1: %\documentclass[aps,prl,amsmath,ams,groupedaddress,twocolumn,showpacs]{revtex4}
2: \documentclass[aps,pra,floatfix,amsmath,ams,preprint,superscriptaddress,showpacs]{revtex4}
3: %\documentclass[aps,pra,floatfix,amsmath,preprint,showpacs,11pt]{revtex4}
4: %\documentclass[aps,floatfix,amsmath,dstroke,amssymb,prlsymb,twocolumn, showpacs]{revtex4}% You should use BibTeX and apsrev.bst for
5: % Choosing a journal automatically selects the correct APS
6: % BibTeX style file (bst file), so only uncomment the line
7: % below if necessary.
8: %\bibliographystyle{apsrev}
9: \usepackage{graphicx}
10: \usepackage{dsfont}
11: %\usepackage{natbib}
12: %\citestyle{nature}
13: \begin{document}
14:
15: \title{The modular multiplication operator and the quantized bakers maps}
16:
17: \author{ Arul Lakshminarayan\footnote{On leave from: Department of Physics, Indian Institute of Technology Madras, Chennai, 600036, India.}}
18: \email[]{arul@physics.iitm.ac.in}
19: \affiliation{Max-Planck-Institut f\"ur Physik komplexer Systeme, N\"othnitzer Stra$\beta$e 38, D-01187 Dresden, Germany}
20: \date{\today}
21: \preprint{IITM/PH/TH/2007/7}
22:
23: %\date{\today}
24: \begin{abstract}
25: The modular multiplication operator, a central subroutine in Shor's factoring algorithm, is shown to
26: be a coherent superposition of two quantum bakers maps when the multiplier is $2$. The
27: classical limit of the maps being completely chaotic, it is shown that there exist perturbations that push the modular multiplication operator into regimes of generic quantum chaos with spectral fluctuations that are those of random matrices. For the initial state of relevance to Shor's algorithm we study fidelity decay due to phase and bit-flip errors in a single qubit and show exponential decay with shoulders at multiples or half-multiples of the order. A simple model is used to gain some understanding of this behavior.
28: %\pacs{05.45.Mt,03.67.Lx}
29: \end{abstract}
30: \pacs{03.67.Lx,05.45.Mt}
31: \maketitle
32:
33: %\newpage
34:
35: \newcommand{\newc}{\newcommand}
36: \newc{\beq}{\begin{equation}}
37: \newc{\eeq}{\end{equation}}
38: \newc{\kt}{\rangle}
39: \newc{\br}{\langle}
40: \newc{\beqa}{\begin{eqnarray}}
41: \newc{\eeqa}{\end{eqnarray}}
42: \newc{\pr}{\prime}
43: \newc{\longra}{\longrightarrow}
44: \newc{\ot}{\otimes}
45: \newc{\rarrow}{\rightarrow}
46: \newc{\h}{\hat}
47: \newc{\bom}{\boldmath}
48: \newc{\btd}{\bigtriangledown}
49: \newc{\al}{\alpha}
50: \newc{\be}{\beta}
51: \newc{\ld}{\lambda}
52: \newc{\sg}{\sigma}
53: \newc{\p}{\psi}
54: \newc{\eps}{\epsilon}
55: \newc{\om}{\omega}
56: \newc{\mb}{\mbox}
57: \newc{\tm}{\times}
58: \newc{\hu}{\hat{u}}
59: \newc{\hv}{\hat{v}}
60: \newc{\hk}{\hat{K}}
61: \newc{\ra}{\rightarrow}
62: \newc{\non}{\nonumber}
63: \newc{\ul}{\underline}
64: \newc{\hs}{\hspace}
65: \newc{\longla}{\longleftarrow}
66: \newc{\ts}{\textstyle}
67: \newc{\f}{\frac}
68: \newc{\df}{\dfrac}
69: \newc{\ovl}{\overline}
70: \newc{\bc}{\begin{center}}
71: \newc{\ec}{\end{center}}
72: \newc{\dg}{\dagger}
73: \newc{\md}{\mbox{mod}}
74: \newc{\prq}{\mbox{PR}_q}
75: \newc{\rt}{\sqrt{2}}
76:
77: \section{Introduction}
78:
79: Given an $M$ dimensional complex Hilbert space consider an orthonormal basis $|m\kt$, $m=0,\ldots,M-1$.
80: The modular multiplication by a number $p$ coprime to $M$ is the unitary permutation operator $U_p$:
81: \beq
82: U_p|m\kt \longrightarrow |m\, p \,( \mbox{mod}\, M)\kt.
83: \eeq
84: Repeated application of $U_p$ is also the modular exponentiation operator and lead to $U_p^k|m \kt =|m \,p^k\, ( \mbox{mod}\, M)\kt$. $U_p$ is periodic, i.e. there exists $k_0$ such that $U_p^{k_0}=I$, or $p^{k_0}= 1(\md M)$. This period though is an irregular function of $M$ for a given $p$ and is the ``multiplicative order'' of $p\, (\mbox{mod}\, M)$ \cite{Mathworld}. This operator is a crucial subroutine of Shor's factoring algorithm \cite{Shor} in which it is required to
85: perform the following operation on a bipartite Hilbert space:
86: \beq
87: |j\kt |1\kt \longrightarrow |j \kt | p^j \md M \kt
88: \eeq
89: where $0\le j \le 2^t-1$ and $M<2^t$. Typically $M$ is a large number and therefore we need
90: to calculate large powers and their residues $\md M$. This is considerably simplified by modular
91: arithmetic \cite{NielsenChuang} and the whole modular exponentiation step can be performed with $\mathcal{O}((\log M)^3)$
92: number of gates \cite{Beckman1996}. Once this is done, a quantum Fourier transform extracts the
93: period $k_0$ with reasonable rate of success. Given the period (order) it is possible to find a factor efficiently,
94: provided that $p_{k_0}$ is even and $p^{k_0/2} \neq -1 (\md M)$ by well-known procedures of number theory,
95: and using only classical computers \cite{NielsenChuang}.
96:
97: Shor's algorithm exploits the polynomial speed of the quantum Fourier transform to find the order
98: and hence offers a polynomially scaling algorithm for factoring numbers. Shor's algorithm has been implemented experimentally \cite{Chuang} although the number so factored is still very small to excite any practical application. The
99: effect of decoherence and gate errors on Shor's algorithm are important considerations and have been addressed by several authors previously \cite{Chuang96,Fowler,Hu,Devitt04,Miquel,Shepel07}. For instance Ref.~\cite{Chuang96}discuss the impact of environmental decoherence on the algorithm, while in \cite{Devitt04} direct detailed simulations have shown that the Shor algorithm is highly sensitive to gate errors, and the effect of static imperfections have been studied recently in\cite{Shepel07}. In this paper we will not directly simulate Shor's algorithm but look closely at the
100: modular multiplication and exponentiation for the special and simplest case $p=2$. We will call
101: $U_2$ as $S$, the shift operator as it performs the simple action of a qubit cyclic shift if $M$ is a power of $2$.
