivanov, O. Lechtenfeld, L. Quevedo, [*Non-singlet Q-deformation of the N=(1,1) gauge multiplet in harmonic superspace*]{}, Nucl. Phys. [**B747**]{} (2006) 1, [hep-th/0510013]{}; A. De Castro, L. Quevedo, [*Non-singlet Q-deformed N=(1,0) and N=(1,1/2) U(1) actions*]{}, ITP-UH-11-06, [hep-th/0605187]{}; T. Araki, K. Ito, A. Ohtsuka, [*N = 2 supersymmetric U(1) gauge theory in noncommutative harmonic superspace*]{}, JHEP [**0401**]{} (2004) 046, [hep-th/0401012]{}; [*Deformed supersymmetry in non(anti)commutative N=2 supersymmetric U(1) gauge theory*]{}, Phys. Lett. [**B606**]{} (2005) 202, [hep-th/0410203]{}; [*Non(anti)commutative N=(1,1/2) supersymmetric U(1) gauge theory*]{}, JHEP [**0505**]{} (2005) 074, [hep-th/0503224]{}. S. Ferrara, E. Ivanov, O. Lechtenfeld, E. Sokatchev, B. Zupnik, [*Non-anticommutative chiral singlet deformation of N=(1,1) gauge theory*]{}, Nucl. Phys. [**B704**]{} (2005) 154, [hep-th/0405049]{}. E. Ivanov, O. Lechtenfeld, B. Zupnik, [*Non-anticommutative deformation of N=(1,1) hypermultiplets*]{}, Nucl. Phys. [**B707**]{} (2005) 69, [hep-th/0408146]{}. E.A. Ivanov, B.M. Zupnik, [*Non-anticommutative deformations of N=(1,1) supersymmetric theories*]{}, Theor. Math. Phys. [**142**]{} (2005) 197, [hep-th/0405185]{}. T. Araki, K. Ito, [*Singlet deformation and non(anti)commutative N=2 supersymmetric U(1) gauge theory*]{}, Phys. Lett. [**B595**]{} (2004) 513, [hep-th/0404250]{}. I.L. Buchbinder, E.A. Ivanov, O. Lechtenfeld, I.B. Samsonov, B.M. Zupnik, [*Renormalizability of non-anticommutative N=(1,1) theories with singlet deformation*]{}, Nucl. Phys. [**B740**]{} (2006) 358. N. Seiberg, E. Witten, [*Electric-magnetic duality, monopole condensation, and confinement in N=2 supersymmetric Yang-Mills theory*]{}, Nucl. Phys. [**B426**]{} (1994) 19. A. Galperin, E. Ivanov, V. Ogievetsky, E. Sokatchev, [*Harmonic superspace: key to N=2 supersymmetric theories*]{}, JETP Lett. [**40**]{} (1984) 912 \[Pisma Zh.Eksp.Teor.Fiz. [**40**]{} (1984) 155\]; A. Galperin, E. Ivanov, S. Kalitzin, V. Ogievetsky, E. Sokatchev, [*Unconstrained N=2 matter, Yang-Mills and supergravity theories in harmonic superspace*]{}, Class. Quant. Grav. [**1**]{} (1984) 469. A. Galperin, E. Ivanov, V. Ogievetsky, E. Sokatchev, [*Harmonic Superspace*]{}, Cambridge Univ. Press, 2001, 306 p. I.L. Buchbinder, E.I. Buchbinder, E.A. Ivanov, S.M. Kuzenko, B.A. Ovrut, [*Effective action of the N=2 Maxwell multiplet in harmonic superspace*]{}, Phys. Lett. [**B412**]{} (1997) 309, [hep-th/9703147]{}; E.I. Buchbinder, I.L. Buchbinder, E.A. Ivanov, S.M. Kuzenko, [*Central charge as the origin of holomorphic effective action in N=2 gauge theory*]{}, Mod. Phys. Lett. [**A13**]{} (1998) 1071, [hep-th/9803176]{}. S. Eremin, E. Ivanov, [*Holomorphic effective action of N=2 SYM theory from harmonic superspace with central charges*]{}, Mod. Phys. Lett. [**A15**]{} (2000) 1859, [hep-th/9908054]{}; I.L. Buchbinder, I.B. Samsonov, [*On holomorphic effective actions of hypermultiplets coupled to external gauge superfields*]{}, Mod. Phys. Lett. [**A14**]{} (1999) 2537, [hep-th/9909183]{}. M. Dine, N. Seiberg, [*Comments on higher derivative operators in some susy field theories*]{}, Phys. Lett. [**B409**]{} (1997) 239, [hep-th/9705057]{}; V. Periwal, R. von Unge, [*Accelerating D-branes*]{}, Phys. Lett. [**B430**]{} (1998), [hep-th/9801121]{}; F. Gonzalez-Rey, M. Roček, [*Nonholomorphic N=2 terms in N=4 SYM: one loop calculation in N=2 superspace*]{}, Phys. Lett. [**B434**]{} (1998) 303, [hep-th/9804010]{}. I.L. Buchbinder, E.I. Buchbinder, S.M. Kuzenko, B.A. Ovrut, [*The Background field method for N=2 superYang-Mills theories in harmonic superspace*]{}, Phys. Lett. [**B417**]{} (1998) 61; I.L. Buchbinder, S.M. Kuzenko, [*Comments on the background field method in harmonic superspace: nonholomorphic corrections in N=4 SYM*]{}, Mod. Phys. Lett. [**A13**]{} (1998) 1623, [hep-th/9804168]{}; E.I. Buchbinder, I.L. Buchbinder, S.M. Kuzenko, [*Nonholomorphic effective potential in N=4 SU(n) SYM*]{}, Phys. Lett. [**B446**]{} (1999) 216, [hep-th/9810239]{}. D.A. Lowe, R. von Unge, [*Constraints on higher derivative operators in maximally supersymmetric gauge theory*]{}, JHEP [**9811**]{} (1998) 014, [hep-th/9811017]{}; I.L. Buchbinder, S.M. Kuzenko, A.A. Tseytlin, [*On low-energy effective action in N=2,4 superconformal theories in four dimensions*]{}, Phys. Rev. [**D62**]{} (2000) 045001, [hep-th/9911221]{}; I.L. Buchbinder, E.A. Ivanov, [*Complete N=4 structure of low-energy effective action in N=4 super Yang-Mills theories*]{}, Phys. Lett. [**B524**]{} (2002) 208, [hep-th/0111062]{}; I.L. Buchbinder, E.A. Ivanov, A.Yu. Petrov, [*Complete low-energy effective action in N=4 SYM: a direct N=2 supergraph calculation*]{}, Nucl. Phys. [**B653**]{} (2003) 64, [hep-th/0210241]{}; A.T. Banin, I.L. Buchbinder, N.G. Pletnev, [*One-loop effective action of N=4 SYM theory in the hypermultiplet sector: low-energy approximation and beyond*]{}, Phys. Rev. [**D68**]{} (2003) 065024, [hep-th/0304046]{}; I.L. Buchbinder, N.G. Pletnev, [*Construction of one-loop N=4 SYM effective action in the harmonic superspace approach*]{}, JHEP [**0509**]{} (2005) 073, [hep-th/0504216]{}. N. Dragon, S.M. Kuzenko, [*The Higgs mechanism in N=2 superspace*]{}, Nucl. Phys. [**B508**]{} (1997) 229, [hep-th/9705027]{}; S.M. Kuzenko, I.N. McArthur, [*Effective action of N=4 superYang-Mills: N=2 superspace approach*]{}, Phys. Lett. [**B506**]{} (2001) 140, [hep-th/0101127]{}; [*Hypermultiplet effective action: N = 2 superspace approach*]{}, Phys. Lett. [**B513**]{} (2001) 213, [hep-th/0105121]{}.
[^1]: [email protected]
[^2]: [email protected]
[^3]: [email protected]
[^4]: Note that we write here (and further) the box operator $\square$ assuming that it is nothing but the Laplacian operator rather than a d’Alambertian one since we deal with the Euclidian space.
[^5]: Note that the one-loop effective action in the Euclidean space is given by $\Gamma=\Tr\ln S^{(2)}_{\phi\bar\phi}$ rather then the Minkowski space expression $\Gamma= i\Tr\ln S^{(2)}_{\phi\bar\phi}\,$. Here $S^{(2)}_{\phi\bar\phi}$ is the second mixed functional derivative of a classical action $S[\phi,\bar\phi]$.
| {
"pile_set_name": "ArXiv"
} | ArXiv |
---
abstract: 'The spectral properties of one exciton trapped in a self-assembled multi-layered quantum dot is obtained using a high precision variational numerical method. The exciton Hamiltonian includes the effect of the polarization charges, induced by the presence of the exciton in the quantum dot, at the material interfaces. The method allows to implement rather easily the matching conditions at the interfaces of the hetero-structure. The numerical method also provides accurate approximate eigenfunctions that enable the study of the separability of the exciton eigenfunction in electron and hole states. The separability, or the entanglement content, of the total wave function allows a better understanding of the spectral properties of the exciton and, in particular, shed some light about when the perturbation theory calculation of the spectrum is fairly correct or not. Finally, using the approximate spectrum and eigenfunctions, the controlled time evolution of the exciton wave function is analyzed when an external driving field is applied to the system. It is found that it is possible to obtain pico and sub-picoseconds controlled oscillations between two particular states of the exciton with a rather low leakage of probability to other exciton states, and with a simple pulse shape.'
author:
- Mariano Garagiola
- Omar Osenda
title: 'Spectral properties and control of an exciton trapped in a multi-layered quantum dot'
---
Introduction {#section:introduction}
============
The calculation of the spectral properties of an exciton trapped in a semiconductor nano-structure is a long standing problem. Even in the case of the most simple model used to study this system, [*i.e.*]{} the one band effective mass approximation (EMA), there is a number of subtle details that it is necessary to take into account. In particular, when the nano-structure consists in a heterogeneous quantum dot, the mismatch between the different material parameters imposes boundary conditions to the particle wave functions and the electrostatic potential between the electron and hole that form the exciton can not be taken as the simple Coulomb potential [@Delerue2004; @Ferreyra(1998)].
The presence of excitons can be easily spotted in the absorption spectrum of different semiconductors. The energy associated to each absorption peak is smaller to the energy difference between the electronic energy levels that lie on the semiconductor valence band and those that lie on the semiconductor conduction band. If the electrons in the semiconductor were independent between them and no many-body effects were present, the absorption energy should be equal to the energy difference between two electronic levels, one located in the conduction band and the other located in the valence band. But, when an electron is promoted from the lower band to the upper one, by radiation absorption for instance, the “hole” produced interacts with the electron, and the binding of the pair gives place to the exciton. The energy difference observed in the absorption spectrum, with respect to the values that correspond to independent electrons, is precisely given by the binding energy between the pair electron-hole [@Cardona(2005); @Bastard(1988)].
An exciton trapped in a quantum dot or in a given nano-structure has very different physical properties compared to a bulk one. Roughly speaking, the radius associated to the exciton is, more or less, of the same magnitude order than the characteristic length of many nano-structures. This can be used to tune the physical trait of interest using the material parameters of the semiconductors employed in the nano-structure, its geometry and its size to tailor different properties. For instance, the mean life time of the exciton can be adjusted depending on the application in mind. Recently, some Quantum Information Processing (QIP) proposals use an exciton trapped in a quantum dot as qubit, in which case the basis states are given by the presence ($\left| 1\right\rangle$), or absence ($\left| 0\right\rangle$), of the exciton [@Troiani(2000); @Biolatti(2000)]. Obviously, for QIP applications the life time of the exciton should be larger than the time needed to operate and control the qubit.
Modeling an exciton in a particular physical situation can be a tricky business, since a plethora of effective Hamiltonians are available to use. Starting from the multi-band many-electrons Hamiltonian, there is a number of ways to derive effective Hamiltonians which usually decompose the one-electron wave function in an “envelop part” and a “Bloch part”. The Bloch part accounts for the underlying periodical lattice structure of the semiconductor and the envelop part for the specific nano-structure under consideration. If the electron is not confined, [*i.e.*]{} belongs to the bulk of the material, the envelop part is reduced to the usual imaginary exponential present in the wave functions allowed by the Bloch theorem. Otherwise, when a nano-structure is present, the effective Hamiltonians for the envelop part resemble a multicomponent Schrödinger-like equations. Probably, the most well know procedure to derive effective-Hamiltonians is the [**k.p**]{} method [@Voon(2009)]. The number of components of the Hamiltonian derived using the [**k.p**]{} method depends on how much information about the semiconductor band structure is incorporated. For instance, the well-known eight-band model is derived using the fact that the valence band levels have orbital angular momentum quantum number $L=1$, the conduction band levels $L=0$, and the electronic spin has $s=1/2$ [@Haug(2004)].
In this work we consider the simplest two-band effective mass approximation model (EMA) for an exciton, trapped in a spherical Type -I device, with a core, well, and barrier structure. The core and the barrier semiconductor compound is exactly the same, while the well semiconductor is characterized by a conduction band whose bottom energy is lower than the corresponding energy of the material that form the core and the external barrier. The gap between the conduction band and the valence band is narrower for the well semiconductor, than the gap for the core one. Consequently, the confinement potential for the electron and the hole are given by the profiles of the conduction and valence bands of the semiconductors that form the device. This model has been studied extensively because, despite its apparent simplicity, allows to obtain accurately the spectrum of different QD accurately in a wide variety of situations. Anyway, there are some features on it that ask for some careful treatment. The first feature that must be handled with care is owed to the model binding potential, since this is assumed as given by the profile of the semiconductor bands, a step-like potential results for both, the electron and the hole. Moreover, since the effective mass of both, the electron and the hole, are assumed as discontinuous position dependent functions, the Hermitian character of the kinetic energy is assured only with appropriate matching conditions for the wave functions at the interfaces between different materials. Finally, the electrostatic potential between the electron and hole can not be taken as the simple Coulomb potential. It has been shown that polarization terms should be included because, again, of the presence of interfaces between materials with different dielectric constants [@Ferreyra(1998)].
It is worth to mention that the study of the properties of excitons confined in quantum dots [@Woggon(1995)] has been tackled using a number of methods, like perturbation theory [@Chang(1998); @Schooss(1994)], the [**k.p**]{} method [@Pokatilov(2001); @Efros(1998)], and with different types of confinement potential as the infinite potential well [@Ferreyra(1998)] or the parabolic one [@Garm(1996)]. The selection of potentials and method are often dictated by the application in mind, that could range from excitonic lasers [@Bimberg(1997); @Bimberg(2005)], through quantum information processing [@Kamada(2004); @Chen(2001); @Biolatti(2002)], up to one-electron transistors [@Ishibashi(2003)] in micro electronics devices. Between the physical phenomena that have received more attention, it can be mentioned the binding energy [@Lelong(1996)], the decoherence [@Calarco(2003); @Bonadeo(1998)A; @Bonadeo(1998)B] and the Stark effect [@Billaud(2009)] among many others.
To study the two-band EMA model described above we calculated its spectrum and eigen-states using a variational approach. The approach allows us to take into account the matching conditions at the interfaces of the device and the (quite) complicated interaction potential between the two parts of the exciton that includes the effects of polarization. As we shall see, our approach allows us to obtain highly accurate approximate results for the spectrum and approximate wave functions that provide information about the entanglement content of the two-particle quantum state. This information gives useful information about the parameter range where the two-particle wave function can be more or less accurately taken as the product of one-particle wave functions. The manuscript is organized as follows: in Section \[section:one-body-model\] the variational method to obtain the spectrum of one or two-particle Hamiltonians is discussed and presented in some detail, In Section \[section:two-particle\] the electron-hole pair EMA Hamiltonian is analyzed paying some attention to the electrostatic problem originated by the geometry and composition of the quantum dot and the binding energy of the electron-hole pair is calculated, in Section \[section:separability\] the separability problem of the exciton wave function is presented and analyzed, in particular it is shown that the characterization through a measure of separability gives a better understanding of the exciton spectral properties. The time evolution of the exciton wave function when an external driving field is applied to the QD is studied in Section \[section:control\]. The study is intended to look for regimes where the external driving allows to switch between only two pre-selected states of the exciton, with a negligible or controllable leakage of probability to other exciton states. Finally, our results are discussed and summarized in Section \[section:conclusions\]
One body models and methods {#section:one-body-model}
===========================
The EMA approximation, when applied to the description of an independent electron (e), or hole (h), is characterized by a number of parameters such as the effective mass of the particle or the energy band gap of the materials from which is made the quantum dot. Otherwise, the Hamiltonian associated to it seems like an ordinary Schrödinger-like Hamiltonian $$\label{eq:one-body-ham}
\mathcal{H}_{e(h)}=-\frac{\hbar^2}{2}\nabla_{e(h)}\left(\frac{1}{m_{e(h)}^*(r_{
e(h)})}\nabla_{e(h)}\right)+
V_{e(h)}(r_{e(h)})+V_{s}(r_{e(h)})$$ where $m_{e(h)}^*(r_{e(h)})$ is the electron (hole) effective mass. For an one-band model, $m_{e}^*$ corresponds to the the effective mass of the electron in the conduction band, while $m_{h}^*$ is taken as the [*light*]{} effective mass of the hole in the valence band. $V_{e(h)}(r_{e(h)})$ is the binding potential for the electron (hole).
Figure \[fig:tipoQD\]a) shows a cartoon of the spherical self assembled quantum dot under consideration. The inner core, of radius $a$, and the outer shell (also called clad) are made of the same compound, let us call it two (2), while the middle layer, of radius $b$, is made of a different compound, let us call it one (1). These kind of hetero structures are termed of Type I or II, if the top energy of the valence band of the middle layer material, $E^1_{top}$, is larger than the top energy of the valence band of material two, $E^2_{top}$, [*i.e.*]{} the hetero structure is of Type I if $E^1_{top} > E^2_{top}$, otherwise is of Type II, as can be seen clearly in Figure \[fig:tipoQD\] b). The cartoon in Figure \[fig:tipoQD\]b) shows the conduction and valence bands profile as a function of the radial distance from the center of the hetero-structure.
![\[fig:tipoQD\] a) The cartoon shows the structure of the spherical self-assembled quantum dot. The different layers correspond to, from the center and going in the radial direction, the core whose radius $a$ and dielectric constant are shown, the potential well, with inner radius $a$ and exterior radius $b$, and the clad. b) The profiles of the conduction band, $E_c(r)$, and of the valence band, $E_v(r)$, for both a Type-I device (top) and a a Type-II devive (bottom). $E_g1$ and $E_g2$ are the gap energies of the potential well material and the core/clad material, respectively.](fig1.eps "fig:") ![\[fig:tipoQD\] a) The cartoon shows the structure of the spherical self-assembled quantum dot. The different layers correspond to, from the center and going in the radial direction, the core whose radius $a$ and dielectric constant are shown, the potential well, with inner radius $a$ and exterior radius $b$, and the clad. b) The profiles of the conduction band, $E_c(r)$, and of the valence band, $E_v(r)$, for both a Type-I device (top) and a a Type-II devive (bottom). $E_g1$ and $E_g2$ are the gap energies of the potential well material and the core/clad material, respectively.](fig2.eps "fig:")
Since the potential in Equation \[eq:one-body-ham\], $V_{e(h)}(r_{e(h)})$, is taken as exactly the conduction band profile (valence band profile) for the electron (hole), it is clear that a Type-I device corresponds to the case where the middle layer acts as a potential well for both particles, so $$\label{eq:potencial}
V_{e(h)}(r)=\left\{\begin{array}{lcl}
V_0^{e(h)}>0& &0<r<a\\
0& &a<r<b\\
V_0^{e(h)}>0& &b<r
\end{array}\right.,$$ and $V_0^{e(h)}=const.$
The potential $V_s$ is owed to the polarization charges produced at the interfaces core/middle layer and middle layer/clad because of the mismatch between their dielectric constants. It can be shown that the auto-polarization potential is given by $$\label{eq:polarizacion}
V_s(r)=\frac{q^2}{2\,
\varepsilon_1}\sum_{l}\frac{1}{(1-pq)}\left(qr^{2l}+\frac{p}{r^{2l+1}}+\frac{pq}
{r}
\right)$$ where $$p=(\varepsilon_1-\varepsilon_2)la^{(2l+1)}/[\varepsilon_2 l+\varepsilon_1
(l+1)],$$ and $$q=(\varepsilon_1-\varepsilon_2)(l+1)b^{-(2l+1)}/[\varepsilon_1 l+\varepsilon_2
(l+1)] .$$
The kinetic energy term in Equation \[eq:one-body-ham\] is written in a symmetric fashion in order to guarantee the Hermitian character of the operator, if it is considered together with the following matching conditions
$$\begin{aligned}
\label{eq:matching}
\psi(r=a^-)&=&\psi(r=a^+) \\ \nonumber
\psi(r=b^-)&=&\psi(r=b^+) \\ \nonumber
\left( \frac{1}{m^*} \frac{d\psi}{dr}\right) |_{r=a^-} &= &\left( \frac{1}{m^*}
\frac{d\psi}{dr}\right) |_{r=a^+} \\ \nonumber
\left( \frac{1}{m^*} \frac{d\psi}{dr}\right) |_{r=b^-} &= &\left( \frac{1}{m^*}
\frac{d\psi}{dr}\right) |_{r=b^+} .\end{aligned}$$
The eigenvalue problem $$\label{eq:eigenvalue-problem}
\mathcal{H} \psi = E \psi ,$$ with $\mathcal{H}$ given by Equation \[eq:one-body-ham\] can be studied using different (an numerous) methods that result in an approximate spectrum and, in some cases, approximate eigenfunctions. Anyway, the matching conditions can not be implemented in a direct way depending on which method is used to tackle the problem. A method that allows to weight adequately the matching conditions is one that incorporates easily the step-like nature of the binding potential, Equation \[eq:potencial\], and the effective mass. The approximate eigenfunctions and eigenvalues analyzed in this work were obtained using B-splines basis sets, which have been used to obtain high accuracy results for atomic, molecular and quantum dot systems. The method has been well described elsewhere, so the only details that we discus here are the relevant ones to understand some of results presented later on.
To use the B-splines basis, the normalized one-electron orbitals are given by $$\label{phi-bs}
\phi_{n}({r}) = C_n \, \frac{B^{(k)}_{n+1}(r)}{r} \,;\;\;n=1,\ldots$$
where $B^{(k)}_{n+1}(r)$ is a B-splines polynomial of order $k$. The numerical results are obtained by defining a cutoff radius $R$, and then the interval $[0,R]$ is divided into $I$ equal subintervals. B-spline polynomials [@deboor] (for a review of applications of B-splines polynomials in atomic and molecular physics, see Reference [@bachau01], for its application to QD problems see Reference [@Ferron2013]) are piecewise polynomials defined by a sequence of knots $t_1=0\leq t_2\leq\cdots \leq t_{2 k+I-1}=R$ and the recurrence relations
$$\label{bs1}
B_{i}^{(1)}(r)\,=\,\left\{ \begin{array}{ll} 1 & \mbox{if}\,t_i\leq r <
t_{i+1} \\
0 &\mbox{otherwise,} \end{array} \right. \,.$$
$$\label{bsrr}
B_{i}^{(k)}(r)\,=\,\frac{r-t_i}{t_{i+k-1}-t_i}\,B_{i}^{(k-1)}(r)\,+\,
\frac{t_{i+k}-r}{t_{i+k}-t_{i+1}}\,B_{i}^{(k-1)}(r)\; (k>1)\,.$$
The standard choice for the knots in atomic physics [@bachau01] is $t_1=\cdots=t_k=0$ and $t_{k+I}=\cdots=t_{2k+I-1}=R$. Because of the matching conditions at the interfaces between the core and the potential well and between the potential well and the clad, it is more appropriate to choose the knots as follows: at the extremes of the interval $\left[0,R\right]$ where the wave function is calculated, there are $k$ knots that are repeated, while at the interfaces $r=a$ and $r=b$ there are $k-3$ knots that are repeated, and in the open intervals $(0,a)$, $(a,b)$ and $(b,R)$ the knots are distributed uniformly [@HaoXue(2002)]
The constant $C_n$ in Equation \[phi-bs\] is a normalization constant obtained from the condition $\left\langle \phi_n |\phi_n \right\rangle =1 $
$$\label{nor-c}
C_n = \frac{1}{\left[ \int_0^{R} \, \left(B^{(k)}_{n+1}(r) \right)^2 \,dr
\right]^{1/2}} \,.$$
Because $B_1(0)\ne0$ and $B_{I+3k-9}(R)\ne0$, we have $N=I+3k-11$ orbitals corresponding to $B_2,\ldots,B_{I+3k-10}$. In all the calculations we used the value $k=5$, and, we do not write the index $k$ in the eigenvalues and coefficients.
![\[fig:fin-vs-inf\] The first one-electron eigenvalues for a infinite potential well (dashed lines) and for a finite potential well (solid lines) as functions of the dimensionless ratio $a/b$. In both cases, $b=31.71$ nm, that is the Bohr radius for an exciton immersed in a HgS matrix. The eigenvalues correspond to zero angular momentum ($\ell=0$) states.](fig3.eps)
To gain some insight about the performance of the B-spline method we first studied the eigenvalue problem of one electron confined in a multi-layered quantum dot as the one depicted in Figure \[fig:tipoQD\]a). The quantum dots whose core, well and clad are made of CdS/HgS/CdS have been extensively studied [@Schooss(1994)], so it is a good starting point to our study. For these materials, the effective masses for the electron in their respective conduction bands are $m_{e,CdS}^*=0.2$, $m_{e,HgS}^*=0.036$, while for the hole in the valence bands the effective masses are given by $m_{h,CdS}^*=0.7$, $m_{h,HgS}^*=0.040$. The dielectric constants are $\varepsilon_{CdS}=\varepsilon_2=5.5$, $\varepsilon_{HgS}=\varepsilon_1=11.36$. Besides, $$\begin{aligned}
\label{eq:band-energies}
E^{CdS}_{bottom} - E^{HgS}_{bottom} & = & 1.35\mbox{eV}, \\
E^{HgS}_{top} - E^{CdS}_{top} & = & 0.9\mbox{eV} .\end{aligned}$$ The other parameters that define the device are the radii $a$ and $b$. To augment the effect of the confinement owed to the binding potential we choose $b$ equal to the Bohr radius of a bulk exciton in HgS [@Chang(1998)], then $a$ can take any value between zero and $b$. This model was studied in [@Ferreyra(1998)] considering the limit of “strong confinement”, [*i.e.*]{} the electron is bound in an infinite potential well. We want to remark that the electro-hole pair behaves very much as a hydrogen-like atom when it is immersed in a semiconductor bulk besides, in many cases, the hole effective mass is much larger than electron one. So, it makes sense to consider the Bohr radius as one of the length scales that characterize the problem.
Figure \[fig:fin-vs-inf\] shows the behavior of the lowest lying approximate one-electron eigenvalues obtained using the B-spline method for a quantum dot with the material parameters listed above, and for one electron bound in an infinite potential well, as functions of the ratio between both radii, $a/b$. The eigenvalues corresponds to eigenfunctions with orbital angular momentum with quantum number $\ell=0$. As can be observed from the Figure, the infinite potential well eigenvalues, that can be obtained exactly, are a pretty good approximation to the quantum dot eigenvalues for small values of $a/b$, but for larger values of $a/b$, or for the excited states, the relative error grows considerably. Since in many works in the literature, the binding energy of the exciton is obtained using the exact solutions of the infinite potential well, the results shown in Figure \[fig:fin-vs-inf\] indicate that it is necessary to proceed with great caution if this approximation were to be used. The eigenvalues obtained using the B-spline method have a relative error of less than $10^{-3}$.
![\[fig:auto-polarization\] The lowest lying one-electro eigenvalues for different radial ($n_r$) and angular momentum ($\ell$) quantum numbers. The eigenvalues were calculated for a finite potential well and the Hamiltonian includes the auto-polarization term in Equation \[eq:polarizacion\] (solid lines). The dashed lines are the corresponding eigenvalues obtained taking out the autopolarization term (or equivalently choosing $\varepsilon_1=\varepsilon_2$). The labels for each curve are the quantum numbers that identify the corresponding eigenstate. ](fig4.eps)
There has been some discussion about the need to include, or not, the self polarization term, Equation \[eq:polarizacion\], in the one electron Hamiltonian, Equation \[eq:one-body-ham\] [@Delerue2004]. Figure \[fig:auto-polarization\] shows the behaviour of the lowest lying energy eigenvalues, for several orbital angular momentum quantum numbers, and with, or without, the auto-polarization term. Is is clear that for $a/b<0.4$, and for all the angular momentum quantum numbers shown, that the auto-polarization term changes the eigenvalues in a, approximately, 10$\%$ , showing that for large quantum dot the auto-polarization contribution can not be ignored. When $a$ grows up to $b$ the effect becomes less and less important. Other way to characterize the phenomenology is the following, if the eigenvalues become independent of $\ell$ for a given radial quantum number $n$, [*i.e.*]{} the energy eigenvalues depends almost only on $n$, then the well potential becomes “small enough” and the auto-polarization term becomes negligible. In this limit, the polarization term and the angular momentum part of the kinetic energy can be treated as perturbations.
It is interesting to point that despite that the hole Hamiltonian is determined by different parameters than the electron one, the behavior of its spectrum is very similar, for that reason we do not include a detailed analysis of it. On the other hand, once the electron and hole approximate eigenfunctions are obtained, a plot of them reveals that both particles are well localized inside the potential well. Besides, the repetition of knots at the interfaces enables that the approximate solutions meet the matching conditions, Equation \[eq:matching\], to a high degree of accuracy.
Two-particle model {#section:two-particle}
==================
The Hamiltonian for an exciton formed by an electron and a hole can be written as $$\label{eq:hamiltoniano-exciton}
\mathcal{H}_{ex}=\mathcal{H}_{e}+\mathcal{H}_{h}+V_c({\mathbf
r}_e,{\mathbf r}_h) ,$$ where $\mathcal{H}_{e}$ and $\mathcal{H}_{h}$ are the one-body Hamiltonians of Equation \[eq:one-body-ham\], and $V_c({\mathbf r}_e,{\mathbf r}_h)$ is the electrostatic potential between the electron and hole. Usually, one is tempted to consider $V_c$ as the usual Coulomb potential between a positive charge and a negative one but, if the exciton is confined to an hetero-structure made of different materials, this approach oversimplifies the situation.
Having in mind the argument of the paragraph above, we consider a better a approximation for the actual electrostatic potential that was suggested by Ferreyra and Proetto [@Ferreyra(1998)]. Since the hole is usually “heavier” than the electron, and that the scenario of most interest occurs when both particle are well localized, it is simpler to consider $V_c$ as the solution to the Poisson equation considering that the hole and electron coordinates are restricted to the potential well, [*i.e.*]{} $$V_c({\mathbf r}_e,{\mathbf r}_h) = q_e q_h \, G({\mathbf r}_e,{\mathbf r}_h) ,$$ where $$\label{eq:green}
\nabla_r^2 G({\mathbf r},{\mathbf r}') = \left\{
\begin{array}{lcl}
-\frac{4\pi}{\varepsilon_1} \delta({\mathbf r}-{\mathbf r}') & \quad &
\mbox{if}\;
a<r<b \\
0 & & \mbox{otherwise}
\end{array}
\right. ,$$
![\[fig:binding-energy\] The binding energy expectation value as a function of the ratio $a/b$. The curves are grouped in two bundles, the higher one was calculated including all the polarization effects originated by the dielectric constants mismatch, $\varepsilon_1\neq\varepsilon_2$ , while the lower one ignores the mismatch and takes $\varepsilon_1 = \varepsilon_2$. In each bundle, the curve at the top (solid line) corresponds to the exciton ground state. As the binding energy is a decreasing function of the exciton eigen-energy, in each bundle the lower and lower lying curves correspond to the first excited state, to the second one, ans so successively. The sudden drop of the curves observed for large enough values of the ratio $a/b$ show where each level reach the “ionization threshold”.](fig6.eps)
So, solving Equation \[eq:green\], it can be shown that the electrostatic potential that gives the interaction between the electron and the hole can be written as
$$\begin{aligned}
\label{eq:electrostatic-potential}
V_c(\mathbf{r}_e,\mathbf{r}_h)&=&q_eq_h\sum_{l,m}Y_{lm}^*(\theta_h,\varphi_h)Y_{
lm }
(\theta_e,\varphi_e) \\ \nonumber
& &\quad \times \frac{4\pi}{\varepsilon_1(2l+1)(1-pq)} \\
&&\quad \times [r_<^l+pr_<^{l+1}][r_>^{-(l+1)}+qr_>^l] ,\end{aligned}$$
where $$\label{eq:p-term}
p=(\varepsilon_1-\varepsilon_2)la^{(2l+1)}/[\varepsilon_2 l+\varepsilon_1
(l+1)] ,$$ $$\label{eq:q-term}
q=(\varepsilon_1-\varepsilon_2)(l+1)b^{-(2l+1)}/[\varepsilon_1 l+\varepsilon_2
(l+1)] ,$$ and $$r_{>(<)}=max(min) \{r_e, r_h\}.$$
The exciton binding energy can be obtained as the expectation value of the electrostatic potential $$\label{eq:binding-energy}
E_{binding}^{\alpha}= -\langle
\psi_{\alpha}|V_c(\mathbf{r}_e,\mathbf{r}_h)|\psi_{\alpha}\rangle,$$ where $|\psi_{\alpha}\rangle$ is an eigenstate of the exciton Hamiltonian, Equation \[eq:hamiltoniano-exciton\].
Figure \[fig:binding-energy\] shows the behavior of the binding energy as a function of the radii ratio $a/b$. For clarity, we restrict the curves shown to those data obtained with eigenfunctions with orbital angular momentum quantum numbers $l_e=l_h=m_e=m_h=0$. The Figure shows two well defined separated sets of curves. The lower set of curves correspond to the binding energy calculated without polarization terms (or equivalently putting $\varepsilon_1=\varepsilon_2$ in Equations \[eq:electrostatic-potential\],\[eq:p-term\] and \[eq:q-term\]). The upper set of curves corresponds to the binding energy calculated considering all the polarization effects ($\varepsilon_1\neq
\varepsilon_2$). The polarization terms, owed to the polarization charges at the interfaces between the different materials, does not change the qualitative behavior of the binding energy but, at least for the parameters of Figure \[fig:binding-energy\], not including them leads to an underestimation of the binding energy of approximately 100% . The inclusion of the polarization terms in the electrostatic potential increases the binding energy since for $\varepsilon_1>\varepsilon_2$ the correction terms in Equation \[eq:electrostatic-potential\], with respect to the Coulomb potential, have all the same sign.
separability of the exciton eigenfunctions {#section:separability}
==========================================
The availability of accurate numerical approximations for the actual exciton eigenfunctions gives the possibility to analyze their separability, in particular in this Section we analyze the behavior of the von Neumann entropy associated to the excitonic quantum state. We will show that some features of the binding energy already described in the precedent Section, can be understood best studying simultaneously both quantities. In particular, the analysis of the separability of the quantum states can shed some light about when the interaction between the exciton components can be treated using perturbation theory. Besides, there are some quantities that determine the strength of the interaction of the exciton with external fields, as the dipole moment, or expectation values that can not be accurately obtained if the correlation between the electron and hole is not taken into account.