102:
103: Quantum algorithms have been studied earlier with a view to see if they had properties of
104: quantum chaotic systems \cite{Braun,Maity}. Recently it was shown that the spectrum of the unitary part of Shor's algorithm, properly desymmetrized had typical random matrix fluctuations \cite{Maity}, indicating that the operator itself may be quantum chaotic. However it was also pointed out that the origin of the chaos is the modular exponentiation part which is akin
105: to nongeneric quantum chaotic systems such as the cat maps \cite{QCat}. Here we make this connection more
106: precise and show that the classical limit of these subroutines is an ad-mixture of two bakers maps. Bakers maps are paradigms of deterministic classical chaos that are as random as a coin toss \cite{LL,Tabor}. The dimensionless inverse Planck constant in the Shor algorithm is the number to be factored and hence the classical limit is reached through a practically important regime.
107: We show that due to the proximity with such operators, there are perturbations that push the modular exponentiation
108: part (and therefore indeed the whole of Shor's algorithm) into regimes of generic quantum chaos. However we are not
109: that much interested in stationary state properties as in time-evolving states, in fact on those states that are
110: used in Shor's algorithm. Hence we study the fidelity of repeated modular multiplication, or the modular exponentiation, and show how the decay depends on the classical limit. We also provide a simple model for the exponential fidelity decay that is exactly solvable and captures the actual behavior reasonably well.
111:
112:
113: \section{The baker and the shift operator}
114:
115: The classical baker's map\cite{LL,ArnAvez,Tabor} $B_c$ is an area-preserving transformation
116: of the unit phase-space square $[0,1)\times [0,1)$ onto itself, which takes a
117: phase-space point $(q,p)$ to $(q',p')$ where $(q'=2q,\, p'=p/2)$ if
118: $0\le q<1/2$ and $(q'=2q-1,\,p'=(p+1)/2)$ if $1/2\le q<1$. The
119: stretching along the horizontal $q$ direction by a factor of two is
120: compensated exactly by a compression in the vertical $p$
121: direction. This is well known to be a fully chaotic system that in a
122: mathematically precise sense is as random as a coin toss\cite{Ornstein}. The
123: area-preserving property makes this map a model of chaotic two-degree
124: of freedom Hamiltonian systems. The lack of a generating Hamiltonian is compensated
125: by the existence of a classical generating function of the canonical transformation $B_c$.
126: The chaos is inferred by expressing a phase space point in the binary representation, if
127: $q=0.a_0a_1a_2\cdots$ and $p=0.a_{-1}a_{-2}a_{-3}\cdots$, where $a_i$
128: are either $0$ or $1$, then $q'=0.a_1a_2\cdots$ and
129: $p'=0.a_{0}a_{-1}a_{-2}a_{-3}\cdots$. Thus the most significant bits
130: of $q$ are lost at the rate of a bit per iteration, leading to an
131: exponential increase in any initial error. The Lyapunov exponent is
132: $\log(2)$ per iteration. This ``left-shift'' is in fact an important
133: mechanism for the generation of Hamiltonian chaos\cite{Tabor}, and in more
134: complicated forms arises generically.
135:
136: This was quantized first by Balazs and Voros\cite{BalVor1,BalVor2}.
137: Quantization in this context is the construction
138: of an appropriate unitary operator that evolves states over one
139: iteration and has the correct classical limit. Symmetries that may be broken
140: on quantization must be restored in this limit. The
141: Hilbert space of states is finite dimensional, has $N$ position and
142: momentum states, denoted by $|q_n\kt$ and $|p_m\kt$. If periodic
143: boundary conditions are assumed, $|q_{n+N}\kt=|q_n\kt$,
144: $|p_{m+N}\kt=|p_m\kt$, this implies that the transformation functions
145: between position and momentum is the discrete Fourier transform:
146: $(F_{N})_{mn}=\br p_m|q_n\kt = \exp[-2 \pi i mn/N]/\sqrt{N}$,
147: $m,n=0,1,2,\ldots,N-1$. Here $N$ is an effective scaled Planck
148: constant as $N=A/h=1/h$, where $A$ is the area of the phase space,
149: here unity. Thus the classical limit is the large $N$ limit. If $B$ is
150: the quantum baker's map, Balazs and Voros required that $\br
151: p_m|B|q_n\kt = \sqrt{2}\, \br p_m|q_{2n}\kt\, =\, (F_{N/2})_{mn}$ if
152: $n$ and $m$ are {\it both} $\le N/2-1$. This is almost like requiring
153: that $B$ takes $|q_n\kt$ to $|q_{2n}\kt$ mimicking the classical
154: stretching action, except that the momentum components above $N/2$
155: are set zero. ($\br p_m|B|q_n \kt=0$ for $p_m\ge N/2$ and $q_n <N/2$).
156: It is also clear from this that $B$ is very close to the action of
157: modular multiplication with $p=2$ \cite{aruljphys}. $N$ is throughout assumed to
158: be an even integer. In fact we will set $N=2L$ and can then consider the
159: quantum baker to act on a Hilbert space of a qubit coupled to an $L$ dimensional
160: systems.
161:
162: A similar argument is made for the second half of the transformation, and remarkably these conditions are
163: consistent and produce an unitary operator which has a broken parity symmetryc \cite{BalVor2}.
164: The classical symmetry being $(q \longrightarrow1-q, \, p\longrightarrow 1-p)$. Saraceno\cite{Saraceno}
165: restored this by imposing anti-periodic boundary conditions, and this leads
166: to the quantum baker's map:
167: \beq B=G_{2L}^{-1}\left( \begin{array}{cc}G_{L} &0
168: \\0 & G_{L} \end{array}\right), \eeq where $(G_{N})_{mn}=\br
169: p_m|q_n\kt = \exp[-2 \pi i (m+1/2)(n+1/2)/N]/\sqrt{N}$.
170: $B$ is an unitary matrix, whose repeated application
171: is the quantum version of the full left-shift of classical chaos.
172: This quantum map has been continued to be studied as it has many properties
173: of generic quantum chaotic systems, including random matrix like
174: spectral fluctuations \cite{BalVor2} and eigenfunction scarring \cite{Saraceno}. It is also amenable to
175: a simple semiclassical periodic orbit sum, and hence has been used in the
176: study of such approximations \cite{SarAlm,Connor}. For $N$ that are powers of $2$ it
177: was found that the Hadamard and related transforms highly simplified the
178: eigenstates and some of them are remarkably well described by the Thue-Morse
179: sequence \cite{Schroeder} and its Fourier transform \cite{MeenArul12,Ermann}. It has also been used in the study of entanglement \cite{ScottEnt} and hypersensitivity of quantum chaos \cite{Scott07}. It is possible to design a quantum
180: circuit for the quantum baker map \cite{SchackQBM} and this been implemented on a NMR quantum computer
181: experimentally \cite{QBMexp}.
182:
183: If one is not mindful of classical
184: symmetries being fully preserved on quantization, there are a large number of possible
185: quantum baker maps \cite{BalVor2,SarVor}. An important class of such ``decorated bakers'' \cite{Voros} are got
186: by embellishing the original bakers map with relative phases in the half-sized Fourier
187: blocks, as well as in the definition of the Fourier transform itself, as done below.
188: In a previous work we have constructed such a decorated quantum
189: bakers map using the shift operator $S$ \cite{aruljphys}. It will be useful to do
190: the converse and construct the shift operator from the quantum
191: bakers map or similar operators. It is well-known that the ``hard'' part of
192: Shor's algorithm is the implementation of the modular exponentiation
193: step. On the other hand the quantum bakers map is implemented with
194: quantum Fourier transforms (QFTs) and this may make the implementation of
195: the shift operator possible with the QFTs. We explicitly show this at least for
196: the case $p=2$. More importantly for us it will enable embedding the shift operator in a larger family of operators which includes maps with well defined classical limits, thereby making the classical limit
197: of modular multiplication explicit.
198:
199: The shift operator we have already defined, however we restate it for clarity as:
200: \beq
201: S|n \kt =|2 n (\md N-1) \kt
202: \eeq
203: with the caveat that $S|N-1\kt =|N-1\kt$. This corresponds to our earlier definition with
204: $M=N-1$ with one more state ($|N-1\kt$) added to the Hilbert space, but which remains fixed, and
205: outside any dynamics we are interested in, but may participate when there are perturbations.
206: Note that since for the bakers map $N$ is an even integer $S$ is unitary.
207: We define a generalized Fourier transform as
208: \beq
209: (F_{N}(\al,\beta))_{nm}=\df{1}{\sqrt{N}} \exp(-2 \pi (n+\al)(m+\beta)/N)
210: \eeq
211: Evaluating the product of the Fourier transform and $S$ we can derive, merely by summing finite geometric
212: series and using elementary properties of exponentials that
213: \beq
214: S=\dfrac{1}{\sqrt{2}} F_{2L}^{-1}(\al,\al) \, \left( \begin{array}{cc} F_{L}(\al,\f{\al}{2})&F_{L}(\al,\f{1+\al}{2})\\
215: e^{-i \pi \al} F_{L}(\al,\f{\al}{2})& -e^{-i \pi \al} F_{L}(\al,\f{\al+1}{2}) \end{array} \right).
216: \eeq
217: Note that the operator $S$ does not depend on the phase $\al$ that appears on the R.H.S.. We use
218: this freedom to break or keep the parity symmetry. A natural and simple choice is $\al=0$, but
219: $\al=1/2$ leads to symmetric operators as explained below. The structure of the above identity
220: allows this to be written as
221: \beq
222: S=\df{1}{\sqrt{2}}\left( B_{2L} \, +\, B'_{2L}\right)
223: \eeq
224: where
225: \beq
226: B_{2L}= F_{2L}^{-1}(\al,\al) \, \left( \begin{array}{cc} F_{L}(\al,\f{\al}{2})&0\\
227: 0& -e^{-i \pi \al} F_{L}(\al,\f{\al+1}{2}) \end{array} \right).
228: \eeq
229: and
230: \beq
231: B'_{2L}= F_{2L}^{-1}(\al,\al) \, \left( \begin{array}{cc} 0 &F_{L}(\al,\f{1+\al}{2})\\
232: e^{-i \pi \al} F_{L}(\al,\f{\al}{2})& 0 \end{array} \right).
233: \eeq
234:
235: \begin{figure}
236: %\includegraphics[width=2.5in,angle=-90]{fig2A.eps}
237: \includegraphics[height=4in]{fig1.eps}
238: \caption{A schematic view of the shift operator as a sum of two bakers maps, one
239: with the usual stacking order of the vertical left half being stretched to the bottom horizontal half ($B_c$-classical, $B_{2L}$-quantum)
240: and the baker with a reverse stacking order ($B^{'}_{c}$-classical, $B'_{2L}$-quantum). }
241: \label{bakers}
242: \end{figure}
243:
244: Thus remarkably the modular multiplication $S$ can be written as a {\it sum} of two
245: unitary operators (with a normalization factor). $B_{2L}$ and $B'_{2L}$ are two {\em quantum baker maps}, of which the former one is the standard one, which we have discussed above. It is well known that such decorated bakers also
246: perform the same classical actions as normal bakers \cite{BalVor2,SarVor}, the classical limit being
247: $L \longrightarrow \infty$. The operator $B'_{2L}$ has not been studied nearly as much,
248: but has recently appeared in a work that uses this to study coupled chaotic systems \cite{Vallejos07}.
249: The classical limit (say $B'_c$) as pointed out in this study corresponds to a different stacking
250: order of the vertical partitions of the bakers map after they have been stretched.
251: Instead of the usual left-half transiting to the bottom-half, it is put in the top-half
252: and the right-half goes into the bottom half. This fixes the lower right-hand corner
253: of the square. Thus $B'_c(q,p)=(2q,(p+1)/2)$ if $q\le 1/2$ and $(2q-1,p/2)$ if $q>1/2$.
254: Again the operator $B'_{2L}$ that appears above differs from the
255: one used earlier in terms of the ``decorations''. Of course these decorations are
256: absolutely crucial so that the two unitary evolutions, which are non-periodic and
257: have random matrix like properties, {\em add} and conspire to produce the simple shift operator that is completely periodic. Previous studies of the classical limit of operators such as $S$ include those
258: of what is called the ``extremal quantum baker map'' \cite{SchackCaves} and it has been suggested that the classical limit corresponds to a ``stochastic classical map'' \cite{Scott,Nonnen}. In Fig.~(\ref{bakers}) we have shown a schematic of the classical bakers maps that on quantization and coherent addition yield the shift operator. Also see Ref.~\cite{aruljphys} for a
259: description and figure of the action of $S$ on Weyl coherent states.
260:
261: That the simple shift operator's can be thought of as a coherent superposition of
262: two quantum chaotic evolutions has been demonstrated above in a particularly simple way.
263: This suggests that there maybe perturbations of the operator $S$ that are generic and may possess
264: random matrix \cite{Mehta} like and other quantum chaotic properties \cite{GutzBook,HaakeBook}.