A well known separability measure of the quantum state is the von Neumann entropy, $S$, which can be calculated as $$\label{eq:von-Neumann-def}
S=-\sum_k \lambda_k \; \ln(\lambda_k) ,$$ where $\lambda_k$ are the eigenvalues of the electron reduced density matrix $\rho^{red}(\mathbf{r}_e,\mathbf{r}^{\prime}_e)$ which, if the quantum state of the exciton is given by a vector state $\psi(\mathbf{r}_e,\mathbf{r}_h)$, then $$\label{eq:reduced-density}
\rho^{red}(\mathbf{r}_e,\mathbf{r}_e^{\prime} ) = \int
\psi^{\star}(\mathbf{r}_e,\mathbf{r}_h) \psi(\mathbf{r}_e',\mathbf{r}_h) \;
d\mathbf{r}_h .$$
![\[fig:von-Neumann-entropy\] a) The von Neummann entropy for the first few eigenstates, the lowest curve corresponds to the ground state eigenstate (label 0), the following to the second excited eigenstate (label 2) and so on. The inset shows a detailed view of the fundamental state von Neumann entropy. b) The binding energy for the ground state calculated following different methods. From top to bottom the curves were obtained with the full Hamiltonian and the B-spline method (solid line), without the auto-polarization terms ([*i.e.*]{} neglecting $V_s(r)$) and the B-spline method (dot-dashed line), again with the full Hamiltonian but using perturbation theory (dashed line) and finally without the auto-polarization terms and using perturbation theory (double-dot dashed line), respectively. For details see the text ](fig7.eps "fig:") ![\[fig:von-Neumann-entropy\] a) The von Neummann entropy for the first few eigenstates, the lowest curve corresponds to the ground state eigenstate (label 0), the following to the second excited eigenstate (label 2) and so on. The inset shows a detailed view of the fundamental state von Neumann entropy. b) The binding energy for the ground state calculated following different methods. From top to bottom the curves were obtained with the full Hamiltonian and the B-spline method (solid line), without the auto-polarization terms ([*i.e.*]{} neglecting $V_s(r)$) and the B-spline method (dot-dashed line), again with the full Hamiltonian but using perturbation theory (dashed line) and finally without the auto-polarization terms and using perturbation theory (double-dot dashed line), respectively. For details see the text ](fig8.eps "fig:")
Figure \[fig:von-Neumann-entropy\]a shows the behavior of the von Neumann entropy as a function of the ratio $a/b$, for the approximate eigenfunctions of the first low lying exciton eigenvalues calculated using the B-spline method. As can be appreciated from the figure, the von Neumann entropy is a monotone decreasing function of the ratio $a/b$, [*i.e.*]{} the electron and hole became more and more independent (separable its wave function) when the radius of the core is increased. This is expected, but even for the ground state the effect of the non-separability of the excitonic wave function has a rather large influence in the binding energy, as can been appreciated in panel b). The von Neumann entropy, on the other hand is larger for the exciton excited states so, at least in principle, any effect related to the non-separability of the exciton wave function should be stronger for the excited states.
Figure \[fig:von-Neumann-entropy\]b shows the behaviour of the groud state binding energy, again as a function of the ratio $a/b$, obtained using different methods. The top curve corresponds to the binding energy calculated with the B-spline approximation while, in decreasing order, the figure also shows the curves that correspond to the binding energy calculated with the B-spline method but without including the auto-polarization terms, besides the binding energy calculated using perturbation theory with and without the auto-polarization terms. The binding energy curve obtained using perturbation theory shows the first order approximation energy calculated using the finite potential well electron and hole eigenfunctions as the unperturbed levels.
At least for this set of parameters and materials, taking into account the auto-polarization terms has less influence than using a method (the B-splines) that takes into account the non-separability of the two-particle wave function. For the ground state the worst case scenario, perturbation theory without auto-polarization, differs from the best one, B-splines plus auto-polarization terms, by about five percent. Of course, since the self-assembled quantum dots offer a huge amount of different combinations of materials, geometries and sizes the quantitative results may change more or less broadly.
Figure \[fig:otro-dot\] shows the binding energy obtained with the same methods that were used to obtain the data in Figure \[fig:von-Neumann-entropy\]b), as describe above, but for an exciton in a different quantum dot. In this case, we considered a quantum dot formed by core/well/barrier structure made of ZnS/CdSe/ZnS. All the necessary parameters to determine the Hamiltonian, effective masses, dielectric constants, etc, can be found in Reference [@Schooss(1994)].
![\[fig:otro-dot\] The ground state exciton binding energy [*vs*]{} $a/b$ for a device with core,well and clad made of ZnS/CdSe/ZnS. The curves are obtained following the same prescription than those shown in Figure \[fig:von-Neumann-entropy\]. All the materials parameters can be found in Reference [@Chang(1998)]. It is easy to appreciate that for this device the influence of the auto-polarization terms is smaller that in the first device analyzed, but the difference between the values obtained using perturbation theory and the B-spline method is larger](fig9.eps)
For the case shown in Figure \[fig:otro-dot\], the influence of the auto-polarization terms is even smaller that in the first case analyzed, Figure \[fig:von-Neumann-entropy\], and the difference between the best and worst scenarios defined above is around seven, eight percent.
control {#section:control}
=======
There are two important applications of excitons in which the speed with which you can go from having to not having a exciton is crucial: the controlled and rapid photon production and the switching between the basis states of the qubit logically associated to the exciton. The physical process is exactly the same, but the motivation and requirements depend on which application is being considered. In this section we are interested in estimating how much we can force an excitonic qubit so that it oscillates periodically between its basis states with the highest possible frequency and a small probability loss to other exciton eigenstates. The need to estimate how fast the qubit can be switched between its basis states comes from the limits imposed by the decoherence sources that are unavoidable and restrict seriously the total operation time in which the qubit would keep its quantum coherence. Moreover, we intend to achieve the switching between the basis states with the simplest control pulse, that is, a sinusoidal one. So, many times it is found reasonable to ask that, if $T_S$ is the switching time and $\tau$ the time scale associated to the decoherence processes present in the physical system, then $T_S\sim 10^{-4} \tau$.
The main source of decoherence in charge qubits is owed to the interaction of the charge carrier, the electron(s) trapped in the quantum dot, and the thermal phonon bath present in the semiconductor matrix. This is the reason why, in many cases, spin qubits are preferred although its control is more complicated and have more longer operation time. Since in the exciton case the strength of the coupling with the phonon bath depends on the difference between the electronic and hole wave functions, it is to be expected that the decoherence rate for qubits based on one exciton would be smaller than for a charge qubit based on one (or more) electrons trapped in a quantum dot. Ideally, when the potential wells for the electron and the hole have exactly the same depth, for equal effective masses, and in the limit of zero interaction, the coupling with the phonon bath almost disappears. In this sense, the separability of the electron-hole wave function provides a good measure to select the parameter region where the coupling of the exciton with the phonon bath is smaller. Consequently, since even for very low temperatures the decoherence produced by the phonon bath imposes a total operation time on the order of a few tens of nanoseconds for qubits based on multi-layered self-assembled quantum dots, then to be considered a putative useful qubit the switching time $T_S$ should be on the order of the picoseconds or sub-picoseconds.
The leaking of probability to other exciton eigenstates when an external driving is applied can be analyzed using the following unitary one-exciton Hamiltonian, which describes the interaction of the electron-hole dipolar moment, $\vec{d}$, with an external periodic field $\vec{E}(t)$ applied to the quantum dot [@Haug(2004)] $$\mathcal{H}_{int}(t)=-\vec{d}\ldotp\vec{E}(t) ,$$ which in second quantization formalism can be written as [@Haug(2004); @Biolatti(2002)] $$\label{interaccion}
\mathcal{H}_{int}(t)=-E(t)\sum_{nm} \left[ \mu_{nm}^* a_{n}^{\dagger}
b_{m}^{\dagger} + h.c\right] ,$$ where $a^{\dagger}_n$ is the creation operator of an electron in the conduction band, $b^{\dagger}_m$ is the creation operator of a hole in the valence band, $n$ and $m$ stand for the corresponding one-particle levels, and $\mu^{\star}_{nm}$ is the matrix element of the dipolar moment operator, given by $$\mu_{nm}=\mu_{bulk}\int \phi_{n}^e(\vec{r})\phi_{m}^h(\vec{r})d^3r,$$ where $\mu_{bulk} $ is the dipolar moment corresponding to an electronic transition from the valence band to the conduction band[@Haug(2004); @Biolatti(2002)] .
All the one- and two-particles quantities needed to determine the parameters in Equation \[interaccion\] and the time evolution of the exciton state can be obtained using the B-spline method described in the preceding Sections. Moreover, since the B-spline method provides a very good approximation to all the exciton bound states the time evolution of the approximate exciton quantum state can be written as a sum over bound states as follows $$\Psi(t) = \sum_i U_i(t) \psi_i ,$$ where the sum runs over the approximate bound states provided by the B-spline method, $\psi_i$, and the time-dependent coefficients $U_(t)$ can be calculated integrating a set of complex coupled ordinary differential equations. The number of ordinary differential equations is determined by how many bound states the B-spline method is able to find. In the cases analyzed from now on we considered up to thirty bound states. The numerical integration of the ordinary differential equation was performed using standard Runge-Kutta algorithms.
Before analyzing the behavior of the exciton time-evolution it is worth to remark that the model allows only one exciton but the electron-hole pair can occupy many different exciton eigenstates and not only those associated to the qubit. Also, the model does not allow the “ionization” of the quantum dot nor the spontaneous recombination of the electron-hole pair in the time scale associated to the periodic external driving.
As $|U_0|^2$ and $|U_1|^2$ are the probability that the exciton is in state $|0\rangle$ or in state $|1\rangle$ respectively, we use the [*leakage*]{}, $L$, to characterize the probability loss that experiments the qubit when the driving $E(t) = E_0 \sin(\omega t )$ is applied. The leakage is defined as $$L = \lim_{n \rightarrow \infty} \frac{1}{nT} \int_{t^{\prime}}^{t^{\prime}+nT}
\; \left( 1 - |U_0|^2 - |U_1|^2\right) \, dt \; ,$$ where $T = 2\pi/\omega$.
 The time evolution of the occupation probabilities $|U_0|^2$ (black solid line) and $|U_1|^2$ (red solid line). The exciton is initialized, in all cases, in the fundamental state, so $|U_0(t=0)|^2 = 1$, and $|U_i(t=0)|^2=0$, $\forall i \neq 0$. From top to bottom, panel a) shows the time evolution for $E_0 = 1\times 10^{-3}$ eV/nm, panel b) for $E_0 = 1\times 10^{-2}$eV/nm and panel c) for $E_0 = 5\times
10^{-2}$eV/nm, respectively. ](fig10.eps)
Form now on, we consider a CdS/HgS/CdS structure with $a=b/2$. Figure \[fig:time-evolution\] shows the behavior of the occupation probabilities $|U_0|^2$ and $|U_1|^2$ as a function of time, for different external field strengths $E_0$. In the three cases shown, the frequency of the external driving is set equal to the resonance frequency $\omega_{res}$ of the exciton ground state. The resonance frequency, $\omega_{res} = (E_0 +
E_{g1})/\hbar$, where $E_0$ is the lowest eigenvalue calculated from the exciton Hamiltonian and $E_{g1}$ is the energy gap of material one, [*i.e.*]{} $E_{g1} = E^{gap}_{HgS}$.
From the different panels in Figure \[fig:time-evolution\], it is possible to appreciate that switching times on the order of picoseconds or less are achievable for driving strengths small with no noticeable leakage. This scenario is further supported by the data shown in Figure \[fig:leakage-vs-w\].
 a) The [*leakage*]{} of probability [*vs*]{} the strength of the driving field. The linear dependency of $L$ with $E_0$, when the axis scale is log-log, can be clearly appreciated. From top to bottom, the Figure shows curves obtained for $\omega=0.9
\omega_{res}$ (green line and symbols), $\omega= \omega_{res}$ (black line and symbols) and $\omega=1.1 \omega_{res}$. The leakage for $\omega=0.9
\omega_{res}$ is smaller than for $\omega = \omega_{res}$ because the qubit does not leaves the fundamental state. b) The leakage as a function of the external driving frequency for three different values of $E_0$. Here we use the same external driving strengths that in Figure \[fig:time-evolution\]. From bottom to top, $E_0 = 1\times 10^{-3}$ eV/nm (black line), $E_0 = 1\times 10^{-2}$eV/nm (red line) and $E_0 = 5\times
10^{-2}$eV/nm (green line).](fig11.eps "fig:")  a) The [*leakage*]{} of probability [*vs*]{} the strength of the driving field. The linear dependency of $L$ with $E_0$, when the axis scale is log-log, can be clearly appreciated. From top to bottom, the Figure shows curves obtained for $\omega=0.9
\omega_{res}$ (green line and symbols), $\omega= \omega_{res}$ (black line and symbols) and $\omega=1.1 \omega_{res}$. The leakage for $\omega=0.9
\omega_{res}$ is smaller than for $\omega = \omega_{res}$ because the qubit does not leaves the fundamental state. b) The leakage as a function of the external driving frequency for three different values of $E_0$. Here we use the same external driving strengths that in Figure \[fig:time-evolution\]. From bottom to top, $E_0 = 1\times 10^{-3}$ eV/nm (black line), $E_0 = 1\times 10^{-2}$eV/nm (red line) and $E_0 = 5\times
10^{-2}$eV/nm (green line).](fig12.eps "fig:")
Figure \[fig:leakage-vs-w\]a) shows the [*leakage*]{} as a function of the strength of the external driving $E_0$ for several driving frequencies $\omega$. The data is shown in a $\log-\log$ scale, and under this assumption the leakage shows a linear behavior for a span of a few magnitude orders. The different curves correspond to different values of the driving frequency, but to analyze the dependency of the leakage with the driving frequency we choose to plot it at fixed values of the driving strength. Figure \[fig:leakage-vs-w\] shoes the behavior of the leakage as a function of the external drving frequency and for the three external driving strengths used in Figure \[fig:time-evolution\]. The different curves show a rich structure, and several spikes which are present in all the curves for the same frequencies. These spikes are owed to the presence of many one-exciton levels that are very close to the ground state energy. It is clear that to obtain low levels of leakage, besides an excellent tuning of the driving frequency, it is necessary to have well resolved exciton energy levels or, in other words, the nano-structure should be designed in such way that the exciton is formed in the non-perturbative regime, [i.e]{} when the binding energy is as large as possible. This seems to advice the use of large potential wells, but there is a trade off to consider since the separation between the one-particle levels diminishes when the characteristic sizes of the potential well are increased. A simple way to enhance the interaction between electron-hole pair, at least in principle, is to choose materials whose combination provides a deep potential well and a low potential-well dielectric constant.
Conclusions and Discussion {#section:conclusions}
==========================
One of the advantages of using the B-spline method is its adaptability to tackle problems with step-like parameters and matching conditions at the interface between different spatial regions. Similarly, the method allows to tackle complicated one or two-particle potentials, with a limited number of adaptations. The interaction potential between the electron and the hole, Equation \[eq:electrostatic-potential\] can be treated modifying the method usually employed with the Coulomb potential [@deboor], since both can be written as expansions in spherical harmonics.
As the analysis of the von Neumann entropy shows, for large QD (or small cores in our case) the perturbation theory calculations give a rather poor approximation for the binding energy. We can be fairly sure that our results, since they are variational, that predict smaller ground state energy for the exciton Hamiltonian, Equation \[eq:hamiltoniano-exciton\], than other methods are more accurate than previous results. This implies that our results predict larger values for the exciton binding energy.
To avoid an excessive leakage of probability, it is mandatory to design a quantum dot such that the two lower states of the exciton are well separated from the other one-exciton states. Choosing materials that allow for a larger interaction between the electron and the hole seems to be an apparent solution.
We would like to acknowledge SECYT-UNC, and CONICET for partial financial support of this project. We acknowledge the fruitful discussions with Dr. César Proetto in the early stages of this work.
[40]{}
C. Delerue and M. Lannoo, [*Nanostructures. Theory and Modelling*]{} (Springer, Berlin, 2004)
J. M. Ferreyra, and C. R. Proetto, Phys. Rev. B, [**57**]{}, 9061 (1998).
M. Cardona, and Y. Y. Peter [*Fundamentals of semiconductors*]{} (Springer-Verlag, Berlin Heidelberg, 2005)
Bastard, G. (1988). [*Wave mechanics applied to semiconductors. Les editions de Physique*]{} (CNRS, Paris, 1988).
F. Troiani, U. Hohenester, and E. Molinari Phys. Rev. B [**62**]{}, R2263(R) (2000).
E. Biolatti, R. C. Iotti, P. Zanardi, and F. Rossi, Phys. Rev. Lett. [**85**]{}, 5647 (2000).
L.C. L. Y. Voon and M. Willatzen, [*The kp method: electronic properties of semiconductors.*]{} (Springer-Verlag, Berlin Heidelberg, 2009)
H. Haug, H., and S. W. Koch, [*Quantum theory of the optical and electronic properties of semiconductors.*]{} [*Fourth Edition*]{} (World Scientific Publishing Co. Pte. Ltd., Singapore, 2004)
U. Woggon and S. V. Gaponenko, Physica Status Solidi (b) [**189**]{}, 285-343 (1995).
K. Chang and J. B. Xia, Phys. Rev. B [**57**]{}, 9780 (1998)
D. Schooss, A. Mews, A. Eychmüller, and H. Weller, Phys. Rev. B, [**49**]{}, 17072 (1994).
E. P. Pokatilov, V. A. Fonoberov, V. M. Fomin, and J. T. Devreese, Phys. Rev. B [**64**]{}, 245329 (2001).
A. L. Efros and M. Rosen, Phys. Rev. B [**58**]{}, 7120 (1998).
T. Garm, T, J. Phys.: Condens. Matter [**8**]{}, 5725 (1996).
Bimberg, D., Kirstaedter, N., Ledentsov, N. N., Alferov, Z. I., Kop’ev, P. S., and Ustinov, V. M. (1997). IEEE Journal of, 3(2), 196-205.
D. Bimberg, Journal of Physics D: Applied Physics [**38**]{}, 2055 (2005). H. Kamada and H. Gotoh, Semiconductor science and technology, [**19**]{}, S392 (2004).
P. Chen, C. Piermarocchi, and L. J. Sham, Phys. Rev. Lett, [**87**]{}, 067401 (2001).
E. Biolatti, I. D’Amico, P. Zanardi, and F. Rossi, Phys. Rev. B, [**65**]{}, 075306 (2002).
K. Ishibashi, M. Suzuki, D. Tsuya, and Y. Aoyagi, Microelectronic engineering [**67**]{}, 749-754 (2003).
T. Calarco,A. Datta, P. Fedichev, E. Pazy, and P. Zoller, Phys. Rev. A [**68**]{}, 012310 (2003).
N. H. Bonadeo, G. Chen, D. Gammon, D. S. Katzer, D. Park, and D. G. Steel, Phys. Rev. Lett. [**81**]{}, 2759 (1998).
N. H. Bonadeo, J. Erland, D. Gammon, D. Park, D. S. Katzer, and D. G. Steel, Science [**282**]{}, 1473-1476 (1998).
P. Lelong and G. Bastard, Solid state communications [**98**]{}, 819-823 (1996).
B. Billaud, M. Picco, and T. T. Truong, J. Phys.: Condens. Matter [**21**]{}, 395302 (2009).
C. de Boor, [*A Practical Guide to Splines*]{} (Springer, New York, 2001)
H. Bachau, E. Cormier, P. Decleva, J. E. Hansen, and F. Martin, Rep. Prog. Phys. [**64**]{}, 1815 (2001).
A. Ferrón, P. Serra and O. Osenda, J, Appl. Phys. [**113**]{}, 134304 (2013)
HaoXue, Q. I. A. O., TingYun, S. H. I., and Baiwen, L. I. Commun. Theor. Phys. [**37**]{},221 (2002)
| {
"pile_set_name": "ArXiv"
} | ArXiv |
---
abstract: 'Local descriptors based on the image noise residual have proven extremely effective for a number of forensic applications, like forgery detection and localization. Nonetheless, motivated by promising results in computer vision, the focus of the research community is now shifting on deep learning. In this paper we show that a class of residual-based descriptors can be actually regarded as a simple constrained convolutional neural network (CNN). Then, by relaxing the constraints, and fine-tuning the net on a relatively small training set, we obtain a significant performance improvement with respect to the conventional detector.'
author:
-
bibliography:
- 'forensic.bib'
title: |
Recasting Residual-based Local Descriptors\
as Convolutional Neural Networks:\
an Application to Image Forgery Detection
---
Introduction {#sec:intro}
============
Images and videos represent by now the dominant source of traffic on the Internet and the bulk of data stored on social media. Nowadays, however, such data may be easily manipulated by malicious attackers to convey some twisted and potentially dangerous messages. Interested areas include politics, journalism, judiciary, even the scientific world. For this reason, in the last few years there has been intense and ever growing activity on multimedia forensics, aiming at developing methods to detect, localize, and classify possible image manipulations.
Typical attacks consist in adding or deleting objects, using material taken from the same image (copy-move) or from other sources (splicing). As a consequence, many researchers have focused on detecting near-duplicates in the image or across a repository of images. More fundamentally, one may be interested in establishing whether the image under analysis is pristine or else has been subjected to some post-processing, since any form of manipulation may raise suspects and suggest deeper inquiry. In fact, most of the times, copy-moves and splicing are accompanied by various forms of elaboration aimed at removing the most obvious traces of editing. These include, for example, resizing, rotation, linear and non-linear filtering, contrast enhancement, histogram equalization and, eventually, re-compression.
A number of papers have been proposed to detect one or the other of such elaborations [@Popescu2005; @Kirchner2008; @Kirchner2010; @Huang2010; @Stamm2010]. These methods, however, are sensitive to just some specific manipulations. A more appealing line of research is to detect [*all*]{} possible manipulations, an approach that has been followed in several papers [@Cao2012; @Verdoliva2014; @Fan2015; @Li2016]. Notably, in the 2013 IEEE Image Forensics Challenge, the most effective techniques for both image forgery detection [@Cozzolino2014a] and localization [@Cozzolino2014b] used this approach, relying on powerful residual-based local descriptors. These features, such as SPAM (subtractive pixel adjacency matrix) [@Pevny2010] or SRM (spatial rich models) [@Fridrich2012], inspired to previous work in steganalysis, are extracted from the so-called residual image. In fact, the noise residual, extracted through some high-pass filtering of the image, contains a wealth of information on the in-camera and out-camera processes involved in the image formation. Such subtle traces, hardly visible without enhancement, may reveal anomalies due to object insertion [@Cozzolino2015b; @Cozzolino2016] or can detect different types of image editing operations [@Verdoliva2014; @Li2016; @Boroumand2017].
Very recently, inspired by impressive results in the closely related fields of computer vision and pattern recognition [@Krizhevsky2012], the multimedia forensics community began focusing on the use of deep learning [@Bengio2013; @LeCun2015], especially convolutional neural networks (CNN) [@Zeiler2014]. Taking advantage of the lesson learnt from SPAM/SRM features, constrained CNN architectures have been proposed both for steganalysis [@Qian2015] and manipulation detection [@Bayar2016], where the first convolutional layer is forced to perform a high-pass filtering. In this paper we show that there is no real contraposition between residual-based features and CNNs. Indeed, these local features can be computed through a CNN with architecture and parameters selected so as to guarantee a perfect equivalence. Once established this result, we go beyond emulation, removing constraints on parameters, and fine-tuning the net to further improve performance. Since the resulting network has a lightweight structure, fine-tuning can be carried out by means of a limited training set, limiting computation time and memory usage. A significant performance gain with respect to the conventional feature is observed, especially in the most challenging situations.
In the following we describe in more detail the residual-based local features (Section 2), recast them as a constrained CNN, to be further trained after removing constraints (Section 3), show experimental results (Section 4), and draw conclusions (Section 5).
Residual-based local descriptors
================================
Establishing which type of processing an image has undergone calls for the ability to detect the subtle traces left by these operations, typically in the form of recurrent micropatterns. This problem has close ties with steganalysis, where weak messages hidden in the data are sought, so it is no surprise that the same tools, residual-based local descriptors, prove successful in both cases. To associate a residual-based feature to an image, or an image block, the following processing chain has been successfully used in steganalysis [@Pevny2010; @Fridrich2012]:
1. extraction of noise residuals
2. scalar quantization
3. computation of co-occurrences
4. computation of histogram
In the following, we describe in some more depth all these steps, taking a specific model out of the 39 proposed in [@Fridrich2012] as running example.
[**Extraction of noise residual.**]{} The goal is to extract image details, in the high-frequency part of the image, which enables the analysis of expressive micropatterns. As the name suggests, this step can be implemented by resorting to a high-pass filter. In [@Fridrich2012] a number of different high-pass filters have been considered, both linear and nonlinear, with various supports. Here, as an example, we focus on a single 4-tap mono-dimensional linear filter, with coefficients ${\bf w} = [1, -3, 3, -1]$. The filter extracts image details along one direction, but is applied also on the image transpose (assuming vertical/horizontal invariance) to augment the available data. Choosing a single filter rather than considering all models proposed in [@Fridrich2012] is motivated not only by the reduced complexity but also by the very good performance observed in the context of image forgery detection [@Cozzolino2014a; @Verdoliva2014; @Li2016].
[**Scalar quantization.**]{} Residuals are conceptually real-valued quantities or, in any case, high-resolution integers, so they must be quantized to reduce cardinality and allow easy processing. In [@Fridrich2012] a uniform quantization is used with an odd number of levels (to ensure that 0 is among the possible outputs). Therefore, the only parameters to set are the number of quantization levels, $L$, and the quantization step $\Delta$. In our example we set $L=3$ and $\Delta=4.5$. [**Co-occurrences.**]{} The computation of co-occurrences is the core step of the procedure. In fact, this is a low-complexity means for taking into account high-order dependencies among residuals and hence gather information on recurrent micropatterns. Following [@Fridrich2012] we compute co-occurrences on $N=4$ pixels in a row, both along and across the filter direction. With these values, two co-occurrence matrices with $3^4=81$ entries are obtained. Of course, the image or block under analysis must be large enough to obtain meaningful estimates. All the co-occurrence $N$-dimensional bins are eventually coded as integers. [**Feature formation.**]{} Counting co-occurrences one obtains the final feature vector describing the image. Neglecting symmetries, our final feature has length equal to 162. The final classification phase is performed by a linear SVM classifier.
Recasting local features as CNN
===============================
We will now show that local residual-based features can be extracted by means of a convolutional neural network. Establishing this equivalence leaves us with a CNN architecture and a set of parameters that are already known to provide an excellent performance for the problem of interest. Then, given this good starting point, we can move a step forward and fine-tune the network through a sensible training phase with labeled data. Note that in this way we will carry out a joint optimization of both the feature extraction process and classification. In the following we will first move from local features to a Bag-of-Words (BoW) paradigm, and then proceed to the implementation by means of Convolutional Neural Networks.
![Basic processing scheme (0) for extracting the single model SRM feature, and equivalent scheme (1) with inverted order of scalar quantization (SQ) and $n$-pixel shifting ($z^{-n}$).[]{data-label="schemaFridrichLD"}](figure/schema00to01.pdf){width="1.0\linewidth"}
From local features to Bag-of-Words
-----------------------------------
In Fig.1 (top) we show the basic processing scheme used to extract the single model SRM feature. Let ${\bf X}$ be the input image[^1], ${\bf R}$ the residual image, and ${\widehat{\bf R}}$ the quantized residual image. To compute the output feature, the input image is high-pass filtered, then the residual image is quantized, and $N$ versions of it are generated, shifted one pixel apart from one another. For each pixel $s$, the values ${\widehat{r}}_{1,s},\ldots,{\widehat{r}}_{N,s}$ are regarded as base-$L$ digits and encoded as a single scalar $i^*_s$, finally, the histogram ${\bf h}$ of this latter image is computed. The scheme at the bottom of Fig.1 is identical to the former except for the inverted order of scalar quantization (SQ) and shifting. This inversion, however, allows us to focus on the two groups of blocks highlighted at the top of Fig.2.
The filter-shifter group can be replaced by a bank of $N$ filters, all identical to one another except for the position of the non-zero weights. So, with reference to our running example, the $n$-th filter will have non-zero weights, \[1, -3, 3, -1\], only on the $n$-th row, and zero weights everywhere else. Turning to the second group, the combination of $N$ scalar quantizers can be regarded as a constrained form of vector quantization (VQ). More specifically, it is a product VQ, since the VQ codebook is obtained as the cartesian product of the $N$ SQ codebooks. On one hand, product quantization is much simpler and faster than VQ. On the other hand, its strong constraints are potentially detrimental for performance. Its $K=L^N$ codewords are forced to lie on a truncated $N$-dimensional square lattice [@Gray1998] and cannot adapt to the data distribution. Many of them will be wasted in empty regions of the feature space, causing a sure, and often severe, loss of performance with respect to unconstrained VQ.
However, the most interesting observation about the new structure at the bottom of Fig.2 is that it implements the Bag-of-Words (or also Bag-of-Features) paradigm. The filter bank extracts a feature vector for each image pixel, based on its neighborhood. These feature are then associated, through VQ, with some template features. Finally, the frequency of occurrence of the latter, computed in the last block, provides a synthetic descriptor of the input image. The fact that filters and vector quantizer are largely sub-optimal impacts only on performance, not on interpretation. Needless to say, they could be both improved through supervised training.
![The cascade of filter and shifters of scheme (1) can be replaced by a bank of filters, while the bank of independent SQ’s + coder can be replaced by product VQ. The resulting scheme (2) fits the Bag-of-Words paradigm.[]{data-label="schemaFridrichLD"}](figure/schema01to02){width="1.0\linewidth"}
![A vector quantizer (left) can be implemented through a bank of filter followed by argmax (right).[]{data-label="schemaLD"}](figure/schemaVQ2){width="0.9\linewidth"}
From Bag-of-Words to CNN
------------------------
We now show that the processing steps of Fig.2 can be all implemented through a CNN. First of all, the bank of linear filters used to extract noise residuals can be replaced by a pure convolutional layer, with neurons computing the residuals as $$r_{n,s} = f(w_{n,s} * x_s + b_n) \hspace{5mm} n=1,\cdots,N$$ with $s$ used for image spatial location and $n$ to identify neurons. The neuron weights coincide with filter coefficients, biases $b_n$ are all set to zero, and the non-linearity $f(\cdot)$ is set to identity. As for the vector quantizer, assuming the usual minimum distance hard-decision rule, it can be implemented by means of a convolutional layer followed by an hard-max layer. Let ${{\bf r}}_s$ be a vector formed by collecting a group of residuals at site $s$, and ${{\bf c}}_k$ the $k$-th codeword of the quantizer. Their squared Euclidean distance, $d^2_{k,s}$, can be expanded as $$\begin{aligned}
||{{\bf r}}_s-{{\bf c}}_k||^2 & = & ||{{\bf r}}_s||^2 + ||{{\bf c}}_k||^2 - 2<{{\bf r}}_s,{{\bf c}}_k> \nonumber\\
& = & ||{{\bf r}}_s||^2 - 2 \epsilon_k - 2<{{\bf r}}_s,{{\bf c}}_k>\end{aligned}$$ with $||\cdot||^2$ and $<\cdot,\cdot>$ indicating norm and inner product, respectively. Hence, neglecting the irrelevant $||{{\bf r}}_s||^2$ term: $$\begin{aligned}
i^*_s & = & {\rm argmin}_{k=1,\cdots,K} \,\, d_{k,s} \nonumber\\
& = & {\rm argmax}_{k=1,\cdots,K} \,\,(<{{\bf r}}_s,{{\bf c}}_k> + \epsilon_k) \nonumber\\
& = & {\rm argmax}_{k=1,\cdots,K} \,\, m_{k,s}\end{aligned}$$ with $m_{k,s}$ interpreted as a matching score between the feature vector at site $s$ and the $k$-th codeword.
![The whole scheme (2) can be converted in the CNN (3). The filter bank is replaced by a convolutional layer, VQ is replaced by convolutional-hardmax layers, the histogram can be computed through an average pooling layer.[]{data-label="schemaLD"}](figure/schema02to03){width="1\linewidth"}
This equivalence is depicted in Fig.3. The matching scores $m_{k,s}$ are computed through a convolutional layer, equipped with $K=L^N$ filters (remember that $L$ is the number of quantization levels), one for each codeword, having weights ${{\bf c}}_k$, bias $\bm{\epsilon}_k$ and, again, an identity as activation function. The best matching codeword is then selected through a hardmax processing.
The evolution of the whole network is shown in Fig.4. As already said, the first filter bank is replaced by a convolutional layer. Then, the VQ is replaced by another convolutional layer followed by a hardmax layer. The former outputs $K$ feature maps, ${\bf M}_k$, with the matching scores. The latter outputs $K$ more binary maps, ${\bf P}_k$, where $p_{k,s}=1$ when the corresponding matching score $m_{k,s}$ is maximum over $k$, and 0 otherwise. Finally, the histogram computation is replaced by an average pooling layer operating on the whole feature maps, that is, $h_k = \sum_s p_{k,s}$. The resulting scheme is shown at the bottom of Fig.4.
![The constrained CNN of scheme (3) extracts the features which feed an external classifier. In scheme (4) this is replaced by an internal fully connected layer, and all constraints are removed. By fine-tuning on training data, all layers can be optimized jointly. The dashed lines are to remind that this net provides only half the feature, a twin net (not shown for clarity) provides the other half.[]{data-label="schemaCnnLD"}](figure/schema03to04){width="1\linewidth"}
This net, actually, computes only half the desired feature, the part based on across-filter co-occurrences. The twin half is computed similarly, and the complete feature is eventually fed to the SVM classifier, as shown in Fig.5 (top). In this network, weights and biases of the convolutional layers are all hard-wired to reproduce exactly the behavior of the residual-based local feature described in Section 2, thereby ensuring the good performance observed in the literature. We now proceed to remove all constraints and allow the net to learn on a suitable training set. First of all, the classifier itself can be implemented as part of the CNN architecture by including a fully connected layer at the end, obtaining the architecture of Fig.5 (bottom). Now, to exploit the full potential of deep learning, all parameters must be optimized by appropriate training, thereby overcoming all the impairing constraints mentioned before. Note that the learning phase allows us not only to optimize all layers, which could be done also in the BoW framework, but to optimize them [*jointly*]{}, taking full advantage of the CNN structural freedom. Moreover, the lightweight architecture of the network is instrumental to achieve good results even in the presence of a limited training set.
ayer with a sft-max layer that approximates it, so as to avoid non-differentiable operators.
However, before proceeding with the training, it is necessary to replace the hard-max layer, with a soft-max layer that approximates it, so as to avoid non-differentiable operators. Given the input vector $\{m_{k,s}, k=1,\ldots,K\}$, the soft-max computes the quantities $$p_{k,s} = e^{\alpha m_{k,s}}/\sum_l e^{\alpha m_{l,s}}$$ With the aim to preserve a close correspondence with the original descriptor, we should choose a very large $\alpha$ parameter, so as to obtain a steep nonlinearity. However, as said before, this is not really necessary, since our goal is only to improve performance. Hence, we select a relatively small value for $\alpha$ in order not to slow down learning. Likewise, to implement minimum distance VQ exactly, the biases in the second convolutional layer should depend on the filter weights, but there is no practical reason to enforce this constraint, and we allow also the biases to adapt freely. Now, the final CNN can be trained as usual with stochastic gradient descent [@Krizhevsky2012] to adapt to the desired task. It should be clear that a number of architectural modifications could be also tested starting from this basic structure, but this goes beyond the scope of the present paper, and will be the object of future research. We must underline that the equivalence between CNN and BoW has been noticed before in the literature, for example in [@Lan2015; @Arandjelovic2016; @Richard2016].
Experimental analysis
=====================
[|l|l|]{}
------------------------------------------------------------------------
Manipulation & Parameters\
------------------------------------------------------------------------
Median Filtering & kernel: 7$\times$7, 5$\times$5, 3$\times$3\
------------------------------------------------------------------------
Gaussian Blurring & st. dev: 1.1, 0.75, 0.5\
------------------------------------------------------------------------
Additive Noise (AWGN) & st. dev.: 2.0, 0.5, 0.25\
------------------------------------------------------------------------
Resizing & scale: 1.5, 1.125, 1.01\
------------------------------------------------------------------------
JPEG Compression & quality factor: 70, 80, 90\
\[tab:type\]
\[tab:perf\]
[|c|c||c|c|c||c|c|c|]{} & &\
& Bayar2016 & SRM+SVM & prop. CNN & Bayar2016 & SRM+SVM & prop. CNN\
& 60 epochs & & 15 epochs & 60 epochs & & 15 epochs\
&
------------------------------------------------------------------------
7x7 & 98.23 & 99.61 & 99.07 & 99.69 & 99.68 & 99.55\
Median Filtering &
------------------------------------------------------------------------
5x5 & 96.66 & 99.67 & 99.47 & 99.78 & 99.68 & 99.60\
&
------------------------------------------------------------------------
3x3 & 94.56 & 99.83 & 99.35 & 99.80 & 99.87 & 99.75\
&
------------------------------------------------------------------------
1.1 & 99.65 & 99.93 & 99.79 & 99.98 & 99.97 & 99.95\
Gaussian Blurring &
------------------------------------------------------------------------
0.75 & 98.52 & 99.90 & 99.82 & 99.94 & 99.77 & 99.93\
&
------------------------------------------------------------------------
0.5 & 83.10 & 87.10 & 95.70 & 94.57 & 87.55 & 96.56\
&
------------------------------------------------------------------------
2.0 & 97.08 & 99.94 & 99.95 & 99.56 & 99.94 & 99.94\
Additive Noise &
------------------------------------------------------------------------
0.5 & 82.93 & 99.37 & 99.36 & 93.83 & 99.34 & 99.66\
&
------------------------------------------------------------------------
0.25 & 51.83 & 85.06 & 88.81 & 80.28 & 84.01 & 90.79\
&
------------------------------------------------------------------------
1.5 & 99.22 & 99.99 & 100.00 & 99.72 & 99.87 & 100.00\
Resizing &
------------------------------------------------------------------------
1.125 & 91.06 & 98.94 & 99.56 & 97.02 & 96.00 & 99.78\
&
------------------------------------------------------------------------
1.01 & 80.51 & 96.01 & 97.81 & 98.44 & 95.11 & 97.20\
&
------------------------------------------------------------------------
70 & 96.04 & 99.99 & 99.99 & 99.43 & 99.99 & 99.94\
JPEG Compression &
------------------------------------------------------------------------
80 & 77.01 & 99.73 & 99.37 & 98.12 & 99.94 & 99.86\
&
------------------------------------------------------------------------
90 & 63.77 & 90.86 & 92.08 & 79.69 & 90.90 & 94.59\
To test the performance of the proposed CNN architecture we carry out a number of experiments with typical manipulations. Our synthetic dataset includes images taken from 9 devices, 4 smartphones (Apple iPhone 4S, Apple iPhone 5s, Huawei P7 mini, Nokia Lumia 925) and 5 cameras (Canon EOS 450D, Canon IXUS 95 IS, Sony DSC-S780, Samsung Digimax 301, Nikon Coolpix S5100). Each device contributes 200 images, and from each image non-overlapping patches of dimension $128\times 128$ are sampled. We select at random 6 devices to form the training set, while the remaining 3 are used as a testing set. Therefore, the patches used for testing come from devices that are never seen in the training phase. For each pristine patch, the corresponding manipulated patch is also included in the set. Overall, our training set comprises a total of just 10800+10800 patches, quite a small number for deep learning applications. We consider 5 types of image manipulation: median filtering, gaussian blurring, AWGN noise addition, resizing, and JPEG compression, with three different settings for each case (see Tab.1) corresponding to increasingly challenging tasks. For example, JPEG compression with quality factor Q=70 is always easily detected, while a quality factor Q=90 makes things much more difficult.