265: We show below that this is indeed the case. Since quantum chaotic operators also are typically sensitive to perturbations
266: \cite{CavesSchack,Peres}, this may have implications for the operation of the Shor algorithm. We partly study this by measuring the fidelity of $S$ to small perturbations and show that the fidelity decays exponentially in time till the order of $N-1$ or half of this. Thereafter it typically shows enhanced rate of decays at multiples of this time, but could also show strong recurrences. Surprisingly a simple analysis when $N$ is a power of 2 captures many of the qualitative features of the more general case.
267:
268:
269: \section{Perturbations of the shift operator and quantum chaos}
270:
271: In terms of operations on the Hilbert spaces of the tensor product ${\cal H}_2 \otimes {\cal H}_L$
272: we may write the shift operator as
273: \beq
274: S= F_{2L}^{-1}(\al,\al)\circ\df{1}{\sqrt{2}} \left( \begin{array}{cc} 1 & 1 \\ e^{-i \pi \al} & -e^{-i \pi \al} \end{array}\right) \otimes I_L \circ
275: \left( \begin{array}{cc} F_L(\al,\f{\al}{2}) &0 \\0 & F_L(\al,\f{1+\al}{2}) \end{array} \right).
276: \eeq
277: Thus the modular exponentiation maybe implemented with QFTs that have suitable phases. However the dimensionalities
278: of the QFTs are not in general powers of 2 and are therefore not the standard ones in use.
279: We choose to perturb the central operator in the above equation, and perturb only the
280: qubit space ${\cal H}_2$. In particular we consider the smooth embedding of the shift
281: operator in the family:
282: \beqa
283: S(\theta; \al,P)&=& F_{2L}^{-1}(\al,\al)\circ \exp(-i \theta P) \df{1}{\sqrt{2}} \left( \begin{array}{cc} 1 & 1 \\ e^{-i \pi \al} & -e^{-i \pi \al} \end{array}\right) \otimes I_L \circ
284: \left( \begin{array}{cc} F_L(\al,\f{\al}{2}) &0 \\0 & F_L(\al,\f{1+\al}{2}) \end{array} \right) \nonumber \\
285: &=&
286: F_{2L}^{-1}(\al,\al) \circ \exp(-i \theta P) \otimes I_L \circ F_{2L}(\al,\al) \, S \, = \,V(\theta) S.
287: \eeqa
288: here $P$ is the perturbing Hermitian operator on the qubit space, and $V(\theta)$ defined through the
289: last equation is the Fourier transform of the perturbation generated by $P$. The operator $S(0;\f{1}{2},P)$
290: is the unperturbed shift operator simply called $S$ so far. The family of operators $S(\theta;\al,P)$ now depends on the phase $\al$ as well, although $S(0;\al,P)$ does not. We display this dependence explicitly as the
291: phase $\al$ does play a crucial role.
292:
293: The operator $S$ has the quantum parity symmetry $R$ : $R|n \kt = |N-n-1\kt$, that is
294: $SR=RS$. If $L$ is a power of 2, then $R$ is simply the product $\otimes ^{2L} \sigma_x$.
295: Perturbations of the shift will therefore in general approximately preserve this symmetry. To analyse
296: random matrix properties it is desirable to completely break a symmetry or preserve it and
297: desymmetrize the operators. Since we want to retain the character of a small perturbation, we
298: first choose to preserve the parity symmetry exactly. We can do this by adopting anti-periodic
299: boundary conditions, $\al=1/2$ and choosing $P=\sigma_x$. This will lead to the family:
300: \beq
301: \label{alhalfX}
302: S(\theta; \f{1}{2},\sigma_x)=F_{2L}^{-1}(\f{1}{2},\f{1}{2}) \left( \begin{array}{cc} \sin(\f{\pi}{4}-\theta) F_{L}(\f{1}{2},\f{1}{4})&
303: \cos(\f{\pi}{4}-\theta) F_{L}(\f{1}{2},\f{3}{4}) \\-i\cos(\f{\pi}{4}-\theta) F_{L}(\f{1}{2},\f{1}{4})&
304: i\sin(\f{\pi}{4}-\theta) F_{L}(\f{1}{2},\f{3}{4})
305: \end{array} \right).
306: \eeq
307: When $\theta=\pm \pi/4$, the operators correspond to the bakers of type $B'_{2L}$ and $B_{2L}$
308: respectively. For other angles it represents a coherent mixture of the two types of baker operator stacking
309: while at $\theta=0$ it is the usual shift operator.
310:
311: We take the even subspace of the spectrum of $S(\theta;\f{1}{2},\sigma_x)$ and show in
312: Fig. ~(\ref{nns}) the nearest neighbor spacing statistics for two case of small angles $\theta$.
313: It is clear that if the perturbation is very small, the rigid, harmonic oscillator like spectrum
314: of $S$ widens into one where there is dominant level repulsion, and for fairly significant
315: perturbations becomes a generic one that belongs to the universality class of the
316: Gaussian Orthogonal Ensemble (GOE) of random matrices \cite{Mehta}, well-known to apply
317: to quantum chaotic systems that have time-reversal symmetry \cite{Bohigas84,HaakeBook}.
318: Previously it was shown that the unitary part of the full Shor algorithm, including the Hadamard and the
319: Fourier transforms, had fluctuations that were of the CUE kind \cite{Maity}. We note here that restricting ourselves
320: to the modular multiplication part with a particular perturbation allows us to preserve the
321: time-reversal that holds for individual quantum bakers maps \cite{BalVor2}.
322:
323: Thus we see that indeed there are perturbations of the shift-operator that are quantum chaotic. There is also a crucial
324: dependence on the number $N$ (or the number to be factored $N-1$). If $N$ were a power of
325: $2$ such as $4096$, instead of $4094$ in Fig. ~(\ref{nns}) there would be much more
326: deviation from the GOE distribution, with a large peak near the origin. This anomalous
327: statistic arises from the extreme degeneracy of the eigenangles when $N$ is a power of 2,
328: and is special. A similar situation where there is an extreme dependence of the statistics
329: of the spectrum on the effective Planck constant $N$ arises in the case of the perturbed
330: cat maps \cite{Esposti05}, and presumably for similar reasons. Earlier it has also been pointed out that perturbing the
331: cat maps slightly so that the sawtooth map arises leads to a rapid restoration of the generic
332: fluctuation characteristics of quantum chaotic systems \cite{arulbalazssaw}. Thus the similarities of the shift map
333: to the quantum chaotic cat maps with their special arithmetic properties is further highlighted here.
334:
335: \begin{figure}
336: \includegraphics[height=5in,angle=-90]{fig2.eps}
337: \caption{The nearest neighbor spacing distribution of even-subspace eigenangles
338: for two perturbations of the shift operator when $N=4094$, the perturbations
339: preserving the parity symmetry. The smooth curve shows the corresponding
340: GOE result of random matrix theory, the Wigner-Dyson distribution.