In the proposed CNN the first convolutional layer includes 4 filters of size $5\times5\times1$ operating on the monochrome input (we use only the green band normalized in $[0,1]$. In the second layer there are 81 filters of size $1\times1\times4$. Filters are initialized as described in Section 3 and the $\alpha$ parameter of soft-max is set to $2^{16}$. The code is implemented in Tensorfow and runs on a Nvidia Tesla P100 with 16GB RAM. We set the learning rate to $10^{-6}$, with decay $5 \cdot 10^{-4}$, batch size 36 and Adam [@Kingma2015] optimization method, using the cross-entropy loss function. Together with the proposed CNN we consider also the basic solution, with the handcrafted feature followed by linear SVM, and the CNN proposed in Bayar2016 [@Bayar2016] based on the use of a preliminary high-pass convolutional layer.
Results in terms of probability of correct decision for each binary classification problem are reported in the left part of Tab.2 (small training set). With “easy” manipulations, [*e.g.*]{} JPEG@70, all methods provide near-perfect results and there is no point in replacing the SRM+SVM solution with something else. In the presence of more challenging attacks, however, the performance varies significantly across methods. After just 15 epochs of fine tuning, the proposed CNN improves over SRM+SVM of about 2 percent points for JPEG compression, resizing, and noising, and more than 8 points for blurring, while median filtering is almost always detected in any case. In the same cases, the CNN architecture proposed in [@Bayar2016] provides worse results, sometimes close to 50%, even after 60 epochs of training. Our conjecture is that a deep CNN is simply not able to adapt correctly with a small training set. In this condition a good hand-crafted feature can work much better. The proposed CNN builds upon this result and takes advantage of the available limited training data to fine-tune its parameters.
To carry out a fair comparison we also considered a case in which a much larger training set is available, comprising 460800 patches, that is more than 20 times larger than before. Results are reported in the right part of Tab.2. As expected the performance does not change much for the SRM+SVM solution, since the SVM needs limited training anyway. For the proposed CNN, some improvements are observed for the more challenging tasks. As an example, for JPEG@90 the accuracy grows from 92.08 to 94.59. Much larger improvements are observed for the network proposed in [@Bayar2016], which closes almost always the performance gap and sometimes outperforms slightly the proposed CNN. Nonetheless in a few challenging cases, like the already mentioned JPEG@90 or the addition of low-power white noise, there is still a difference of more than 10 percent point with our proposal. It is also interesting to compare the two adjacent columns, proposed CNN at 15 epochs and the CNN architecture proposed in [@Bayar2016] at 60 epochs, which speak clearly in favor of the first solution, in terms of both complexity and performance.
[cccc]{} {width="0.22\linewidth"} {width="0.22\linewidth"} {width="0.22\linewidth"} {width="0.22\linewidth"}\
\
{width="0.22\linewidth"} {width="0.22\linewidth"} {width="0.22\linewidth"} {width="0.22\linewidth"}
The ability to reliably classify small patches may be very valuable in the presence of spatially localized attacks. This is the case of image copy-move or splicing, where only a small part of the image is tampered with. In these cases, a descriptor computed on small patches can more reliably detect manipulations, and even localize the forgery by working in sliding-window modality. Fig.6 shows two examples of forgery localization with a slightly blurred splicing and a resized copy-move, respectively. In both cases using the proposed CNN in sliding-window modality, a sharp heat map is obtained, allowing for a precise localization of the forgery. The SRM+SVM solution also provides good results, but the heat map is more fuzzy, with a higher risk of false alarms. It is worth underlining again that the images used for these tests come from cameras that did not contribute to the training set.
Conclusions
===========
Residual-based descriptors have proven extremely effective for a number of image forensic applications. Improving upon the current state of the art, however, is slow and costly, since the design of better hand-crafted features is not trivial. We showed that a class of residual-based features can be regarded as compact constrained CNNs. This represent a precious starting point to exploit the huge potential of deep learning, as testified by the promising early results. However, this is only a first step, and there is much room for improvements, especially through new architectural solutions. This will be the main focus of future work.
[^1]: We use capital boldface for images, lowercase boldface for vectors, and simple lowercase for scalars. The value of image ${\bf X}$ at spatial site $s$, will be denoted by $x_s$.
| {
"pile_set_name": "ArXiv"
} | ArXiv |
@addtoreset
.3in
[**Double and cyclic $\lambda$-deformations and their canonical equivalents**]{}
0.35in
[**George Georgiou,$^1$ Konstantinos Sfetsos**]{}$^2$ and [**Konstantinos Siampos**]{}$^3$ 0.1in
0.1in [*${}^1$Institute of Nuclear and Particle Physics,\
National Center for Scientific Research Demokritos,\
Ag. Paraskevi, GR-15310 Athens, Greece* ]{} 0.1in
[*${}^2$Department of Nuclear and Particle Physics,\
Faculty of Physics, National and Kapodistrian University of Athens,\
Athens 15784, Greece\
*]{} 0.1in
[*${}^3$Albert Einstein Center for Fundamental Physics,\
Institute for Theoretical Physics / Laboratory for High-Energy Physics,\
University of Bern, Sidlerstrasse 5, CH3012 Bern, Switzerland* ]{}
0.1in
[`g`[email protected], [email protected], [email protected]]{}
0.1in
**Abstract**
We prove that the doubly $\lambda$-deformed $\sigma$-models, which include integrable cases, are canonically equivalent to the sum of two single $\lambda$-deformed models. This explains the equality of the exact $\beta$-functions and current anomalous dimensions of the doubly $\lambda$-deformed $\sigma$-models to those of two single $\lambda$-deformed models. Our proof is based upon agreement of their Hamiltonian densities and of their canonical structure. Subsequently, we show that it is possible to take a well defined non-Abelian type limit of the doubly-deformed action. Last, but not least, by extending the above, we construct multi-matrix integrable deformations of an arbitrary number of WZW models.
20 pt
Introduction and results {#introduction-and-results .unnumbered}
========================
A new class of integrable theories based on current algebras for a semi-simple group was recently constructed [@Georgiou:2016urf]. The starting point was to consider two independent WZW models at the same positive integer level $k$ and two distinct PCM models which were then left-right asymmetrically gauged with respect to a common global symmetry. The models are labeled by the level $k$ and two general invertible matrices $\lambda_{1,2}$. For certain choices of $\lambda_{1,2}$ integrability is retained [@Georgiou:2016urf]. This idea can be generalized to include integrable deformations of exact CFTs on symmetric spaces. This construction is reminiscent to the one for single $\lambda$-deformations [@Sfetsos:2013wia; @Hollowood:2014rla; @Hollowood:2014qma].
Subsequently, the quantum properties of the aforementioned multi-parameter integrable deformations were studied in [@Georgiou:2017aei], by employing a variety of techniques. One of the main results of that work was that the running of the couplings $\lambda_1$ and $\lambda_2$, as well as the anomalous dimensions of current operators depend only on one of the couplings, either $\lambda_1$ or $\lambda_2$ and are identical to those found for single $\lambda$-deformations [@Kutasov:1989dt; @Gerganov:2000mt; @Itsios:2014lca; @Sfetsos:2014jfa; @Georgiou:2015nka; @Georgiou:2016iom; @Georgiou:2016zyo]. These rather unexpected results seek for a simple explanation. The purpose of this work is to demonstrate that they are due to the fact that the doubly deformed models are canonically equivalent to the sum of two single $\lambda$-deformations, one with deformation matrix being $\lambda_1$ and the other with deformation matrix $\lambda_2$. Recall that all known forms of T-duality, i.e., Abelian, non-Abelian and Poisson–Lie T-duality can be formulated as canonical transformations in the phase space of the corresponding two-dimensional $\sigma$-models [@Curtright:1994be; @Alvarez:1994wj; @Lozano:1995jx; @Sfetsos:1996xj; @Sfetsos:1997pi]. Moreover, it has been shown in various works that the running of couplings is preserved under these canonical transformations even though the corresponding $\sigma$-models fields are totally different [@Balazs:1997be; @Balog:1998br; @Sfetsos:1998kr; @Sfetsos:1999zm; @Sfetsos:2009dj]. All of the above strongly hint towards the validity of our assertion, which of course we will prove.
The plan of the paper is as follows: In section \[canonicalsection\], after a brief review of the single and doubly $\lambda$-deformed models and of their [*non-perturbative*]{} symmetries, we will show that the doubly deformed models are canonically equivalent to the sum of two single $\lambda$-deformations. In section \[newnonabeliansection\], we will present the type of non-Abelian T-duality that is based on the doubly deformed $\sigma$-models of [@Georgiou:2016urf]. Finally, in section \[cycliclambdasection\], we will construct multi-matrix [*integrable*]{} deformations of an arbitrary number of independent WZW models by performing a left-right asymmetric gauging for each one of them but in such a way that the total classical gauge anomaly vanishes. This happens if these models are forced to obey the cyclic symmetry property or if they are infinitely many, resembling in structure either a closed or an infinitely open spin chain. Their action can be thought of as the all-loop effective action of several independent WZW models for $G$ all at level $k$, perturbed by current bilinears mixing the different WZW models with nearest neighbour-type interactions. These models are also canonically equivalent to a sum of single $\lambda$-deformed models with appropriate couplings. Furthermore, we will argue that the Hamiltonian of these new models maps to itself under an inversion of all couplings $\lambda_i \mapsto \lambda_i^{-1}, \,\,i=1,...,n$ accompanied generically by non-local redefinitions of the group elements involved when $n=3,4,\dots$. This symmetry, which in the special cases where $n=1,2$ simplifies to the one reviewed in section \[canonicalsection\], is in accordance with the fact that the $\b$-functions and anomalous dimensions of currents are again given by the same expressions as in the case of the single $\lambda$-deformed model.
Review and canonical equivalence {#canonicalsection}
================================
Single $\lambda$-deformed $\s$-models {#singlelambdalkslsl}
-------------------------------------
The construction of the single $\lambda$-deformed $\sigma$-model starts by considering the sum of a gauged WZW and a PCM for a group $G$, defined with group elements $g$ and $\tilde g$, respectively and next gauging the global symmetry [@Sfetsos:2013wia] $$g\mapsto\Lambda^{-1}g\Lambda\,,\quad \tilde g\mapsto\Lambda^{-1}\tilde g\ .$$ This is done by introducing gauge fields $A_\pm$ in the Lie-algebra of $G$ transforming as $$A_\pm\mapsto\Lambda^{-1}A_\pm\Lambda-\partial_\pm\Lambda\,.$$ The choice $\tilde g=\mathbb{I}$ completely fixes the gauge and the gauged fixed action reads \[kdldjkdfkj\]
& S\_[k,]{}(g\_;A\_) =S\_k(g) + [k]{} \^2 ( A\_- \_+ g g\^[-1]{} - A\_+ g\^[-1]{} \_- g\
& + A\_- g A\_+ g\^[-1]{} -A\_+ \^[-1]{} A\_-) ,
where $S_k(g)$ is the WZW model. The $A_{\pm}$’s are non-dynamical and their equations of motion read \[kshdg1\] \_+ g g\^[-1]{} = (ł\^[-T]{}-) A\_+ ,g\^[-1]{} \_- g = - (ł\^[-1]{}-) A\_-, with $\nabla_\pm g=\partial_\pm g-[A_\pm,g]$. Solving them in terms of the gauge fields we find $$\label{sklldldd}
A_+=i\left(\l^{-T}-D\right)^{-1}J_+\,,\quad A_-=-i\left(\l^{-1}-D^T\right)^{-1}J_-\,,$$ where \[hg3\] J\^a\_+ = - i (t\_a \_+ g g\^[-1]{}) ,J\^a\_- = - i (t\_a g\^[-1]{}\_- g ) . D\_[ab]{}= (t\_a g t\_b g\^[-1]{}) , where $t_a$’s are Hermitian representation matrices obeying $[t_a,t_b]=if_{abc}t_c$, so that the structure constants $f_{abc}$ are real. We choose the normalization such that $\Tr(t_at_b)=\d_{ab}$.
Using into one finds the action [@Sfetsos:2013wia] S\_[k,ł]{}(g)= S\_k(g)+ [k]{} \^2[Tr]{}(J\_+(ł\^[-1]{}- D\^T)\^[-1]{} J\_-) . \[laact\] For small elements of the matrix $\lambda$ this action becomes $$S_{k,\l}(g) = S_k(g) + {k\ov \pi}
\int \text{d}^2\s\, {\rm Tr}\left(J_{+}\l J_{-}\right) + \cdots \ .$$ Hence represents the effective action of self-interacting current bilinears of a single WZW model. The action has the remarkable [*non-perturbative*]{} symmetry [@Itsios:2014lca; @Sfetsos:2014jfa] k-k ,łł\^[-1]{} , gg\^[-1]{} . \[duallli\] As in the case of gauged WZW models [@Bowcock:1988xr], we define the currents $\mathcal{J}_\pm$
\[singleAffine\] \_[+]{} =\_+ g g\^[-1]{} + A\_[+]{} - A\_[-]{} ,\_- = - g\^[-1]{} \_- g+ A\_[-]{} - A\_[+]{} ,
The above form for the $\mathcal{J}_{\pm}^a$’s when rewritten in terms of phase space variables of the $\sigma$-model action, assumes the same form as the currents $J_\pm^a$ of the WZW action. Hence, they satisfy two commuting current algebras as in [@Witten:1983ar] \[singleBowcock\] {\_\^a , \_\^b} = f\_[abc]{} \_\^c \_[’]{} \_[ab]{} ’\_[’]{},\_[’]{}=(-’) . Moreover using we can rewrite as \[singleSthroughA\] \_+ =ł\^[-T]{} A\_[+]{} - A\_[-]{} ,\_- = ł\^[-1]{} A\_[-]{} - A\_[+]{} . Inversely \[singleAthroughS\]
&A\_[+]{} = h\^[-1]{} ł\^T (\_+ + ł\_-) , A\_[-]{} = h\^[-1]{} ł(\_- + ł\^T \_+) ,\
&h=-ł\^Tł,h=-łł\^T,
assuming that the matrix $\l$ is such that $h,\tilde h$ are positive-definite matrices. To obtain the Poisson algebra in the base of $A_\pm$ we use , and .
To study the Hamiltonian structure of the problem we need to define its phase space [@Hollowood:2014rla; @Hollowood:2014qma]. This is given in terms of the currents $\mathcal{J}_{\pm}$, the gauge fields $A_\pm$ and the associated momenta $P_\pm$ to $A_\pm$. The $\mathcal{J}_{\pm}$ obey two commuting current algebras and have vanishing Poisson brackets with $A_\pm$ and $P_\pm$ $$\{P_\pm^a(\s),A^b_\mp(\s')\}=\d^{ab}\d(\s-\s')\,.$$ Furthermore, since the $A_{\pm}$’s are non-dynamical their associated momenta $P_\pm$ vanish. This introduces two primary constraints $$\varphi_1=P_+\approx0\,,\quad \varphi_2=P_-\approx0\, .$$ Their time-evolution gives rise to the secondary constraints $$\varphi_3=\mathcal{J}_+ -\l^{-T} A_{+} + A_{-}\approx0\,,\quad
\varphi_4=\mathcal{J}_- - \l^{-1} A_{-} +A_{+}\approx0\ .$$ Time evolution generates no further constraints. The $\varphi_i$’s with $i=1,2,3,4$, turn out to be second class constraints, since the matrix of their Poisson brackets is invertible in the deformed case. Finally, the Hamiltonian density of the single $\lambda$-deformed model before integrating out the gauge fields takes the form [@Sfetsos:2013wia; @Bowcock:1988xr] $$\begin{split}
\mathcal{H}_{\text{single}}&=\frac{k}{4\pi}{\rm Tr}\left(\mathcal{J}_+\mathcal{J}_++\mathcal{J}_-\mathcal{J}_-\right.
\left. +4(\mathcal{J}_+A_{-}+\mathcal{J}_-A_{+})\right.\\
&\left. +2(A_{+}-A_{-})(A_{+}-A_{-})\right.
\left.-4A_{+}(\lambda_1^{-1}-\mathbb{I}) A_{-}\right),
\end{split}$$ or equivalently through , in terms of $A_{\pm}$’s .
Doubly $\lambda$-deformed $\s$-models {#doublyldeformed}
-------------------------------------
The action defining the doubly deformed models depends on two group elements $g_i\in G,\,\,\,i=1,2$ and is given by the deformation of the sum of two WZW models $S_k(g_1)$ and $S_k(g_2)$ as [@Georgiou:2016urf]
& S\_[k,\_1,\_2]{}(g\_1,g\_2) = S\_k(g\_1) + S\_k(g\_2)\
&+ [k]{} \^2 [Tr]{}{ (
[cc]{} J\_[1+]{} & J\_[2+]{}
) (
[cc]{} Ł\_[21]{}\_1 D\_2\^T\_2 & Ł\_[21]{}\_1\
Ł\_[12]{}\_2 & Ł\_[12]{} \_2 D\_1\^T\_1\
) (
[c]{} J\_[1-]{}\
J\_[2-]{}
)} ,
\[defactigen\] where $$\label{doubleLambdas}
\L_{12}= (\mathbb{I} - \lambda_2 D_1^T \lambda_1 D_2^T)^{-1}\ ,\quad
\L_{21}= (\mathbb{I} - \lambda_1 D_2^T \lambda_2 D_1^T)^{-1}\ .$$ The matrices $D_{ab}$ and the currents $J^a_{\pm}$ are defined in . When a current or the matrix $D$ has the extra index $1$ or $2$ this means that one should use the corresponding group element in its definition. The action has the [*non-perturbative*]{} symmetry [@Georgiou:2016urf] k -k , \_1 \_1\^[-1]{} ,\_2 \_2\^[-1]{} , g\_1g\_2\^[-1]{},g\_2g\_1\^[-1]{} , \[symmdual\] which is similar to . For small elements of the matrices $\lambda_i$’s the action becomes $$S_{k,\lambda_1,\lambda_2}(g_1,g_2) = S_k(g_1) + S_k(g_2) + {k\ov \pi}
\int \text{d}^2\s\ {\rm Tr}(J_{1+}\lambda_1 J_{2-} + J_{2+}\lambda_2 J_{1-}) + \cdots\ .$$ Hence represents the effective action of two WZW models mutually interacting via current bilinears. Similarly to we define the currents[^1]^,^[^2]
\[Affine\] &\^[(1)]{}\_[+]{} =\_+ g\_1 g\_1\^[-1]{} + A\^[(1)]{}\_+ - A\^[(1)]{}\_- , \^[(1)]{}\_[-]{} = - g\_1\^[-1]{} \_- g\_1+ A\^[(2)]{}\_[-]{} - A\^[(2)]{}\_[+]{} ,\
&\^[(2)]{}\_[+]{} =\_+ g\_2 g\_2\^[-1]{} + A\^[(2)]{}\_[+]{} - A\^[(2)]{}\_[-]{} , \^[(2)]{}\_[-]{} =- g\_2\^[-1]{} \_- g\_2 + A\^[(1)]{}\_- - A\^[(1)]{}\_+ .
These currents obey two commuting copies of current algebras [@Georgiou:2016urf] \[Bowcock\] {\_\^[(i)a]{} , \_\^[(i)b]{}} = f\_[abc]{} \_\^[(i)c]{} \_[’]{} \_[ab]{} ’\_[’]{},i=1,2 , which encode the canonical structure of the theory. The action does not depend on derivatives of $A^{(i)}_\pm\,, i=1,2$, so that as in subsection \[singlelambdalkslsl\], their equations of motion are second class constraints [@Georgiou:2016urf] \[kshdg2\]
&\_+ g\_1 g\_1\^[-1]{} = (\_1\^[-T]{}-) A\^[(1)]{}\_+ ,g\_1\^[-1]{} \_- g\_1 = - (\_2\^[-1]{}-) A\^[(2)]{}\_[-]{},\
&\_+ g\_2 g\_2\^[-1]{} = (\_2\^[-T]{}-) A\^[(2)]{}\_[+]{} ,g\_2\^[-1]{} \_- g\_2 = - (\_1\^[-1]{}-) A\^[(1)]{}\_-,
determining the gauge fields in terms of the group elements similarly to [(\[sklldldd\])]{} (for the precise expressions we refer to [@Georgiou:2016urf]). Then rewrites as \[SthroughA\]
&\^[(1)]{}\_[+]{} = \_1\^[-T]{} A\^[(1)]{}\_[+]{} - A\^[(1)]{}\_[-]{} , \^[(1)]{}\_[-]{} = \_2\^[-1]{} A\^[(2)]{}\_- - A\^[(2)]{}\_+ ,\
&\^[(2)]{}\_[+]{} = \_2\^[-T]{} A\^[(2)]{}\_[+]{} - A\^[(2)]{}\_[-]{} , \^[(2)]{}\_[-]{} = \_1\^[-1]{} A\^[(1)]{}\_- - A\^[(1)]{}\_+ .
Equivalently the gauge fields in terms of the dressed currents are given by \[AthroughS\]
&A\^[(1)]{}\_+ = h\_1\^[-1]{} \_1\^T (\^[(1)]{}\_[+]{} + \_1 \^[(2)]{}\_[-]{}) , A\^[(1)]{}\_- = h\_1\^[-1]{} \_1 (\^[(2)]{}\_[-]{} + \_1\^T \^[(1)]{}\_[+]{}) ,\
&A\^[(2)]{}\_[+]{} = h\_2\^[-1]{} \_2\^T (\^[(2)]{}\_[+]{} + \_2 \^[(1)]{}\_[-]{}) , A\^[(2)]{}\_[-]{} = h\_2\^[-1]{} \_2 (\^[(1)]{}\_[-]{} + \_2\^T \^[(2)]{}\_[+]{}) ,\
&h\_i=-\_i\^T\_i,h\_i=-\_i\_i\^T,i=1,2.
To obtain the Poisson algebra in the base of $A^{(1)}_{\pm}$ and $A^{(2)}_{\pm}$ we use , and . As a corollary one can easily show that $\{A^{(1)}_\pm,A^{(2)}_\pm\}=0$, for all choices of signs and for generic coupling matrices $\lambda_{1,2}$. The Hamiltonian density of our system before integrating out the gauge fields takes the form [@Georgiou:2016urf] $$\begin{split}
\mathcal{H}_{\text{doubly}}&=
\frac{k}{4\pi}{\rm Tr}\left\{\mathcal{J}^{(1)}_{+}\mathcal{J}^{(1)}_{+}
+\mathcal{J}^{(1)}_{-}\mathcal{J}^{(1)}_{-}+\mathcal{J}^{(2)}_{+}\mathcal{J}^{(2)}_{+}+
\mathcal{J}^{(2)}_{-}\mathcal{J}^{(2)}_{-}\right.\\
&\left. +4(\mathcal{J}^{(1)}_{+}A^{(1)}_{-}+\mathcal{J}^{(2)}_{+}A^{(2)}_{-}+\mathcal{J}^{(1)}_{-}A^{(2)}_{+}+
\mathcal{J}^{(2)}_{-}A^{(1)}_{+})\right.\\
&\left. +2(A^{(1)}_{+}-A^{(1)}_{-})(A^{(1)}_{+}-A^{(1)}_{-})+2(A^{(2)}_{+}-A^{(2)}_{-})(A^{(2)}_{+}-A^{(2)}_{-})\right.\\
&\left.-4A^{(1)}_{+}(\lambda_1^{-1}-\mathbb{I})A^{(1)}_{-}-4A^{(2)}_{+}(\lambda_2^{-1}-\mathbb{I})A^{(2)}_{-}\right\}\
\end{split}$$ and can be rewritten through in terms of $A^{(i)}_{\pm}$ and $\lambda_i$ as \[Hdensity0\] .
The fact that the Hamiltonian density is the sum of two terms one depending on $A^{(1)}_{\pm}$ and the other on $A^{(2)}_{\pm}$ combined with the fact that the currents $\mathcal{J}^{(i)}_{\pm}, \,\,i=1,2,$ obey two commuting copies of the current algebra of the single $\lambda$-deformed model shows that the doubly deformed models are canonically equivalent to the sum of two single $\lambda$-deformed models, one with coupling $\lambda_1$ and the other with coupling $\lambda_2$. The relations defining the canonical transformation are given by $$\label{can-def}
A^{(1)}_{\pm} =\tilde A^{(1)}_{\pm} \ ,\quad A^{(2)}_{\pm}=\tilde A^{(2)}_{\pm} \ ,$$ where the gauge fields without the tildes correspond to the doubly deformed models and depend on $(\lambda_1,\lambda_2;g_1,g_2)$, while the tilded gauge fields correspond to the canonically equivalent sum of two single $\lambda$-deformed models the first of which depends on $(\lambda_1;\tilde g_1)$ only while the second depends on $(\lambda_2;\tilde g_2)$.
Furthermore, the gauge fields of should be considered as functions of the coordinates parametrising the group elements and their conjugate momenta. We may write relations involving world-sheet derivatives of the various group elements by using and . As in all canonical transformation involving canonical variables as well as their momenta, the relation between the $g_i$’s and the $\tilde g_i$’s is a non-local one.
A comment is in order concerning the $\eta$-deformed models [@Klimcik:2002zj; @Klimcik:2008eq; @Delduc:2013fga; @Delduc:2013qra; @Arutyunov:2013ega] which are closely related to the single $\lambda$-deformed ones via Poisson–Lie T-duality [@KS95a] and an appropriate analytic continuation of the coordinates and the parameters [@Klimcik:2015gba; @Klimcik:2016rov; @Vicedo:2015pna; @Hoare:2015gda; @Sfetsos:2015nya] $${
\lambda\mapsto \frac{i E-\eta\mathbb{I}}{i E+\eta\mathbb{I}}\ ,
}$$ where $E$ is an arbitrary constant matrix. Poisson–Lie T-duality can also be formulated as a canonical transformation [@Sfetsos:1996xj; @Sfetsos:1997pi] and therefore there is a chain of canonical transformations from doubly $\lambda$-deformed, to two single $\lambda$-deformed and to $\eta$-deformed models. It would be interesting to formulate the canonical transformation via a duality invariant action similarly perhaps to the case of Poisson–Lie T-duality in [@Klimcik:1995dy].
There is an important observation for further use in section \[cycliclambdasection\]. The Hamiltonian density has the following [*non-perturbative*]{} symmetry \[npsym\] k -k, \_i ł\^[-1]{}\_i, A\^[(i)]{}\_+ ł\^[-T]{}\_i A\^[(i)]{}\_+, A\^[(i)]{}\_- ł\^[-1]{}\_i A\^[(i)]{}\_-,i=1,2. In other words $\mathcal{H}_{\text{doubly}}$ maps to itself under . By using this implies the following transformation for the group elements $g_1$ and $g_2$ \[npsym-curr\] \^[(1)]{}\_[+]{} -\^[(2)]{}\_[-]{} , \^[(2)]{}\_[+]{} -\^[(1)]{}\_[-]{} , \^[(1)]{}\_[-]{} -\^[(2)]{}\_[+]{}, \^[(2)]{}\_[-]{} -\^[(1)]{}\_[+]{}. Since the currents $\mathcal{J}^{(i)}_{\pm}, \,\,i=1,2$, depend both on the group elements and their derivatives, the transformation can be viewed as a non-local transformation at the level of the group elements. In the special cases of the single and doubly $\lambda$-deformed theories the symmetry and can be realized locally simply by a mapping of group elements, i.e. and . Indeed, it is not difficult to check that and imply for the gauge fields the transformation . The situation is slightly different for the generic cyclic models constructed below in section \[cycliclambdasection\] which can have arbitrarily many group elements.
Doubly-deformed models and non-Abelian T-duality {#newnonabeliansection}
================================================
It is has been known that the action admits the non-Abelian T-dual limit that involves taking $k\to \infty$, whereas simultaneously taking the matrix $\lambda$ and the group element $g$ to the identity [@Sfetsos:2013wia]. Specifically, if we let $$\l= \mathbb{I} - {E\ov k}\ ,\qq g=\mathbb{I} + i {v\ov k}\ ,\qq k\to \infty\ ,$$ where $E$ is a constant matrix and $v=v_a t^a$, then the action becomes $$S(v,E)={1\ov \pi} \int \text{d}^2\s\, \Tr\left(\del_+ v (E+f)^{-1} \del_- v\right) \ ,$$ where $f$ is a matrix with elements $f_{ab}= f_{abc} v^c$. This is the non-Abelian T-dual of the PCM action with general coupling matrix $E$ $$S_{\rm PCM}(g,E)= -{1\ov \pi}\int \text{d}^2\s\, \Tr
\left(g^{-1}\del_+ g\, E\, g^{-1}\del_-g\right)\ ,$$ with respect to the global symmetry $g\mapsto \L\, g$, $\L\in G$. The above limit is well defined when is taken on the $\b$-function for $\lambda$, as well as on the anomalous dimensions of various operators in the theory. In the case of doubly $\lambda$ or even multiple/cyclic $\lambda$-deformations (see section \[cycliclambdasection\]) we have shown in particular that, the $\b$-functions and current anomalous dimensions are the same with those of two or more simple $\lambda$-deformations. Hence, it is expected that it should be possible to take a well defined non-Abelian type limit in the action . This is not necessarily simple since a suitable limit involves the two group elements.
In the following we focus on the most interesting case in which the matrices $\lambda_i$, $i=1,2$ are isotropic, i.e. $(\lambda_i)_{ab}= \lambda_i\, \d_{ab}$. It is convenient to use the group element $\mathcal{G}=g_1 g_2$ and also rename $g_2$ by $g$. Then employing the Polyakov–Wiegmann identity [@Polyakov:1983tt], the action , using also , takes the form \[rewr\] && S\_[k,\_1,\_2]{}(,g)= S\_k() + [k]{} \^2 ((1-\_2)g\^[-1]{}\_+ g ([D]{}-\_1) g\^[-1]{}\_-g\
&& -(1-\_2)g\^[-1]{}\_+ g \_- \^[-1]{} + \_1 (1-\_2) \^[-1]{}\_+ g\^[-1]{}\_-g + \_1\_2 \^[-1]{}\_+ \^[-1]{}\_-) ,where: $\Sigma=\left(\lambda_1\lambda_2 \mathbb{I} -{\cal D}\right)^{-1}$ and ${\cal D}=D(\mathcal{G})=D(g_1)D(g_2)$. Next we take the limit \[dsh11\] \_i=1-[\_i\^2k]{} ,i=1,2 ,= + i [vk]{} ,k . After some algebra we find that becomes $$\begin{split}
&
S_{\k_1^2,\k_2^2}(v,g) = -{1\ov \pi}\int \text{d}^2\s\, \Tr\Big( \k_2^2 g^{-1} \del_+ g g^{-1}\del_- g
\\
&\quad + \big(i \del_+ v - \k_2^2 g^{-1} \del_+ g\big)\big((\k_1^2+\k_2^2)\mathbb{I}+f\big)^{-1}
\big(i \del_- v + \k_2^2 g^{-1} \del_- g\big)\Big) \ .
\end{split}$$ It can be shown that this action is the non-Abelian T-dual of $$S = -{1\ov \pi} \int \text{d}^2\s\, \Tr\Big( \k_1^2 \tilde g^{-1} \del_+ \tilde g
\tilde g^{-1} \del_- \tilde g
+\k_2^2 (g^{-1}\del_+ g -\tilde g^{-1}\del_+ \tilde g)
(g^{-1}\del_- g -\tilde g^{-1}\del_- \tilde g)\Big)\ ,$$ with respect to the global symmetry $\tilde g \mapsto \L \tilde g$, $\L\in G$. [ Note that, if we define the new group element $\widetilde{\mathcal{G}}=g\tilde g^{-1}$ one may write the previous action as S= -[1]{} \^2 (\_1\^2 g\^[-1]{} \_+ g g\^[-1]{} \_- g + \_2\^2 \^[-1]{} \_+ \^[-1]{} \_- ) , which is the sum of two independent PCM actions for a group $G$.]{} The previous group element redefinition introduces interactions between them.
Finally consider a limit in which only $\lambda_2$ tends to one, whereas $\lambda_1$ stays inactive. Then, has to be modified as $$\lambda_2=1-{\k_2^2\ov k}\ ,\quad \mathcal{G}= \mathbb{I} + i {v\ov \sqrt{k}}\ ,\quad k\to \infty\ ,$$ in order for to stay finite. In particular, this becomes S\_[\^2]{}(v,g)=[12]{} [1+\_11-\_1]{} \^2 (\_+ v\_-v)- [\_2\^2]{} \^2 (g\^[-1]{}\_+ g g\^[-1]{}\_- g) , representing $\dim G$ free bosons and a PCM model for a group $G$. This is consistent with the limit of the $\beta$-functions for $\lambda_1$ and $\lambda_2$ (see, eqs. (2.6) and (2.7) in [@Georgiou:2017aei]). In this limit, the constant $\lambda_1$ does not run since it can be absorbed into a redefinition of the $v$’s. Also the coupling constant $\k_2^2$ obeys the same RG flow equation appropriate for the PCM model and its non-Abelian T-dual, since these models are canonically equivalent.
It would be very interesting to explore physical applications in an AdS/CFT context of this version of non-Abelian T-duality along the lines and developments of [@Sfetsos:2010uq; @Itsios:2012zv; @Itsios:2013wd; @Lozano:2016kum; @Lozano:2016wrs; @Itsios:2016ooc; @Lozano:2017ole] (for a partial list of works in this direction). Prototype examples this can be applied are the backgrounds $\text{AdS}_3\times \text{S}^3 \times \text{S}^3 \times \text{S}^1$ and $\text{AdS}_5\times \text{S}^5$.
Cyclic $\lambda$-deformations {#cycliclambdasection}
=============================
In this section we construct a class of multi-parameter deformations of conformal field theories of the WZW type Consider $n$ WZW models and $n$ PCMs for a group $G$, [ defined with group elements $g_i$ and $\tilde g_i$, respectively.]{} We would like to gauge the global symmetry $$g_i \mapsto \L_i^{-1}\, g_i\, \L_{i+1}\ ,\quad \tilde g_i \mapsto \L^{-1}_i \tilde g_i\ ,
\quad i=1,2,\dots , n,$$ with the periodicity condition $\L_{n+1}=\L_1$ implied. We introduce gauge fields $A^{(i)}_{\pm}$ in the Lie-algebra of $G$ transforming as $$A^{(i)}_{\pm} \mapsto \L_i^{-1} A^{(i)}_{\pm} \L_i - \L_i^{-1}\del_\pm \L_i\ ,\quad i=1,2,\dots ,n\ .$$ In this way we have a periodic chain of interacting models each one of which separately is gauge anomalous by a term independent of the group elements. The full model has no gauge anomaly since these cancel among themselves (the chain may be open as long as it is infinite long). The details are quite similar to those for the $n=2$ case [@Georgiou:2016urf], so that we omit them here.
The choice $\tilde g_i =\mathbb{I}$, $i=1,2,\dots , n $ completely fixes the gauge and is consistent with the equations of motion for the group elements $\tilde g_i$ of the PCMs which are automatically satisfied. Then, the gauged fixed action becomes \[action333\]
& S\_[k,\_i]{}({g\_i;A\^[(i)]{}\_}) = \_[i=1]{}\^n S\_k(g\_i) + [k]{} \^2\_[i=1]{}\^n ( A\^[(i)]{}\_- \_+ g\_i g\^[-1]{}\_i - A\^[(i+1)]{}\_+ g\_i\^[-1]{} \_- g\_i\
& + A\^[(i)]{}\_- g\_i A\^[(i+1)]{}\_+ g\_i\^[-1]{} -A\^[(i)]{}\_+ \_i\^[-1]{} A\^[(i)]{}\_-) ,
where the index $i$ is defined modulo $n$. The equations of motion with respect to the $A^{(i)}_{\pm}$’s are given by $$\lambda_i^T D_i A^{(i+1)}_+ - A^{(i)}_+ =- i \lambda_i^T J^{(i)}_{+}\ ,\quad
\lambda_{i+1} D_i^T A^{(i)}_- - A^{(i+1)}_{-} = i \lambda_{i+1} J^{(i)}_{-}\ .$$ Solving them we find that $$\begin{split}
A^{(1)}_{+}=i (\mathbb{I}- x_1 x_2\cdots x_n )^{-1} \sum_{i=1}^n x_1 x_2 \cdots x_{i-1} \lambda_i^T J^{(i)}_{+}\,,\quad
x_i=\lambda_i^T D_i\ .
\end{split}$$ The rest can be obtained by cyclic permutations. Plugging the latter into we find that the on-shell action reads $$\begin{aligned}
\label{defactigenn}
&& S_{k,\lambda_i}(\{g_i\})= {k\ov 12\pi} \int \Tr(g_1^{-1} \text{d}g_1)^3
+ {k\ov \pi} \int \text{d}^2\s\,{\rm Tr}\Bigg(\ha J^{(1)}_{+} D_1
{\mathbb{I} +x_1^T x_n^T x_{n-1}^T \cdots x_2^T\ov \mathbb{I} - x_1^T x_n^T x_{n-1}^T \cdots x_2^T} J^{(1)}_{-} \nonumber
\\ &&
+ \sum_{i=2}^n J^{(i)}_{+} \lambda_i x_{i-1}^T\cdots x_2^T (\mathbb{I} -x_1^T x_n^T x_{n-1}^T \cdots x_2^T)^{-1} J^{(1)}_{-}\Bigg)
\ +\ {\rm cyclic\ in}\ 1,2,\dots , n\ ,\end{aligned}$$ where we have separated the Wess–Zumino term from the WZW model action. For small values of the matrices we have that \[fhh22\] S\_[k,\_i]{}({g\_i}) = \_[i=1]{}\^n S\_k(g\_i) + [k]{} \_[i=1]{}\^n \^2 [Tr]{}(J\^[(i+1)]{}\_+ \_[i+1]{} J\^[(i)]{}\_-) + [O]{}(ł\^2) , representing $n$ distinct WZW models interacting by mutual current bilinears, for which is the all loop, in the $\lambda_i$’s, effective action.
We would like to stress that the $n=2$ is significantly different with respect to higher $n$’s. Firstly, the [*non-perturbative*]{} symmetry $\lambda_i\mapsto\lambda_i^{-1}$ and $k\mapsto-k$, is seemingly realized at a local level for the group elements only when $n=2$, see (also for $n=1$, see ). For higher values of $n$ the group elements need to be transformed non-locally by using $\mathcal{J}^{(i)}_{\pm} \mapsto -\mathcal{J}^{(i+1)}_{\mp}$, with $n+1 \equiv 1$. There are exceptions to this. In particular, if all $\lambda_i$ are equal and isotropic, i.e. $\lambda_i=\l \mathbb{I}$, then this duality-type symmetry is k-k ,ł , g\_1g\_2\^[-1]{} , g\_ng\_3\^[-1]{} ,g\_[n-1]{} g\_4\^[-1]{} , , that is the group elements are paired up as above. For odd $n$ one group element simply gets inverted. Despite the fact that the symmetry can not be realized locally for the generic case it is still powerful enough to constrain the $\b$-functions and current correlation functions of the cyclic model to have the same values as those of the single $\lambda$-deformations.
A second remark concerns the form of the action when one of the coupling matrices vanishes. Consider this action for $n=2$ and $n=3$ when $\lambda_1=0$ while the other coupling matrices stay general $$\begin{split}
&S_{k,0,\lambda_2}(g_1,g_2)=\sum_{i=1}^2S_k(g_i)+\frac{k}{\pi}\,\int\text{d}^2\s\, {\rm Tr}\left(J_+^{(2)}\lambda_2J_-^{(1)}\right)\,,\\
&S_{k,0,\lambda_2,\lambda_3}(g_1,g_2,g_3)=\sum_{i=1}^3S_k(g_i)+\frac{k}{\pi}\,\int\text{d}^2\s\,{\rm Tr}\left(J_+^{(2)}\lambda_2J_-^{(1)}+
J_+^{(3)}\lambda_3J_-^{(2)}+J_+^{(2)}\lambda_3D_2^T\lambda_2J_-^{(1)}\right)\,.
\end{split}$$ When $n=2$ the exact expression matches the approximate one in , while for $n=3$ the last term couples the three WZW models and it is quadratic in the $\lambda$’s.
Algebra and Hamiltonian
-----------------------
Here we provide the proof that the $\sigma$-model action is integrable for specific choices of the matrices $\lambda_i$, $i=1,2,\dots , n$. In particular, we will show that it is integrable for all choices of the deformation matrices $\lambda_i$ which, separately, give an integrable $\lambda$-deformed model. These include the isotropic $\lambda$ for semi-simple group and symmetric coset, the anisotropic $SU(2)$ and the $\lambda$-deformed Yang–Baxter model [@Sfetsos:2013wia; @Sfetsos:2014lla; @Sfetsos:2015nya; @Hollowood:2014rla; @Hollowood:2014qma].
It is equivalent and more convenient to work with the gauged fixed action before integrating out the gauge fields. Varying the gauged fixed action with respect $A^{(i)}_-$ and $A^{(i+1)}_+$ we find the constraints \_+ g\_i g\_i\^[-1]{} = (\_i\^[-T]{}-) A\^[(i)]{}\_+ ,g\_i\^[-1]{} \_- g\_i = - (\_[i+1]{}\^[-1]{}-) A\^[(i+1)]{}\_[-]{} , \[dggd\] respectively. Varying with respect to $g_i$ we obtain that \[eqg1g2\] \_-(\_+ g\_i g\_i\^[-1]{})= F\_[+-]{}\^[(i)]{} ,\_+( g\_i\^[-1]{}\_- g\_i) = F\_[+-]{}\^[(i+1)]{} , which are in fact equivalent and where $F_{+-}^{(i)}=\del_+ A^{(i)}_- - \del_- A^{(i)}_+ - [A^{(i)}_+,A^{(i)}_-]$.
Substituting into we obtain after some algebra that
\[eomAinitial1\] &\_+ A\^[(i)]{}\_- - \_i\^[-T]{} \_- A\^[(i)]{}\_+ = \[\_i\^[-T]{} A\^[(i)]{}\_+,A\^[(i)]{}\_-\] ,\
& \_i\^[-1]{} \_+ A\^[(i)]{}\_- -\_-A\^[(i)]{}\_+=\[A\^[(i)]{}\_+,\_i\^[-1]{}A\^[(i)]{}\_-\] .
Hence the equations of motion split into $n$ identical sets which are seemingly decoupled even though the $A^{(i)}_{\pm}$ depend on all group elements $g_i$ and coupling matrices $\l_i$, $i=1,2,\dots , n$. Moreover, each set is the same one that one would have obtained had [we]{} performed the corresponding analysis for the $\lambda$-deformed action . Working along the lines of subsection \[doublyldeformed\]; Eqns. – we find (for $n=2$ this was performed in detail in [@Georgiou:2016urf])
\[Bowcockn\] &{\_\^[(i)a]{} , \_\^[(i)b]{}} = f\_[abc]{} \_\^[(i)c]{} \_[’]{} \_[ab]{} ’\_[’]{},\
&\^[(i)]{}\_[+]{} =\_i\^[-T]{} A\^[(i)]{}\_+ - A\^[(i)]{}\_- , \^[(i)]{}\_[-]{} = \_[i+1]{}\^[-1]{} A\^[(i+1)]{}\_[-]{} - A\^[(i+1)]{}\_[+]{}
and as a consequence $
\{A^{(i)}_{\pm},A^{(j)}_{\pm}\}=0$, for $i\neq j$, for all choices of signs and for generic coupling matrices $\lambda_{i}$. Hence, all choices for matrices known to give rise to integrability for the $\lambda$-deformed models provide integrable models here as well with independent conserved changes. The Hamiltonian density of the system in terms of $A^{(i)}_\pm$ and $\lambda_i$ is \[Hdensitycyclic\] . Using the above we generalize the result of subsection \[doublyldeformed\], that the cyclic $\lambda$-deformed [models]{} are canonically equivalent to $n$ single $\lambda$-deformed $\sigma$-model. The relations which define the canonical transformation are given by: $
A^{(i)}_{\pm} =\tilde A^{(i)}_{\pm}\ ,\quad i=1,2,\dots, n\ ,
$ where the gauge fields without the tildes correspond to the cyclic deformed models and depend on $(\lambda_1,\dots,\lambda_n;g_1,\dots,g_n)$, while those with tildes correspond to the canonically equivalent sum of $n$ single $\lambda$-deformed models each one depending on $(\lambda_i;\tilde g_i)$.
RG flows and currents anomalous dimensions {#rg-flows-and-currents-anomalous-dimensions .unnumbered}
------------------------------------------
Similar to the case with $n=2$ considered in [@Georgiou:2017aei], the expression can be used to argue that the RG flow equations of the $n$ coupling matrices $\lambda_i$ for the cyclic model as well as the currents anomalous dimensions are the same with those obtained for the single $\lambda$-deformations model [@Itsios:2014lca; @Sfetsos:2014jfa; @Appadu:2015nfa]. The basic reason is that the various interaction terms have regular OPE among themselves so that correlations functions involving currents factorize to those of $n$ single $\lambda$-deformed models. This is also in agreement with the fact that the cyclic model is canonically equivalent to $n$ single $\lambda$-deformations. Furthermore we mention without presenting any details that using the analysis performed in [@Appadu:2015nfa; @Georgiou:2017aei] we have explicitly checked the above claim for the cases of $n$ isotropic couplings for general groups and symmetric spaces.
Acknowledgements {#acknowledgements .unnumbered}
----------------
K. Sfetsos would like to thank the Physics Division, National Center for Theoretical Sciences of the National Tsing-Hua University in Taiwan and the Centre de Physique Théorique, École Polytechnique for hospitality and financial support during initial stages of this work. G. Georgiou and K. Siampos acknowledge the Physics Department of the National and Kapodistrian University of Athens for hospitality. Part of this work was developed during HEP 2017: Recent Developments in High Energy Physics and Cosmology in April 2017 at the U. of Ioannina.
[1]{}
G. Georgiou and K. Sfetsos, [*A new class of integrable deformations of CFTs*]{}, JHEP [**1703**]{} (2017) 083, [arXiv:1612.05012 \[hep-th\]](https://arxiv.org/abs/1612.05012).
K. Sfetsos, [*Integrable interpolations: From exact CFTs to non-Abelian T-duals*]{}, Nucl. Phys. [**B880**]{} (2014) 225, [arXiv:1312.4560 \[hep-th\]](http://arxiv.org/abs/arXiv:1312.4560).
T. J. Hollowood, J. L. Miramontes and D. M. Schmidtt, [*Integrable Deformations of Strings on Symmetric Spaces*]{}, JHEP [**1411**]{} (2014) 009, [arXiv:1407.2840 \[hep-th\]](http://arxiv.org/abs/1407.2840).
T. J. Hollowood, J. L. Miramontes and D. Schmidtt, [*An Integrable Deformation of the $AdS_5 \times S^5$ Superstring*]{}, J. Phys. [**A47**]{} (2014) 49, 495402, [arXiv:1409.1538 \[hep-th\]](http://arxiv.org/abs/1409.1538).
G. Georgiou, E. Sagkrioti, K. Sfetsos and K. Siampos, [*Quantum aspects of doubly deformed CFTs*]{}, Nucl. Phys. [**B919**]{} (2017) 504, [arXiv:1703.00462 \[hep-th\]](https://arxiv.org/abs/1703.00462).
K. Sfetsos and K. Siampos, [*Gauged WZW-type theories and the all-loop anisotropic non-Abelian Thirring model*]{}, Nucl. Phys. [**B885**]{} (2014) 583, [arXiv:1405.7803 \[hep-th\]](http://arxiv.org/abs/arXiv:1405.7803).
D. Kutasov, [*String Theory and the Nonabelian Thirring Model*]{},\
[Phys. Lett. [**B227**]{} (1989) 68](http://www.sciencedirect.com/science/article/pii/0370269389912859).
B. Gerganov, A. LeClair and M. Moriconi, [*On the beta function for anisotropic current interactions in 2-D*]{}, Phys. Rev. Lett. [**86**]{} (2001) 4753, [hep-th/0011189](http://arxiv.org/abs/hep-th/0011189).
G. Itsios, K. Sfetsos and K. Siampos, [*The all-loop non-Abelian Thirring model and its RG flow*]{}, Phys. Lett. [**B733**]{} (2014) 265, [arXiv:1404.3748 \[hep-th\]](http://arxiv.org/abs/1404.3748).
G. Georgiou, K. Sfetsos and K. Siampos, [*All-loop anomalous dimensions in integrable $\lambda$-deformed $\sigma$-models*]{}, Nucl. Phys. [**B901**]{} (2015) 40, [arXiv:1509.02946 \[hep-th\]](http://arxiv.org/abs/1509.02946).
G. Georgiou, K. Sfetsos and K. Siampos, [*All-loop correlators of integrable $\lambda$-deformed $\sigma$-models*]{}, Nucl. Phys. [**B909**]{} (2016) 360, [1604.08212 \[hep-th\]](http://arxiv.org/abs/arXiv:1604.08212).
G. Georgiou, K. Sfetsos and K. Siampos, [*$\lambda$-deformations of left-right asymmetric CFTs*]{}, Nucl. Phys. [**B914**]{} (2017) 623, [arXiv:1610.05314 \[hep-th\]](https://arxiv.org/abs/1610.05314).
T. Curtright and C. K. Zachos, [*Currents, charges, and canonical structure of pseudodual chiral models*]{}, Phys. Rev. [**D49**]{} (1994) 5408, [hep-th/9401006](https://arxiv.org/abs/hep-th/9401006).
E. Alvarez, L. Alvarez-Gaume and Y. Lozano, [*A Canonical approach to duality transformations*]{}, Phys. Lett. [**B336**]{} (1994) 183, [hep-th/9406206](https://arxiv.org/abs/hep-th/9406206).
Y. Lozano, [*Non-Abelian duality and canonical transformations*]{}, Phys. Lett. [**B355**]{} (1995) 165, [hep-th/9503045](https://arxiv.org/abs/hep-th/9503045).
K. Sfetsos, [*Poisson–Lie T duality and supersymmetry*]{}, Nucl. Phys. Proc. Suppl. [**B56**]{} (1997) 302, [hep-th/9611199](https://arxiv.org/abs/hep-th/9611199).
K. Sfetsos, [*Canonical equivalence of nonisometric sigma models and Poisson–Lie T duality*]{}, Nucl. Phys. [**B517**]{} (1998) 549, [hep-th/9710163](https://arxiv.org/abs/hep-th/9710163).
L. K. Balazs, J. Balog, P. Forgacs, N. Mohammedi, L. Palla and J. Schnittger, [*Quantum equivalence of sigma models related by non-Abelian duality transformations*]{} Phys. Rev. [**D57**]{} (1998) 3585, [hep-th/9704137](https://arxiv.org/abs/hep-th/9704137).
J. Balog, P. Forgacs, N. Mohammedi, L. Palla and J. Schnittger, [*On quantum T duality in sigma models*]{}, Nucl. Phys. [**B535**]{} (1998) 461, [hep-th/9806068](https://arxiv.org/abs/hep-th/9806068).
K. Sfetsos, [*Poisson-Lie T duality beyond the classical level and the renormalization group*]{}, Phys. Lett. [**B432**]{} (1998) 365, [hep-th/9803019](https://arxiv.org/abs/hep-th/9803019).
K. Sfetsos, [*Duality invariant class of two-dimensional field theories*]{}, Nucl. Phys. [**B561**]{} (1999) 316, [hep-th/9904188](https://arxiv.org/abs/hep-th/9904188).
K. Sfetsos and K. Siampos, [*Quantum equivalence in Poisson–Lie T-duality*]{}, JHEP [**0906**]{} (2009) 082, [arXiv:0904.4248 \[hep-th\]](https://arxiv.org/abs/0904.4248).
P. Bowcock, [*Canonical Quantization of the Gauged Wess–Zumino Model*]{}, [Nucl. Phys. [**B316**]{} (1989) 80](http://www.sciencedirect.com/science/article/pii/0550321389903878).
E. Witten, [*Nonabelian Bosonization in Two-Dimensions*]{},[Commun. Math. Phys. [**92**]{} (1984) 455](http://link.springer.com/article/10.1007%2FBF01215276).
C. Klimčík, [*YB sigma models and dS/AdS T-duality*]{}, JHEP [**0212**]{} (2002) 051,\
[hep-th/0210095](http://arxiv.org/abs/hep-th/hep-th/0210095).
C. Klimčík, [*On integrability of the YB sigma-model*]{}, J. Math. Phys. [**50**]{} (2009) 043508, [arXiv:0802.3518 \[hep-th\]](http://arxiv.org/abs/0802.3518).
F. Delduc, M. Magro and B. Vicedo, [*On classical $q$-deformations of integrable sigma-models*]{}, JHEP [**1311**]{} (2013) 192, [arXiv:1308.3581 \[hep-th\]](http://arxiv.org/abs/1308.3581).
F. Delduc, M. Magro and B. Vicedo, [*An integrable deformation of the $AdS_5 \times S^5$ superstring action*]{}, Phys. Rev. Lett. [**112**]{} (2014) 5, 051601, [arXiv:1309.5850 \[hep-th\]](http://arxiv.org/abs/1309.5850).
G. Arutyunov, R. Borsato and S. Frolov, [*S-matrix for strings on $\eta$-deformed $AdS_{5} \times S^5$*]{}, JHEP [**1404**]{} (2014) 002, [arXiv:1312.3542 \[hep-th\]](http://arxiv.org/abs/1312.3542).
.
B. Vicedo, [*Deformed integrable $\sigma$-models, classical $R$-matrices and classical exchange algebra on Drinfel’d doubles*]{}, J. Phys. A: Math. Theor. [**48**]{} (2015) 355203,\
[arXiv:1504.06303 \[hep-th\]](http://arxiv.org/abs/1504.06303).
B. Hoare and A. A. Tseytlin, [*On integrable deformations of superstring sigma models related to $AdS_n \times S^n$ supercosets*]{}, [Nucl. Phys. [**B897**]{} (2015) 448,]{}\
[arXiv:1504.07213 \[hep-th\]](http://arxiv.org/abs/1504.07213).
K. Sfetsos, K. Siampos and D. C. Thompson, [*Generalised integrable $\lambda$- and $\eta$-deformations and their relation*]{}, Nucl. Phys. [**B899**]{} (2015) 489,\
[arXiv:1506.05784 \[hep-th\]](http://arxiv.org/abs/1506.05784).
C. Klimčík, [*$\eta$ and $\lambda$ deformations as ${\cal E}$-models*]{}, [ Nucl. Phys. [**B900**]{} (2015) 259]{},\
[arXiv:1508.05832 \[hep-th\]](http://arxiv.org/abs/1508.05832).
C. Klimčík, [*Poisson–Lie T-duals of the bi-Yang–Baxter models*]{}, Phys. Lett. [**B760**]{} (2016) 345, [arXiv:1606.03016 \[hep-th\]](https://arxiv.org/abs/1606.03016).
C. Klimcik and P. Severa, [*Poisson-Lie T duality and loop groups of Drinfeld doubles*]{} Phys. Lett. [**B372**]{} (1996) 65 [hep-th/9512040.](https://arxiv.org/abs/hep-th/9512040)
K. Sfetsos and D. C. Thompson, [*On non-abelian T-dual geometries with Ramond fluxes*]{}, Nucl. Phys. [**B846**]{} (2011) 21, [arXiv:1012.1320 \[hep-th\]](http://arxiv.org/abs/1012.1320).
G. Itsios, C. Nunez, K. Sfetsos and D. C. Thompson, [*On Non-Abelian T-Duality and new N=1 backgrounds*]{}, Phys. Lett. [**B721**]{} (2013) 342, [arXiv:1212.4840 \[hep-th\]](http://arxiv.org/abs/1212.4840).
G. Itsios, C. Nunez, K. Sfetsos and D.C. Thompson, [*Non-Abelian T-duality and the AdS/CFT correspondence:new N=1 backgrounds*]{}, Nucl. Phys. [**B873**]{} (2013) 1,\
[arXiv:1301.6755 \[hep-th\]](http://arxiv.org/abs/1301.6755).
Y. Lozano and C. Nunez, [*Field theory aspects of non-Abelian T-duality and $ \mathcal{N}$ = 2 linear quivers*]{}, JHEP [**1605**]{} (2016) 107, [arXiv:1603.04440 \[hep-th\]](http://arxiv.org/abs/1603.04440).
Y. Lozano, N. T. Macpherson, J. Montero and C. Nunez, [*Three-dimensional $ \mathcal{N}$ = 4 linear quivers and non-Abelian T-duals*]{}, JHEP [**1611**]{} (2016) 133,\
[arXiv:1609.09061 \[hep-th\]](http://arxiv.org/abs/1609.09061).
G. Itsios, C. Nunez and D. Zoakos, [*Mesons from (non) Abelian T-dual backgrounds*]{}, JHEP [**1701**]{} (2017) 011, [arXiv:1611.03490 \[hep-th\]](http://arxiv.org/abs/1611.03490).
Y. Lozano, C. Nunez and S. Zacarias, [*BMN Vacua, Superstars and Non-Abelian T-duality*]{}, [arXiv:1703.00417 \[hep-th\]](http://arxiv.org/abs/1703.00417).
A. M. Polyakov and P. B. Wiegmann, [*Theory of Nonabelian Goldstone Bosons*]{},\
[Phys. Lett. [**B131**]{} (1983) 121](http://www.sciencedirect.com/science/article/pii/0370269383911048?via%3Dihub).
K. Sfetsos and K. Siampos, [*The anisotropic $\lambda$-deformed $SU(2)$ model is integrable*]{}, Phys. Lett. [**B743**]{} (2015) 160, [arXiv:1412.5181 \[hep-th\]](https://arxiv.org/abs/1412.5181).
C. Appadu and T. J. Hollowood, [*Beta function of k deformed ${\text AdS}_{5} \times \mathcal{S}^5$ string theory*]{}, JHEP [**1511**]{} (2015) 095, [arXiv:1507.05420 \[hep-th\]](http://arxiv.org/abs/arXiv:1507.05420).
[^1]: To conform with the notation of the current work we have renamed the gauged fields $(A_\pm,B_\pm)$ of [@Georgiou:2016urf] by $(A^{(1)}_{\pm},A^{(2)}_{\pm})$.
[^2]: The various covariant derivatives are defined according to the transformation properties of the object they act on. For instance $$\nabla_\pm g_1= \del_\pm g_1 -A^{(1)}_{\pm} g_i + g_iA^{(2)}_{\pm}\,,\quad
\nabla_\pm (\nabla_\mp g_1 g_1^{-1})=
\del_\pm(\nabla_\mp g_1 g_1^{-1}) -[A^{(1)}_\pm,\nabla_\mp g_1 g_1^{-1}]\,.$$
| {
"pile_set_name": "ArXiv"
} | ArXiv |
---
abstract: 'The present paper formulates and solves a problem of dividing coins. The basic form of the problem seeks the set of the possible ways of dividing coins of face values $1,2,4,8,\ldots$ between three people. We show that this set possesses a nested structure like the Sierpinski-gasket fractal. For a set of coins with face values power of $r$, the number of layers of the gasket becomes $r$. A higher-dimensional Sierpinski gasket is obtained if the number of people is more than three. In addition to Sierpinski-type fractals, the Cantor set is also obtained in dividing an incomplete coin set between two people.'
address: 'Department of Physics, Faculty of Science and Engineering, Chuo University, Kasuga, Bunkyo, Tokyo 112-8851, Japan'
author:
- Ken Yamamoto
title: Fractal patterns related to dividing coins
---
iterated function system, coin dividing, Sierpinski gasket, Cantor set
Introduction {#sec1}
============
Money is a mathematical system so familiar to us. It is a very good instance of combinatorics and discrete mathematics.
Telser [@Telser] considered the problem of optimal currency, and deduced that the optimal currency system consists of denominations with face values of $1, 3, 9,\ldots, 3^{m-1}$. This result comes from the problem of Bachet which seeks the smallest number of weights so that they can weigh any integer quantity on a two-pan balance. Many actual currency systems take their average multiples close to three: $2.8$ by Wynne [@Wynne] and $2.60$ by Tschoegl [@Tschoegl]. This fact is surprising because most currencies are established based on the decimal system, which is not compatible with Telser’s power-of-three theory, and because individual customs and culture of each country are reflected to its currency.
One of the classical and famous problems is the Frobenius coin problem, which seeks the largest amount of money that cannot be made using only coins of specified denominations [@RamirezAlfonsin2005]. The solution to this problem is called the Frobenius number. For only two types of coins of denominations $a_1$ and $a_2$ which are relatively prime, the Frobenius number is given by $a_1a_2-a_1-a_2$ [@Sylvester]. However, closed expressions are not known for more than two types of coins, and the problem was found to be NP-hard [@RamirezAlfonsin1996].
Another classical problem about a coin is the change-making problem. It asks how a given amount of money can be made with the least number of coins of given denominations. This problem is a variant of the knapsack problem [@Kellerer], and is also an NP-hard problem [@Lueker]. The greedy algorithm [@Magazine] and other heuristic methods such as dynamic programming [@Martello] give the optimal solution in some cases. A coin system is said to be [*canonical*]{} if the greedy algorithm works correctly [@Pearson], and almost all currencies in the world are arranged to be canonical. Even with a canonical coin system, which is easy to study, the change-making process possesses a rich mathematical structure. If we repeatedly pay money so that the number of coins in the purse after each payment is minimized, a fractal pattern is obtained from a sequence of change amounts [@Yamamoto2012; @Yamamoto2013].
The present article develops a generation mechanism of a fractal set in the form of dividing coins between people. The problem is very simple in appearance, but the possible ways of division form a nontrivial fractal pattern like the Sierpinski gasket and Cantor set. The basic situation, discussed in [§\[sec2\]]{}, is that three people divide coins of face values $1,2,4,\ldots,2^{m-1}$, and the Sierpinski gasket appears as the attractor. The emergence of the Sierpinski gasket has been previously reported [@Yamamoto], but in the present article we rigorously formulate the fractal structure and diverse generalizations in the framework of iterated function systems. In addition to the Sierpinski gasket, the Cantor set also appears in a specific case.
The notion of fractal was invented by Mandelbrot [@Mandelbrot] to measure the morphology of natural objects. Fractal theory has been applied to various phenomena such as in physics, economics, and biology. In naive description, a fractal object is created by defining a starting shape (an [*initiator*]{}) and replacing each part with another shape called a [*generator*]{}, ad infinitum [@Feder]. Mathematically, a fractal is defined using an iterated function system (IFS for short). Here we briefly review construction of the Sierpinski gasket and Cantor set (see Barnsley [@Barnsley] for detail). Let ${\bm{p}}_1,{\bm{p}}_2,{\bm{p}}_3$ be three points in ${\mathbb{R}}^d$ which are not collinear. Let three contraction maps $f_1,f_2,f_3:{\mathbb{R}}^d\to{\mathbb{R}}^d$ be given by $f_i({\bm{x}})=({\bm{x}}+{\bm{p}}_i)/2$ for $i=1,2,3$. The set of contraction maps $\{f_1,f_2,f_3\}$ is called an IFS. We let $\mathcal{H}({\mathbb{R}}^d)$ denote the collection of all nonempty compact subsets of ${\mathbb{R}}^d$.
It is well known that $\mathcal{H}({\mathbb{R}}^d)$ is a complete metric space with Hausdorff metric [@Barnsley]. The IFS $\{f_1,f_2,f_3\}$ defines a map $F:\mathcal{H}({\mathbb{R}}^d)\ni K
\mapsto f_1(K)\cup f_2(K)\cup f_3(K)\in\mathcal{H}({\mathbb{R}}^d)$ which is a contraction on $\mathcal{H}({\mathbb{R}}^d)$. The Sierpinski gasket ${\triangle}(\in\mathcal{H}({\mathbb{R}}^d))$ is defined as the unique fixed point of $F$ which obeys ${\triangle}=F({\triangle})=f_1({\triangle})\cup f_2({\triangle})\cup f_3({\triangle})$; existence and uniqueness are guaranteed by the contraction mapping theorem (or the Banach fixed-point theorem). The fixed point ${\triangle}$ also satisfies ${\triangle}=\lim_{m\to\infty}F^m(K)$ for any $K\in\mathcal{H}({\mathbb{R}}^d)$; in this sense, ${\triangle}$ is called the [*attractor*]{} of the IFS. The three points ${\bm{p}}_1$, ${\bm{p}}_2$, and ${\bm{p}}_3$ are at the three corners of ${\triangle}$. In a similar way, the Cantor set between ${\bm{p}}_1$ and ${\bm{p}}_2$ is the attractor of an IFS $\{g_1,g_2:{\mathbb{R}}^d\to{\mathbb{R}}^d\}$ where $g_i({\bm{x}})=({\bm{x}}+{\bm{p}}_i)/3$ for $i=1,2$.
Formulation of the problem and a basic solution {#sec2}
===============================================
This paper treats of a problem related to dividing a set of coins by some “players”. In order to specify the set of coins, we introduce a notation $$\begin{pmatrix}
v_1 & \cdots & v_m\\ c_1 & \cdots & c_m
\end{pmatrix},$$ where $v_1,\ldots,v_m$ is the list of face values of the coins, and $c_i$ is the number of coins of face value $v_i$. For a coin set $S=\begin{pmatrix}v_1 & \cdots & v_m\\ c_1 & \cdots & c_m\end{pmatrix}$, we define ${|S|}$ as the total amount of coins in $S$, given by $${|S|}:=\sum_{i=1}^m v_i c_i.$$
For example, the set of coins $$S=\begin{pmatrix}1 & 10\\ 2 & 1\end{pmatrix}$$ consists of two pennies (1 cent coins) and one dime (10 cent coin). When two players $A$ and $B$ divide these coins, the division of these coins is expressed by a pair $(n_A, n_B)$ of money amounts that $A$ and $B$ respectively receive. There are six ways to divide the coins in this case: $$(n_A,n_B)=(0,12), (1,11), (2,10), (10,2), (11,1), (12,0).$$ Note that we admit cases in which some players receive no coins. In this paper, we mainly focus on the coin-dividing problem between three players.
We start from the most basic form of the problem. Let us suppose a “binary” currency system $$S_{2,m}=\begin{pmatrix}1 & 2 & 4 &\cdots& 2^{m-1}\\
1& 1& 1&\cdots& 1\end{pmatrix},$$ and three players $A$, $B$, and $C$ divide these coins. Each way of division is represented by lining each player’s share as the triplet $(n_A,n_B,n_C)$, which defines a point in a three-dimensional space. We study the structure of the possible points $(n_A,n_B,n_C)$.
This problem can be described by weights. If there are $m$ weights of $1, 2,\ldots, 2^{m-1}$ grams, and one separates them into three groups (admitting one or two null groups), then we study the shape of the possible groupings. The equivalence of coins and weights resembles Telser’s theory of optimal currency (see the beginning of [§\[sec1\]]{}).
Formally, an orthonormal basis $\{{\bm{e}}_A,{\bm{e}}_B,{\bm{e}}_C\}$ is taken so that $(n_A, n_B, n_C)=n_A{\bm{e}}_A+n_B{\bm{e}}_B+n_C{\bm{e}}_C$. By definition each division satisfies $n_A+n_B+n_C=2^m-1$ and $n_A,n_B,n_C\ge0$. Hence, all the possible points $(n_A,n_B,n_C)$ are confined in a triangular area spanned by $(2^m-1){\bm{e}}_A$, $(2^m-1){\bm{e}}_B$, and $(2^m-1){\bm{e}}_C$. Let ${\triangle}_m (\subset{\mathbb{R}}^3)$ denote the set of the points of possible division $(n_A,n_B,n_C)$. For the simplest case $m=1$, three players divide only one coin of face value 1. The possible ways of division are $(n_A,n_B,n_C)=(1,0,0)$, $(0,1,0)$, and $(0,0,1)$, and they form an equilateral triangle in the three dimensional space (Fig. \[fig1\](a)): $${\triangle}_1=\{(1,0,0), (0,1,0), (0,0,1)\}
=\{{\bm{e}}_A, {\bm{e}}_B, {\bm{e}}_C\}.$$
![ (a) The set ${\triangle}_1$ consists of three points corresponding to the basis vectors ${\bm{e}}_A$, ${\bm{e}}_B$ and ${\bm{e}}_C$, which form an equilateral triangle (the dashed lines). (b) ${\triangle}_2$ is composed of three copies of ${\triangle}_1$ placed at the three corners. (c) ${\triangle}_7$ looks like the Sierpinski gasket. []{data-label="fig1"}](fig1b.eps "fig:") ![ (a) The set ${\triangle}_1$ consists of three points corresponding to the basis vectors ${\bm{e}}_A$, ${\bm{e}}_B$ and ${\bm{e}}_C$, which form an equilateral triangle (the dashed lines). (b) ${\triangle}_2$ is composed of three copies of ${\triangle}_1$ placed at the three corners. (c) ${\triangle}_7$ looks like the Sierpinski gasket. []{data-label="fig1"}](fig1c.eps "fig:")
Next we take into account the coin of face value 2 to advance to $m=2$. If the player $A$ receives this coin, the set of possible division is obtained by considering who receives the coin of face value 1. This set is written as $\{(3,0,0), (2,1,0),(2,0,1)\}={\triangle}_1+2{\bm{e}}_A$, which is translation of ${\triangle}_1$. Similar expressions hold for $B$ and $C$, so ${\triangle}_2$ is given by $${\triangle}_2=({\triangle}_1+2{\bm{e}}_A)\cup
({\triangle}_1+2{\bm{e}}_B)\cup
({\triangle}_1+2{\bm{e}}_C).$$ That is, ${\triangle}_2$ consists of three copies of ${\triangle}_1$, each of which is placed at one of the three corners (see Fig. \[fig1\](b) for reference). Similarly, by considering who receives the coin of face value $2^{m-1}$, ${\triangle}_m$ is inductively given by $${\triangle}_m=({\triangle}_{m-1}+2^{m-1}{\bm{e}}_A)\cup
({\triangle}_{m-1}+2^{m-1}{\bm{e}}_B)\cup
({\triangle}_{m-1}+2^{m-1}{\bm{e}}_C).
\label{eq:induction}$$ For each $m\ge2$, three copies of ${\triangle}_{m-1}$ are placed triangularly to form ${\triangle}_m$. By setting ${\triangle}_0=\{(0,0,0)\}$ for convenience, Eq. holds also for $m=1$.
Intuitively, ${\triangle}_m$ achieves self-similarity as $m$ increases. Actually, ${\triangle}_7$, shown in Fig. \[fig1\] (c), is almost indistinguishable from the genuine Sierpinski gasket. The connection between ${\triangle}_m$ and the Sierpinski gasket is formulated as follows. We set three contraction maps $f_A,f_B,f_C:{\mathbb{R}}^3\to{\mathbb{R}}^3$ given by $$f_P({\bm{x}})=\frac{{\bm{x}}+{\bm{e}}_P}{2}\quad (P=A,B,C).