341: }
342: \label{nns}
343: \end{figure}
344:
345:
346: \section{Fidelity decay}
347:
348: We turn to non-stationary properties, as indeed the Shor algorithm is the result of time evolution of
349: a particular initial state which corresponds to the state $|1\kt$. The algorithm requires requires finding
350: the states $|x^j \, \mod(N-1)\kt$ for $x$ coprime to $N-1$. As stated previously we take $x=2$ throughtout,
351: and we now study how gate errors would proliferate in time. In particular we study the fidelity
352: \beq
353: f(t)= |\br 1 |S^{-t} S^t(\theta; \al,P) |1\kt |^2\, =\, |\br 2^t\, \mbox{mod}(N-1)|S^t(\theta; \al,P)|1\kt |^2
354: \eeq
355: In this section choose $\alpha=0$ for simplicity and note that there is a weakly broken
356: parity symmetry as a result of this.
357:
358: \subsection{Case: $P=\sigma_x$}
359:
360: The first case we take will be a rather special one wherein
361: the perturbation is ineffective: the bit-flip $P=\sigma_x$. In this case we have that
362: \beq
363: S(\theta; 0,\sigma_x)=F_{2L}^{-1}(0,0) \left( \begin{array}{cc} \f{1}{\rt}e^{-i \theta} F_{L}(0,0)&
364: \f{1}{\rt}e^{i \theta} F_{L}(0,\f{1}{2}) \\ \f{1}{\rt}e^{-i \theta} F_{L}(0,0)&
365: \f{1}{\rt}e^{i \theta} F_{L}(0,\f{1}{2})
366: \end{array} \right).
367: \eeq
368: and $f(t)=1$ for all time $t$. Note that there are only phases multiplying the Fourier blocks, and therefore
369: the classical limit of this family of operators is the same as that of the simple shift operator: a coherent
370: sum of two baker maps. However, the reason the fidelity is unity for all time is due to a rather intriguing if
371: simply verifiable identity. We will write $F_{2L}$ for $F_{2L}(0,0)$ below. The perturbation operator is:
372: \beq
373: V(\theta)=F_{2L}^{-1}\left(\exp(-i \theta \sigma_x) \otimes I_L\right)F_{2L}=I_L\otimes \exp(-i \theta \sigma_z).
374: \eeq
375: The last equality is an identity valid for all integer $L$. This in turn simply follows from the identity:
376: \beq
377: \label{identX}
378: F_{2L}^{-1} \left( \sigma_x \otimes I_L\right)F_{2L}= \left( I_L \otimes \sigma_z\right)
379: \eeq
380: which maybe directly verified. Note that since $\sigma_x\otimes I_L$ is a circulant matrix it has
381: to be diagonalized by the Fourier transform, and since the eigenvalues are $\pm 1$ these are
382: the only possible diagonal entries. It is also easily verified that this has the structure of $L$ repetitions
383: of $(1,-1)$ pairs which are the diagonal entries of $\sigma_z$. In some sense the Fourier transform is
384: simultaneously performing a bit reversal and a ninety degree rotation in qubit space. However note that
385: for this identity to be true we do not require that $L$ be a power of $2$. We state here that similar
386: identities do {\it not} hold for the other two Pauli matrices, but there are approximations that we will state
387: further ahead. Since $\sigma_z$ merely changes the phase of the state (in standard basis) it follows
388: that the fidelity $f(t)=1$ always.
389:
390: \subsection{Case: $P=\sigma_y$}
391:
392: In this case we see that the classical limit is altered by the perturbation. The Hadamard transform
393: in the qubit space is further rotated around the $y$-axis in spin space and the final operator
394: is similar to that used in Eq.~(\ref{alhalfX}), which we recall is for the case when the phase
395: $\al=1/2$ and for a $\sigma_x$ perturbation.
396: \beq
397: \label{alzeroY}
398: S(\theta; 0,\sigma_y)=F_{2L}^{-1} \left( \begin{array}{cc} \sin(\f{\pi}{4}-\theta) F_{L}(0,0)&
399: \cos(\f{\pi}{4}-\theta) F_{L}(0,\f{1}{2}) \\ \cos(\f{\pi}{4}-\theta) F_{L}(0,0)&
400: -\sin(\f{\pi}{4}-\theta) F_{L}(0,\f{1}{2})
401: \end{array} \right).
402: \eeq
403: Therefore this case is the closest to the parity preserving case we have already discussed
404: and shown the sharp transitions to features of a quantum chaotic spectrum. In Fig.~({\ref{fid00y})
405: we plot the fidelity $f(t)$ for a set of $N$ values that are close to 256. We notice immediately that
406: although the $N$ values are as close as can be ($N$ must be even) the fidelity decays in qualitatively
407: different manners. Except for a very short-time scale the decays are different and one sees a prominent
408: ``shoulder'' in each of the curves at which the fidelity starts to decay even faster.
409:
410: The easiest case to discern this in the figure is for $N=254$ when the shoulder occurs at $t=110$.
411: It is quite easy to numerically relate the time at which this occurs to the multiplicative order of $2$ modulo $N-1$,
412: referred to henceforth loosely as simply the order of $N-1$. Recall that this is the smallest number $k_0(N-1)$ such
413: that $2^{k_0(N-1)}=1 \;\mbox{modulo} (N-1)$. We are guaranteed that such a number exists because $N-1$ is
414: an odd integer. Indeed $k_0(253)=110$, while $k_0(255)=8$, $k_0(251)=50$, and $k_0(249)=82$. Thus the fidelity
415: shows a shoulder either exactly at $t=k_0(N-1)$ or at $t=k_0(N-1)/2$, the first case is observed for
416: $N=256$ and $N=254$ while the latter is the case for $N=252$ and $250$. It is significant then that for
417: numbers of larger orders the fidelity can decay considerably even for small perturbations. Note that $ \theta=0.05$
418: in the figure which roughly translates to a $0.51:0.49$ mixture of the two types of bakers, while a $0.5:0.5$ ``mixture''
419: will be the unperturbed shift operator. The larger the order is the higher powers of $S$ must be calculated and
420: the higher chance of the fidelity to be lowered. It is interesting that the objective of the Shor algorithm
421: namely finding the order already appears in the fidelity as a critical time.
422: \begin{figure}
423: \includegraphics[height=5in,angle=-90]{fig3.eps}
424: \caption{The fidelity decay for four neighboring values of $N$. The perturbation is $P=\sigma_y$, the
425: phase $\alpha=0$, and $\theta=0.05$.}
426: \label{fid00y}
427: \end{figure}
428:
429: We can gain a qualitative understanding of these behaviors with a surprisingly simple model. Consider the
430: case when $N$ is a power of $2$ say $N=2^M$ and let the perturbation be
431: \beq
432: V(\theta)=I_L\otimes \exp(-i \theta \sigma_x).