\label{eq:fP}$$ As stated in [§\[sec1\]]{}, the attractor ${\triangle}\in\mathcal{H}({\mathbb{R}}^3)$ of an IFS $\{f_A,f_B,f_C\}$ is the Sierpinski gasket spanned between ${\bm{e}}_A$, ${\bm{e}}_B$, and ${\bm{e}}_C$. On the other hand, multiply $2^{-m}$ to Eq. to get $$\begin{aligned}
2^{-m}{\triangle}_m
&=\left(\frac{2^{-(m-1)}{\triangle}_{m-1}+{\bm{e}}_A}{2}\right)\cup
\left(\frac{2^{-(m-1)}{\triangle}_{m-1}+{\bm{e}}_B}{2}\right)\cup
\left(\frac{2^{-(m-1)}{\triangle}_{m-1}+{\bm{e}}_C}{2}\right)\\
&=f_A(2^{-(m-1)}{\triangle}_{m-1})\cup
f_B(2^{-(m-1)}{\triangle}_{m-1})\cup f_C(2^{-(m-1)}{\triangle}_{m-1})\\
&=F(2^{-(m-1)}{\triangle}_{m-1}),\end{aligned}$$ where $F$ is the contraction map on $\mathcal{H}({\mathbb{R}}^3)$ induced from the IFS. That is, adding the $(m+1)$st coin to the coin set $S_{2,m}$ is directly represented as the action of $F$. Hence $2^{-m}{\triangle}_m=F^m({\triangle}_0)$. Because ${\triangle}_0=\{(0,0,0)\}\in\mathcal{H}({\mathbb{R}}^3)$, it follows from the contraction mapping theorem that $${\triangle}=\lim_{m\to\infty}F^m({\triangle}_0)=\lim_{m\to\infty}2^{-m}{\triangle}_m.$$ This is the result which connects the coin-dividing problem and the Sierpinski gasket.
Generalization of the coin system
=================================
In this section we consider a generalized coin-dividing problem where the coin system is not powers of two. We give the set of coins $$S_{r,m}=\begin{pmatrix}1 & r & r^2 &\cdots& r^{m-1}\\
r-1& r-1& r-1&\cdots& r-1\end{pmatrix},$$ where $r$ is an integer equal to or greater than two, and ${|S_{r,m}|}=r^m-1$.
Before working on the problem of $S_{r,m}$, we revise Eq. to derive a useful expression. By means of the result ${\triangle}_1=\{{\bm{e}}_A, {\bm{e}}_B, {\bm{e}}_C\}$, Eq. is rewritten as $${\triangle}_m
=\bigcup_{{\bm{q}}\in{\triangle}_1}({\triangle}_{m-1}+2^{m-1}{\bm{q}})$$ for $m\ge2$. A similar formula is valid for $S_{r,m}$, and hence the set ${\triangle}_{r,m}$ of the possible points corresponding to $S_{r,m}$ is inductively given by $$\begin{aligned}
{\triangle}_{r,1}
&=\{q_A{\bm{e}}_A+q_B{\bm{e}}_B+q_C{\bm{e}}_C|
q_A,q_B,q_C\in{\mathbb{N}}\cup\{0\}, q_A+q_B+q_C =r-1\}, \label{eq:triangle_r1}\\
{\triangle}_{r,m}&=\bigcup_{{\bm{q}}\in{\triangle}_{r,1}}
({\triangle}_{r,m-1}+r^{m-1}{\bm{q}}).
\label{eq:triangle_rm}\end{aligned}$$ The set ${\triangle}_{r,1}$ consists of $r(r+1)/2$ points which form a triangle, and similarly ${\triangle}_{r,m}$ consists of $r(r+1)/2$ copies of ${\triangle}_{r,m-1}$ arranged triangularly.
By multiplying $r^{-m}$ to Eq. , we have $$r^{-m}{\triangle}_{r,m}=\bigcup_{{\bm{q}}\in{\triangle}_{r,1}}
f_{{\bm{q}}}(r^{-(m-1)}{\triangle}_{r,m-1}),$$ where $f_{{\bm{q}}}:{\mathbb{R}}^3\ni{\bm{x}}\mapsto({\bm{x}}+{\bm{q}})/r\in{\mathbb{R}}^3$ is a contraction map. By the contraction mapping theorem, $r^{-m}{\triangle}_{r,m}$ converges to an attractor of IFS $\{f_{{\bm{q}}}\vert {\bm{q}}\in{\triangle}_{r,1}\}$, as $m\to\infty$. This attractor is the Sierpinski gasket with $r$ layers. We depict the coin-dividing plot corresponding to $r=3$, $4$, and $5$ in Fig. \[fig2\]. The plot is essentially the same as a class of Pascal-Sierpinski gaskets if $r$ is prime— the Pascal-Sierpinski gasket of order $p$ is obtained by coloring the number not divisible by $p$ in Pascal’s triangle [@Holter].
![ The sets of possible points $(n_A,n_B,n_C)$ of the coin set $S_{3,4}$ (a), $S_{4,3}$ (b), and $S_{5,3}$ (c). They resemble the Sierpinski gasket, but the number of layers is equal to $r$. []{data-label="fig2"}](fig2a.eps "fig:"){height="40mm"} ![ The sets of possible points $(n_A,n_B,n_C)$ of the coin set $S_{3,4}$ (a), $S_{4,3}$ (b), and $S_{5,3}$ (c). They resemble the Sierpinski gasket, but the number of layers is equal to $r$. []{data-label="fig2"}](fig2b.eps "fig:"){height="40mm"} ![ The sets of possible points $(n_A,n_B,n_C)$ of the coin set $S_{3,4}$ (a), $S_{4,3}$ (b), and $S_{5,3}$ (c). They resemble the Sierpinski gasket, but the number of layers is equal to $r$. []{data-label="fig2"}](fig2c.eps "fig:"){height="40mm"}
More generally, we further generalize the coin set $$S_{r,c,m}=\begin{pmatrix} 1 & r & \cdots & r^{m-1}\\
c & c & \cdots & c\end{pmatrix},$$ and the corresponding set ${\triangle}_{r,c,m}$ of possible division of coins. Obviously, the point set ${\triangle}_{r,c,m}$ lies on the plane given by $n_A+n_B+n_C={|S_{r,c,m}|}=c(r^m-1)/(r-1)$. Furthermore, the coin-dividing problem is symmetric with respect to the three players $A$, $B$, and $C$. Hence ${\triangle}_{r,c,m}$ is invariant under the action of any permutation of $(A,B,C)$. That is, ${\triangle}_{r,c,m}$ has left-right symmetry and three-fold rotational symmetry.
The inductive formula of $S_{r,c,m}$ is $${\triangle}_{r,c,m}
=\bigcup_{{\bm{q}}\in{\triangle}_{c+1,1}}
({\triangle}_{r,c,m-1}+r^{m-1}{\bm{q}}),$$ where ${\triangle}_{c+1,1}$ is given by Eq. . The structure of the attractor is classified into three cases according to the values of $r$ and $c$.
1. If $c<r-1$, some money amounts less than ${|S_{r,c,m}|}$ cannot be made by picking out coins from $S_{r,c,m}$. In fact, the money amount $r-1$ needs $r-1$ coins of face value 1, but there are only $c$ ($<r-1$) coins of face value 1 in $S_{r,c,m}$. Reflecting this, the attractor becomes totally disconnected.
2. If $c=r-1$, the attractor is the Sierpinski gasket with $r$ layers. It belongs to the class of finitely-ramified fractals, which means that any subset of the fractal can be disconnected by removing a finite number of points.
3. If $c>r-1$, some money amounts can be made by more than one way; for example, the money amount $r$ can be made by one coin of face value $r$ or by $r$ coins of face value 1. The attractor becomes infinitely ramified.
Figure \[fig3\] shows the structural difference by the magnitude of $c$ and $r-1$.
![ Illustration of ${\triangle}_{r,c,m}$ with $c=3$, $m=3$, and $r=5,4,3$. Three types of fractals appear according to $c$ and $r-1$. (a) $r=5$ ($c<r-1$). The pattern is totally disconnected. (b) $r=4$ ($c=r-1$). The pattern is finitely ramified. (c) $r=3$ ($c>r-1$). The pattern is infinitely ramified. []{data-label="fig3"}](fig3a.eps "fig:"){height="40mm"} ![ Illustration of ${\triangle}_{r,c,m}$ with $c=3$, $m=3$, and $r=5,4,3$. Three types of fractals appear according to $c$ and $r-1$. (a) $r=5$ ($c<r-1$). The pattern is totally disconnected. (b) $r=4$ ($c=r-1$). The pattern is finitely ramified. (c) $r=3$ ($c>r-1$). The pattern is infinitely ramified. []{data-label="fig3"}](fig3b.eps "fig:"){height="40mm"} ![ Illustration of ${\triangle}_{r,c,m}$ with $c=3$, $m=3$, and $r=5,4,3$. Three types of fractals appear according to $c$ and $r-1$. (a) $r=5$ ($c<r-1$). The pattern is totally disconnected. (b) $r=4$ ($c=r-1$). The pattern is finitely ramified. (c) $r=3$ ($c>r-1$). The pattern is infinitely ramified. []{data-label="fig3"}](fig3c.eps "fig:"){height="40mm"}
The Hausdorff dimension of the corresponding attractor is easily derived for $c\le r-1$. The whole pattern can be decomposed into the $(c+1)(c+2)/2$ smaller copies of itself scaled by a factor $r$, so the similarity dimension $D_{r,c}$, depending on $r$ and $c$, is given by $$D_{r,c}=\frac{\ln((c+1)(c+2)/2)}{\ln r}.$$ Since $(c+1)(c+2)/2<r^2$ for any $r\ge2$ and $c\le r-1$, we have $D_{r,c}<2$. In particular, $D_{2,1}=\ln3/\ln2$ ($\approx1.585$) is the Hausdorff dimension of the ordinary Sierpinski gasket. We recall here that $(c+1)(c+2)/2$ is the number of contraction maps of the IFS, and $1/r$ is the contraction ratio of each map in the IFS. For $c>r-1$, the dimension is not calculated easily because of overlap. The Hausdorff dimension of an overlapping Sierpinski gasket has been derived only in restricted situations [@Broomhead].
In the case of $c>r-1$, a point of ${\triangle}_{r,c,m}$ can correspond to more than one ways of coin division. We can decompose ${\triangle}_{r,c,m}$ into subsets according to this multiplicity. Figure \[fig4\] shows an example of this decomposition where $(c,r,m)=(3,3,4)$. The highest multiplicity in this case is nine, so ${\triangle}_{r,c,m}$ splits into nine subsets, each of which has a fractal shape. A subset becomes sparse in high multiplicity, and we only present the lowest four subsets in the figure. We expect that this multi-level structure gives insight to analyses of an overlapping Sierpinski gasket.
![ Decomposition of ${\triangle}_{3,3,4}$ according to the multiplicity of each point. The subsets of multiplicity 1, 2, 3, and 4 are shown. []{data-label="fig4"}](fig4.eps)
For reference, we consider the coin system of the US dollar: 1, 5, 10, 25, 50 cent coins. We choose a set of coins as $$S_{\mathrm{cent}}=\begin{pmatrix}
1 & 5 & 10 & 25 & 50\\
4 & 1 & 2 & 1 & 1
\end{pmatrix},$$ so that each money amount from 1 to ${|S_{\mathrm{cent}}|}=104$ cents can be made by some coins of $S_{\mathrm{cent}}$. The set of possible division is shown in Fig. \[fig5\], which is obtained by enumerating all division directly. A hierarchical structure can be seen but not perfectly. Breaking of the hierarchical structure is due to the property that some money amounts can be made in two ways, e.g., $25=25\times1=5\times1+10\times2$ and $50=50\times1=5\times1+10\times2+25\times1$.
![ The coin-dividing set of the coin $S_{\mathrm{cent}}$. A hierarchical structure breaks a little. []{data-label="fig5"}](fig5.eps)
Generalization of the number of players
=======================================
In this section, the number of players is changed. For simplicity, we mainly deal with the binary coin set $S_{2,m}$ as in the most basic type of the problem stated in [§\[sec2\]]{}.
We start from the coin dividing between four players $A$, $B$, $C$, and $D$. Each way of division, written as a quadruplet $(n_A, n_B, n_C, n_D)$, gives a point in a four-dimensional space, and we represent the point as $(n_A, n_B, n_C, n_D)
=n_A{\bm{e}}_A+n_B{\bm{e}}_B+n_C{\bm{e}}_C+n_D{\bm{e}}_D$. Let us denote by ${\triangle}_m^{(4)}$ the point set corresponding to the possible division of the coins. The superscript “$(4)$” represents explicitly the number of players.
By using an argument similar to that in [§\[sec2\]]{}, ${\triangle}_m^{(4)}$ is determined inductively as $$\begin{aligned}
{\triangle}_1^{(4)}&=\{{\bm{e}}_A, {\bm{e}}_B, {\bm{e}}_C, {\bm{e}}_D\},\\
{\triangle}_m^{(4)}&=({\triangle}_{m-1}^{(4)}+2^{m-1}{\bm{e}}_A)\cup
({\triangle}_{m-1}^{(4)}+2^{m-1}{\bm{e}}_B)\cup
({\triangle}_{m-1}^{(4)}+2^{m-1}{\bm{e}}_C)\cup
({\triangle}_{m-1}^{(4)}+2^{m-1}{\bm{e}}_D).\end{aligned}$$ The first equation signifies that ${\triangle}_1^{(4)}$ consists of four points corresponding to who receives the coin of face value 1. ${\triangle}_1^{(4)}$ forms the vertices of a regular tetrahedron whose sides are $\|{\bm{e}}_A-{\bm{e}}_B\|=\|{\bm{e}}_A-{\bm{e}}_C\|
=\cdots=\|{\bm{e}}_C-{\bm{e}}_D\|=\sqrt{2}$. The second equation signifies that ${\triangle}_m^{(4)}$ is made up of four subsets corresponding to who receives the coin of face value $2^{m-1}$, which generates a nested structure of tetrahedra.
${\triangle}_m^{(4)}$ is a point set in the four-dimensional space, but it lies on a three-dimensional affine hyperplane given by $n_A+n_B+n_C+n_D=2^m-1$. Hence we can visualize ${\triangle}_m^{(4)}$ by taking a three-dimensional coordinate suitably. Setting three vectors ${\bm{u}}_1:={\bm{e}}_A-{\bm{e}}_D$, ${\bm{u}}_2:={\bm{e}}_B-{\bm{e}}_D$, and ${\bm{u}}_3:={\bm{e}}_C-{\bm{e}}_D$, we get $$\begin{aligned}
n_A{\bm{e}}_A+n_B{\bm{e}}_B+n_C{\bm{e}}_C+n_D{\bm{e}}_D
&=n_A({\bm{e}}_A-{\bm{e}}_D)+n_B({\bm{e}}_B-{\bm{e}}_D)
+n_C({\bm{e}}_C-{\bm{e}}_D)+(2^m-1){\bm{e}}_D\\
&=n_A{\bm{u}}_1+n_B{\bm{u}}_2+n_C{\bm{u}}_3+(2^m-1){\bm{e}}_D.\end{aligned}$$ That is, $(n_A,n_B,n_C,n_D)\in\operatorname{span}\{{\bm{u}}_1,{\bm{u}}_2,{\bm{u}}_3\}+(2^m-1){\bm{e}}_D$. One can find an orthonormal basis $\{{\bm{v}}_1,{\bm{v}}_2,{\bm{v}}_3\}$ of subspace $\operatorname{span}\{{\bm{u}}_1, {\bm{u}}_2, {\bm{u}}_3\}$ by the Gram-Schmidt process, as $${\bm{v}}_1=\frac{{\bm{u}}_1}{\sqrt{2}},\quad
{\bm{v}}_2=\frac{2{\bm{u}}_2-{\bm{u}}_1}{\sqrt{6}},\quad
{\bm{v}}_3=\frac{3{\bm{u}}_3-{\bm{u}}_2-{\bm{u}}_1}{2\sqrt{3}}.$$ Therefore, we have an isometric embedding $${\triangle}_m^{(4)}\ni(n_A,n_B,n_C,n_D)\mapsto
\left(\frac{2n_A+n_B+n_C}{\sqrt{2}},\frac{3n_B+n_C}{\sqrt{6}},
\frac{2n_C}{\sqrt{3}}\right)
\in{\mathbb{R}}^3.
\label{eq:GramSchmidtmap}$$ The $i$-th component on the right-hand side is the scalar product of $n_A{\bm{u}}_1+n_B{\bm{u}}_2+n_C{\bm{u}}_3$ and ${\bm{v}}_i$ ($i=1,2,3$). In Fig. \[fig6\] (a), we illustrate ${\triangle}_m^{(4)}$ with $m=6$ embedded in a three-dimensional space by using mapping . ${\triangle}_m^{(4)}$ is a finite approximation of the three-dimensional Sierpinski gasket or so-called Sierpinski tetrahedron, which is intuitively obtained by iterating the removal process shown in Fig. \[fig6\] (b). By analogy with the results in the previous section, the change of a coin system affects the number of layers of the Sierpinski tetrahedron.
![ (a) A three-dimensional illustration of ${\triangle}_m^{(4)}$ with $m=6$ using mapping . It forms a Sierpinski tetrahedron. (b) The iteration scheme of the Sierpinski tetrahedron. []{data-label="fig6"}](fig6a.eps)
The coin-dividing problem between $s(>4)$ players is considered as well. Taking an orthonormal basis $\{{\bm{e}}_1,\ldots,{\bm{e}}_s\}$ of an $s$-dimensional space, we have a recursive equation of ${\triangle}_m^{(s)}$ $${\triangle}_1^{(s)}=\{{\bm{e}}_1,\ldots,{\bm{e}}_s\},\quad
{\triangle}_m^{(s)}
=\bigcup_{\sigma=1}^s({\triangle}_{m-1}^{(s)}+2^{m-1}{\bm{e}}_\sigma).$$ ${\triangle}_m^{(s)}$ becomes $(s-1)$-dimensional Sierpinski gasket as $m$ increases, but we cannot visualize such a high-dimensional object.
Of course, we can treat of the coin-dividing problem between one or two players. For the dividing by one player $A$, there is only one [*trivial*]{} dividing of which all coins go to $A$; therefore, ${\triangle}_m^{(1)}$ is a one-point set. As for the dividing between two players $A$ and $B$, the solution is ${\triangle}_m^{(2)}=\{(n_A,n_B)|n_A=0,1,\ldots,2^m-1, n_A+n_B=2^{m}-1\}$, consisting of $2^m$ points which are equally spaced. The inductive relation in this case is $${\triangle}_m^{(2)}=({\triangle}_{m-1}^{(2)}+2^{m-1}{\bm{e}}_A)\cup
({\triangle}_{m-1}^{(2)}+2^{m-1}{\bm{e}}_B),$$ which is converted to $$2^{-m}{\triangle}_m^{(2)}=f_A(2^{-(m-1)}{\triangle}_{m-1})\cup f_B(2^{-(m-1)}{\triangle}_{m-1}),$$ where $f_A$ and $f_B$ are given in Eq. . It is easy to find that the attractor of IFS $\{f_A,f_B\}$ is a line segment $\{(n_A,n_B)| n_A\ge0, n_B\ge0, n_A+n_B=1\}$. In conclusion, nontrivial fractal pattern is not observed in these two cases. However, we explain in the next section that coin-dividing between two players with a suitable coin set can generate the Cantor set.
Cantor set between two players
==============================
In the previous section, no interesting fractal structure emerges in coin dividing between two players. We see in this section that the Cantor set is obtained if the coin set is not complete.
Let us study on trial the coin-dividing problem by two players $A$ and $B$ with the set of coins $$S_{3,1,m} =
\begin{pmatrix}1 & 3 & 9 &\cdots& 3^{m-1}\\
1 & 1 & 1&\cdots& 1\end{pmatrix}.$$ Some money amounts, e.g., 2, 5, 6, 7, and 8, cannot be made using these coins— this [*incompleteness*]{} is a different property from $S_{2,m}(=S_{2,1,m})$. We let ${\mathcal{C}}_m$ denote the set of pairs $(n_A, n_B)$ of dividing coins $S_{3,1,m}$.
The inductive relation of ${\mathcal{C}}_m$ is $${\mathcal{C}}_m=({\mathcal{C}}_{m-1}+3^{m-1}{\bm{e}}_A)
\cup({\mathcal{C}}_{m-1}+3^{m-1}{\bm{e}}_B),$$ and is equivalently $$3^{-m}{\mathcal{C}}_m=\left(\frac{3^{-(m-1)}{\mathcal{C}}_{m-1}+{\bm{e}}_A}{3}\right)
\cup\left(\frac{3^{-(m-1)}{\mathcal{C}}_{m-1}+{\bm{e}}_B}{3}\right)
=g_A(3^{-(m-1)}{\mathcal{C}}_{m-1})\cup g_B(3^{-(m-1)}{\mathcal{C}}_{m-1}),$$ where $g_A({\bm{x}})=({\bm{x}}+{\bm{e}}_A)/3$ and $g_B({\bm{x}})=({\bm{x}}+{\bm{e}}_B)/3$ are contraction maps. The set $3^{-m}{\mathcal{C}}_m$ converges to the attractor of IFS $\{g_A, g_B\}$, and as stated in [§\[sec1\]]{}, it is the Cantor set spanned between ${\bm{e}}_A$ and ${\bm{e}}_B$. In addition, ${\mathcal{C}}_m$ can be regarded as a subset of ${\triangle}_{3,1,m}$ formed by the points of $n_C=0$, or the intersecting points of ${\triangle}_{3,1,m}$ and the $(n_A,n_B)$ plane (see Fig. \[fig7\] for reference).
Let us study more about the coin-dividing problem and the Cantor set. We introduce a mapping $\varphi_m:{\mathcal{C}}_m\ni(n_A,n_B)\mapsto 2\cdot3^{-m}n_A\in [0,1]$. $\varphi_m$ is the composite mapping of a projection $\pi:(x,y)\mapsto x$ and a scaling function $\rho_m:x\mapsto 2x/3^m$. By definition, money amount $n_A$ is written as $n_A=\sum_{k=0}^{m-1}\chi_k 3^k$, where $\chi_k\in\{0,1\}$ is the indicator of whether the coin of face value $3^k$ goes to $A$ or not. Thus, $$\varphi_m({\mathcal{C}}_m)
=\left\{\left.\sum_{k=0}^{m-1}\frac{2\chi_k}{3^{m-k}}
\right|\chi_k\in\{0,1\}\right\}.$$ $\varphi_m({\mathcal{C}}_m)$ consists of numbers within $[0,1]$ whose base-3 representations are up to $m$ digits with entirely 0s and 2s, and formally $\lim_{m\to\infty}\varphi({\mathcal{C}}_m)\subset[0,1]$ consists of numbers whose base-3 representation are entirely 0s and 2s. As is known well [@Peitgen], this is identical with another definition of the Cantor set.
![ (a) The possible division of $S_{3,1,4}$. For ease of display, two axes are chosen to $(2n_A,2n_B)$. (b) The intersecting points of ${\triangle}_{3,1,4}$ and the $(n_A,n_B)$-plane is ${\mathcal{C}}_4$. []{data-label="fig7"}](fig7a.eps "fig:") ![ (a) The possible division of $S_{3,1,4}$. For ease of display, two axes are chosen to $(2n_A,2n_B)$. (b) The intersecting points of ${\triangle}_{3,1,4}$ and the $(n_A,n_B)$-plane is ${\mathcal{C}}_4$. []{data-label="fig7"}](fig7b.eps "fig:")
Generalization to the coin set $$S_{r,1,m}=\begin{pmatrix}1 & r & r^2 &\cdots& r^{m-1}\\
1 & 1 & 1&\cdots& 1\end{pmatrix}$$ is straightforwardly. By way of the inductive relation for the set ${\mathcal{C}}_{r,m}$ of division, $r^{-m}{\mathcal{C}}_{r,m}$ converges to a Cantor-like fractal whose Hausdorff dimension is $$D_r=\frac{\ln2}{\ln r}.$$ In particular, $D_3=\ln2/\ln3\approx0.631$ is the dimension of the Cantor set. Moreover, $D_2=\ln2/\ln2=1$ is consistent with the result that the attractor of coin-dividing of $S_{2,m}(=S_{2,1,m})$ becomes a line segment, as stated at the end of the previous section.
Conclusion
==========
In this article, we have treated of the [*coin-dividing problem*]{} which seeks the set of the possible division of a set of coins. By considering an appropriate scaling limit, the set of points in dividing of coins $S_{r,c,m}$ converges to a fractal set as $m$ tends to infinity. This result follows from a remarkable property that increment of the coin types $m$ represents the action of an iterated function system. The parameters $r$ and $c$ of $S_{r,c,m}$ are related to the contraction ratio and the number of maps of the iterated function system, respectively. Depending on the magnitude of $c$ and $r-1$, the coin-dividing fractal belongs to one of three classes: totally disconnected if $c<r-1$, finitely ramified if $c=r-1$, and infinitely ramified if $c>r-1$. In particular, we have obtained the Sierpinski gasket when three players divide $S_{2,1,m}$, and the Cantor set when two players divide $S_{3,1,m}$.
Acknowledgments {#acknowledgments .unnumbered}
===============
The author is very grateful to Dr. Makoto Katori for fruitful and instructive discussion. The author is also grateful to Dr. Yoshihiro Yamazaki and Dr. Jun-ichi Wakita for their comments and discussion.
[99]{} L. G. Telser, Optimal denominations for coins and currency, Econ. Lett. 49 (1995) 425–427. M. A. Wynne, More on optimal denominations for coins and currency, Econ. Lett. 55 (1997) 221–225. A. E. Tschoegl, The optimal denomination of currency, J. Money, Credit and Banking 29 (1997) 546–554. J. L. Ramírez-Alfonsín, The Diophantine Frobenius Problem, Oxford University Press, Oxford, 2005. J. J. Sylvester, Question 7382, Mathematical Questions from the Educational Times 41 (1884) 21. J. L. Ramírez-Alfonsín, Complexity of the Frobenius problem, Combinatorica 16 (1996) 143–147. H. Kellerer, U. Pferschy, and D. Pisinger, Knapsack Problems, Springer, Berlin, 2004. G. S. Lueker, Two NP-complete problem in nonnegative integer programming, Report 178, Computer Science Laboratory, Princeton University, 1975. M. J. Magazine, G. L. Nemhauser, and L. E. Trotter, Jr., When the greedy solution solves a class of knapsack problems, Oper. Res. 23 (1975) 207–217. S. Martello and P. Toth, An exact algorithm for large unbounded knapsack problems, Oper. Res. Lett. 9 (1990) 15–20. D. Pearson, A polynomial-time algorithm for the change-making problem, Oper. Res. Lett. 33 (2005) 231–234. K. Yamamoto and Y. Yamazaki, Fractal behind coin-reducing payment, Chaos, Solitons & Fractals 45 (2012) 1058–1066. K. Yamamoto and Y. Yamazaki, Multifractal aspects of an efficient change-making process, Fractals 21 (2013) 1350014. K. Yamamoto, Emergence of the Sierpinski gasket in coin-dividing problems, J. Stat. Phys. 152 (2013) 534–540. B. B. Mandelbrot, The Fractal Geometry of Nature, WH Freeman, San Francisco, 1982. J. Feder, Fractals, Plenum, New York, 1988. M. F. Barnsley, Fractals Everywhere, Academic Press, Boston, 1993. T. W. Gamelin and M. A. Mnatsakanian, Arithmetic based fractals associated with Pascal’s triangle, Publ. Mat. 49 (2005) 329–349. N. S. Holter, A. Lakhtakia, V. K. Varadan, V. V. Varadan, and R. Messier, On a new class of planar fractals: the Pascal-Sierpinski gaskets, J. Phys. A: Math. Gen. 19 (1986) 1753–1759. D. Broomhead, J. Montaldi, and N. Sidorov, Golden gaskets: variations on the Sierpinski sieve, Nonlinearity 17 (2004) 1455–1480. H.-O. Peitgen, H. Jürgens, and D. Saupe, Chaos and Fractals: New Frontiers of Science, Springer, New York, 2004.
| {
"pile_set_name": "ArXiv"
} | ArXiv |
---
abstract: 'Markets composed of stocks with capitalization processes represented by positive continuous semimartingales are studied under the condition that the market excess growth rate is bounded away from zero. The following examples of these markets are given: [*i*]{}) a market with a singular covariance matrix and instantaneous relative arbitrage; [*ii*]{}) a market with a singular covariance matrix and no arbitrage; [*iii*]{}) a market with a nonsingular covariance matrix and no arbitrage; [*iv*]{}) a market with a nonsingular covariance matrix and relative arbitrage over an arbitrary time horizon.'
bibliography:
- 'math.bib'
---
**Variations on an example of Karatzas and Ruf**
Robert Fernholz[^1]
For $n\in\N$ and $T\in(0,\infty)$, consider a market composed of stocks with capitalization processes $X_1,\ldots,X_n$ represented by positive continuous semimartingales defined on $[0,T]$. give an example of such a market in which the capitalization processes are martingales, the market covariance matrix is singular, and the excess growth rate $\g^*_\m$ of the market portfolio is bounded away from zero. The condition that the capitalization processes are martingales is of interest because arbitrage is not possible in a market with martingale capitalization processes (see, e.g., ). The condition that $\g^*_\m$ is bounded away from zero is of interest because this implies the existence of relative arbitrage over long enough time horizons (see ).
Here we present five variations on the example of . The first variation is an example of a market with a singular covariance matrix and instantaneous relative arbitrage; the second is a market with a singular covariance matrix and martingale capitalization processes; the third is a market with a nonsingular covariance matrix and martingale capitalization processes; and the fourth is a market with a nonsingular covariance matrix and relative arbitrage over an arbitrary time horizon. In these four examples, the market capitalization process is a martingale; in the fifth example this condition is relaxed.
A market is [*strongly nondegenerate*]{} if the eigenvalues of its covariance matrix are bounded away from zero. In all the examples we consider here the markets are [*diverse*]{}, i.e., the market weights are all bounded away from one. In a diverse, strongly nondegenerate market, relative arbitrage exists over an arbitrary time horizon (see Section 8 of ), so the capitalization processes cannot be martingales. Hence, in our third example, nonsingularity of the market covariance matrix cannot be strengthened to strong nondegeneracy, since arbitrage is not possible in a market with martingale capitalization processes.
[**Definition 1.**]{} For $T>0$, a market defined on $[0,T]$ has [*relative arbitrage*]{} if there exist portfolios $\nu$ and $\eta$ with value processes $Z_\nu$ and $Z_\eta$ such that $$\begin{aligned}
\P\big[Z_\nu(T)/Z_\eta(T)&\ge Z_\nu(0)/Z_\eta(0)\big]=1,\\
\P\big[Z_\nu(T)/Z_\eta(T)&> Z_\nu(0)/Z_\eta(0)\big]>0.\end{aligned}$$
[**Definition 2.**]{} For $T>0$, a market defined on $[0,T]$ has [*instantaneous relative arbitrage*]{} if there exist portfolios $\nu$ and $\eta$ with value processes $Z_\nu$ and $Z_\eta$ such that for any $t_1,t_2\in[0,T]$ with $t_1<t_2$, $$\P\big[Z_\nu(t_2)/Z_\eta(t_2)> Z_\nu(t_1)/Z_\eta(t_1)\big]=1.$$
[**Example 1.**]{}
Here we give an example of a market with $\g^*_\m$ bounded away from zero and a singular covariance matrix. In this example the market weight processes are confined to a circle, and this generates instantaneous relative arbitrage.
Let $a\in(0,1/3)$ be a real constant, and consider the time horizon $[0,T]$. Let $(W,\th)$ be a 2-dimensional Brownian motion with the usual filtration $\F$, and let $(X_1,X_2,X_3)$ be the market defined by $$X_i(t)=e^{W(t)-t/2}\Big(\frac{1}{3}+a\cos\big(\th(t)+(i-1)2\p/3\big)\Big),\qquad 0\le t\le T,$$ for $i=1,2,3$. The market capitalization process $X$ is given by $$\begin{aligned}
X(t)&=X_1(t)+X_2(t)+X_3(t)\\
&= e^{W(t)-t/2},\end{aligned}$$ so the market weights are $$\label{0}
\m_i(t)=\frac{X_i(t)}{X(t)}=\frac{1}{3}+a\cos\big(\th(t)+(i-1)2\p/3\big),$$ for $i=1,2,3$. Hence, $0<\m_i(t)<2/3$, for $i=1,2,3$, and the market is diverse. Since for all $x\in\R$, $$\label{00}
\sum_{i=1}^3\sin^2\big(x+(i-1)2\p/3\big) = \sum_{i=1}^3\cos^2\big(x+(i-1)2\p/3\big) = \frac{3}{2}.$$ we have $$\label{1}
\m_1^2(t)+\m_2^2(t)+\m_3^2(t)=\frac{1}{3}+\frac{3a^2}{2}<\half,$$ so the points $(\m_1(t),\m_2(t),\m_3(t))\in\R^3$ lie on the intersection of the plane $$x_1+x_2+x_3=1$$ with the sphere of radius $\sqrt{1/3+3a^2/2}$ centered at the origin. This intersection is a circle of radius $\sqrt{3a^2/2}<\sqrt{1/6}$ centered at $(1/3,1/3,1/3)$, and this circle lies in the (open) simplex $$\D^3 \eqdef \big\{x\in\R^3: x_1+x_2+x_3 = 1, x_i>0\big\}.$$
From we see that for $i=1,2,3$, $$d\m_i(t)= -a\sin\big(\th(t)+(i-1)2\p/3\big)d\th(t) - \frac{a}{2} \cos\big(\th(t)+(i-1)2\p/3\big)dt,\as,$$ so $$d\brac{\m_i}_t = a^2\sin^2\big(\th(t)+(i-1)2\p/3\big)dt,\as,$$ and $$\begin{aligned}
\t_{ii}(t)&=\frac{1}{\m^2_i(t)}\frac{d\brac{\m_i}_t}{dt}\notag\\
&= \frac{a^2\sin^2\big(\th(t)+(i-1)2\p/3\big)}{\m^2_i(t)},\as\label{2}\end{aligned}$$
Consider the portfolio generating function $$\S(x)=\big(x_1^2+x_2^2+x_3^2\big)^{1/2}.$$ This function generates the portfolio $\p$ with weights $$\label{2.1}
\p_i(t)=\frac{\m_i^2(t)}{\m_1^2(t)+\m_2^2(t)+\m_3^2(t)}=\frac{\m_i^2(t)}{1/3+3a^2/2}$$ and with value function $Z_\p$ that satisfies $$d\log\big(Z_\p(t)/Z_\m(t)\big)=d\log \S(\m(t))-\g^*_\p(t)\,dt,\as,$$ where $Z_\m$ is the market value process (see , Example 3.1.9). Since implies that $\S(\m(t))$ is constant, this reduces to $$\label{3}
d\log\big(Z_\p(t)/Z_\m(t)\big)=-\g^*_\p(t)\,dt,\as$$ Since this has no stochastic component, the relative variance $\t_{\p\p}(t)$ vanishes for all $t\in[0,T]$, so $$\begin{aligned}
\g^*_\p(t) &= \half\Big(\sum_{i=1}^3 \p_i(t)\t_{ii}(t)-\t_{\p\p}(t)\Big)\notag\\
&= \half\sum_{i=1}^3 \p_i(t)\t_{ii}(t)\notag\\
&= \frac{a^2}{2/3+3a^2}\sum_{i=1}^3\sin^2\big(\th(t)+(i-1)2\p/3\big)\notag\\
&= \frac{3a^2}{4/3+6a^2}>0,\label{33}\end{aligned}$$ by , , and . From and it follows that in this market there is instantaneous relative arbitrage.
On the other hand, from we have $$\begin{aligned}
\m_i(t)\t_{ii}(t)&=\frac{a^2\sin^2\big(\th(t)+(i-1)2\p/3\big)}{\m_i(t)}\\
&> \frac{3a^2\sin^2\big(\th(t)+(i-1)2\p/3\big)}{2},\as,\end{aligned}$$ for $i=1,2,3$, since $0<\m_i(t)<2/3$, so implies that $$\g^*_\m(t)=\half\sum_{i=1}^3\m_i(t)\t_{ii}(t)> \frac{9a^2}{8},\as,$$ for $t\in[0,T]$.
Of the two processes that drive this market, $\th$ generates circular motion in the plane of $\D^3$ about the point $(1/3,1/3,1/3)\in\R^3$, and $W$ generates radial motion from the origin in the positive orthant. The motion due to $W$ does not lie in the plane of $\D^3$, so the rank of the market covariance matrix will be two.
In the next example, we shall show that the same conditions as in this example, i.e., $\g^*_\m$ bounded away from zero and market covariance matrix of rank two, can result in a market with no arbitrage.
[**Example 2.**]{}
Here we give an example of a market with $\g^*_\m$ bounded away from zero, a singular covariance matrix, and martingale capitalization processes. In this example the market weight processes are confined to an expanding circle, similarly to , in which the market weight processes are confined to an expanding annulus. In this example there is no arbitrage, since arbitrage is not possible in a market with martingale capitalization processes.
Let $a\in(0,1/3)$ be a real constant and consider the time horizon $[0,T]$ with $T<-2\log (3a)$. Let $(W,\th)$ be a 2-dimensional Brownian motion with the usual filtration $\F$, and let $(X_1,X_2,X_3)$ be the market defined by $$X_i(t)=e^{W(t)-t/2}\Big(\frac{1}{3}+ae^{t/2}\cos\big(\th(t)+(i-1)2\p/3\big)\Big),\qquad 0\le t\le T,$$ for $i=1,2,3$. In this case, $e^{W(t)-t/2}$ and $e^{t/2}\cos\big(\th(t)+(i-1)2\p/3\big)$ are independent $\F$-martingales, so the $X_i$ are also $\F$-martingales. As a result, arbitrage cannot exist in to this market.
The market capitalization process $X$ is given by $$\begin{aligned}
X(t)&=X_1(t)+X_2(t)+X_3(t)\\
&= e^{W(t)-t/2},\end{aligned}$$ so the market weights are $$\m_i(t)=\frac{X_i(t)}{X(t)}=\frac{1}{3}+ae^{t/2}\cos\big(\th(t)+(i-1)2\p/3\big),$$ for $i=1,2,3$. In this case, $$d\m_i(t)=- ae^{t/2}\sin\big(\th(t)+(i-1)2\p/3\big)d\th(t),\as,$$ so the $\m_i$ are $\F$-martingales.
As in Example 1, we have $0<\m_i(t)\le 1/3+ae^{t/2}<2/3$, for $i=1,2,3$, so the market is diverse. It follows from that $$\m_1^2(t)+\m_2^2(t)+\m_3^2(t)=\frac{1}{3}+\frac{3a^2 e^t}{2}<\half,\qquad 0\le t\le T,$$ so the points $(\m_1(t),\m_2(t),\m_3(t))\in\R^3$ lie on the intersection of the plane $$x_1+x_2+x_3=1$$ with the sphere of radius $\sqrt{1/3+3a^2e^t/2}$ centered at the origin. This intersection is a circle of radius $\sqrt{3a^2e^t/2}<\sqrt{1/6}$ centered at $(1/3,1/3,1/3)$, and this circle lies in the simplex $\D^3$.
Similarly to , for $i=1,2,3$, we have $$\label{4}
\t_{ii}(t) = \frac{a^2e^t\sin^2\big(\th(t)+(i-1)2\p/3\big)}{\m^2_i(t)},\as,$$ so $$\begin{aligned}
\m_i(t)\t_{ii}(t)&=\frac{a^2e^t\sin^2\big(\th(t)+(i-1)2\p/3\big)}{\m_i(t)}\\
&> \frac{3a^2e^t\sin^2\big(\th(t)+(i-1)2\p/3\big)}{2},\as,\end{aligned}$$ since $0<\m_i(t)<2/3$. Hence, it follows from that $$\g^*_\m(t)=\half\sum_{i=1}^3\m_i(t)\t_{ii}(t)> \frac{9a^2e^t}{8}\ge \frac{9a^2}{8},\as,$$ for $t\in[0,T]$.
Of the two martingales that drive this market, $\th$ generates circular motion in the plane of $\D^3$ about the point $(1/3,1/3,1/3)\in\R^3$, and $W$ generates radial motion from the origin in the positive orthant. The motion due to $W$ does not lie in the plane of $\D^3$, so the rank of the market covariance matrix will be two.
In the next example, we shall perturb the current model radially in the plane of $\D^3$, and this will result in a nonsingular market covariance matrix. This perturbation will not alter the martingale structure of the capitalization processes, so the resulting market will still not permit arbitrage.
[**Example 3.**]{}
Here we give an example of a market with $\g^*_\m$ bounded away from zero, a nonsingular covariance matrix, and martingale capitalization processes. In this example the market weight processes are confined to an expanding annulus, as in , where the market covariance matrix was singular. Although the market covariance matrix is nonsingular in this example, the market cannot be strongly nondegenerate, since strong nondegeneracy in a diverse market implies the existence of relative arbitrage over an arbitrary time horizon (see Section 8 of ), and arbitrage is not possible with martingale capitalization processes.
Let $a\in(0,1/9)$ be a real constant and consider the time horizon $[0,T]$ with $T<-2\log(9a)$. Let $(W,\th,B)$ be a 3-dimensional Brownian motion with the usual filtration $\F$, and let $(X_1,X_2,X_3)$ be the market defined by $$X_i(t)=e^{W(t)-t/2}\Big(\frac{1}{3}+\ph(t)e^{t/2}\cos\big(\th(t)+(i-1)2\p/3\big)\Big),\qquad 0\le t\le T,$$ for $i=1,2,3$, where $\ph$ is a continuous $\F$-martingale driven by $B$ such that $a<\ph(t)<3a$. We shall establish the precise structure of the process $\ph$ below. Since $e^{W(t)-t/2}$, $e^{t/2}\cos\big(\th(t)+(i-1)2\p/3\big)$, and $\ph$ are all independent $\F$-martingales, the $X_i$ will also be $\F$-martingales. As a result, arbitrage cannot exist in to this market.
The market capitalization process $X$ is given by $$\begin{aligned}
X(t)&=X_1(t)+X_2(t)+X_3(t)\\
&= e^{W(t)-t/2},\end{aligned}$$ and the market weights are $$\m_i(t)=\frac{X_i(t)}{X(t)}=\frac{1}{3}+\ph(t)e^{t/2}\cos\big(\th(t)+(i-1)2\p/3\big),$$ for $i=1,2,3$. In this case, $$d\m_i(t)=- \ph(t)e^{t/2}\sin\big(\th(t)+(i-1)2\p/3\big)d\th(t)+e^{t/2}\cos\big(\th(t)+(i-1)2\p/3\big)d\ph(t),\as$$ so the $\m_i$ are all $\F$-martingales.
As in Examples 1 and 2, we have $0<\m_i(t)< 1/3+3ae^{t/2}<2/3$, a.s., for $i=1,2,3$, so the market is a.s. diverse. It follows from that $$\m_1^2(t)+\m_2^2(t)+\m_3^2(t)=\frac{1}{3}+\frac{3\ph^2(t) e^t}{2}<\half,\as,\qquad 0\le t\le T,$$ so the points $(\m_1(t),\m_2(t),\m_3(t))\in\R^3$ lie on the intersection of the plane $$x_1+x_2+x_3=1$$ with the sphere of radius $\sqrt{1/3+3\ph^2(t)e^t/2}$ centered at the origin. This intersection is a circle of radius $\sqrt{3\ph^2(t)e^t/2}<\sqrt{1/6}$ centered at $(1/3,1/3,1/3)$, and this circle lies in the simplex $\D^3$.
Now, define the process $\ph$ by $$\label{4.1}
\ph(t)=2a+\psi(t),$$ where $$\label{4.2}
\psi(t)=\intt \big(a^2-\psi^2(s)\big)dB(s),$$ for $t\in[0,T]$, and $B$ is the Brownian motion introduced above. In this case, $\psi$ is a continuous $\F$-martingale and $$\label{5}
\P\big[-a<\psi(t)<a, t\in[0,T]\big]=1.$$ The process $\psi$ was suggested by Ioannis Karatzas. The structural details of this process can be verified in , Proposition 5.5.22(d) and Theorem 5.5.29. This process $\psi$ is a linear diffusion with state space $(-a,a)$ and visits, with positive probability, any given neighborhood $U\subset (-a,a)$ during any time interval $(0,\d]$, for $\d>0$ (see Theorem 4.8 of , and ). It follows that $\ph$ is also a continuous $\F$-martingale with $a<\ph(t)<3a$, a.s., and $$\label{6}
d\brac{\ph}_t=d\brac{\psi}_t= \big(a^2-\psi^2(t)\big)^2dt,\as$$
Similarly to and , we have for $i=1,2,3$, $$\label{5.1}
\t_{ii}(t) = \frac{\ph^2(t)e^t\sin^2\big(\th(t)+(i-1)2\p/3\big)+e^t\cos^2\big(\th(t)+(i-1)2\p/3\big)\big(a^2-\psi^2(t)\big)^2}{\m^2_i(t)},\as,$$ so $$\begin{aligned}
\m_i(t)\t_{ii}(t)&=\frac{\ph^2(t)e^t\sin^2\big(\th(t)+(i-1)2\p/3\big)+e^t\cos^2\big(\th(t)+(i-1)2\p/3\big)\big(a^2-\psi^2(t)\big)^2}{\m_i(t)}\\
&> \frac{3\ph^2(t)e^t\sin^2\big(\th(t)+(i-1)2\p/3\big)+3e^t\cos^2\big(\th(t)+(i-1)2\p/3\big)\big(a^2-\psi^2(t)\big)^2}{2},\as,\end{aligned}$$ since $0<\m_i(t)<2/3$. Hence, from we have $$\g^*_\m(t)=\half\sum_{i=1}^3\m_i(t)\t_{ii}(t)> \frac{9\ph^2(t)e^t+9e^t\big(a^2-\psi^2(t)\big)^2}{8}> \frac{9a^2}{8},\as,$$ for $t\in[0,T]$.
Of the three martingales that drive this market, $\th$ generates circular motion in the plane of $\D^3$ about the point $(1/3,1/3,1/3)\in\R^3$, $\ph$ generates radial motion from the point $(1/3,1/3,1/3)$ in the plane of $\D^3$, and $W$ generates radial motion from the origin in the positive orthant. It follows from and that $d\brac{\ph}_t/dt>0$, a.s., so these three movements span $\R^3$, and the market covariance matrix will be nonsingular. However, since $\big(a^2-\psi^2(t)\big)$ can be arbitrarily small, the market will not be strongly nondegenerate.
In the next example, we shall show that the same conditions as in this example, i.e., $\g^*_\m$ bounded away from zero and nonsingular market covariance matrix, can result in a market with relative arbitrage over an arbitrary time horizon.
[**Example 4.**]{}
Here we give an example of a market with $\g^*_\m$ bounded away from zero and a nonsingular covariance matrix. In this example the market weight processes range throughout a stationary annulus, and this time homogeneity generates relative arbitrage over an arbitrary time horizon.
Let $a\in(0,1/9)$ be a real constant and consider the time horizon $[0,T]$. Let $(W,\th,B)$ be a 3-dimensional Brownian motion with the usual filtration $\F$, and let $(X_1,X_2,X_3)$ be the market defined by $$X_i(t)=e^{W(t)-t/2}\Big(\frac{1}{3}+\ph(t)\cos\big(\th(t)+(i-1)2\p/3\big)\Big),\qquad 0\le t\le T,$$ for $i=1,2,3$, where $\ph$ is the martingale defined in and , with $a<\ph(t)< 3a$, a.s.
The market capitalization process $X$ is given by $$\begin{aligned}
X(t)&=X_1(t)+X_2(t)+X_3(t)\\
&= e^{W(t)-t/2},\end{aligned}$$ and the market weights are $$\label{12}
\m_i(t)=\frac{X_i(t)}{X(t)}=\frac{1}{3}+\ph(t)\cos\big(\th(t)+(i-1)2\p/3\big),$$ for $i=1,2,3$. In this case, $$\begin{gathered}
d\m_i(t)=- \ph(t)\sin\big(\th(t)+(i-1)2\p/3\big)d\th(t)\\+\cos\big(\th(t)+(i-1)2\p/3\big)d\ph(t)-\frac{\ph(t)}{2}\sin\big(\th(t)+(i-1)2\p/3\big)dt,\as\end{gathered}$$
As in Examples 1, 2, and 3, we have $0<\m_i(t)<1/3+3a<2/3$, a.s., for $i=1,2,3$, so the market is a.s. diverse. It follows from that $$\m_1^2(t)+\m_2^2(t)+\m_3^2(t)=\frac{1}{3}+\frac{3\ph^2(t) }{2},\qquad 0\le t\le T,$$ so $$\frac{1}{3}+\frac{3a^2}{2}< \m_1^2(t)+\m_2^2(t)+\m_3^2(t)<\frac{1}{3}+\frac{27a^2}{2}<\half,\as,$$ so the points $(\m_1(t),\m_2(t),\m_3(t))\in\R^3$ lie on the plane $$x_1+x_2+x_3=1$$ between the spheres of radius $\sqrt{1/3+3a^2/2}$ and $\sqrt{1/3+27a^2/2}$ centered at the origin. This set is the open annulus $$\label{13}
A=\big\{(x_1,x_2,x_3)\in\D^3: 3a^2/2<(x_1-1/3)^2+(x_2-1/3)^2+(x_3-1/3)^2<27a^2/2\big\},$$ which lies between the circles of radius $\sqrt{3a^2/2}$ and $3\sqrt{3a^2/2}<\sqrt{1/6}$ centered at $(1/3,1/3,1/3)$, both of which lie within the simplex $\D^3$.
Similarly to , we have for $i=1,2,3$, $$\t_{ii}(t) = \frac{\ph^2(t)\sin^2\big(\th(t)+(i-1)2\p/3\big)+\cos^2\big(\th(t)+(i-1)2\p/3\big)\big(a^2-\psi^2(t)\big)^2}{\m^2_i(t)},\as,$$ so $$\begin{aligned}
\m_i(t)\t_{ii}(t)&=\frac{\ph^2(t)\sin^2\big(\th(t)+(i-1)2\p/3\big)+\cos^2\big(\th(t)+(i-1)2\p/3\big)\big(a^2-\psi^2(t)\big)^2}{\m_i(t)}\\
&> \frac{3\ph^2(t)\sin^2\big(\th(t)+(i-1)2\p/3\big)+3\cos^2\big(\th(t)+(i-1)2\p/3\big)\big(a^2-\psi^2(t)\big)^2}{2},\as,\end{aligned}$$ since $0<\m_i(t)<2/3$. Hence, from we have $$\g^*_\m(t)=\half\sum_{i=1}^3\m_i(t)\t_{ii}(t)> \frac{9\ph^2(t)+9\big(a^2-\psi^2(t)\big)^2}{8}> \frac{9a^2}{8},\as,$$ for $t\in[0,T]$.
Of the three processes that drive this market, $\th$ generates circular motion in the plane of $\D^3$ about the point $(1/3,1/3,1/3)\in\R^3$, $\ph$ generates radial motion from the point $(1/3,1/3,1/3)$ in the plane of $\D^3$, and $W$ generates radial motion from the origin in the positive orthant. It follows from and that $d\brac{\ph}_t/dt>0$, a.s., so these three movements span $\R^3$, and the market covariance matrix will be nonsingular. However, since $\big(a^2-\psi^2(t)\big)$ can be arbitrarily small, the market will not be strongly nondegenerate.
The structure of the market weights given by results in a type of time homogeneity for $t>0$, and this produces relative arbitrage over the (arbitrary) time horizon $[0,T]$. If $A\subset\D^3$ is the annulus defined above in , then for any $t\in(0,T]$, $$\P\big[\m(t)\in A\big]=1,$$ and for any open subset $U\subset A$, $$\P\big[\m(t)\in U\big]>0$$ (see Theorem 4.8 of , and ). Hence, for any portfolio generating function $\S$ and for any $t\in(0,T]$, $$\text{ess inf}\big\{\S(\m(t))\big\}=\inf\big\{\S(x):x\in A\big\}.$$ This implies that the essential infimum of $\S(\m(t))$ is invariant over $(0,T]$, so it follows from Proposition 1 of that relative arbitrage exists in this market.
[**Example 5.**]{}
The behavior of the market in Examples 1 through 4 does not depend of the fact that the process $e^{W(t)-t/2}$ is a martingale, since the relevant structure is determined by the market weights alone. Indeed, the market capitalization process $e^{W(t)-t/2}$ could be replaced by any positive continuous semimartingale $\kappa$ that is independent of $\ph$ and $\th$, and the market weights $\m_i$ will remain unchanged. Consider, for instance, Example 3, where $$X_i(t)=e^{W(t)-t/2}\Big(\frac{1}{3}+\ph(t)e^{t/2}\cos\big(\th(t)+(i-1)2\p/3\big)\Big),\qquad 0\le t\le T.$$ In this case, the market model would become $$X_i(t)=\kappa(t)\Big(\frac{1}{3}+\ph(t)e^{t/2}\cos\big(\th(t)+(i-1)2\p/3\big)\Big),\qquad 0\le t\le T,$$ for $i=1,2,3$. Of course, the filtration $\F$ would have to be adjusted accordingly, but after that, all the analysis would remain the same. In the simplest case, the process $\kappa$, which represents the total capitalization of the market, could be set identically equal to one, and the capitalization processes $X_i$ would be the same as the market weight processes $\m_i$. This can be done in all four of the previous examples.
[^1]: INTECH, One Palmer Square, Princeton, NJ 08542. [email protected]. The author thanks René Carmona, Ioannis Karatzas, and Johannes Ruf for their invaluable comments and suggestions regarding this research.
| {
"pile_set_name": "ArXiv"
} | ArXiv |
---
abstract: |
In this paper, we take cosmological constant as a thermodynamical pressure and it’s conjugate quantity as a thermodynamical volume. This expression help us to investigate the phase transition and holographic heat engine. So, in order to have Van der Waals fluid behaviour in Horava -Lifshitz (HL) black hole, we modified the solution of such black hole with some cosmology ansatz. Also from holographic heat engine, we obtain Carnot efficiency for the HL black hole. The phase transition of the system lead us to investigate the stability condition for the corresponding black hole. In that case, we show that the stability exist only in special region of black hole.\
[**Keywords:**]{} Horava-Lifshitz black hole; Van der Waals fluid; Phase transition; Carnot efficiency.
author:
- |
[Kh. Jafarzade $^{a}$ [^1] , J. Sadeghi $^{a}$[^2]]{}\
$^a$\
title: '**The phase transition and holographic in modified Horava -Lifshitz black hole**'
---
Introduction
============
Bekenstain, Hawking and Bardeen compared the laws of black hole mechanics with the ordinary thermodynamics. They used the appropriate quantities as temperature, entropy and energy and etc and clarified all laws of black hole. In that case, they have shown that all laws of black hole mechanics are identical to the laws of the ordinary thermodynamics and hence the black hole is considered as a thermal system \[1-2\].\
In recent years, the thermodynamic properties and critical behaviour of various black holes investigated by Ref.s. \[3-7\]. But this topic will be attractive when we study black hole thermodynamic in the presence of a negative cosmological constant. The most important example is AdS black holes with various phase transitions. The physics of such black holes has been improved by the AdS/CFT correspondence. The primary study phase transition of AdS black hole is started by Hawking-Page phase transition and it demonstrated phase transition between the Schwarzschild AdS black hole and thermal AdS space \[8\]. On the other hand, one of the interesting results is investigation of phase transition Reissner-Nordstrom-AdS (RN-AdS) black hole. The study of charged AdS black holes indicated that they have behaviour like as liquid-gas system. In this case, the cosmological constant and it’s conjugate quantity treat as thermodynamic pressure and thermodynamic volume respectively.\
The influence of different topologies on the $P-V$ criticality of black hole can not be ignored in the extended phase space \[6,9-11\]. In this regard some efforts have been made such as study of thermodynamic of topological black hole solutions in Einstein gravity \[12,13\], Einstein-Gauss- Bonnet gravity, dilaton gravity \[14-16\] and Lovelock gravity \[17\]. The investigation of charged AdS black holes in Einstein gravity indicated that the critical behavior exists only for a black hole with horizon of spherical topology. But for planner and hyperbolic cases there is not any critical behavior. Similar investigation has been made for charged black hole in Gauss- Bonnet gravity and the obtained results was similar to charged AdS black holes in Einstein gravity. The research of black hole solutions in Horava-Lifshitz gravity will be interesting and also is different to Einstein gravity theory \[18-23\]. Because, the phase transition of charged (uncharged) topological black holes in Horava-Lifshitz gravity only take place in $k=-1$ (hyperbolic horizon) in the nonextended phase space. This consequence does not satisfy in the extended phase space. In this case, for all horizon topologies there is not any physical critical point for the uncharged and charged cases. Therefore, topological black hole in Horava-Lifshitz gravity like BTZ black hole, have not any phase transition in the extended phase space. So, by looking to the two black holes as BTZ and Horava-Lifshitz, one can say that the existence of cosmological constant in the black hole metric is not sufficient for the existence $P-V$ criticality. Therefore, the problem of phase transition in HL black hole give us motivation to modify the solution of such system with suitable ansatz. In that case it’s thermodynamic quantities match exactly with Van der Waals fluid equation state.\
Here, we study the P-V criticality in classical mechanic. It may be interesting to investigate such behaviour in quantum mechanic . From quantum approach, one can consider effect of thermal fluctuations where they interpret as quantum effect. Thermal fluctuations arise due to quantum fluctuations in the geometry of space-time and they appear in the black hole entropy as logarithmic term \[24-27\]. But it’s not our purpose in this paper. Here, we study thermodynamic black hole from the classical point of view.\
So, we are going to recall some classical thermodynamic relations which are necessary for the investigation of thermodynamic and phase transition of black hole.\
The quantities which constitute base of black hole thermodynamics are as follows, $$T=\frac{\kappa\hbar}{2\pi
k_{B}c}~~~~,~~~~S_{H}=\frac{A}{4L^{2}_{PL}}~~~~,~~~~E=Mc^{2},$$ where $\kappa$, $L_{PL}=(\frac{\hbar G}{c^{3}})^{\frac{1}{2}}$, $ A$ and $E$ are the surface gravity, the Planck length, area of event horizon and energy (mass) of black hole respectively. One can extend this classical subject with definition of pressure and volume. The cosmological constant is associated by thermodynamic pressure as \[6,7\], $$P=-\frac{\Lambda}{8\pi},$$ and it’s thermodynamic conjugate of pressure is interpreted as a black hole thermodynamic volume, $$V=\bigg( \frac{\partial M}{\partial P}\bigg)_{S,...}.$$ The cosmological constant can be changed in the first law of black hole thermodynamic, so in this case the first law in the extended phase space will be as, $$dM=TdS+VdP+...~.$$ In extended phase space, the mass is not related to the internal energy $U$ of the system, but it is related by enthalpy which is given by, $$M=H\equiv U+PV.$$ Here we note that, when $P$ is constant (the cosmological constant is not allowed to vary), the relation (4) reduces to the standard first law of thermodynamics in the “non-extended” phase space.
By using relation (2) and (3), we can write black hole equation of state $P=P(V,T)$ and compare to the corresponding fluid. In that case, we identify the black hole and fluid quantities as $T\sim T_{f}$, $V\sim V_{f}$ and $P\sim P_{f}$ \[28\].\
On the other hand, the Van der Waals equation of state is modification of the ideal gas which includes the nonzero size and attraction between corresponding molecules. It is used to describe basic qualitative features of the liquid-gas phase transition, $$kT=(P+\frac{a}{\upsilon^{2}})(\upsilon -b),$$ here $\upsilon = V/N$ is the specific volume of the fluid, $P$, $T$ and $k$ are pressure, temperature and the Boltzmann constant respectively. The constant $b>0$ indicate the nonzero size of the molecules of fluid and the attraction between them is indicated by $a>0$. After studying of phase transition, it is interesting to define classical cycles for black holes like usual thermodynamic systems. It means that when small black hole translate to large black hole we need again the large black hole translate to small black hole, in other words the system must be back to primary state. This definition interpret as holographic heat engine which is mentioned for first time by \[29-31\]. We are going to investigate this subject at the last section of this paper. So, we organize the corresponding paper as follows. In Sect.2, we will review the solution of black hole in Horava-Lifshitz gravity. In Sect.3, In order to have phase transition we will modify Lu-Mei-Pop (LMP) solution of HL black hole. In that case, we show that the corresponding stress energy tensor does not obey any of the energy conditions everywhere outside of the horizon, but the weak condition will be satisfied in near horizon. Then we investigate the stability of solution and determine stable and unstable phase at the end of this section. In Sect.4, we investigate holography stuff and exhibit suitable cycle for corresponding heat engine. Also we arrange special condition to have Carnot efficiency. Finally, in section 5 we have some results and conclusion.
Horava-Lifshitz black hole solution
===================================
One of the interesting kinds of black holes is the Horava-Lifshitz, which may be regarded as a $UV$ candidate for general relativity. The HL is a non-relativistic renormalizable theory of gravity at a Lifshitz point. It provides an interesting framework to study the connection between ordinary gravity and string theory. It is expected that the HL black hole solutions asymptotically become Einstein gravity solutions. Also it is found that a Schwarzschild AdS solution can be realized in infrared modified HL gravity theory \[23,32,33\].\
The four-dimensional gravity action of HL theory is given by the following expression \[34-36\], $$\begin{aligned}
S_{HL}=\int
dtd^{3}x\sqrt{g}N\bigg(\frac{2}{\kappa^{2}}(K_{ij}K^{ij}-\lambda
K^{2})+\frac{\kappa^{2}\mu^{2}(\Lambda_{W}R-3\Lambda_{W}^{2})}{8(1-3\lambda)}+\frac{\kappa^{2}\mu^{2}(1-4\lambda)}{32(1-3\lambda)}R^{2}\bigg)\nonumber
\\&& \hspace{-155mm}-\int
dtd^{3}x\sqrt{g}N\bigg(\frac{\kappa^{2}\mu^{2}}{8}R_{ij}R^{ij}+\frac{\kappa^{2}\mu}{2\omega^{2}}\epsilon^{ijk}R_{il}\nabla_{j}R^{l}_{k}-\frac{\kappa^{2}}{2\omega^{4}}C_{ij}C^{ij}\bigg),\end{aligned}$$ where $\kappa^{2}, \Lambda_{W},\omega,\lambda$ and $\mu$ are constant parameters and $C_{ij}$ is Cotton tensor, $$C^{ij}=\epsilon^{ikl}\nabla_{k}\bigg(R^{j}_{l}-\frac{1}{4}R\delta^{j}_{l}\bigg),$$ and $K_{ij}$ is the extrinsic curvature, $$K_{ij}=\frac{1}{2N}(g_{ij}-\nabla_{i}N_{j}-\nabla_{j}N_{i}),$$ where $N_{i}$ and $N$ are shift and lapse functions respectively. Also the cosmological constant is given by the following relation, $$\Lambda=\frac{3}{2}\Lambda_{W}.$$ The HL black hole metric is, $$ds^{2}=-f(r)dt^{2}+\frac{dr^{2}}{f(r)}+r^{2}d\Omega^{2},$$ where $f (r)$ is given by \[37-39\], $$f(r)=k+(\omega-\Lambda_{W})r^{2}-\sqrt{(r(\omega(\omega-2\Lambda_{W})r^{3}+\beta))},$$ and $\beta$ is integration constant. There are two limits of HL black hole which is following. The first one is Kehagius-Sfetsos (KS) solution which is obtained by $\Lambda_{W}=0$ and $\beta=4\omega M$ \[40\]. In that case the equation (12) reduces to the following expression, $$f_{ks}(r)=k+\omega r^{2}-\omega r^{2}\sqrt{1+\frac{4M}{\omega
r^{3}}}.$$ The second one is Lu-Mei-Pop (LMP) solution which is obtained by $\omega=0$ and $\beta=-\frac{\alpha^{2}}{\Lambda_{W}}$, where $\alpha$ related to the black hole mass $(\alpha = aM)$ \[41\]. In that case the equation (12) reduces to the following expression, $$f_{LMP}(r)=k-\Lambda_{W}r^{2}-\alpha\sqrt{\frac{r}{-\Lambda_{W}}}.$$ The charged topological black holes in Horava-Lifshitz gravity were investigated by Ref.s \[22,23\]. When the dynamical coupling constant $\lambda$ is set to one, the corresponding metric is given by, $$f(r)=k+x^{2}-\sqrt{c_{0} x-\frac{q^{2}}{2}},$$ where $x=\sqrt{-\Lambda}r$ and $\Lambda=(-\frac{3}{l^{2}})$ corresponds to the negative cosmological constant. The physical mass and the charge ($Q$) corresponding to the black hole solution are respectively given by, $$M=\frac{\kappa^{2}\mu^{2}\Omega_{k}\sqrt{-\Lambda}}{16}c_{0}~~~;~~~Q=\frac{\kappa^{2}\mu^{2}\Omega_{k}\sqrt{-\Lambda}}{16}q,$$ where $c_{0}$ and $q$ are the integration constants, $\Omega_{k}$ is the volume of the two dimensional Einstein space and $\kappa$ and $\mu$ are the constant parameters of the theory. The event horizon can be obtained by $f(r_{+}) = 0$ where for Eq.(15) is as follows, $$x^{4}+Ax^{2}+Bx+c=0,$$ where the coefficients are as: $A=2k$, $B=-c_{0}$ and $C=(k^{2}+\frac{q^{2}}{2})$. Applying the Descarte method \[42\] in equation (17), we find the following solution, $$\begin{aligned}
x_{1}=\frac{-a-\sqrt{-(a^{2}+2A+\frac{2B}{a})}}{2},~~~x_{2}=\frac{-a+\sqrt{-(a^{2}+2A+\frac{2B}{a})}}{2},\nonumber
\\&&
\hspace{-125mm}x_{3}=\frac{a-\sqrt{-(a^{2}+2A-\frac{2B}{a})}}{2},~~~x_{4}=\frac{a+\sqrt{-(a^{2}+2A-\frac{2B}{a})}}{2},\end{aligned}$$ with $$\begin{aligned}
a=\pm\bigg[\bigg(-\frac{q}{2}-\frac{1}{2}\sqrt{\frac{27q^{2}+4p^{3}}{27}}
\bigg)^{\frac{1}{3}}+
\bigg(-\frac{q}{2}+\frac{1}{2}\sqrt{\frac{27q^{2}+4p^{3}}{27}}-\frac{2A}{3}
\bigg)^{\frac{1}{3}}\bigg]^{\frac{1}{2}}\nonumber
\\&&\hspace{-138mm}b=\frac{1}{2}\bigg(a^{2}+A+\frac{B}{a}\bigg),~~~~~~~~~~~~~c=\frac{1}{2}\bigg(a^{2}+A-\frac{B}{a}\bigg)\nonumber
\\&&\hspace{-138mm}p=-4C-\frac{A^{2}}{3},~~~~~~~~~~~~~~~~~~~~q=\frac{27A}{3}\bigg(A^{2}+36C\bigg)+C.\end{aligned}$$ Here the horizon corresponds to a positive choice of solutions, $$x_{h}=\frac{a+\sqrt{-(a^{2}+2A-\frac{2B}{a})}}{2}.$$ The black hole entropy, Hawking temperature and heat capacity at constant pressure determined by thermodynamical relation \[22,43,44\], $$S=\int\frac{dM}{T}~~~,~~~T_{H}=\frac{\kappa}{2\pi}=\frac{1}{4\pi}\bigg(\frac{\partial
f}{\partial r}\bigg)_{r=r_{h}}~~~,~~~C_{P}=\frac{\partial
M}{\partial T}=\frac{T}{(\frac{\partial T}{\partial S})_{P}}.$$ By using these relations, one can obtain thermodynamical quantities of (LMP) solution Eq.(14) as follows, $$S=\frac{8\pi}{a}\sqrt{-\Lambda_{W}r_{h}}~~,~~T_{H}=\frac{1}{8\pi
r_{h}}(-k-3\Lambda_{W}r_{h}^{2})~~,~~C_{P}=\frac{4\pi}{a}\sqrt{-\Lambda_{W}r_{h}}\frac{k+3\Lambda_{W}r_{h}^{2}}{-k+3\Lambda_{W}r_{h}^{2}}.$$ The heat capacity at constant pressure has a discontinuity at $r_{h}=\sqrt{\frac{k}{3\Lambda_{W}}}$. This shows that there is a second-order phase transition at this point. The cosmological constant is negative in AdS space, therefore phase transition exits only in $k=-1$. This is completely different to Einstein theory, where we have phase transition for only $k=1$ case. Also thermodynamical quantities for solution (15) are given by, $$\begin{aligned}
T_{H}=\frac{\sqrt{-\Lambda}(3x_{h}^{4}+2kx_{h}^{2}-k^{2}-\frac{Q^{2}}{2})}{8\pi
x_{h}(k+x_{h}^{2})},\nonumber
\\&&\hspace{-72mm}S=\frac{x_{h}^{2}}{4}+\frac{k}{2}\ln
x_{h}+S_{0},\nonumber
\\&&\hspace{-72mm}C_{P}=\frac{(k+x_{h}^{2})^{2}(3x_{h}^{4}+2kx_{h}^{2}-k^{2}-\frac{Q^{2}}{2})}{6x_{h}^{6}+14kx_{h}^{4}+10k^{2}x_{h}^{2}+2k^{3}+Q^{2}(k+3x_{h}^{2})}.\end{aligned}$$ The heat capacity suffers a discontinuity at $x_{h}=x_{c}$ in $k=-1$. Then this solution has a phase transition in nonextended phase space for hyperbolic horizon. The critical behavior of charged topological black holes in Horava-Lifshitz gravity in the extended phase space was studied in \[23\] and no physical critical point was found in all the cases $(k=\pm1,0)$.
Modification of Horava-Lifshitz black hole solution
===================================================
Here the most important thing is investigation of the thermodynamical and hydrodynamical properties Horava-Lifshitz black hole. Also, the phase transition and Van der Waals fluid behaviour for such black hole play important role in holography and AdS/CFT. So, for this reason we first investigated the usual Horava-Lifshitz black hole solution and observed that there are not such properties. Therefor we take advantage from some ansatz and energy condition and modify the usual Horava-Lifshitz black hole to new form of solution which is modification of Horava-Lifshitz black hole solution. In that case we see this new form of HL are satisfied by phase transition and Van der Waals fluid properties. So, in this section, we are going to modify the equation (15) for the uncharge case, where thermodynamics properties of black hole exactly coincide with the given fluid equation of state. In order to modify such metric background, first we give following ansatz for the $f(r)$, $$\begin{aligned}
ds^{2}=-f(r)dt^{2}+\frac{dr^{2}}{f(r)}+r^{2}d\Omega^{2}\nonumber
\\&& \hspace{-67mm} f(r)=-\Lambda r^{2}-2\sqrt{Mr}-h(r,P).\end{aligned}$$ In order to simplify the corresponding results, we set $\frac{\kappa^{2}\mu^{2}\Omega_{k}}{4}=1$.
Now we are going to determine the function $h(r,P)$ which has to be satisfied with the equation of motion. So, generally we assume that the above metric is a solution of the HL gravity with given energy momentum source from HL lagrangian. In order to the energy-momentum source be physically acceptable it should be satisfied by certain following conditions:\
1)The energy density be positive.\
2)The energy density conquest to pressure.\
These conditions are known as energy conditions, e.g. \[45\], which is given by $$\begin{aligned}
Week: ~~~\rho\geq0~~~~,~~~\rho+p_{i}\geq0\nonumber \\&&
\hspace{-65mm}
Strong:~~~\rho+\sum_{i}p_{i}\geq0~~~,~~~\rho+p_{i}\geq0\nonumber
\\&& \hspace{-65mm}
Dominant:~~~\rho\geq\mid p_{i}\mid,\end{aligned}$$ where $\rho$ and $p_{i}$ are energy density and principal pressure respectively. Also, $\rho$ and $p_{i}$ clarified by the following relations, $$\begin{aligned}
\rho=-p_{1}=\frac{1-f-r f'}{8\pi r^{2}}+P,\nonumber \\&&
\hspace{-59mm} p_{2}=p_{3}=\frac{r f''+2f'}{16\pi r}-P.\end{aligned}$$ After determining $f(r)$, we examine the corresponding energy conditions. The mass of the black hole $(M)$ is related to the horizon radius $r_{+}$ according to the following expression, $$\sqrt{M}=\frac{1}{2\sqrt{r_{+}}}(-\Lambda r_{+}^{2}-h(r_{+},P)).$$ On the other hand the black hole temperature is given by the following equation, $$T=\frac{1}{4\pi}f'(r_{+})=3Pr_{+}+
\frac{h(r_{+},P)}{8Pr_{+}}-\frac{h'(r_{+},P)}{4\pi}.$$
As a mentioned above, we have to construct a metric which exactly matches thermodynamics properties of black hole to the thermodynamic of Van der Waals fluid. For this purpose, we compare the temperature of $T$ in equation (28) with (6) and use $\upsilon=3r_{+}$, $$3Pr_{+}+
\frac{h(r_{+},P)}{8Pr_{+}}-\frac{h'(r_{+},P)}{4\pi}-P\upsilon+b\upsilon-\frac{a}{\upsilon}+\frac{ab}{\upsilon^{2}}=0,$$ Here we use the following ansatz \[28,46\], $$h(r,P)=A(r)-PB(r).$$ By substituting the equation (30) into equation (29), $B(r)$ and $A(r)$ will be as, $$B(r)=-8b\pi r+C_{1}\sqrt{r}.$$ and $$A(r)=\frac{8a\pi}{3}-\frac{8ab\pi}{27r}+C_{2}\sqrt{r}.$$ Here $C_{1}$ and $C_{2}$ are integration constants and also $C_{2}$ complectly has a mass dimension and $C_{1}$ is mass over pressure. By replacing $A(r)$ and $B(r)$ in the relation (24) we have, $$f(r)=-\frac{8a\pi}{3}+\frac{8ab\pi}{27r}-C\sqrt{Mr}+\Lambda
br-\Lambda r^{2},$$ where $C$ is a constant in terms of $C_{1}$ and $C_{2}$. Since $a$ is positive, we conclude that the topology of horizon is hyperbolic. Without loss of generality we can set $a=\frac{3}{8\pi}$. We can determine event horizon by using $f(r_{+}=0)$ and draw its variations with respect to $b$ and $C$ parameters. As we see in figure (1), the event horizon increases with $b$ parameter linearly but it decreases by increasing $C$ and will be zero for very large values of $C$.