433: \eeq
434: Note that this has the same structure as the perturbation from the previous case, but is not the
435: true perturbation in this one. Then the initial state is $|0\cdots01\kt$ and it is clear that
436: for $t\le M$:
437: \beq
438: \left(V(\theta) S\right)^t|0\cdots01\kt=I_{2^{M-t}}\otimes \exp(-i \theta \sigma_x) \otimes \exp(-i \theta \sigma_x) \cdots
439: \otimes \exp(-i \theta \sigma_x) |0\cdots01\kt.
440: \eeq
441: and $f(t)=|\cos^2(\theta)|^{t}$. Thus the initial fidelity decay is in fact exponential with a rate
442: $-\log(|\cos^2(\theta)|)$. However beyond $t=M$ there is an additional error adding up and so
443: $f(t)=|\cos^2(\theta)|^{2M-t} |\cos^2(2\theta)|^{t-M}$ for $M<t\le 2M$. Thus in this range of time
444: the fidely decays about four times as fast. We can write in general for this model that
445: \beq
446: \label{analform}
447: f(t)=|\cos^2(r \theta)|^{(r+1)M-t} |\cos^2((r+1)\theta)|^{t-rM}
448: \eeq
449: where $r=[t/M]$ and $[x]$ is the integer part of $x$.
450: We have shown in Fig.~(\ref{fid00y256}) how good an approximation this can be for the case 2 situation when
451: $N$ is indeed a power of 2 (or the number we want to factor is one less than a power of 2). There is even good
452: quantitative agreement.
453: \begin{figure}
454: \includegraphics[height=4in,angle=-90]{fig4a.eps}
455: \includegraphics[height=4in,angle=-90]{fig4b.eps}
456: \caption{The comparison of the fidelity decay for the $\sigma_y$ perturbation ($\alpha=0$, and $\theta=0.05$)
457: with the analytical estimate in Eq.~(\ref{analform}) from an approximate model for two cases of $N$}
458: \label{fid00y256}
459: \end{figure}
460: On the other hand when $N=254$, ($253$ has a high order of $110$) the approximate formula is only qualitatively correct
461: as seen in Fig.~(\ref{fid00y256}).
462:
463: The model works reasonably well because
464: \beq
465: \label{identexpY}
466: F_{2L}^{-1} \left( \exp(-i \theta \sigma_y ) \otimes I_L\right)F_{2L} \approx \left( I_L \otimes \exp(-i \theta \sigma_x)\right),
467: \eeq
468: which follows from
469: \beq
470: \label{identY}
471: F_{2L}^{-1} \left( \sigma_y \otimes I_L\right)F_{2L} \approx \left( I_L \otimes \sigma_x\right),
472: \eeq
473: so that exact perturbation operator which would have been the L.H.S. of Eq.~(\ref{identexpY}) is approximated by its
474: R.H.S. which we have used above. This is the counterpart of Eq.~(\ref{identX}), however here this is only an approximation.
475:
476: This ``model'' or approximation does not explain the appearance of half of the periods for
477: some values of $N$, such as for $252$ and $250$ above. Indeed when $N$ is a power of $2$ we
478: will always observe the first shoulder at the order of $N-1$. This is in fact the result of the possibility that
479: there exists an integer $k'$ such that $2^{k'}\, \mbox{mod}\, (N-1)=-1$, which implies that $k'=k_0/2$.
480: If there exists such an integer then we must, according to Shor's algorithm, choose a different integer (other than 2) to find its order of. That is we cannot use the order of $2$ to find a factor of $N-1$, which is the ultimate objective. For our analysis, this situation implies that
481: \beq
482: \label{spowhalf}
483: S^{k_0(N-1)/2}\equiv R'=\left(\begin{array}{ccc}1&&\\&R_{N-2}&\\&&1\end{array}\right)
484: \eeq
485: where $R_{N-2}$ is the parity operator with $1$ along its secondary diagonal and zero elsewhere.
486: That is $S^{k_0(N-1)/2}$ is almost the parity operator except that $0$ and $N-1$ instead of being
487: interchanged are fixed by $S$, and hence all its powers. To clarify Eq.~(\ref{spowhalf}) may or
488: may not hold depending on if $2^{k_0(N-1)/2}=-1\, \mbox{mod}\, (N-1)$ or not. For instance this
489: is never the case if $N$ is a power of 2. We note that this simplified model has also been considered
490: by \cite{Scott07} recently to show that exponential fidelity decay does not necessarily mean an
491: hypersensitivity to perturbations. However in the context of this paper it is interesting that the
492: model works approximately even when $N$ is not a power of $2$ and preliminary results indicate
493: that there is hypersensitivity to perturbations as well \cite{Unpublished}.
494:
495: In general for an initial state $|\psi_0\kt$ we have that
496: \beq
497: f(t)=|\br \psi_0|V_t V_{t-1}\cdots V_1|\psi_0\kt|^2
498: \eeq
499: where $V_l$ is the perturbation in the interaction picture: $V_l=S^{-l}VS^{l}$. Thus always $V_{k_0(N-1)}=V$
500: but in the situation where Eq.~(\ref{spowhalf}) holds we have that $V_{k_0(N-1)/2}=R'^{-1}VR'$. Thus it is
501: clear that these times are special for the fidelity as seen in the numerical calculations as well. While these arguments
502: along with the approximate model gives a fair understanding of the decay, it is not
503: complete and a more detailed analysis of the product above must be carried out, which
504: the author is unable to provide.
505: There is a rather large literature surrounding the so-called Loschmidt echo \cite{FidelReview}, or fidelity, wherein quantum chaotic systems have been subjected to a small perturbation on reversal. The current discussion is in fact closely related,
506: but previous work has naturally concentrated on the generic case of a non-degenerate operator that is perturbed. In the case
507: of the shift operator, it can be highly degenerate, as well as completely periodic, thereby making
508: it ``non-generic''. It can be compared again to the quantum chaotic cat maps that are also periodic and
509: degenerate in general. Smooth perturbations of this for instance of the type that has been studied
510: before \cite{PertCat} could produce fidelity decays of a similar character that we have noted here.
511:
512: \subsection{Case: $P=\sigma_z$}
513: In this case
514: \beq
515: S(\theta; 0,\sigma_z)=F_{2L}^{-1}(0,0) \left( \begin{array}{cc} \f{1}{\rt}e^{-i \theta} F_{L}(0,0)&
516: \f{1}{\rt}e^{-i \theta} F_{L}(0,\f{1}{2}) \\ \f{1}{\rt}e^{i \theta} F_{L}(0,0)&
517: \f{1}{\rt}e^{i \theta} F_{L}(0,\f{1}{2})
518: \end{array} \right).
519: \eeq
520: Note that there seems to be only a minor change, namely those of signs of phases in the Fourier blocks,
521: compared to the first case, and also the classical limit still remains unaltered by the phase-flip
522: perturbation. However, the fidelity does decay even due to the ``quantum perturbation'' and seems
523: to be of a similar character to that observed when $P=\sigma_y$, namely the previous case. The differences
524: start showing up sharply in the case when Eq.~(\ref{spowhalf}) holds, namely when $k_0(N-1)$ is such that
525: $2^{k_0(N-1)}=-1\, \mbox{mod}\, (N-1)$. It appears to be generically the case that beyond this time there
526: are large oscillations reminiscent of fidelity decay in near-integrable systems \cite{Sankar}. This is illustrated in Fig.~(\ref{fid00z}) for values of $N$. In the cases when $N=252$ and $250$ Eq.~(\ref{spowhalf}) holds and we see that
527: beyond time of half the order there are regular oscillations with this period.
528:
529: That we must expect a fidelity decay is due to a counterpart of the approximation used in the
530: previous case, namely
531: \beq
532: \label{identexpZ}
533: F_{2L}^{-1} \left( \exp(-i \theta \sigma_z ) \otimes I_L\right)F_{2L} \approx \left( I_L \otimes \exp(-i \theta \sigma_y)\right),
534: \eeq
535: which follows from
536: \beq
537: \label{identZ}
538: F_{2L}^{-1} \left( \sigma_z \otimes I_L\right)F_{2L} \approx \left( I_L \otimes \sigma_y\right),
539: \eeq
540: which is a result of combining the identity in Eq.~(\ref{identX}) and the approximation in
541: Eq.~(\ref{identY}).
542: \begin{figure}
543: \includegraphics[height=5in,angle=-90]{fig5.eps}
544: \caption{The fidelity decay for four neighboring values of $N$. The perturbation is $P=\sigma_z$, the
545: phase $\alpha=0$, and $\theta=0.05$.}
546: \label{fid00z}
547: \end{figure}
548: Thus this final case of perturbation we consider is sort of intermediate between cases A and B, however for
549: practical purposes it is closer to case B, as the time behavior beyond the time of the order or half the order
550: is not likely to be of interest from the point of view of the Shor algorithm. The approximate formula in
551: Eq.~(\ref{analform}) continues to be approximately good for those $N$ for which the condition in Eq.~(\ref{spowhalf})
552: does not hold, and for those for which it does, it is approximately good till time of half the order at which
553: the oscillations begin.
554:
555: \section{Discussion}
556:
557: We have studied three archetypal perturbations, phase-flip, bit-flip and a combination therefore,
558: that are possible in the critical part of Shor's
559: algorithm, namely the modular multiplication or exponentiation part. We have confined ourselves to the simplest
560: possible case when the multiplier is 2, when these perturbations can be interpreted in terms
561: of coherent superpositions of quantum bakers maps, whose classical limits are completely chaotic,
562: and are models of randomness. Thus we have shown that there are generic perturbations of the
563: modular exponentiation operator that will qualify as ``quantum chaotic''. We have shown this by
564: computing the nearest-neighbor spacing statistics and seeing that it is of the type expected of
565: random matrices. More pertinent to the algorithm itself we have studied the fidelity decay
566: that occurs with the relevant initial state and shown that for the three types of perturbations there
567: are three possible fidelity decay behaviors. This can be interpreted in terms of the fact that
568: some perturbations alter the classical limit while some do not, as well as in arising
569: from some identities (one exact and one approximate) that involve the Pauli spin matrices and the Fourier transform, which
570: while the author has not seen before, are completely elementary and likely to be known and useful already.
571: A simple model of the fidelity decay is afforded by these identities that describes surprisingly well the
572: exponential decay in time punctuated by shoulders at times related to the order. An exact solution of the problem seems unlikely, and semiclassical analysis cumbersome due to the fact that the modular exponentiation (when the multiplier is 2) is essentially the {\it sum} of two unitary operators with well defined classical limits.
573:
574: The precise impact of the exponential fidelity decay on the functioning of the algorithm remains to be seen. Such an
575: study for the case of static imperfections was recently carried out \cite{Shepel07}. We have been
576: primarily interested in pointing to the deep and exact relationship between the modular exponentiation
577: part of the Shor algorithm and the quantization of an archetypal model of classical deterministic
578: chaos, namely the bakers map. If we had larger multipliers that 2, as will generally be the case, it is
579: reasonable to expect that these will be related to generalized bakers maps with more than 2 partitions \cite{BalVor2}. The
580: number of possible stacking are also more, but it is completely conceivable that once again there
581: are perturbations to the modular exponentiation operator that are quantum chaotic and close to the
582: quantization of such bakers. That such bakers will have larger Lyapunov exponents and have greater
583: classical randomness may make the quantum operators even more susceptible to such gate errors;
584: however this is at the moment mere speculation.
585:
586: \acknowledgments{It is a pleasure to thank Steven Tomsovic and Alfredo M Ozorio de Almeida for
587: discussions}
588:
589: \begin{thebibliography}{99}
590: \bibitem{Mathworld} Weisstein, Eric W. "Multiplicative Order." From MathWorld--A Wolfram Web Resource. http://mathworld.wolfram.com/MultiplicativeOrder.html
591: \bibitem{Shor} P. W. Shor, {\it Proc. 35$^{th}$ Annual Symposium on Foundations of Computer Science}
592: (IEEE Press, 1994 USA).
593: \bibitem{NielsenChuang} M. A. Nielsen and I. L. Chuang, {\it Quantum Computation and Quantum
594: Information} (Cambridge Univ. Press, 2000 Cambridge).
595: \bibitem{Beckman1996}D. Beckman, A. N. Chari, S. Devabahaktuni, and J. Preskill, Phys. Rev. A {\bf 54}, 1034 (1996).
596: \bibitem{Chuang} L. K. M. Vandersypen, {\em et. al.}, Phys. Rev. Lett. {\bf 85}, 5452 (2000); Nature {\bf 414}, 883 (2001).
597: \bibitem{Chuang96}I.L. Chuang, R. Laflamme, P.W. Shor, W.H. Zurek, Science, 270:1633, (1995)
598: \bibitem{Fowler}A.Fowler and L.C.L Hollenberg, Phys. Rev. A. {\bf 70}, 032329, (2004).
599: \bibitem{Hu}X. Hu F. Nori L.F. Wei, X. Li, Phys. Rev. A, 71:022317, (2005).
600: \bibitem{Devitt04}S. J. Devitt, A. G. Fowler, and L. C. L. Hollenberg, Quantum Inf. Comput. {\bf 6}, 616 (2006).
601: \bibitem{Miquel} C. Miquel, J.P. Paz, R. Parazzo, Phys. Rev. A {\bf 54}, 2605, (1996).