Now by rearranging $f(r)$, we can investigate the energy condition. So, by using relation (26), we obtain $\rho$ and $p_{i}$, $$\begin{aligned}
\rho=-p_{1}=\frac{1}{16\pi r}(\frac{b}{3r^{2}}+\frac{1}{r}+8b\pi
P-8\pi Pr)\nonumber
\\&&
\hspace{-85mm}
p_{2}=p_{3}=\frac{1}{16\pi r}(-\frac{b}{12r^{2}}+\frac{3}{4r}-10b\pi
P+26\pi Pr)\nonumber
\\&&
\hspace{-85mm}\rho+p_{2}=\frac{1}{16\pi r}(\frac{b}{4r^{2}}+\frac{7}{4r}-2b\pi
P+18\pi Pr).\end{aligned}$$
We display the energy density $\rho$ (Solid and Dash dot curves) and the quantity $\rho+P_{2}$ (Dot curve) with respect to $r_{+}$. We always have $\rho+P_{1}=0$, but for the arbitrary pressure and everywhere outside of the horizon, we have $\rho+P_{2}\geq0$. Also note that if the pressure be sufficiently small ($P=0.001$), the energy condition $\rho\geq0$ will be satisfy near horizon, which is illustrated by Fig (2).
In HL black hole with comparing to thermodynamical system we have some phase transition. In order to find the stability and instability phase transition we have to investigate the corresponding subject. As we know the stability of system from phase transition point of view play important role in some phenomena in particle physics and also cosmology system. So, for this reasons we consider stability condition for the corresponding system. In that case, we need two quantities as Gibbs free energy and heat capacity which play important role for the study of stability system. When the Gibbs free energy is negative $(G<0)$, the system has a global stability. For the local stability equilibrium in thermodynamic system requires that $C_{P}\geq C_{V}\geq0$ \[47,48\]. In that case, the heat capacity at constant pressure and constant volume is given by, $$C_{V}=\frac{\partial H}{\partial T}\bigg|_V=T\frac{\partial
S}{\partial T}\bigg|_V~~~~;~~~~C_{P}=\frac{\partial H}{\partial
T}P=T\frac{\partial S}{\partial T}\bigg|_P.$$ The temperature and the entropy can be obtained by equation (21), which are related by the first law of thermodynamics. $$T=\frac{1}{4\pi}f'(r_{+})=\frac{1}{4\pi}(\frac{1}{2r_{+}}-\frac{b}{6r_{+}^{2}}-4b\pi
P+12\pi Pr_{+}).$$ We emphasis that the area formula for entropy breaks down in case of higher derivative gravity especially HL theory \[43\]. For this reason we leave this formula and employ the following first law of thermodynamics and obtain the corresponding entropy as, $$S=\int\frac{dM}{T}=\frac{8\pi}{C^{2}}(-\ln(r_{+})-\frac{b}{9r_{+}}-8b\pi
Pr_{+}+4\pi Pr_{+}^{2})+S_{0}.$$ In the above equation, $S_{0}$ is the integration constant. By using Eqs. (2) and (3), the thermodynamic volume can be calculated by, $$V=\frac{16\pi}{C^{2}}(r_{+}-b)(-1+\frac{b}{9r_{+}}-8b\pi Pr_{+}+8\pi
Pr_{+}^{2}).$$ As we see in equation (38) the volume depend on $r_{+}$ and $P$. Then one can say, the heat capacity at constant volume vanishes, $$C_{V}=T\bigg(\frac{\partial S}{\partial P}\frac{\partial P}{\partial
T}+\frac{\partial S}{\partial r_{+}}\frac{\partial r_{+}}{\partial
T}\bigg)=0.$$ One can obtain $C_{P}$ as, $$C_{P}=\frac{8\pi r_{+}}{C^{2}}\frac{(1-\frac{b}{3r_{+}}-8b\pi
Pr_{+}+24\pi Pr_{+}^{2})(-\frac{1}{r_{+}}+\frac{b}{9r_{+}^{2}}-8b\pi
P+8\pi Pr_{+})}{(-1+\frac{2b}{3r_{+}}+24\pi Pr_{+}^{2})}.$$ In figure (3), We draw $C_{P}$ with respect to $r_{+}$ for two values of $b$ parameter ($b=0.01$ and $b=0.5$).As we know, if the heat capacity is divergent then the system will has a phase transition. We notice that the heat capacity dose not diverge for small values of $C$ then there is not the second-order phase transition. As we see the heat capacity has a discontinuity for small values of $b$. But by increasing this parameter ($b$) the discontinuity points increases which can be decreased by increasing $C$. For phase 1, the heat capacity is always positive which means that this phase is thermodynamically stable. On the other hand for phase 2 the heat capacity is negative therefor it is an unstable phase.
In order to discuss the global stability of the black hole, we need to calculate the Gibbs free energy which is given by, $$G=E-TS+PV,$$ and $$G=\frac{1}{C^{2}r_{+}}(1-\frac{b}{9r_{+}}A_{1}-4\pi r_{+}^{2}A_{2}
)+\frac{64\pi^{2}}{C^{2}}Tr_{+}\ln(r_{+}),$$ where $$\begin{aligned}
A_{1}=1+\frac{2b}{9r_{+}}+48b\pi Pr_{+}-268\pi Pr_{+}^{2}\nonumber
\\&&
\hspace{-74mm}
A_{2}=5-24b\pi P r_{+} +8\pi Pr_{+}^{2}.\end{aligned}$$
In figure (4), we display the Gibbs free energy with respect to $r_{+}$. As we see in figure, for large values of $C$ the system has an instability. Here, we note that the range of stability increases by decreasing $C$ and $b$.
The corresponding holographic heat engine
=========================================
In this section, we intend to consider holographic heat engine for the black hole solution. We start with equation of state of black hole $P(V,T)$ which is relation between temperature, horizon radius, other parameter of black hole as a cosmological constant. By using the $P(V,T)$ function one can define an engine as a closed path in $P-V$ plane. This lead us to have input and output heat as $Q_{H}$ and $Q_{C}$ respectively. The total mechanical work is $W=Q_{H}-Q_{C}$ and also the efficiency of the heat engine is determined by relation $\eta=\frac{W}{Q_H}=1-\frac{Q_C}{Q_H}$.\
The properties of the heat engine depend on the equation of state and choice of path in $P-V$ plane. For a heat engine which works between two temperatures $T_{H}$ and $T_{C}$ the maximum efficiency will be Carnot efficiency \[29-31\].\
The Carnot cycle is made from two isothermal and two adiabatic processes. During isothermal expansion (constant temperature $T_{H}$) the system absorbs heat of $Q_{H}$ and during isothermal compression (constant temperature $T_{C}$) loses heat of $Q_{C}$.\
The corresponding work done in this cycle is, $$W=\oint PdV=\oint TdS=(T_{H}-T_{C})(S_{2}-S_{1}).$$ And total amount of absorbed thermal energy will be as, $$Q_{H}=T_{H}(S_{2}-S_{1}).$$ The efficiency $\eta$ is defined by, $$\eta=\frac{W}{Q_H}=1-\frac{T_C}{T_H}.$$ This relation show that the higher than Carnot efficiency violate the second law of thermodynamics. Of course there is other method to connect two isotherms $T_{H}$ and $T_{C}$ which are isochoric paths. The formed cycle is called the Stirling cycle, which has an efficiency less than the Carnot efficiency.\
For static black holes the Carnot cycle and the Stirling cycle are identical. Because in such black holes the entropy and volume both depend on the horizon radius, so isochores and adiabatic are the same.\
Now we study the efficiency of heat engine for modified solution and investigate under what conditions, our cycle has the Carnot efficiency. As we noticed in pervious section, the heat capacity at constant volume vanishes for our solution. This is useful because it can define another cycle in $P-V$ plane, as a rectangle which is composed of isobars and isochores, see fig (5).
The work done along the isobars is, $$W=(V_{2}-V_{1})(P_{1}-P_{4}).$$
The heat flows occur along to the top and bottom of the figure. Here, one can determine the inflow of heat with the upper isobar, which is $Q_{H}$ , $$Q_H=\int_{T_1}^{T_2} C_P(P_1,T)dT.$$ In order to calculate the integral, we need to write the heat capacity in terms of $T$. So, this is complicated integral and one can not solve directly because both $C_{P}$ and $T$ are function of $r_{+}$. In that case, we take a high temperature limit for simplicity of calculation. In this limit we have, $$r_{+}\sim \frac{T}{3P},$$ and $$C_P\sim \frac{64\pi^{2}}{9PC^{2}}T^{2},$$ then $$Q_H= \frac{64\pi^{2}}{27PC^{2}}(T_{2}^{3}-T_{1}^{3}).$$ Also in this limit the thermodynamical volume is, $$V= \frac{16\pi}{C^{2}}\bigg(\frac{8\pi T^{3}}{27P^{2}}-\frac{16b\pi
T^{2}}{9P}+\frac{(8b^{2}\pi P-1)T}{3P}-\frac{b^{2}P}{3T}\bigg).$$
We substitute equation (52) in equation (47) we have,
$$W=\frac{16\pi}{C^{2}}(P_{1}-P_{4})\bigg(\frac{8\pi}{27P^{2}}(T_{2}^{3}-T_{1}^{3})-\frac{16b\pi}{9P}(T_{2}^{2}-T_{1}^{2})+\frac{(8b^{2}\pi
P-1)}{3P}(T_{2}-T_{1})-\frac{b^{2}P}{3}(\frac{1}{T_{2}}-\frac{1}{T_{1}})\bigg).$$
Therefore, the efficiency of this heat engine is,
$$\eta=\eta_{C}(1+\beta),$$
where $\beta$, $$\beta=1-12b\pi\frac{(T_{2}^{2}-T_{1}^{2})}{(T_{2}^{3}-T_{1}^{3})}+\frac{9P(8b^{2}\pi
P-1)}{4\pi}\frac{(T_{2}-T_{1})}{(T_{2}^{3}-T_{1}^{3})}-\frac{9b^{2}P^{3}}{4\pi}\frac{(T_{2}^{-1}-T_{1}^{-1})}{(T_{2}^{3}-T_{1}^{3})}~.$$
Eq. (55) is satisfied by the condition $-1<\beta\leq0$. And if $\beta=0$ we will have the carnot efficiency. Figures (6) and (7) show the behavior of $\beta$ with respect to temperature $T_{2}$ for different values of $T_{1}$. From figures (6) and (7), we see that the system has efficiency in a certain range of $T_{1}$. And the efficiency increases by reducing $T_{1}$ in this corresponding certain range. Also the system will have the Carnot efficiency for two specific values of $T_{2}$ . As we see from figure (6), by increasing parameter $b$, the $T_{1}$ also increases. By comparing figures (6) and (7), we notice that by increasing the pressure, both temperatures $T_{1}$ and $T_{2}$ are increasing.
Conclusion
==========
As we know the investigation of phase structure of Horava-Lifshitz black hole has shown that the phase transition exists only for hyperbolic topology of the horizon in the non-extended phase space. This result shown that Horava-Lifshitz theory is not exactly identical to Einstein theory. Because in Einstein theory exists no phase transition for k=0, -1 but exist in case of k=1. As we have seen, with considering phase structure of HL black hole in the extended phase space, there is not any phase transition. For this reason\
in this paper, we modified the solution of the Horava-Lifshitz black hole for uncharged case. We have shown that the modified solution dose not satisfy any of three standard energy condition everywhere outside of horizon. But for small pressure in near horizon the modified solution satisfied by weak energy condition. We investigated the stability and instability with certain thermodynamical parameters. We noticed for small values of $b$ and $C$ parameters the corresponding system has a global stability, but to have a phase transition the $C$ parameter must be large. Also we saw that the divergence points of heat capacity increase by increasing $b$ and $C$ parameters. Finally, we studied the heat engine and observed that the system has a Carnot efficiency at special conditions. For small values of pressure and $b$ parameter the system had a maximum efficiency in range of very small $T_{1}$ , but this temperature increased by increasing $b$ parameter. We also noticed by increasing pressure the range of temperature $T_{2}$ increased.
[4]{} J. D. Bekenstein, Phys. Rev. D **7** (1973) 2333.
J. M. Bardeen, B. Carter and S. W. Hawking, Commun. Math. Phys. **31** (1973) 161.
Y. S. Myung, Y. W. Kim and Y. J. Park, Thermodynamics and phase transitions in the Born-Infeld-anti-de Sitter black holes, Phys. Rev. D **78** (2008) 084002. Y. S. Myung, Phase transition between the BTZ black hole and AdS space, Phys. Lett. B **638** (2006) 515-518. B. M. N. Carter and I. P. Neupane, Thermodynamics and stability of higher dimensional rotating Kerr-AdS black holes, Phys. Rev. D **72** (2005) 043534. D. Kubiznak and R. B. Mann, P-V criticality of charged AdS black holes, J. High Energy Phys. **078** (2012) 033, arXiv:1205.0559 \[hep-th\].
S. Gunasekaran, D. Kubiznak and R. B. Mann, Extended phase space thermodynamics for charged and rotating black holes and Born-Infeld vacuum polarization, J. High Energy Phys. **11** (2012) 110, arXiv:1208.6251 \[hep-th\]. S. Hawking and D. N. Page, Thermodynamics of black holes in anti-deSitter space, Commun. Math. Phys. **87** (1983) 577. J. X. Mo and W. B. Liu, P-V Criticality of Topological Black Holes in Lovelock-Born-Infeld Gravity, Eur. Phys. J. C **74** (2014) 2836. S. Chen, X. Liu and C. Liu, P-V criticality of AdS black hole in f(R) gravity, Chin. Phys. Lett. **30** (2013) 060401.
S. H. Hendi and Z. Armanfard, Extended phase space thermodynamics and P-V criticality of charged black holes in Brans-Dicke theory, Gen. Relativ. Gravit. **47** (2015) 125.
D. Birmingham, Topological Black Holes in Anti-de Sitter Space, Class. Quantum Grav. **16** (1999) 1197-1205.
R. G. Cai and A. Wang, Thermodynamics and stability of hyperbolic charged black holes, Phys. Rev. D **70** (2004) 064013.
R. G. Cai, S. P. Kim and B. Wang, Ricci flat black holes and Hawking-Page phase transition in Gauss-Bonnet gravity and dilaton gravity, Phys. Rev. D **76** (2007) 024011. R. G. Cai, L. M. Cao, L. Li and R. Q. Yang, P-V criticality in the extended phase space of Gauss-Bonnet black holes in AdS space, J. High Energy Phys. **1309** (2013) 005. R. Zhao, H. Zhao, M. S. Ma and L. C. Zhang, On the critical phenomena and thermodynamics of charged topological dilaton AdS black holes, Eur. Phys. J. C **73** (2013) 2645.
R. G. Cai, A note on thermodynamics of black holes in Lovelock gravity, Phys. Lett. B **582** (2004) 237 -242.
R. G. Cai, L. M. Cao and N. Ohta, Topological black holes in Horava-Lifshitz gravity, phys. Rev. D **80** (2009) 024003.
R. G. Cai and N. Ohta, Horizon thermodynamics and gravitational field equations in Horava-Lifshitz gravity, phys. Rev. D **81** (2010) 084061.
R. G. Cai, L. M. Cao and N. Ohta, Thermodynamics of black holes in Horava-Lifshitz gravity, phys. Lett. B **679** (2009) 504-509.
Y. S. Myung and Y. W. Kim, Thermodynamics of Horava-Lifshitz black holes, Eur. Phys. J. C **68** (2010) 265-270.
B. R. Majhi and D. Roychowdhury, Phase transition and scaling behavior of topological charged black holes in Horava-Lifshitz gravity, Class. Quantum Gravity **29** (2012) 245012.
J. X. Mo, No P-V criticality for charged topological black holes in Horava-Lifshitz gravity, Astrophys Space Sci **356** (2015) 319-325.
N. M. Duran, A. F. Vargas, P. H. Restrepo and P. Bargueno, Simple regular black hole with logarithmic entropy correction, Eur. Phys. J. C **76** (2016) 559. A. Pourdarvish, J. Sadeghi, H. Farahani and B. Pourhassan, Thermodynamics and Statistics of Godel Black Hole with Logarithmic Correction, Int. J. Theor. Phys. **52** (2013) 3560.
A. Mandal and R. Biswas, Thermodynamic products in Horava-Lifshitz gravity with logarithmic correction of entropy, Astrophys Space Sci **361** (2016) 39. J. Sadeghi, B. Pourhassan and M. Rostami, P-V criticality of logarithm-corrected dyonic charged AdS black holes, phys. Rev. D **94** (2016) 064006.
A. Rajagopal, D. Kubiznak and R. B. Mann, Van der Waals black hole, Phys. Lett. B **08** (2014) 054, arXiv:1408.1105 \[gr-qc\].
C. V. Johnson, Holographic Heat Engines, Class. Quant. Grav. **31** (2014) 205002 , arxiv: 1404.5982 \[hep-th\]. C. V. Johnson, Born-Infeld AdS black holes as Heat Engine, arxiv: 1512.01746 \[hep-th\].
C. V. Johnson, Gauss-Bonnet Black Holes and Holographic Heat Engine Beyond Large N, arxiv: 1511.08782 \[hep-th\].
I. Radinschi, F. Rahaman and A. Banerjee, On the energy of Horava-Lifshitz black holes, Int. J. Theor. Phys. **50** (2011) 2906.
Y. S. Myung and Y. W. Kim, Thermodynamics of Horava-Lifshitz black holes, Eur. Phys. J. C **68** (2010) 265.
R. G. Cai, L. M. Cao and N. Ohta, Topological Black Holes in Horava-Lifshitz Gravity, Phys. Rev. D **80** (2009) 024003.
H. Lu, Jianwei Mei and C. N. Pope, Solutions to Horava Gravity, Phys. Rev. Lett. **103** (2009) 091301.
H. W. Lee, Y. W. Kim and Y. S. Myung, Slowly rotating black holes in the Horava-Lifshitz gravity, Eur. Phys. J. C **70** (2010) 367.
J. Sadeghi and B. Pourhassan, Particle acceleration in Horava-Lifshitz black holes, Eur. Phys. J. C **72** (2012) 1984
M. I. Park, The black hole and cosmological solutions in IR modified Horava gravity, J. High Energy Phys. **0909** (2009) 123, arXiv:0905.4480 \[hep-th\].
I. Radinschi, F. Rahaman and A. Banerjee, On the energy of Horava-Lifshitz black holes, Int. J. Theor. Phys. **50** (2011) 2906-2916, arXiv:1012.0986 \[gr-qc\]. A. Kehagias and K. Sfetsos, The black hole and FRW geometries of non-relativistic gravity, Phys. Lett. B **678** (2009) 123 -127.
H. Lu, J. Mei and C. N. Pope, Solutions to Horava gravity, Phys. Rev. Lett. **103** (2009) 091301, arXiv: 0904.1595 \[hep-th\].
R. Descartes, Discours de la Methode et Premiere Meditation, ed. Hachette et cie, 1881, 156 pages.
J. Sadeghi, K. Jafarzade and B. Pourhassan, Thermodynamical Quantities of Horava-Lifshitz Black Hole, Int. J. Theor. Phys. **51** (2012) 3891. R. Biswas and S. Chakraborty, Black Hole Thermodynamics in Horava Lifshitz Gravity and the Related Geometry, Astrophys. Space Sci. **332** (2011) 193, arXiv:1104.3719 \[gr-qc\].
E. Poisson, A Relativist’s Toolkit, Cambridge University Press, 2004. T. Delsate and R. Mann, Van Der Waals Black Holes in d dimensions, JHEP **02** (2015) 070.
M. S. Ma and R. Zhao, Stability of black holes based on horizon thermodynamics, phys. Lett. B **751** (2015) 278, arxiv: 1511.03508 \[gr-qc\].
H. B. Callen, Thermodynamics and an Introduction to Thermostatistics, John Wiley and Sons, New York, NY, USA, 1985.
[^1]: Email: [email protected]
[^2]: Email: [email protected]
| {
"pile_set_name": "ArXiv"
} | ArXiv |
---
abstract: 'Experimental data suggest that some classes of spiking neurons in the first layers of sensory systems are electrically coupled via gap junctions or ephaptic interactions. When the electrical coupling is removed, the response function (firing rate [*vs.*]{} stimulus intensity) of the uncoupled neurons typically shows a decrease in dynamic range and sensitivity. In order to assess the effect of electrical coupling in the sensory periphery, we calculate the response to a Poisson stimulus of a chain of excitable neurons modeled by $n$-state Greenberg-Hastings cellular automata in two approximation levels. The single-site mean field approximation is shown to give poor results, failing to predict the absorbing state of the lattice, while the results for the pair approximation are in good agreement with computer simulations in the whole stimulus range. In particular, the dynamic range is substantially enlarged due to the propagation of excitable waves, which suggests a functional role for lateral electrical coupling. For probabilistic spike propagation the Hill exponent of the response function is $\alpha=1$, while for deterministic spike propagation we obtain $\alpha=1/2$, which is close to the experimental values of the psychophysical Stevens exponents for odor and light intensities. Our calculations are in qualitative agreement with experimental response functions of ganglion cells in the mammalian retina.'
author:
- 'Lucas S. Furtado'
- Mauro Copelli
title: 'Response of electrically coupled spiking neurons: a cellular automaton approach'
---
[^1]
Introduction
============
Unveiling how neuronal activity represents and processes sensory information remains a very difficult problem, despite theoretical and experimental efforts undertaken by neuroscientists for the last several decades (for a recent review, see [@Kreiman04]). In this broad context, relatively little attention has been devoted to the question of how organisms cope with the several orders of magnitude spanned by the intensities of sensory stimuli [@Cleland99]. This astonishing ability is most easily revealed in humans by classical results in psychophysics [@Stevens]: the perception of the intensity of a given stimulus is experimentally shown to depend on the stimulus intensity $r$ as $\sim \log(r)$ (Weber-Fechner law) or $\sim
r^{\alpha}$ (Stevens law), where the Stevens exponent $\alpha$ is typically $<1$. Those laws have in common the fact that they are [*response functions with broad dynamic range*]{}, i.e. they map several decades of stimuli into a single decade of response.
One would like to understand how this broad dynamic range is physically achieved by neuron assemblies. Recent experimental evidence suggests that electrical coupling among neurons in the early layers of sensory systems plays an essential role in weak stimulus detection. Deans et al. [@Deans02] showed that electrical coupling is present in the mammalian retina via gap junctions (ionic channels that connect neighboring cells). In particular, the spiking response of ganglion cells to light stimulus changes dramatically when the gap junctions are genetically knocked out: both sensitivity and dynamic range are reduced [@Deans02].
Another example comes from the olfactory system. The spiking response of isolated olfactory sensory neurons (OSNs) to varying odorant concentration usually presents a narrow dynamic range [@Rospars00; @Rospars03]. This is in contrast with the response observed in the next layers of the olfactory bulb: both the glomerular [@Friedrich97; @Wachowiak01] and mitral cell [@Duchamp-Viret90a] responses present a broader dynamic range than the OSNs. In this case, the tightly packed unmyelinated axons of OSNs in the olfactory nerve are believed to interact electrically via ephaptic interactions [@Lowe03] (i.e. mediated by current flow through the extracellular space), as shown by Bokil et al. [@Bokil01]. In particular, their results indicate that a spike in a single axon can evoke spikes in all other axons of the bundle, suggesting that some computation is performed prior to the glomerular layer.
Motivated by these results, previous papers have shown through numerical simulations that electrical coupling among neurons indeed changes the response function in a way that is consistent with experimental results. Due to the coupling, stimuli generate excitable waves which propagate through the neuron population. The interplay between wave creation and wave annihilation leads to a nonlinear amplification of the spiking response, increasing the sensitivity at low input levels [*and*]{} enhancing the dynamic range [@Copelli02; @Copelli05a]. In one dimension, the robustness of the mechanism is attested by the diversity of models employed: either the biophysically realistic Hodgkin-Huxley equations [@Koch; @Copelli02], a lattice of nonlinear coupled maps [@Kuva01; @Copelli04a; @Copelli05a], or the Greenberg-Hastings cellular automata (GHCA) [@Copelli02; @Copelli05a] yield qualitatively similar results. The same phenomenon has recently been observed in simulations with the two-dimensional GHCA [@Copelli05b].
In this paper we calculate the response of excitable GHCA model neurons [@Greenberg78], where the bidirectional (electrical) coupling is modeled by a probability $p$ of spike transmission. While the uncoupled case $p=0$ can be exactly solved, the coupled case $p>
0$ is handled within two mean field approximations, namely at the single-site and pair levels. The aim is to shed light on the analytical behavior of the response function for the one-dimensional case, therefore building on previous efforts which have relied entirely on numerical simulations.
Our focus on the [*response*]{} of a [*continuously driven*]{} spatially extended excitable system should be carefully confronted with other recent studies, where the main interest has been on phase transitions between an excitable and a self-sustained collective state. For instance, the SIRS model of epidemics in hypercubic lattices has been recently investigated under the mean field and pair approximations [@Joo04b]. In those contagion models, stationary self-sustained activity becomes stable for sufficiently strong connection among neighbors, a behavior which has been shown to be universal under very general assumptions [@Dodds04]. Similar results have been obtained for a variety of neuronal models, including collective responses to a localized transient stimulus [@Hasegawa03a; @Roxin04], as well as the emergence of sustained activity in complex networks [@Roxin04; @Netoff04].
While interesting in its own, the framework of stable-unstable collective transitions does not seem particularly suited for our modeling purposes. To account for sensory responses, the employed GHCA model is an excitable system which always returns to its absorbing state in the absence of stimulus, there are no phase transitions. The refractory period of the GHCA model neurons is absolute (unlike, say, reaction-diffusion lattices), mimicking the deterministic behavior of continuous-time systems like the Hodgkin-Huxley equations or integrate-and-fire models [@Koch]. The only source of stochasticity of the model regards the firing of the neurons. Stimuli can come from spiking neighbors (with probability $p$) or from an “external” source, which is modeled by a Poisson process and represents sensory input. Therefore, in the limit $p=1$ the dynamics is that of a [*deterministic*]{} excitable lattice being [*stochastically stimulated*]{}, which casts the problem into the framework of probabilistic cellular automata [@Odor04].
The paper is organized as follows. In section \[model\], the GHCA rules are described; section \[uncoupled\] contains the exact calculations for the response of uncoupled neurons, while in sections \[coupled\] and \[coupledPAIR\] results for the coupled case are discussed in the mean field and pair approximations, respectively. Our concluding remarks are presented in section \[conclusion\].
\[model\]The model
==================
In the $n$-state GHCA model [@Greenberg78] for excitable systems, the instantaneous membrane potential of the $i$-th cell ($i=1,\ldots,L$) at discrete time $t$ is represented by $x_i(t)\in\{0,1,\ldots,n-1\}$, $n\geq 3$. The state $x_i(t)=0$ denotes a neuron at its resting (polarized) potential, $x_i(t)=1$ represents a spiking (depolarizing) neuron and $x_i(t)=2,\ldots,n-1$ account for the afterspike refractory period (hyperpolarization). We employ the simplest rules of the automaton: if $x_i(t)=0$, then $x_i(t+1)=1$ only if there is a supra-threshold stimulus at site $i$; otherwise, $x_i(t+1)=0$. If $x_i(t)\geq 1$, then $x_i(t+1)=(x_i(t)+1)\text{mod
}n$, regardless of the stimulus. In other words, the rules state that a neuron only spikes if stimulated, after which it undergoes an absolute refractory period before returning to rest.
Whether the neurons are isolated or coupled is implicit in the definition of the supra-threshold stimulus. We assume [*external*]{} supra-threshold stimuli to be a Poisson process with rate $r$ (events per second). Hence at each time step an external stimulus arrives with probability $$\label{lambda}
\lambda(r) = 1-e^{-r\tau}$$ per neuron. Notice that $\tau=1$ ms corresponds to the approximate duration of a spike and is the time scale adopted for the time step of the model. The number of states $n$ therefore controls the duration of the refractory period (which corresponds to $n-2$, in ms). In the biological context, $r$ could be related for example with the concentration of a given odorant presented to an olfactory epithelium [@Rospars00], or the light intensity stimulating a retina [@Deans02]. We shall refer to $r$ as the stimulus rate or intensity.
When electrically coupled, neurons at rest can also be stimulated by their neighbors. We define $p$ and $q$ as the probabilities that a resting neuron spikes as a consequence of transmission (ionic current flow) from respectively one or two spiking neighbors \[see Eq. (\[eq:p1coupled\])\]. We keep $p$ and $q$ as two independent parameters in most calculations to show the robustness of some asymptotic results. In the simulations, we concentrate on the more physically intuitive choice of $q = 1 - (1-p)^2$, where the contributions from two spiking neighbors are independent.
Let $P_{t}^{(i)}(k)$ be the probability that the $i$-th neuron is in state $k$ at time $t$. Since the dynamics of the refractory state is deterministic, the equations for $k\geq 2$ are simply
$$\begin{aligned}
\label{eq:refrac}
P_{t+1}^{(i)}(2) & = & P_{t}^{(i)}(1) \nonumber \\
P_{t+1}^{(i)}(3) & = & P_{t}^{(i)}(2) \nonumber \\
\ & \vdots & \ \nonumber \\
P_{t+1}^{(i)}(n-1) & = & P_{t}^{(i)}(n-2) \; .\end{aligned}$$
To describe the coupling among first neighbors, let $P_{t}^{(i)}(k,l,m)$ be the joint probability that sites $i-1$, $i$ and $i+1$ are respectively in the states $k$, $l$ and $m$ at time $t$. Following the definitions of $\lambda$, $p$ and $q$ above, the equation for $P_{t}^{(i)}(1)$ thus becomes
$$\begin{aligned}
\label{eq:p1coupled}
P_{t+1}^{(i)}(1) & = & [1-(1-\lambda)(1-q)]P_{t}^{(i)}(1,0,1) \nonumber \\
& & + [1-(1-\lambda)(1-p)]\left(\sum_{k\neq 1}^{n-1}P_{t}^{(i)}(1,0,k)
\right. \nonumber \\
&& \left. + \sum_{k\neq 1}^{n-1} P_{t}^{(i)}(k,0,1)\right) \nonumber \\
& & + \lambda\sum_{k\neq1}^{n-1}\sum_{l\neq 1}^{n-1} P_{t}^{(i)}(k,0,l) \; .\end{aligned}$$
Finally, the dynamics for $P_{t}^{(i)}(0)$ can be obtained by the normalization condition $$\label{eq:norm}
\sum_{k=0}^{n-1}P_{t}^{(i)}(k) = 1 \; ,\forall t,i\; ,$$ which completes the set of equations for one-site probabilities.
It is reasonable to assume homogeneity in the system when $L\to\infty$, so we can drop the superscript $(i)$ in Eqs. (\[eq:refrac\]-\[eq:norm\]) and in what follows. We also expect isotropy (right-left symmetry) in the probabilities: $P_{t}(k,l) = P_{t}(l,k)$, $P_{t}(k,l,m)=P_{t}(m,l,k)$ etc. Recalling the normalization condition $\sum_{j_1=0}^{n-1}P_{t}(j_1,j_2,\ldots,j_m) = P_{t}(j_2,\ldots,j_m)$, one can rewrite Eq. (\[eq:p1coupled\]) as
$$\begin{aligned}
\label{eq:p1}
P_{t+1}(1) & = & \lambda P_{t}(0) + 2p(1-\lambda)P_{t}(1,0) \nonumber\\
& &+ (1-\lambda)(q-2p)P_{t}(1,0,1)\; .\end{aligned}$$
The stationary value of any joint probability will be denoted by omitting the subscript $t$, thus $P(\bullet)\equiv \lim_{t\to\infty}
P_{t}(\bullet)$. We start by solving Eqs. (\[eq:refrac\]) and (\[eq:norm\]) in the stationary state, which together yield
$$\label{eq:p1p0}
P(0) = 1 - (n-1)P(1)\; ,$$
a result which is exact and holds $\forall p,q$.
We are interested in obtaining the behavior of $P(1)$ as a function of $\lambda$ (or $r$). Note that $P(1)$ coincides with the average firing rate per neuron (measured in spikes per ms, according to the choice of $\tau$) in the limit $L,t\to\infty$. In simulations, firing rates have been calculated by division of the total number of spikes in the chain by $LT$, where $T \sim {\cal O}(10^5)$ and $L \sim {\cal O}(10^5)$ were the typical number of time steps and model neurons employed [@Copelli02]. We define $F(\lambda)\equiv P(1)$ as the response function of the system.
Due to the absolute nature of the refractory period, the maximum firing rate of the model neurons is $F_{max}\equiv 1/n$, a result which is easily obtained $\forall p,q$ by setting $\lambda=1$ in Eqs. (\[eq:p1\]) and (\[eq:p1p0\]). The dynamic range $\delta_{\lambda}$ of the response curve $F(\lambda)$ follows the definition commonly employed in biology [@Firestein93; @Rospars00]: $$\label{eq:dr}
\delta_{\lambda} = 10\log_{10}\left(\frac{\lambda_{0.9}}{\lambda_{0.1}}\right)\; ,$$ where $\lambda_{x}$ satisfies $$\label{eq:lambdmax}
F(\lambda_{x})=xF_{max}\; .$$ The dynamic range is therefore the number of decibels of input which are mapped into the $\simeq 9.5~\text{dB}$ of output comprised in the $[0.1F_{max},0.9F_{max}]$ interval (see Fig. \[fig:respostalinlog\]). In the biological context of the model, it measures the ability of the system to discriminate different orders of magnitude of stimulus intensity. We will show below that if one chooses to calculate $\delta_{r}$ using $r_{x}\equiv
-\tau^{-1}\ln(1 - \lambda_{x})$ instead of $\lambda_{x}$ in Eq. \[eq:dr\], results are essentially unchanged.
\[uncoupled\]Uncoupled neurons
==============================
The uncoupled case $p=q=0$ can be exactly solved by taking the limit $t\to\infty$ in Eq. (\[eq:p1\]) which, together with Eq. (\[eq:p1p0\]), yields
$$\label{eq:linsat}
P(1) = f(\lambda) = \frac{\lambda}{1 + (n-1)\lambda}\; .$$
This linear saturating response is depicted for $n=3$ (Figs. \[fig:campomedio\]-\[fig:respostalinlog\]) and $n=10$ (Figs. \[fig:campomedio\]-\[fig:pares\]), in complete agreement with simulations. It belongs to the family of Hill functions defined by $H_{\alpha}(x)\equiv Cx^\alpha/(x_0^\alpha+x^\alpha)$, where the Hill exponent in this case is $\alpha=1$.
The dynamic range can be promptly calculated: $\delta_{\lambda}(n) =
10\log_{10} \left\{ \left[ 1+9n \right] / \left[ 1+n/9 \right]
\right\}$ and $\delta_{r}(n) = 10\log_{10} \left\{ \ln\left[ 1+9/n
\right] / \ln\left[ 1+1/(9n) \right] \right\}$, both of which rapidly converge to $10\log_{10}(81)\simeq 19$ dB for moderate values of $n$ (see lower curves in Fig. \[fig:faixadinamica\]). As we shall see, the electrical coupling can lead to dynamic ranges typically twice as large.
![\[fig:campomedio\]Response curves for (a) $n=3$ and (b) $n=10$ automata: simulations (symbols) and mean field approximation \[lines, according to Eq. (\[eq:p1ss\])\]. From bottom to top, $p=0$, $0.3$, $0.6$ and $1$, $q=1-(1-p)^2$. In the simulations, standard deviations over 10 runs are smaller than symbol sizes, so error bars are omitted in all figures. Notice the negative slope and multi-valuedness of the single-site approximation for $p>1/2$ and $\lambda\leq 0$.](Fig1.eps){width="0.8\columnwidth"}
![\[fig:pares\]Response curves for (a) $n=3$ and (b) $n=10$ automata: simulations (symbols) and pair approximation \[lines, according to Eqs. (\[eq:2\]-\[eq:1\])\]. From bottom to top, $p=0$, $0.3$, $0.6$ and $1$, $q=1-(1-p)^2$. The pair approximation eliminates the small-$\lambda$ anomalies of the single-site solution, yielding excellent agreement with simulations for the extreme cases $p=0$ and $p=1$.](Fig2.eps){width="0.8\columnwidth"}
\[coupled\]Coupled Neurons: Mean Field Approximation
====================================================
As can be seen in Eq. (\[eq:p1\]), $P_t(1)$ depends on two- and three-site probabilities, and in general $k$-site probabilities depend on up to $(k+2)$-site probabilities. The dynamical description of the system thus requires an infinite hierarchy of equations. The mean field approximation at the single-site level corresponds to the simplest truncation of this hierarchy, and consists in discarding the influence of all neighbors in the conditional probabilities [@Atman03], thus $P_{t}(j_1|j_2,\ldots,j_m) \approx
P_{t}(j_1)$, which leads to $$\label{eq:singlesite}
P_{t}(j_1,\ldots,j_m) \approx \prod_{k=1}^{m}P_{t}(j_k) \; .$$ In this approximation, Eq. (\[eq:p1\]) becomes $$\begin{aligned}
\label{eq:p1ss}
P_{t+1}(1) & \approx & P_{t}(0)\left\{ \lambda + 2p(1-\lambda)P_{t}(1)
\right. \nonumber \\
&& \left. +(q-2p)(1-\lambda)P_{t}(1)^2 \right\}\; ,\end{aligned}$$ which, together with Eq. (\[eq:p1p0\]), can be used to eliminate $P(0)$ and render $P(1)=F(\lambda)$ implicitly through the relation
$$\label{eq:lambf}
\lambda \approx \frac{(1-2p)F + (2pn-q)F^2 +
(n-1)(q-2p)F^3}{[1-(n-1)F][1-2pF+(2p-q)F^2]} \; .$$
As a consistency check, notice that setting $p=q=0$ in Eq. (\[eq:lambf\]) recovers Eq. (\[eq:linsat\]) (in other words, mean field is exact for the uncoupled case, as it should). However, for $0<p,q\leq 1$, $F(\lambda)$ as given by Eq. (\[eq:lambf\]) yields in general a poor agreement with numerical simulations, as can be seen in Fig. \[fig:campomedio\] for different values of $p$. When $\lambda \simeq 0$, Eq. (\[eq:lambf\]) predicts $F\simeq
\lambda/(1-2p)$, which leads to obviously nonphysical results for $p\geq 1/2$ (see leftmost part of Fig. \[fig:campomedio\]). In particular, $F(\lambda)$ is multi-valued, leading to $\lim_{\lambda\to
0^{+}} F \neq 0$. The mean field result therefore suggests a transition to an ordered state at $\lambda=0$ which is simply forbidden by the automaton rules [@lewis00]. By generalizing Eq. (\[eq:p1ss\]), this failure to predict the absorbing state of the system can in fact be extended to regular lattices with coordination $z$, where the single-site approximation yields $F\stackrel{\lambda\to 0}{\simeq} \lambda/(1-pz)$. Since this level of approximation is clearly not satisfactory for the calculation of the dynamic range, a refinement is needed.
\[coupledPAIR\]Coupled Neurons: Pair Approximation
==================================================
The pair approximation consists in keeping the influence of only one neighbor in the conditional probabilities [@Atman03], thus $P_{t}(j_1|j_2,\ldots,j_m) \approx P_{t}(j_1|j_2)$. In this case $m$-site probabilities are reduced to combinations of up to two-site probabilities. In particular, three- and four-site probabilities become [@Atman03]
\[eq:pair\] $$\begin{aligned}
P(k,l,m) & \approx & \frac{P(k,l)P(l,m)}{P(l)} \\
P(j,k,l,m) & \approx & \frac{P(j,k)P(k,l)P(l,m)}{P(k)P(l)} \; .\end{aligned}$$
It is therefore possible to rewrite Eq. (\[eq:p1\]) in this approximation:
$$\label{eq:p1pair}
P_{t+1}(1) \approx \lambda P_{t}(0) + (1-\lambda)P_{t}(1,0)\left[2p
+ (q-2p)\frac{P_{t}(1,0)}{P_{t}(0)}\right]\; .$$
Eq. (\[eq:p1pair\]), on its turn, depends on $P_{t}(1,0)$, whose evolution can be exactly obtained (up to homogeneity and isotropy assumptions):
$$\begin{aligned}
\label{eq:p10}
P_{t+1}(1,0) & = & \lambda P_{t}(n-1,0) + p(1-\lambda)P_{t}(n-1,0,1)
\nonumber \\
& & + \lambda(1-\lambda)P_{t}(0,0) \nonumber \\
&& + p(1-\lambda)(1-2\lambda)P_{t}(1,0,0) \nonumber \\
&& - p^2(1-\lambda)^2 P_{t}(1,0,0,1) \; . \nonumber \\\end{aligned}$$
With the help of the pair approximation in Eqs. (\[eq:pair\]), Eq. (\[eq:p10\]) becomes
$$\begin{aligned}
\label{eq:p10pair}
P_{t+1}(1,0) & \approx & P_{t}(n-1,0) \left[\lambda +
p(1-\lambda)\frac{P_{t}(1,0)}{P_{t}(0)}\right] \nonumber \\
&& + (1-\lambda)P_{t}(0,0)\left[\lambda +
p(1-2\lambda)\frac{P_{t}(1,0)}{P_{t}(0)} \right. \nonumber \\
&& \left. - p^2(1-\lambda)\frac{P_{t}(1,0)^2 }{P_{t}(0)^2} \right]\; .\end{aligned}$$
Since $P_{t}(j,0)$ depends on $P_{t}(j-1,0)$ and $P_{t}(j-1,n-1)$, and $P_{t}(0,0)$ depends, among others, on $P_{t}(n-1,n-1)$, all the equations for two-site probabilities are in principle required for the dynamical description of the system. Together with the equations for single-site probabilities, they form a $(n^2+3n)/2$-dimensional map whose stationary stable solution can be analytically studied. While the Appendix contains details of the derivation of those equations, we discuss the main results below.
The main point to be noted is that the calculation of the stationary state presents additional difficulties when $n\geq 4$. In that case, the pair probabilities $P(j,0)$ with $2\leq j \leq n-2$ have the same stationary value, but differ from $P(n-1,0)$. In particular, for $p=q=1$ one obtains $P(j,0)=0$ \[$2\leq j \leq n-2$, see Eq. (\[eq:J\])\], which in turn leads to many other vanishing probabilities and gives the deterministic case a sparse stationary matrix \[see Eqs. (\[eq:p21\]), (\[eq:pj1\]) and (\[eq:p1j\])\]. Those terms do not exist for the $n=3$ case, which makes its analysis considerably simpler. In either case, for $n\geq 3$ one obtains the reasonable result $P(n-1,0) \approx P(1,0)$, the l.h.s. (r.h.s.) being associated to the end (beginning) of an excitable wave front \[see Eq. (\[eq:pdelta\])\]. Combining these results, a normalization condition and the linearity of Eq. (\[eq:p10pair\]) in $P_{t}(0,0)$, we obtain (see Appendix):
$$\begin{aligned}
\label{eq:2}
&& P(0) - P(1,0)\left\{2 + (n-3)\left[\frac{(1-p)P(0)+(p-q)P(1,0)}{P(0)-pP(1,0)}\right] \right\}
\nonumber \\
&\approx & \frac{P(1,0)P(0)[P(0)-pP(1,0)]}{\lambda P(0)^2 +
p(1-2\lambda)P(0)P(1,0) - p^2(1-\lambda)P(1,0)^2}\; ,\end{aligned}$$
which is valid $\forall n\geq 3$. Consider now the stationary state of Eqs. (\[eq:p1p0\]) and (\[eq:p1pair\]). They can be combined in a quadratic equation for $P(1,0)$, yielding $$\begin{aligned}
\label{eq:1}
&& (2p-q) P(1,0) \approx G_{\pm}(P(0)) \nonumber \\
&& \equiv pP(0) \pm \sqrt{\frac{P(0)\left\{P(0)[(n-1)p^2+2p-q +
\lambda(n-1)(2p-p^2-q)]+(q-2p)\right\}}{(n-1)(1-\lambda)}}\; .\end{aligned}$$
Since $P(1,0)$ must vanish $\forall p,q$ in the limit $\lambda\to 0$, $G_{-}$ is the only acceptable solution.
![\[fig:respostalinlog\]Linear-log plot of the response curve for $n=3$ automata with $p=q=1$ (filled circles), $p=0.5$ and $q=1-(1-p)^2$ (open triangles), and $p=q=0$ (open circles). Lines correspond to the pair approximation. Horizontal lines are $F=0.1F_{max}$ and $F=0.9F_{max}$, vertical lines are $\lambda=\lambda_{0.1}$ and $\lambda=\lambda_{0.9}$ and arrows illustrate the dynamic range $\delta_{\lambda}$ \[Eq. (\[eq:dr\])\] for $p=0$ and $p=1$. The dynamic range of a chain of neurons with deterministic spike propagation is about twice as large as that of its uncoupled counterpart.](Fig3.eps){width="0.99\columnwidth"}
![\[fig:faixadinamica\]Dynamic ranges (triangles for $\delta_\lambda$, circles for $\delta_r$) as a function of the number of states of the GHCA, obtained from the stationary solution of the pair approximation. Open (filled) symbols correspond to the $p=q=0$ ($p=q=1$) case. Inset: $\delta_\lambda$ as a function of $p$ for $n=10$ for simulations (dashed line) and pair approximation (solid line). In spite of the underestimation of the response observed in Fig. \[fig:pares\], the pair approximation is able to reproduce the behavior of the dynamic range as a function of the probability of spike transmission. ](Fig4.eps){width="0.99\columnwidth"}
![\[fig:respostaloglog\]Log-log plot of the response curve for $p=q=1$. Pair approximation (solid lines) and simulations (symbols) follow a power law ($\alpha=1/2$) for weak stimuli, while finite size effects lead to a linear response $F\simeq L\lambda$ (dotted lines) for $\lambda\lesssim\lambda_c(L)$.](Fig5.eps){width="0.99\columnwidth"}
The solution of Eqs. (\[eq:2\]) and (\[eq:1\]) determines $P(0)$ as a function of $\lambda$. Instead of numerically solving them, we iterate the $(n^2+3n)/2$-dimensional map involving the one- and two-site probabilities for each value of $\lambda$ until it converges to its stationary state. Despite the growing number of equations with $n$, this method has the advantage of avoiding unstable fixed points [@Atman03] \[Eqs. (\[eq:2\]) and (\[eq:1\]) can have more than one solution\]. Once $P(0)$ is known, the response $P(1) =
F(\lambda)$ is obtained via Eq. (\[eq:p1p0\]).
\[p=1\]Deterministic spike propagation ($p=1$)
----------------------------------------------
Ordinary differential equations (ODEs) are the standard modeling tool in computational neuroscience. This is due to the fact that, despite the stochastic nature of the opening and closing of individual ionic channels, a neuron containing a large number of such channels can very often be extremely well described by a deterministic dynamics [@Koch] (an approach which has been established since the seminal work of Hodgkin and Huxley [@HH52]). In the present context, it is therefore important to address the case $p=1$. This limit is consistent with a variety of scenarios in which, in addition to the dynamics of individual neurons, spike transmission is also well described by deterministic behavior. Specifically regarding our present study, deterministic spike transmission due to electrical coupling has previously been employed in the literature to model axo-axonal interactions both via ephaptic interactions (e.g. in the olfactory nerve [@Bokil01]) and gap junctions (e.g. in the hippocampus [@Traub99b; @lewis00]). This is in contrast with, say, dendro-dendritic gap junctions or chemical synapses (in the latter case, synaptic transmission can sometimes be as low as 10% due to the inherent stochasticity in the process of neurotransmitter release [@Kandel; @Koch]), where the $p=1$ limit can hardly be expected to apply. As we shall see in the following, in addition to its biological relevance, the response function for $p=1$ also has a different characteristic exponent which will help us understand the limiting behavior for $p\lesssim 1$.
Figure \[fig:pares\] shows the excellent agreement between the pair approximation and the simulations when $p=q=1$. One observes that the response is particularly enhanced in the low stimulus range. This feature is best seen in the logarithmic scale of Fig. \[fig:respostalinlog\]: in comparison with the uncoupled case $p=0$, the effect of the electrical interaction is to increase the sensitivity of the response for more than a decade, leading to a dramatic rise of the dynamic range.
For each value of $n$, we can thus obtain the stationary response $F(\lambda)$ and the dynamic ranges $\delta_\lambda$ and $\delta_r$ in the pair approximation. Even though the response curve changes considerably for varying $n$ (since $F$ is bounded by $F_{max} = 1/n$, see Fig. \[fig:pares\]), the dynamic range levels off smoothly, as can be seen in Fig. \[fig:faixadinamica\]. For increasing $n$, the dynamic range of the $p=q=1$ case approaches twice the value for the uncoupled case. The fact that this result holds for both $\delta_r$ and $\delta_\lambda$ can be understood on the basis of the low-stimulus amplification, which plays the central role in the phenomenon: in this regime $\lambda$ is approximately linear in $r$. Should one choose a different relationship $\lambda(r)$, $\delta_r$ would obviously have different values, but the drastic enhancement in the response due to the electrical coupling would not be affected.
In order to understand the low-stimulus amplification induced by the coupling, we have analyzed Eqs. (\[eq:2\]-\[eq:1\]) when $\lambda
\simeq 0$. Inspection of Fig. \[fig:pares\] and previous numerical simulations [@Copelli02] suggest that $P(1)\simeq
C\lambda^\alpha$, with $\alpha<1$. This ansatz can be inserted into Eqs. (\[eq:p1p0\]) and (\[eq:2\]-\[eq:1\]) for general $p$ and $q$, yielding $\alpha=1/2$ [*and*]{} $p=1$ as solutions. Deterministic spike propagation therefore leads to a power law response $$\label{eq:sqrt}
F(\lambda) \stackrel{\lambda\to 0}{\simeq} \sqrt{2\lambda}\; ,$$ a result that holds $\forall n,q$, as should be expected. This power law suggests a Hill function with $\alpha=1/2$, which is an excellent approximation for $F(\lambda)$ in the whole $\lambda$ interval when $n$ is large. This result explains the doubling of the dynamic range as compared to the uncoupled case and is reminiscent of reaction-diffusion processes modeled by lattice gases [@Stinchcombe93; @Grynberg94; @Oliveira99; @Mendonca98] and partial differential equations [@Ohta05]. Since the Hill function can be regarded as a saturating Stevens law, it is interesting to note that the experimental values of the Stevens exponents for light and smell intensities are respectively $\alpha\simeq 0.5$ and $\alpha\simeq 0.6$ [@Stevens].
Let us now consider a chain with [*finite*]{} $L$ and a very small value of $\lambda$ such that a single external stimulus occurs in a given time interval. In this case, the deterministic nature of the propagation would lead to $L$ spikes in the chain, while a single spike would be observed if the neurons were uncoupled. One would thus have $F\simeq Lf$, and since $f\stackrel{\lambda\to 0}{\simeq}\lambda$ \[from Eq. (\[eq:linsat\])\] we obtain $F\stackrel{\lambda\to
0}{\simeq}L\lambda$. This corresponds to a linear regime where excitable waves do not interact. If one increases $\lambda$, waves will start annihilating each other, leading to the power law response of Eq. (\[eq:sqrt\]), as can be clearly seen in Fig. \[fig:respostaloglog\]. For a given system size $L$, there is therefore a crossover value $\lambda_c(L)\simeq 2/L^2$ from a linear to a nonlinear response. In an infinite chain, there is no linear response since for any nonzero stimulus rate two excitable waves will inevitably interact.
To assess the finite size effects in the biological context of the model, we notice that the dynamic range will be affected only if $\lambda_c(L) \gtrsim \lambda_{0.1}$, that is, for $L\lesssim 20n$. For neurons with refractory periods of the order of tens of ms, neuronal assemblies with $L \gtrsim 10^{3-4}$ should therefore be well approximated by the limit $L\to\infty$, as can be seen in Fig. \[fig:faixadinamicaL\]. It is important to emphasize, however, that even small chains dominated by finite size effects still possess dynamic ranges which are significantly larger than those of the uncoupled case. For $\lambda_{0.1} \lesssim \lambda_c(L)$, the dynamic range increases approximately logarithmically with the total number of connected neurons, a result which holds for regular lattices in any dimension [@Copelli05b].
![\[fig:faixadinamicaL\]Dynamic range as a function of the system size $L$ for $p=q=1$. Lines are just guides to the eye.](Fig6.eps){width="0.99\columnwidth"}
\[pneq1\]Probabilistic spike propagation ($p\neq 1$)
----------------------------------------------------
For $p\neq 1$, communication between spiking and resting neurons may eventually fail. This provides us with the simplest test under which the robustness of the mechanism for dynamic range enhancement can be checked. From the biological point of view, this regime could be useful for modeling networks of neurons connected by chemical synapses, for instance.
We start the analysis of the $p\neq 1$ case by noticing in Figs. \[fig:pares\] and \[fig:respostalinlog\] that the agreement between simulations and the pair approximation is better than the mean field results (specially in the low-stimulus region), but certainly not so good as in the extreme cases $p=0$ and $p=1$. This inevitably affects the estimation of the dynamic range via the stationary state of the pair approximation (see below), but nonetheless allows us to understand qualitatively how the response changes as $p$ varies.
As pointed out in the preceding section, the dynamic range is enhanced for $p=1$ primarily due to the low-stimulus amplification associated to the propagation of excitable waves. As opposed to the deterministic case, however, for $p\neq 1$ a single excitable wave traveling in an infinite chain initially at rest will eventually die out. We should therefore expect a qualitative change in the response function for $\lambda\simeq 0$. This is indeed confirmed by reinserting the ansatz $P(1)\simeq C\lambda^\alpha$ in Eqs. (\[eq:p1p0\]) and (\[eq:2\]-\[eq:1\]) without the constraint $\alpha<1$. In this case, the linear behavior suggested by the plots in Fig. \[fig:pares\] is easily confirmed:
![\[fig:respostalogpneq1\]Log-log plot of the response curve: pair approximation (solid lines) and simulations (symbols) with $q=1-(1-p)^2$ and $n=3$. For $p\lesssim 1$, there is a crossover between $\alpha=1$ and $\alpha=1/2$. The horizontal arrow shows $0.1F_{max}$.](Fig7.eps){width="0.99\columnwidth"}
$$\label{eq:1+p}
F(\lambda) \stackrel{\lambda\to 0}{\simeq}
\left(\frac{1+p}{1-p}\right)\lambda \; ,$$
which is again valid $\forall n,q$. Therefore, the low-stimulus response for $p<1$ is governed by $\alpha=1$, which is confirmed by the simulations displayed in Fig \[fig:respostalogpneq1\]. Interestingly, such a change in exponent for $p<1$ seems to be absent from reaction-diffusion models in lattice gases [@Stinchcombe93; @Grynberg94; @Oliveira99; @Mendonca98] as well as partial differential equations [@Ohta05].
Thanks to the growing coefficient in Eq. (\[eq:1+p\]), for $p\lesssim 1$ the proximity to the transition that occurs at $p=1$ produces a crossover in the response from a linear to a square root behavior, dismissing the suspicion that a larger exponent might severely deteriorate the enhancement of the dynamic range (see Fig. \[fig:respostalogpneq1\]). In particular, notice that, for $p\lesssim 1$, $\alpha=1/2$ is the dominant exponent at $F=0.1F_{max}$, which is used to calculate the dynamic range (see horizontal arrow in Fig. \[fig:respostalogpneq1\]). This explains the smooth monotonic increase in $\delta_\lambda$ with $p$, as shown in the inset of Fig. \[fig:faixadinamica\], even though the exponent changes discontinuously at $p=1$. On the one hand, we observe that deterministic spike propagation ($p=1$) is certainly not essential for the enhancement of the dynamic range, in the sense that any $p>0$ yields a better response than uncoupled neurons. On the other hand, it is interesting to point out that, as $p$ is varied from 0 to 1, the increase in dynamic range is particularly pronounced for $p\gtrsim
0.9$. This is in agreement with the conjecture that the reliability of electrical coupling among spiking neurons could indeed play a significant role in early sensory processing.
\[conclusion\]Concluding remarks
================================
We have calculated the collective response to a Poisson stimulus of a chain of electrically coupled excitable neurons modeled by $n$-state Greenberg-Hastings cellular automata. The single-site mean field approximation has been shown to give poor results, failing to predict the absorbing state of the lattice in the absence of stimulus for $p\geq 1/2$. The pair approximation yields a response curve which agrees reasonably well with simulations [*in the whole stimulus range*]{}. It is interesting to remark that the agreement is particularly good when $p=q=1$, a deterministic regime in which the GHCA lattice mimics a system of coupled nonlinear ODEs. This reinforces an interesting perspective in the context of computational neuroscience: the possibility of applying techniques from nonequilibrium statistical mechanics to the study of spatially extended nonlinear systems.
The enhancement of the dynamic range in the presence of electrical coupling is due to low-stimulus amplification. For uncoupled neurons ($p=0$) the response is governed by the Hill exponent $\alpha=1$, leading to a dynamic range of $\sim 19$dB. For coupled neurons this value can be doubled in the limit $p=q\to 1$, when the Hill exponent becomes $\alpha=1/2$. This value is close to Stevens exponents observed in psychophysical experiments of smell and light intensities. For $0<p<1$, the exponent remains $\alpha=1$, but the dynamic range increases smoothly, which can be understood on the basis of the crossover behavior observed in the response function for $p\lesssim
1$.
In the context of experiments at the cellular level, the enhancement of the dynamic range associated with an increase in sensitivity is also observed in both the olfactory [@Wachowiak01] and visual [@Deans02] systems. While the dynamic range of OSNs (the neurons which perform the initial transduction) is about $\sim
10$dB [@Rospars00; @Rospars03], the glomeruli (the next processing layer) have dynamic ranges at least twice as large [@Wachowiak01]. It remains to be investigated experimentally whether this enhancement is indeed due to ephaptic interactions among the unmyelinated OSN axons in the olfactory nerve.
![\[fig:deans\]Experimental response curves (normalized firing rate vs. light intensity) of retinal on-center ganglion cells in linear-log (main plot) and log-log (inset) scales (data extracted from Fig. 6 of Ref. [@Deans02]). Filled (open) circles are for WT (Cx36-KO) mice, solid (dashed) lines show the results of the pair approximation, thus $L\to\infty$, with $p=q=1$ ($p=q=0$). Upper curves are for $n=10$, lower curves are for $n=3$. The dot-dashed line corresponds to simulations with $n=10$, $p=q=1$ and $L=20$.](Fig8.eps){width="0.99\columnwidth"}
Stronger experimental support for our conjecture on the role of electrical interactions is available for the mammalian retina. Deans et al. [@Deans02] have measured the firing rates of on-center ganglion cells for varying light intensity (measured in isomerized molecules of rhodopsin per rod per second, or Rh\*/rod/s). The response curves have been obtained for both wild type (WT) mice as well as mice in which the expression of the protein connexin36 (responsible for the gap junction intercellular channels) has been genetically knocked out (Cx36-KO). The difference in the response curves can be seen in Fig. \[fig:deans\]. They present the same qualitative behavior of the curves shown in Fig. \[fig:respostalinlog\], exhibiting an increase in dynamic range in the presence of electrical coupling: 14dB for Cx36-KO and 23dB for WT, values which are of the same order as those of Fig. \[fig:faixadinamicaL\]. In particular, the exponent of the “coupled” (WT) case is $\alpha\simeq 0.58$ (see inset), which is slightly larger than what is obtained in the pair approximation.
The quantitative agreement between the analytical and experimental curves is limited. On the one hand, the theoretical $n=3$ curve can provide a good fit of the Cx36-KO data for $p=q=0$, while the coupled case $p=q=1$ does not adjust well to the WT data. For $n=10$ and $p=q=1$, on the other hand, the WT data are well matched by simulations with a finite $L=20$ system (staying below the $L\to\infty$ pair approximation), but for $p=q=0$ the same $n=10$ automata are unable to give a good fit of the Cx36-KO data. The difficulties of a quantitative match are not surprising: the retina is organized in layers which have, to first order, a two-dimensional structure, signal processing from the photoreceptors to the ganglion cells involves a complex intermediate neuronal circuit (with bipolar, horizontal and amacrine cells [@Shepherd]), and individual neurons themselves can have subtle dynamical properties (such as adaptation, for instance). All these properties are clearly absent from our simple one-dimensional CA model. Yet it correctly predicts the reduction in the dynamic range of a neuronal system which loses electrical coupling among its cells.
In order to have a quantitative agreement between experimental and theoretical curves, additional modeling efforts are needed which incorporate specific details of the system under consideration. However, the response of simple models of excitable media remains an important subject to be studied, precisely because they have the potential to reveal simple mechanisms and scaling relations [@Ohta05] whose robustness can thereafter be subjected to further testing in experiments and more detailed models. In this context, the simple Greenberg-Hastings CA strikes an interesting balance, on the one hand capturing essential features of collective neuronal dynamics, while on the other hand lending itself to analytical techniques borrowed from nonequilibrium statistical mechanics.
The authors would like to thank O. Kinouchi, S. G. Coutinho, A. C. Roque, R. F. Oliveira, R. Publio, M. J. de Oliveira and an anonymous referee for useful discussions and comments. LSF is supported by UFPE/CNPq/PIBIC. MC acknowledges support from Projeto Enxoval (UFPE), FACEPE, CNPq and special program PRONEX.
\[ap:pair\]The equations for two-site probabilities
===================================================
Dynamics
--------
In all derivations below, homogeneity and isotropy are assumed. The sign “$\approx$” denotes that the equality holds in the pair approximation \[Eqs. (\[eq:pair\])\]. We start by writing down the equation for $P_{t}(0,0)$, which holds $\forall n\geq 3$:
$$\begin{aligned}
P_{t+1}(0,0) & = & P_{t}(n-1,n-1) \nonumber \\
&& + 2(1-\lambda)\left[P_{t}(n-1,0) - pP_{t}(1,0,n-1) \right] \nonumber \\
& & + (1-\lambda)^2 \left[ P_{t}(0,0) - 2pP_{t}(1,0,0) \right. \nonumber \\
&& \left. + p^2P_{t}(1,0,0,1) \right] \nonumber \\
\label{eq:p00t} \ & \approx & P_{t}(n-1,n-1) \nonumber \\
&& + 2(1-\lambda)P_{t}(n-1,0)\left[1-p\frac{P_{t}(1,0)}{P_{t}(0)} \right]
\nonumber \\
& & + (1-\lambda)^2P_{t}(0,0)\left[ 1-2p\frac{P_{t}(1,0)}{P_{t}(0)}
\right. \nonumber \\
&& \left. + p^2\frac{P_{t}(1,0)^2}{P_{t}(0)^2} \right] .
%% \right. \nonumber \\
%% && \left. \end{aligned}$$
The dynamics for two-site probabilities in the refractory period obey a simple recursive rule due to the deterministic evolution of the automata:
$$\label{eq:pjk}
P_{t+1}(j,k) = P_{t}(j-1,k-1) \; , \; \; 2\leq j,k \leq n-1\; .$$
On the one hand, diagonal terms $P_{t}(j,j)$ with $j \geq 2$ recursively depend on $P_{t}(1,1)$, whose dynamics can be written as follows:
$$\begin{aligned}
P_{t+1}(1,1) & = & \lambda^2P_{t}(0,0) +
2p\lambda(1-\lambda)P_{t}(1,0,0) \nonumber \\
&& + p^2(1-\lambda)^2P_{t}(1,0,0,1)\nonumber \\
\label{eq:p11}\ & \approx & P_{t}(0,0) \left[\lambda^2 +
2p\lambda(1-\lambda)\frac{P_{t}(1,0)}{P_{t}(0)} \right. \nonumber \\
&& + \left. p^2(1-\lambda)^2\frac{P_{t}(1,0)^2}{P_{t}(0)^2}\right]\; .\end{aligned}$$
Off-diagonal terms, on the other hand, ultimately depend on $P_{t}(j,1)$. For $j=2$, the equation is simply
$$\begin{aligned}
\label{eq:p21}
P_{t+1}(2,1) & = & (\lambda+p-p\lambda)P_{t}(1,0) \nonumber \\
&& + (1-\lambda)(q-p)P_{t}(1,0,1) \nonumber \\
&\approx & P_{t}(1,0)\left\{ \lambda + (1-\lambda)\left[ p
\vphantom{\frac{P_{t}(1,0)}{P_{t}(0)}} \right. \right. \nonumber \\
&& \left.\left.
+ (q-p)\frac{P_{t}(1,0)}{P_{t}(0)} \right] \right\}
\; ,\end{aligned}$$
while for $j\geq 3$ one has
$$\begin{aligned}
\label{eq:pj1}
P_{t+1}(j,1) & = & \lambda P_{t}(j-1,0) + p(1-\lambda)P_{t}(j-1,0,1)\nonumber \\
\ & \approx & P_{t}(j-1,0)\left[\lambda + p(1-\lambda)\frac{P_{t}(1,0)}{P_{t}(0)} \right]\; .\end{aligned}$$
Finally, one needs equations for $P_{t}(j,0)$, $j\geq 2$ \[recall Eq. (\[eq:p10pair\]) for $P_{t}(1,0)$\]. Like in Eq. (\[eq:p21\]), the case $j=2$ must be considered separately: $$\begin{aligned}
\label{eq:p20}
P_{t+1}(2,0) & = & P_{t}(1,n-1) + (1-\lambda)(1-p)P_{t}(1,0) \nonumber \\
&& + (1-\lambda)(p-q)P_{t}(1,0,1) \nonumber \\
&\approx & P_{t}(1,n-1)+(1-\lambda)P_{t}(1,0)\left[(1-p) \vphantom{\frac{P_{t}(1,0)}{P_{t}(0)}}\right. \nonumber \\
&& \left. +(p-q)\frac{P_{t}(1,0)}{P_{t}(0)}
\right]\; .\end{aligned}$$ For $j\geq 3$, on the other hand, one immediately obtains
$$\begin{aligned}
\label{eq:pj0}
P_{t+1}(j,0) & = & P_{t}(j-1,n-1) \nonumber \\
&& + (1-\lambda)\left[P_{t}(j-1,0) - pP_{t}(j-1,0,1) \right] \nonumber \\
\ & \approx & P_{t}(j-1,n-1) \nonumber \\
&& + (1-\lambda)P_{t}(j-1,0)\left[1 - p\frac{P_{t}(1,0)}{P_{t}(0)} \right]\; ,\end{aligned}$$
which completes the set of all pair equations. Upon iteration of Eqs. (\[eq:refrac\],\[eq:norm\],\[eq:p1pair\],\[eq:p10pair\],\[eq:p00t\]-\[eq:pj0\]), normalization conditions properly imposed in the initial conditions are naturally preserved. To determine the response function $P(1)=F(\lambda)$, we wait until the $(n^2+3n)/2$-dimensional map reaches a stationary state for each value of $\lambda$. We describe below how the analysis of the stationary state can be reduced to just two equations \[Eqs. (\[eq:2\]-\[eq:1\])\].
Stationary state
----------------
We start by handling the case $n>4$. In the stationary state, the first term on the r.h.s. of Eq. (\[eq:pj0\]) becomes, via recursive iterations of Eq. (\[eq:pjk\]), $$\label{eq:p1j}
P(j-1,n-1) = P(1,1+n-j), \; \forall j\geq 3\; .$$ The above result can on its turn be further developed by means of Eq. (\[eq:pj1\]) as long as $1+n-j\geq 3$, rendering the stationary state of Eq. (\[eq:pj0\]): $$\begin{aligned}
P(j,0) &\approx& P(n-j,0)\left[\lambda+p(1-\lambda)\frac{P(1,0)}{P(0)}\right] \nonumber \\
&& +(1-\lambda)P(j-1,0)\left[1 - p\frac{P(1,0)}{P(0)} \right]\; , \nonumber \\
&& 3\leq j \leq n-2\; .\end{aligned}$$ Notice that we have a nonhomogeneous set of $n-4$ linear equations for $x_j \equiv P(j,0)$: $x_j \approx ax_{n-j} + (1-a)x_{j-1}$, where $a\equiv \lambda + p(1-\lambda)P(1,0)/P(0)$ and $x_2 = P(2,0)$ accounts for the nonhomogeneity in the equations for $x_3$ and $x_{n-2}$. The solution of these equations is simply $x_{n-2} \approx x_{n-1} \approx
\ldots \approx x_{3} \approx x_{2}$, as can be checked by inspection. The combination of Eqs. (\[eq:p20\]) and (\[eq:pj1\]) in the stationary state, on the other hand, leads to
$$\begin{aligned}
\label{eq:J}
P(j,0) & \approx & J\left[P(1,0),P(0)\right] \nonumber \\
& \equiv &
P(1,0)\left[\frac{(1-p)P(0)+(p-q)P(1,0)}{P(0)-pP(1,0)}\right]\; , \nonumber \\
&& 2\leq j \leq n-2 \; . \end{aligned}$$
One therefore obtains
$$\begin{aligned}
\label{eq:xj}
P(n-2,0) & \approx & P(n-3,0) \approx \ldots \approx P(2,0) \nonumber \\
&\approx & J\left[P(1,0),P(0)\right]\; .\end{aligned}$$
Finally, notice that $P(n-1,0)$ can be obtained by combination of Eqs. (\[eq:pj0\]), (\[eq:xj\]) and (\[eq:p21\]): $$\label{eq:pdelta}
P(n-1,0) \approx P(1,0)\; ,$$ which completes the proof for $n>4$. For $n=4$, it suffices to invoke Eqs. (\[eq:p20\]) and (\[eq:pj1\]) to show that $P(2,0) \approx
J\left[P(1,0),P(0)\right]$. With this result, Eq. (\[eq:pdelta\]) holds for $n\geq 4$. Finally, for $n=3$, Eqs. (\[eq:p20\]) and (\[eq:p21\]) together also lead to Eq. (\[eq:pdelta\]).
Invoking the normalization condition $P_{t}(0)=\sum_{j=0}^{n-1}P_{t}(j,0)$, one can deduce that, on the one hand, $$\label{eq:p00}
P(0,0) = P(0) - 2P(1,0) - (n-3)J\left[P(1,0),P(0)\right]\; .$$ On the other hand, in the stationary state Eq. (\[eq:p10pair\]) depends linearly on $P(0,0)$, so it can be inverted, yielding \[after substitution of Eq. (\[eq:pdelta\])\] $P(0,0)$ as a function of $P(1,0)$ and $P(0)$. Equaling this function to Eq. (\[eq:p00\]), $P(0,0)$ is eliminated and one obtains Eq. (\[eq:2\]).
[36]{} natexlab\#1[\#1]{}bibnamefont \#1[\#1]{}bibfnamefont \#1[\#1]{}citenamefont \#1[\#1]{}url \#1[`#1`]{}urlprefix\[2\][\#2]{} \[2\]\[\][[\#2](#2)]{}
, ****, ().
, ****, ().
, ** (, ).
, , , , , ****, ().
, , , , ****, ().
, , , , ****, ().
, ****, ().
, ****, ().
, , , ****, ().
, ****, ().
, , , , , ****, ().
, , , , ****, ().
, , , , ****, ().
, ** (, , ).
, , , , , ****, ().
, , , ****, ().
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, , , ****, ().
, , , , , ****, ().
, ****, ().
, , , ****, ().
, , , ****, ().
, ****, ().
, ****, ().
, , , , ****, ().
, , , eds., ** (, ).
, , , ****, ().
, , , ****, ().
, ****, ().
, ****, ().
, ****, ().
, ed., ** (, ).
[^1]: Corresponding author
| {
"pile_set_name": "ArXiv"
} | ArXiv |