602: \bibitem{Shepel07}Ignacio García-Mata, Klaus M. Frahm, and Dima L. Shepelyansky, Phys. Rev. A {\bf 75}, 052311 (2007).
603: \bibitem{Braun} D. Braun, Phys. Rev. A {\bf 65}, 042317 (2002).
604: \bibitem{Maity} K. Maity and Arul Lakshminarayan, Phys. Rev. E {\bf 74}, 035203(R) (2006).
605: \bibitem{QCat} J. H. Hannay, and M. V. Berry, Physica {\bf D1}, 267 (1980);
606: B. Eckhardt, J. Phys. A: Math. Gen. {\bf 19}, 1823 (1986); J. P. Keating, Nonlinearity {\bf 4}, 493 (1991).
607: \bibitem{LL} A. J. Lichtenberg and M. A. Lieberman, {\it Regular and Chaotic Dynamics},
608: (Springer, New York, 1992).
609: \bibitem{ArnAvez} V. I. Arnold and A. Avez, {\it Ergodic Problems of
610: Classical Mechanics} (WA Benjamin, New York, 1968).
611: \bibitem{Tabor} M. Tabor, {\it Chaos and Integrability in Nonlinear Dynamics} (John Wiley and Sons, New York, 1989).
612: \bibitem{Ornstein} D. S. Ornstein, Science {\bf 243}, 182 (1989).
613: \bibitem{BalVor1} N. L. Balazs, and A. Voros, Europhys. Lett. {\bf 4}, 1089 (1987).
614: \bibitem{BalVor2} N. L. Balazs, and A. Voros, Ann. Phys. {\bf 190}, 1 (1989).
615: \bibitem{aruljphys} Arul Lakshminarayan, J. Phys. A: Math. Gen. {\bf 38}, L597 (2005).
616: \bibitem{Saraceno}M. Saraceno, Ann. Phys.(N.Y.) {\bf 199}, 37 (1990).
617: \bibitem{SarAlm} A. M. Ozorio de Almeida, and M. Saraceno, Ann. Phys., (N.Y.) {\bf 210}, 1 (1991).
618: \bibitem{Connor}P.W. O'Connor, S. Tomsovic and E.J. Heller, J. Stat. Phys. {\bf 68},131 (1992); Physica {\bf D 55},
619: 340 (1992).
620: \bibitem{MeenArul12}N. Meenakshisundaram and Arul Lakshminarayan, Phys. Rev. E {\bf 71} 065303(R) (2005); J. Phys. A: Math. Gen. {bf 39}, 11205 (2006).
621: \bibitem{Schroeder}M. R. Schroeder {\it Number Theory in Science and Communication}
622: (Springer-Verlag, Berlin, 1986).
623: \bibitem{Ermann} Leonardo Ermann, and Marcos Saraceno, Phys. Rev. E {\bf 74}, 046205 (2006).
624: \bibitem{ScottEnt} A. J. Scott, Carlton M. Caves, J. Phys. A {\bf 36}, 9553 (2003); Romulo F. Abreu, and Raul O. Vallejos,
625: e-print: quant-ph/0603261.
626: \bibitem{Scott07}A. J. Scott, Todd A. Brun, Carlton M. Caves, and Ruediger Schack, J. Phys. A {\bf 39}, 13405 (2006).
627: \bibitem{SchackQBM}Ruediger Schack, Phys.Rev. A {\bf 57}, 1634 (1998); Todd A. Brun, and Ruediger Schack, Phys.Rev. A {\bf 59}, 2649 (1999).
628: \bibitem{QBMexp} Y. S. Weinstein, S. Lloyd, J. V. Emerson and D. G. Cory, Phys. Rev. Lett. {\bf 89}, 157902 (2002).
629: \bibitem{SarVor}M. Saraceno, and A. Voros, Physica {\bf D79}, 206 (1994).
630: \bibitem{Voros} This terminology is due to the quantum baker Andr\'e Voros.
631: \bibitem{Vallejos07} Raul O. Vallejos, P. R. del Santoro, and A. M. Ozorio de Almeida, e-print: quant-ph/0603212.
632: \bibitem{SchackCaves} R. Schack and C.M. Caves, Appl. Algebra Engg, Comm. Comput.
633: {\bf 10}, 305 (2000).
634: \bibitem{Scott} Mark M. Tracy and A. J. Scott, J. Phys. A: Math. Gen. {\bf 35}, 8341 (2002).
635: \bibitem{Nonnen}S. Nonnenmacher and M. Zworski, J. Phys. A: Math. Gen. {\bf 38}, 10683 (2005).
636: \bibitem{Mehta} M. L. Mehta, {\it Random Matrices}, (3rd ed. New York: Academic Press, 1991).
637: \bibitem{GutzBook} Gutzwiller, M. C. {\it Chaos in Classical and Quantum Mechanics} (Springer, New York, 1990).
638: \bibitem{HaakeBook}F. Haake, {\it Quantum Signatures of Chaos}, (Springer, Berlin, 1991).
639: \bibitem{CavesSchack} R. Schack, G. M. D'Ariano, and C. M. Caves, Phys. Rev. E {\bf 50}, 972 (1994);
640: R. Schack and C. M. Caves, Phys. Rev. E {\bf 53}, 3257 (1996).
641: \bibitem{Peres} A. Peres in {\it Quantum Chaos}, eds. H. A. Cerdeira, R. Ramaswamy, M. C. Gutzwiller,
642: and G. Casati (World Scientific, Singapore, 1991).
643: \bibitem{Bohigas84}O. Bohigas, M. J. Giannoni, and C. Schmit, Phys. Rev. Lett. {\bf 52}, 1 (1984).
644: \bibitem{Esposti05} M. Degli Esposti, B. Winn, J. Phys.A: Math.Gen. {\bf 38}, 1 (2005).
645: \bibitem{arulbalazssaw}A. Lakshminarayan, N. L. Balazs, Chaos, Solitons and Fractals (Osaka, Japan) {\bf 5}, 1169 (1995).
646: \bibitem{Unpublished}N. Meenakshisudaram, Unpublished.
647: \bibitem{FidelReview}See review article: Thomas Gorin, Tomaz Prosen, Thomas H. Seligman, Marko Znidaric, Phys. Rep. {\bf 435}, 33 (2006)
648: \bibitem{Sankar}R. Sankaranarayanan, Arul Lakshminarayan, Phys. Rev. E {\bf 68}, 036216 (2003)
649: \bibitem{BerryCat} J. H. Hannay and M. V. Berry, Physica D {\bf 1}, 267 (1980);
650: J. P. Keating, Nonlinearity {\bf 4}, 309 (1991).
651: \bibitem{PertCat}M. Basilio de Matos and A. M. Ozorio de Almeida, Ann. Phys. {\bf 237}, 46 (1995).
652: \end{thebibliography}
653:
654:
655: \end{document}
656: