1: \documentclass{article}
2:
3: \usepackage{amsmath}
4: \usepackage{amsthm}
5: \usepackage{amssymb}
6: \usepackage{graphicx}
7: \usepackage{psfrag}
8: \usepackage{framed}
9:
10: \newcommand{\tr}{\operatorname{tr}}
11: \newcommand{\conv}{\operatorname{conv}}
12: \newcommand{\aut}{\operatorname{Aut}}
13: \newcommand{\poly}{\operatorname{poly}}
14: \newcommand{\conc}{\operatorname{conc}}
15: \newcommand{\spanv}{\operatorname{span}}
16: \newcommand{\spanvs}{\operatorname{span}}
17: \newcommand{\gr}{\operatorname{gr}}
18: \newcommand{\VEC}{\operatorname{VEC}}
19: \newcommand{\OP}{\operatorname{OP}}
20: \newcommand{\rank}{\operatorname{rank}}
21: \newcommand{\uinvnorm}{|\kern-2pt|\kern-2pt|}
22: \newcommand{\mnorm}[1]{|\!|\!|#1|\!|\!|_2}
23:
24: \theoremstyle{plain}
25: \newtheorem{theorem}{Theorem}[section]
26: \newtheorem{conjecture}[theorem]{Conjecture}
27: \newtheorem{lemma}[theorem]{Lemma}
28: \newtheorem{proposition}[theorem]{Proposition}
29: \newtheorem{corollary}[theorem]{Corollary}
30:
31: \theoremstyle{definition}
32: \newtheorem{definition}[theorem]{Definition}
33: \newtheorem{example}[theorem]{Example}
34: \newtheorem{problem}[theorem]{Problem}
35:
36: \theoremstyle{remark}
37: \newtheorem{remark}[theorem]{Remark}
38:
39: \bibliographystyle{amsalphaxxx}
40:
41: \begin{document}
42:
43:
44: \title{{\sf\bfseries Quantum Algorithms and Covering Spaces}}
45:
46: \author{{\sf Tobias J.\ Osborne\footnote{\texttt{T.J.Osborne@bristol.ac.uk}}}\\ {\sf School of Mathematics}\\ {\sf University of Bristol}\\ {\sf University Walk}\\ {\sf Bristol BS8 1TW}\\ {\sf United
47: Kingdom} \and {\sf Simone
48: Severini\footnote{\texttt{ss54@york.ac.uk}. Present address: Department of Mathematics and Department of Computer Science, University of York, York YO10 5DD, UK}}\\
49: {\sf
50: Department of Computer Science}\\ {\sf University of Bristol}\\
51: {\sf Merchant Venturers' Building}\\ {\sf Bristol BS8 1UB}\\ {\sf
52: United Kingdom}}
53: \date{{\sf \today}}
54: \maketitle
55:
56: \begin{abstract}
57: It's been recently demonstrated that quantum walks on graphs can
58: solve certain computational problems faster than any classical
59: algorithm. Therefore it is desirable to quantify those purely
60: combinatorial properties of graphs which quantum walks take
61: advantage of and try and separate them from those properties due
62: to the encoding of the problem. In this paper we isolate the
63: combinatorial property responsible (at least in part) for the
64: computational speedups recently observed. We find that
65: continuous-time quantum walks can exploit the \emph{covering
66: space} property of certain graphs. We formalise the notion of
67: graph covering spaces. Then we demonstrate that a quantum walk on
68: a graph $Y$ which covers a smaller graph $X$ can be equivalent to
69: a quantum walk on the smaller graph $X$. This equivalence occurs
70: only when the walk begins on certain initial states,
71: \emph{fibre-constant states}, which respect the graph covering
72: space structure. We illustrate these observations with walks on
73: Cayley graphs; we show that walks on fibre-constant initial states
74: for Cayley graphs are equivalent to walks on the induced Schreier
75: graph. We also consider the problem of constructing efficient gate
76: sequences simulating the time evolution of a continuous-time
77: quantum walk. We argue that if
78: $Y\xrightarrow{\pi_{N}}X_{N}\xrightarrow{\pi_{N-1}}X_{N-1}
79: \xrightarrow{\pi_{N-2}}\cdots\xrightarrow{\pi_{1}}X_1$ is a tower
80: of graph covering spaces satisfying certain uniformity and growth
81: conditions then there exists an efficient quantum gate sequence
82: simulating the walk. For the case of the walk on the $m$-torus
83: graph $T^m$ on $2^n$ vertices we construct a gate sequence which
84: uses $O(\poly(n))$ gates which is independent of the time $t$ the
85: walk is simulated for (and so the sequence can simulate the walk
86: for \emph{exponential} times). We argue that there exists a wide
87: class of nontrivial operators based on quantum walks on graphs
88: which can be measured efficiently using phase estimation.
89: Interestingly, measuring these operators won't be unitarily
90: equivalent to the quantum fourier transform. Finally, motivated by
91: our results we introduce a new general class of computational
92: problems, {\sf HiddenCover}, which includes a variant of the
93: general hidden subgroup problem as a subclass. We argue that
94: quantum computers ought to be able to utilise covering space
95: structures to efficiently solve problems from {\sf HiddenCover}.
96: \end{abstract}
97:
98: \section{Introduction}
99: There is a growing belief that quantum computers can solve certain
100: computational problems exponentially faster than any classical
101: computer. Strong evidence for this belief comes in the form of
102: Shor's algorithm \cite{shor:1994a} for factorisation, and the
103: graph traversal algorithm of Childs \emph{et.\ al.}\
104: \cite{childs:2003a}.
105:
106: Despite the spectacular success of the known quantum algorithms we
107: believe that there is still only a fairly rudimentary
108: understanding of the properties of quantum mechanics which are
109: useful for computational speedups. We ascribe this poor
110: understanding to at least two causes: (i) we are unsure what sorts
111: of problems might be amenable to quantum-computational speed up;
112: and (ii) even if we firmly believed that an efficient quantum
113: algorithm existed for a problem, it is hard to come up with this
114: putative algorithm because it is difficult (at least for us) to
115: reason within the traditional quantum computing model --- the
116: quantum circuit model.
117:
118: All quantum algorithms are traditionally expressed in the quantum
119: circuit model (see \cite{nielsen:2000a} and \cite{preskillnotes}
120: for a detailed description of the quantum circuit model). The
121: quantum circuit model assigns a unit cost to certain elementary
122: quantum gates, such as, for example, {\sc cnot}, $H$, and $T$ (the
123: $\pi/8$ phase gate). This is by no means the only way to express
124: quantum algorithms. Three other computing paradigms are
125: polynomially equivalent to the quantum circuit model: the quantum
126: Turing machine model \cite{bernstein:1997a}, the one-way quantum
127: computer \cite{briegel:2001b}, and the adiabatic evolution model
128: (which has recently been shown to be polynomially equivalent to
129: the quantum circuit model in \cite{aharonov:2004a}).
130:
131: Why might one want to consider computational paradigms other than
132: the quantum circuit model? The reason is that there are conceptual
133: peculiarities with the quantum circuit model which make it hard to
134: work with when designing algorithms the traditional way. Firstly,
135: there is a back-action that quantum superposition induces in gates
136: like the {\sc cnot} where there can be a ``backwards'' flow of
137: quantum information for certain initial states. Secondly, the
138: existence of quantum entanglement seriously confuses the causal
139: structure of quantum circuits because we can think of a Bell pair
140: as sending a qubit backwards in time! However, in favour of the
141: quantum circuit model is the fact that exponentially many degrees
142: of freedom can be summarised succinctly.
143:
144: Some of the negative features of the quantum circuit model are
145: amply addressed in the quantum walk model of computation. Quantum
146: walks, or the quantum dynamics of a free particle hopping on a
147: graph, are an exciting new paradigm for quantum computing. (For a
148: review of quantum walks see \cite{kempe:2003b} and references
149: therein.) The attractiveness of quantum walks is that they provide
150: an extremely intuitive way to \emph{visualise} exponentially many
151: quantum degrees of freedom. In contrast to the quantum circuit
152: model, in a quantum walk there is a clear physical picture of
153: \emph{where} information is flowing and a definite notion of
154: \emph{cause and effect}. The price we pay for these intuitive
155: features is that the exponentially ($2^n$ for $n$ qubits) many
156: degrees of freedom of $n$ qubits translate to exponentially
157: vertices in the graph. Additionally, there is no clear way to
158: translate a quantum walk efficiently into the quantum circuit
159: model. Despite these difficulties we believe that quantum walks
160: provide an attractive methodology in the quantum algorithm
161: designer's toolkit because they appeal to the geometrical
162: intuitions.
163:
164: The growing number of quantum-walk algorithms might be seen as
165: evidence for the conceptual utility of the quantum walk model. We
166: mention three recent algorithms: (i) the quantum walk search
167: algorithm \cite{shenvi:2003a}, \cite{childs:2003c}, and
168: \cite{ambainis:2004a}; (ii) the graph-traversal algorithm of
169: Childs \emph{et.\ al.}\ \cite{childs:2003a}; and (iii) the element
170: distinctness algorithm of Ambainis \cite{ambainis:2003a}, and
171: relatives \cite{childs:2003b, magniez:2003a, szegedy:2004b}.
172:
173: What unites the quantum walk algorithms? (And, more ambitiously,
174: all quantum algorithms?) Let's concentrate on the results of
175: \cite{childs:2002a, childs:2003a, moore:2002a, kempe:2003a}, which
176: are based on the continuous-time quantum walk. (Note that the
177: quantum walks on the hypercube \cite{moore:2002a, kempe:2003a} and
178: the graphs in \cite{childs:2002a} don't provide the speedups for
179: the solution to any algorithmic problem.) The results in these
180: papers appear to be related phenomena (they all take advantage of
181: ``column spaces''), however, this relationship has not yet, to the
182: best of our knowledge, been quantified fully. In this paper we
183: identify and generalise the combinatorial property of these graphs
184: which leads to small hitting times. We believe this is a key
185: ingredient underlying quantum speedups of hitting times. The
186: combinatorial property we isolate is that all of the graphs walked
187: on in these papers are covering spaces for much smaller graphs.
188:
189: \subsection*{Quantum computers and covering spaces}
190:
191: \begin{figure*}
192: \begin{center}
193: \psfrag{pi}{$\pi$} %
194: \includegraphics{cover1.eps}
195: \caption{An example of a covering space. In this case the circle
196: $X=S^1$ is covered by another circle $Y$ which has twice the
197: circumference of $X$ under the projection $\pi$. Note that the
198: inverse projection of any point $x\in X$ is a finite set
199: consisting of two points $a$ and $b$ in the covering space $Y$.
200: One can think of this covering space as a subset of the Riemann
201: surface for $f(z)=\sqrt{z}$, which is a covering space for
202: $\mathbb{C}$.}\label{fig:cover}
203: \end{center}
204: \end{figure*}
205:
206: What is a covering space? Suppose that $X$ and $Y$ are
207: arcwise-connected and locally arcwise-connected topological
208: spaces, respectively. Then $(Y,\pi)$ is said to be a covering
209: space of $X$ if $\pi:Y\rightarrow X$ is a surjective continuous
210: map with every $x\in X$ having an open neighborhood $U$ such that
211: every connected component of $\pi^{-1}(U)$ is mapped
212: homeomorphically onto $U$ by $\pi$. The preimage of a point in $X$
213: is called a \emph{fibre} of $\pi$. An example covering space is
214: shown in figure~\ref{fig:cover}.
215:
216: In this paper we will show that a continuous-time quantum walk on
217: a graph $Y$ which is a covering space for another graph $X$ (in
218: the natural topology) is isomorphic to a quantum walk on $X$,
219: under certain specific circumstances. This equivalence occurs when
220: the walk starts on \emph{fibre-constant} quantum states.
221: Additionally, we will argue, and show in some cases, that if there
222: is a tower of covering spaces
223: \begin{equation}
224: Y\xrightarrow{\pi_{N}}X_{N}\xrightarrow{\pi_{N-1}}X_{N-1}
225: \xrightarrow{\pi_{N-2}}\cdots\xrightarrow{\pi_{1}}X_1
226: \end{equation}
227: then there is an efficient gate sequence (i.e.\ using
228: $O(\poly(n))$ elementary gates) for the quantisation of $Y$. The
229: idea we exploit is to recursively and hierarchically construct the
230: gate sequence from the ``elementary gate'' $U(X_1)$ and the
231: specification of how $U(X_1)$ lives in $U(X_2)$.
232:
233: We believe that these results are not specific to continuous-time
234: quantum walks, but rather \emph{indicative} of a general
235: principle.
236:
237: We formulate this principle in the following way: consider some
238: collection of mathematical objects like the class of simple
239: graphs, or something with more structure, like the class of finite
240: fields. Let $U(X)$ be a quantisation scheme for this class of
241: objects, by which we mean a way to associate a unitary matrix
242: $U(X)$ acting on a finite-dimensional Hilbert space
243: $\mathcal{H}(X)$ with any object $X$. Suppose, further, that $Y$
244: is a covering space $\pi:Y\rightarrow X$ for another object $X$.
245: If the quantisation scheme $U(X)$ is ``sufficiently well-behaved''
246: then $U(Y)$ should be related to $U(X)$ according to a map, the
247: \emph{pull-back} $\pi^*:U(X)\rightarrow U(Y)$. If, further, $Y$ is
248: determined from $X$ in a sufficiently simple way, then it should
249: be possible to construct $U(Y)$ from $U(X)$ according to this
250: specification. Mimicking the recursive and hierarchical
251: construction for gate-sequences for continuous-time quantum walks
252: we expect that if
253: \begin{equation}
254: Y\xrightarrow{\pi_{N}}X_{N}\xrightarrow{\pi_{N-1}}X_{N-1}
255: \xrightarrow{\pi_{N-2}}\cdots\xrightarrow{\pi_{1}}X_1
256: \end{equation}
257: is a tower of covering spaces obeying some uniformity and growth
258: conditions then there should be an efficient gate sequence for the
259: quantisation $U(X)$.
260:
261: We retroactively argue that there is an example of where this
262: philosophy has already proven successful. In this example we take
263: as our space of mathematical objects the category
264: $\mathbf{GrpFin}$ of finite groups. We take the quantisation
265: scheme to be the contravariant functor which associates the
266: quantum fourier transform with every finite group. (For an
267: introduction to category theory see \cite{maclane:1998a}.) For
268: certain towers of subgroups, i.e.\ those which are
269: \emph{polynomially uniform} \cite{moore:2003a}, there is an
270: efficient hierarchical scheme to construct the quantisation
271: $U(Y)$. Here we are taking the ``is a subgroup'' relation to be
272: the covering space relation in this category. To accord this
273: exactly with the definition of covering space given earlier
274: requires the introduction of an appropriate topology on a finite
275: group. This is something we will avoid, preferring only this
276: qualitative argument --- clearly the analogy isn't perfect.
277:
278: What sort of problems could these supposed efficient quantisations
279: solve? We will argue that they could be used to solve a problem we
280: call the \emph{Hidden Covering Space} problem which includes the
281: hidden subgroup problem as a subclass. Roughly speaking, if there
282: is a function $f$ on $Y$ which is periodic on some object $X$
283: covered by $Y$, then a quantum computer should be able to identify
284: the structure $X$ efficiently.
285:
286: The general properties of covering spaces are extremely
287: tantalising and suggestive. We hope to convince the reader that
288: the results we have found concerning graph covering spaces and
289: continuous-time quantum walks are indicative of a much larger
290: framework. We believe that many quantum algorithms exhibiting an
291: exponential separation between classical and quantum complexity
292: could exist. We argue that the properties of quantisations of
293: covering spaces may provide many opportunities to design such new
294: algorithms. This is not least because covering spaces are
295: pervasive in mathematics, from number fields and algebraic
296: surfaces to geometric group theory, and, of course, Galois theory.
297:
298: The outline of this paper is as follows. We begin in
299: \S\ref{sec:graphquant} by describing the quantisation scheme we
300: study in the remainder of this paper, the continuous-time quantum
301: walk. In \S\ref{sec:graphcover} we review the theory of graph
302: covering spaces and the heat kernel for graphs. We then apply
303: these results in \S\ref{sec:indqw} to demonstrate that quantum
304: walks which begin on fibre-constant states are isomorphic to
305: quantum walks on smaller graphs. We illustrate our results in
306: \S\ref{sec:examples} for the hypercube and Cayley graphs. In
307: \S\ref{sec:efficient} we utilise the covering-space properties of
308: the $m$-torus graph on $2^n$ vertices to construct an efficient
309: gate sequence for their quantisation. Motivated by our results we
310: introduce, in \S\ref{sec:hiddencover}, a new class of problem,
311: {\sf HiddenCover}, which quantum computers may be able to solve
312: efficiently. We also solve the hidden cover problem for the
313: $m$-torus graph which, incidentally, provides a (philosophically)
314: different way to solve the abelian hidden subgroup problem.
315:
316: \section{Quantisations of Graphs: the Continuous-Time Quantum
317: Walk}\label{sec:graphquant}
318:
319: In this section we introduce the \emph{continuous-time quantum
320: walk}, which is a quantisation scheme for the class of simple
321: graphs. For further details about the graph-theoretic notation and
322: terminology we use in this section and the rest of this paper see
323: \cite{biggs:1993a, cvetkovic:1995a, chung:1997a}.
324:
325: Let us begin by defining the main objects of our study. By a
326: \emph{weighted graph} $Y$ we mean a \emph{vertex set} $V=V(Y)$
327: with an associated \emph{weight function} $w: V\times
328: V\rightarrow\mathbb{R}^+$ ($\mathbb{R}^+$ denotes the real numbers
329: $x\in\mathbb{R}$ such that $x\ge0$) which satisfies
330: \begin{equation}
331: w(u,v) = w(v,u).
332: \end{equation}
333: If $w(u,v)>0$ then we refer to $\{u,v\}$ as an \emph{edge} of $Y$,
334: and we say that $u$ and $v$ are \emph{adjacent}. By a \emph{simple
335: graph} we mean the special situation where $w(u,v)$ is either $0$
336: or $1$ and $w(u,u)=0$ for all $u\in V$.
337:
338: We define the \emph{degree} $d_v$ for a vertex $v$ to be
339: \begin{equation}
340: d_v = \sum_{v\in V} w(u,v).
341: \end{equation}
342: A graph is \emph{regular} if all the degrees are the same.
343:
344: The \emph{Laplacian} of a weighted graph $Y$ on $n$ vertices and
345: weight function $w$ is the $n\times n$ matrix $\triangle$ given by
346: \begin{equation}
347: \triangle_{u,v}\triangleq\begin{cases}d_v - w(v,v)\quad & \text{if
348: $u=v$},\\-w(u,v)\quad &\text{if $u$ and $v$ are adjacent,} \\
349: 0\quad &\text{otherwise,}
350: \end{cases}
351: \end{equation}
352: where $u$ and $v$ are two arbitrary vertices in $V$.
353:
354: We now consider introduce the (formal) complex vector space
355: $\mathcal{H}(Y)$ spanned by the vectors $|u\rangle$, $u\in V$,
356: which we take to be orthonormal under the natural inner product:
357: $\langle u|v \rangle = \delta_{u,v}$. Of course this vector space
358: is isomorphic to $\mathbb{C}^n$. We think of the vector space
359: $\mathcal{H}(Y)$ as the space of all complex-valued functions $f$
360: on the finite set $V(Y)$, where for each vector $|f\rangle =
361: \sum_{u\in V(Y)}f_u |u\rangle \in \mathcal{H}(Y)$ we define the
362: function $f:V(Y)\rightarrow \mathbb{C}$ by $f(u) = \langle u |
363: f\rangle$.
364:
365: We define the \emph{adjacency matrix} of a weighted graph $Y$ to
366: be the operator
367: \begin{equation}
368: A(Y) \triangleq \sum_{u,v \in V(Y)} w(u,v) |u\rangle \langle v|.
369: \end{equation}
370:
371: The Laplacian, acts in a natural way on $\mathcal{H}(Y)$ as
372: \begin{equation}
373: \triangle = \sum_{u\in V(Y)} (d_u-w(u,u)) |u\rangle\langle u| -
374: \sum_{\substack{u,v\in V(Y)\\ u\sim v}} w(u,v) |u\rangle\langle
375: v|,
376: \end{equation}
377: where we use the notation $u\sim v$ to mean that $u$ is adjacent
378: to $v$ and $u\not= v$. For a specific vector $|f\rangle
379: =\sum_{u\in V(Y)}f_u |u\rangle$ we have
380: \begin{equation}
381: \triangle|f\rangle = \sum_{\substack{u,v\in V(Y) \\ u\sim v}}
382: (f(u)-f(v))w(u,v)|u\rangle.
383: \end{equation}
384: It is worth noting that the Laplacian can be written $\triangle =
385: D(Y) - A(Y)$, where $D(Y) \triangleq \sum_{u,v \in V(Y)} w(u,v)
386: |u\rangle \langle u|$.
387:
388:
389: We define the \emph{heat equation} and \emph{Schr\"odinger
390: equation} for $Y$ to be the differential equations
391: \begin{equation}\label{eq:heat}
392: \frac{\partial}{\partial \tau} |\psi(\tau)\rangle =
393: -\triangle|\psi(\tau)\rangle,
394: \end{equation}
395: and
396: \begin{equation}\label{eq:schroe}
397: i\frac{\partial}{\partial t} |\psi(t)\rangle =
398: \triangle|\psi(t)\rangle,
399: \end{equation}
400: respectively. Note that the Schr\"odinger equation is equivalent
401: to the heat equation with the replacement $\tau = it$.
402:
403: The Laplacian $\triangle$ for a graph on $n$ vertices is a
404: symmetric matrix and so we can write its spectral decomposition
405: \begin{equation}
406: \triangle = \sum_{j=0}^{n-1} \lambda_j |\lambda_j\rangle \langle
407: \lambda_j|.
408: \end{equation}
409: We often refer to an eigenstate of $\triangle$ as a \emph{harmonic
410: eigenfunction}.
411:
412: The spectral decomposition of the Laplacian allows us to define
413: the \emph{heat kernel} and \emph{propagator}
414: \begin{equation}
415: H(\tau) = \sum_{j=0}^{n-1} e^{-\lambda_j\tau} |\lambda_j\rangle
416: \langle \lambda_j|,
417: \end{equation}
418: and
419: \begin{equation}
420: U(t) = \sum_{j=0}^{n-1} e^{-i\lambda_jt} |\lambda_j\rangle \langle
421: \lambda_j|,
422: \end{equation}
423: respectively. Note that $U(t) = H(i\tau)$. Using the the heat
424: kernel and propagator we can solve the heat and Schr\"odinger
425: equations (\ref{eq:heat}) and (\ref{eq:schroe}) with initial state
426: $|\psi(0)\rangle$ by defining $|\psi(\tau)\rangle =
427: H(\tau)|\psi(0)\rangle$ and $|\psi(t)\rangle =
428: U(t)|\psi(0)\rangle$:
429: \begin{equation}
430: \begin{split}
431: \frac{\partial}{\partial \tau} |\psi(\tau)\rangle &=
432: \frac{\partial H(\tau)}{\partial \tau}
433: |\psi(0)\rangle\\
434: &= -\triangle H(\tau)|\psi(0)\rangle.
435: \end{split}
436: \end{equation}
437: The result for the propagator follows by substituting $\tau = it$.
438:
439: \begin{definition}
440: A \emph{continuous-time quantum walk} on a weighted graph $Y$ is
441: the propagator $U[Y](t)$ associated with the Laplacian
442: $\triangle[Y]$ for $Y$.
443: \end{definition}
444:
445: \section{Graph Covering Spaces and the Heat Kernel}\label{sec:graphcover}
446: In this section we review the theory of graph covering spaces and
447: the heat kernel for weighted graphs. For an introduction to graph
448: covering spaces see \cite{biggs:1993a} and \cite{godsil:1980a}.
449: For further results in this area see \cite{chung:1999a} and
450: \cite{stark:1996a, stark:1999a, terras:1999a, stark:2000a,
451: terras:2002a}. We follow \cite{chung:1999a} closely for most of
452: this section. The principle result of this section is that the
453: spectrum of a graph $X$ covered by another graph $Y$ is contained
454: in the spectrum of $Y$. The eigenvectors of $X$ also induce
455: eigenvectors in $Y$.
456:
457: We begin with some definitions.
458: \begin{definition}\label{def:graphcover}
459: Suppose we have two graphs $X$ and $Y$ with weight functions
460: $w_X(x,y)$ and $w_Y(u,v)$, respectively. We say $Y$ is a
461: \emph{covering space}\footnote{Note that our definition of
462: (unramified) graph covering spaces accords with the standard
463: definition given in the introduction when we take for the topology
464: of the graphs $X$ and $Y$ the \emph{weak topology}. See
465: \cite{hatcher:2002a} and \cite{stark:1999a, stark:2000a} for
466: further discussion and details.} for $X$ (alternatively, $X$ is
467: \emph{covered} by $Y$) if there is a set map $\pi:V(Y)\rightarrow
468: V(X)$ satisfying the following two properties:
469: \begin{enumerate}
470: \item There is a $\mu\in \mathbb{R}^+$ (the positive real numbers
471: including $0$), called the \emph{index} of $\pi$, such that for
472: $u$, $v\in V(X)$ the equality holds:
473: \begin{equation}
474: \sum_{\substack{x\in\pi^{-1}(u)\\
475: y\in\pi^{-1}(v)}} w_Y(x,y) = \mu \sqrt{|\pi^{-1}(u)||\pi^{-1}(v)|}
476: w_X(u,v).
477: \end{equation}
478: \item For $x$, $y\in V(Y)$ with $\pi(x)=\pi(y)$ and $v\in V(X)$,
479: we have
480: \begin{equation}
481: \sum_{z\in\pi^{-1}(v)} w_Y(z,x) = \sum_{z'\in\pi^{-1}(v)}
482: w_Y(z',y).
483: \end{equation}
484: \end{enumerate}
485: \end{definition}
486:
487: We now translate Definition~\ref{def:graphcover} into a more
488: useful statement about the adjacency matrices of $Y$ and $X$. In
489: doing so we make contact with the theory of equitable partitions
490: \cite{biggs:1993a}.
491:
492: We begin by considering an arbitrary projection operator
493: $\pi:V(Y)\rightarrow V(X)$ and define the corresponding
494: \emph{pull-back} operator $P:\mathcal{H}(Y)\rightarrow
495: \mathcal{H}(X)$,
496: \begin{equation}\label{eq:projproj}
497: P = \sum_{\substack{u \in V(X) \\ x\in \pi^{-1}(u)}}
498: \frac{1}{\sqrt{|\pi^{-1}(u)|}} |u\rangle \langle x|.
499: \end{equation}
500: (It is easy to establish the identity $(P^\dag P)^2=P^\dag P$ via
501: multiplication, which shows that $P^\dag P$ is a projector.
502: Similarly, one can show that $PP^\dag = I_{\mathcal{H}(X)}$. This
503: implies that $P^\dag$ is an isometric imbedding
504: $P^\dag:\mathcal{H}(X)\rightarrow\mathcal{H}(Y)$.) What are the
505: conditions that $P$ must satisfy in order that $\pi$ be a legal
506: covering map? We answer this question in two steps, by translating
507: parts (i) and (ii) of Definition~\ref{def:graphcover} into
508: conditions on $P$.
509:
510: Using the operator $P$ we define the \emph{quotient matrix}
511: $A_\pi(X) \triangleq P A(Y) P^\dag$, which we want to identify
512: with the adjacency matrix of $X$. Expanding this expression and
513: utilising Definition~\ref{def:graphcover} we obtain
514: \begin{equation}
515: \begin{split}
516: A_\pi(X) &= \sum_{\substack{u \in V(X) \\ l\in
517: \pi^{-1}(u)}}\sum_{\substack{v \in V(X) \\ m\in
518: \pi^{-1}(v)}}\sum_{x,y \in V(Y)}
519: \frac{w_Y(x,y)}{\sqrt{|\pi^{-1}(u)||\pi^{-1}(v)|}} |u\rangle
520: \langle
521: l|x\rangle \langle y|m\rangle \langle v| \\
522: &= \sum_{\substack{u \in V(X) \\ l\in
523: \pi^{-1}(u)}}\sum_{\substack{v \in V(X) \\ m\in \pi^{-1}(v)}}
524: \frac{w_Y(l,m)}{\sqrt{|\pi^{-1}(u)||\pi^{-1}(v)|}}|u\rangle\langle
525: v| \\
526: &=\sum_{\substack{u \in V(X)}}\sum_{\substack{v \in V(X)}} \mu
527: w_X(u,v)|u\rangle\langle v|=\mu A(X).
528: \end{split}
529: \end{equation}
530: This expression shows us that the adjacency matrix $w_Y(u,v)$ for
531: $Y$ satisfies the condition (i) of Definition~\ref{def:graphcover}
532: to be a covering space for the graph $X$ with adjacency matrix
533: $w_X(u,v)$. Note that part (i) of Definition~\ref{def:graphcover}
534: is really saying that $Y$ is a covering space for any graph which
535: has as its adjacency matrix a scalar multiple of $PA(Y)P^\dag$.
536:
537: In order to establish the matrix version of part (ii) of
538: Definition~\ref{def:graphcover} we need to make use of the
539: following result of Godsil and McKay \cite{godsil:1980a}, which is
540: the fundamental connection between $P$ and graph covering spaces.
541: We include a proof for completeness.
542: \begin{lemma}[Godsil and McKay \cite{godsil:1980a}]\label{lem:godsil1}
543: The graph $Y$ is a covering space for $X$ if and only if
544: $PA(Y)=A(X)P$.
545: \end{lemma}
546: \begin{proof}
547: For a vertex $v\in V(Y)$, $1\le j\le |V(X)|$ define $h_{vj}$ to be
548: the sum of the weights of the edges connecting vertices in
549: $\pi^{-1}(u_j)$ with $v$, where $u_j\in V(X)$. That is, $h_{vj} =
550: \sum_{x\in\pi^{-1}(u_j)} w_1(v,x)$. For $1\le j\le |V(X)|$, $v\in
551: V(Y)$, we have
552: \begin{equation}\label{eq:pay}
553: \langle u_j|PA(Y)|v\rangle =
554: \frac{1}{\sqrt{\pi^{-1}(u_j)}}\sum_{x\in \pi^{-1}(u_j)} w_1(x,v) =
555: \frac{h_{vj}}{\sqrt{\pi^{-1}(u_j)}}
556: \end{equation}
557: and
558: \begin{equation}\label{eq:axp}
559: \langle u_j|A(X)P|v\rangle =
560: \frac{w_2(u_j,u_k)}{\sqrt{\pi^{-1}(u_k)}},
561: \end{equation}
562: where $v\in\pi^{-1}(u_k)$.
563:
564: By comparing (\ref{eq:pay}) and (\ref{eq:axp}) we note that if
565: $PA(Y) = A(X)P$, then $h_{vj}$ equals
566: $\sqrt{\frac{|\pi^{-1}(u_j)|}{|\pi^{-1}(u_k)|}}w_2(u_j,u_k)$ for
567: all $v\in \pi^{-1}(u_j)$ and so $\pi$ satisfies part (ii) of
568: definition~\ref{def:graphcover}.
569:
570: For the converse, suppose $\pi$ satisfies the conditions of
571: Definition~\ref{def:graphcover}. We see that
572: \begin{equation}
573: PA(Y) = PA(Y)P^\dag P = A(X) P.
574: \end{equation}
575: \end{proof}
576:
577: It is well known that the spectrum of a graph $Y$ which is a
578: covering space for $X$ is determined, in part, by the spectrum of
579: $X$. Let's make this a bit more precise. Suppose we have a
580: harmonic eigenfunction of $|f\rangle$ of $X$ with eigenvalue
581: $\lambda$. We can \emph{lift} or \emph{pull back} this
582: eigenfunction to a harmonic eigenfunction $|g\rangle$ of $Y$, by
583: defining for each vertex $l\in Y$, $\langle l|g\rangle =
584: \langle\pi(l)|f\rangle/\sqrt{\pi^{-1}(l)}$. Equivalently,
585: $|g\rangle = P^\dag|f\rangle$.
586:
587: We use the notation $|\pi^{-1}(u)\rangle \triangleq
588: 1/\sqrt{|\pi^{-1}(u)|}\sum_{l\in\pi^{-1}(u)} |l\rangle$ and write
589: \begin{equation}
590: |g\rangle = \sum_{l\in V(X)} f(l) |\pi^{-1}(l)\rangle.
591: \end{equation}
592: It follows from Definition~\ref{def:graphcover} that $|g\rangle$
593: is a harmonic eigenfunction for $Y$.
594:
595: \begin{lemma}\label{lem:eveclift}
596: Let $Y$ be a covering space for $X$ with projection map $\pi$.
597: Then for any eigenvector $|f\rangle$ and scalar $\lambda$,
598: $A(X)|f\rangle=\lambda|f\rangle$ if and only if
599: $A(Y)P^\dag|f\rangle = \lambda P^\dag|f\rangle$.
600: \end{lemma}
601: \begin{proof}
602: If $A(X)|f\rangle = \lambda|f\rangle$, then $P^\dag A(X)|f\rangle
603: = \lambda P^\dag|f\rangle$ and so $A(Y)P^\dag|f\rangle = \lambda
604: P^\dag|f\rangle$, by lemma~\ref{lem:godsil1}. If
605: $A(Y)P^\dag|f\rangle = \lambda P^\dag|f\rangle$, then
606: $PA(Y)P^\dag|f\rangle=\lambda PP^\dag|f\rangle$ and so
607: $A(X)|f\rangle = \lambda|f\rangle$.
608: \end{proof}
609:
610: We define the Laplacian $\triangle(X)$ for the covered graph $X$
611: via $\triangle(X) \triangleq PD(Y)P^\dag - PA(Y)P^\dag.$
612:
613: The main result for this section is the following lemma relating
614: the heat kernels and propagators of a graph $X$ and an arbitrary
615: covering space $Y$.
616: \begin{lemma}\label{lem:heatcover}
617: Suppose $Y$ is a covering space for $X$. Let $H[Y](\tau)$ and
618: $H[X](\tau)$ and $U[Y](t)$ and $U[X](\tau)$ denote the heat
619: kernels and propagators for $Y$ and $X$, respectively. Then we
620: have
621: \begin{equation}
622: H[X](\tau) = PH[Y](\tau)P^\dag, \quad \text{and} \quad U[X](t) =
623: PU[Y](t)P^\dag.
624: \end{equation}
625: \end{lemma}
626: \begin{proof}
627: We establish the lemma for the heat kernel. The result for the
628: propagator follows from substituting $\tau = it$.
629:
630: Note first that $A(X)^r=PA(Y)^rP^\dag$, which follows from a
631: simple induction using lemma~\ref{lem:godsil1}.
632:
633: Expanding the heat kernel in an absolutely convergent power series
634: in $\tau$ gives:
635: \begin{equation}
636: \begin{split}
637: H[X](\tau) &= \sum_{j = 0}^\infty \frac{(-\tau PA(Y)P^\dag)^j}{j!}\\
638: &= \sum_{j = 0}^\infty P\frac{(-\tau A(Y))^j}{j!}P^\dag =
639: PH[Y](\tau)P^\dag.
640: \end{split}
641: \end{equation}
642: \end{proof}
643:
644: Because we are living in finite Hilbert spaces, where everything
645: is well-behaved, we don't have to worry about the convergence of
646: series. For this reason, knowing the heat kernel $H(\tau)$ is
647: equivalent to knowing the propagator $U(t)$. We take advantage of
648: this fact by only stating all our results about the propagator in
649: terms of the heat kernel in \emph{imaginary time}.
650:
651:
652: \section{Induced Quantum Walks --- the Quotient Walk}\label{sec:indqw}
653: In this section we show that if a graph $Y$ is a covering space
654: for another graph $X$ with projection $\pi:V(Y)\rightarrow V(X)$
655: then a quantum walk on $Y$ which begins on a state \emph{constant}
656: on fibres of $\pi$ is isomorphic to a quantum walk on $X$. This
657: result provides an insight into the hitting-time speedups observed
658: by Kempe \cite{kempe:2003a} and Childs \emph{et.\ al.}\
659: \cite{childs:2002a, childs:2003a}.
660:
661: \begin{definition}
662: Let $Y$ be a covering space for $X$ with projection
663: $\pi:V(Y)\rightarrow V(X)$. We say that a quantum state
664: $|\psi\rangle \in \mathcal{H}(Y)$ is \emph{fibre-constant} for
665: $\pi$ if $|\psi\rangle$ can be expressed as
666: \begin{equation}
667: |\psi\rangle = \sum_{u\in V(X)} c_{u} |\pi^{-1}(u)\rangle,
668: \end{equation}
669: where we are using the notation
670: \begin{equation}
671: |\pi^{-1}(u)\rangle = P^\dag|u\rangle=
672: \frac{1}{\sqrt{|\pi^{-1}(u)|}}\sum_{x\in\pi^{-1}(u)}|x\rangle
673: \end{equation}
674: of the previous section. We write $|\psi\rangle =
675: P^\dag|\phi\rangle$, where $|\phi\rangle = \sum_{u\in V(X)}
676: c_u|u\rangle$.
677: \end{definition}
678:
679: What is the time evolution of a fibre-constant state
680: $|\psi\rangle$ for covering map $\pi$? The following simple lemma
681: shows us that it is isomorphic to a walk on $X$.
682: \begin{lemma}\label{lem:coverwalk}
683: Suppose $Y$ is a covering space for $X$ with covering map $\pi$.
684: Let $|\psi\rangle = P^\dag|\phi\rangle$ be a fibre-constant state.
685: Then the time evolution of $|\psi(t)\rangle$ on $Y$ is isomorphic
686: to the time evolution of $|\phi\rangle$ on $X$.
687: \end{lemma}
688: \begin{proof}
689: Let $H[Y](\tau)$ be the heat kernel for $Y$. The time evolution of
690: $|\psi(\tau)\rangle$ on $Y$ is given by
691: \begin{equation}
692: \begin{split}
693: |\psi(\tau)\rangle &= H[Y](\tau) P^\dag|\phi\rangle \\
694: &= P^\dag H[X](\tau)|\phi\rangle \quad\text{by
695: lemma~\ref{lem:heatcover}}.
696: \end{split}
697: \end{equation}
698: This equation makes it clear that the time-evolution of
699: $|\psi(\tau)\rangle$ is determined from that of $|\phi\rangle$ on
700: $X$.
701: \end{proof}
702:
703: \section{Examples: the Hypercube $Q^n$ and Cayley and Schreier
704: Graphs}\label{sec:examples}
705:
706: We now illustrate this result for the hypercube and the Cayley and
707: Schreier graphs of a finite group.
708:
709: \begin{definition}
710: Let $G$ be a group, and let $S\subset G$ be a set of group
711: elements such that: (1) The identity element $e\not\in S$; (2) If
712: $x\in S$, then $x^{-1}\in S$. The \emph{Cayley graph} $X(G,S)$
713: associated with $G$ and $S$ is then defined as the simple graph
714: having one vertex associated with each group element and directed
715: edges $(g, h)$ whenever $gh^{-1}\in S$. (It is easy to check that
716: with our definition of the generating set this graph is
717: well-defined.)
718: \end{definition}
719:
720: \begin{example}
721: We think of the hypercube graph $Q_n$ of dimension $n$ as a Cayley
722: graph of the the abelian group $(\mathbb{Z}/2\mathbb{Z})^n$, the
723: $n$-fold direct product of $\mathbb{Z}/2\mathbb{Z}$. Specifically,
724: we write $Q_n =$ $ X((\mathbb{Z}/2\mathbb{Z})^n, \{e_1, e_2,
725: \ldots, e_n\})$, where $e_j = (0,\ldots,1,\ldots,0)$ is the unit
726: vector with a one in the $j$th entry. Note that $|V(Q_n)| = 2^n$.
727: It is easily verified that $Q_n$ is a covering for the weighted
728: path $P^Q_n$ of size $|V(P^Q_n)| = n+1$ with adjacency matrix
729: \begin{equation}\label{eq:qnadj}
730: A(P^Q_n) = \left[ \begin{matrix} 0 & \sqrt{n} & 0 & \cdots & 0\\
731: \sqrt{n} & 0 & \sqrt{2(n-1)} & \cdots & 0 \\
732: 0 & \sqrt{2(n-1)} & 0 &
733: \cdots & 0 \\
734: \vdots & & & \ddots & \sqrt{n}\\
735: 0 & 0 & \cdots & \sqrt{n} & 0
736: \end{matrix}\right].
737: \end{equation}
738: The projection map $\pi:V(Q_n)\rightarrow V(P^Q_n)$ maps the
739: collection of vertices of $V(Q_n)$ of hamming weight $j$ to the
740: single vertex $v_j\in V(P^Q_n)$ at position $j$.
741: \end{example}
742:
743: \begin{remark}
744: Because $Q_n$ is vertex-transitive (i.e.\ the automorphism group
745: $\aut(Q_n)$ acts transitively on $V(Q_n)$) a quantum walk on $Q_n$
746: beginning on the state concentrated on any single vertex is
747: equivalent to a walk on the weighted path $P^Q_n$ beginning at the
748: first vertex. (Note that all Cayley graphs $Y$ are
749: vertex-transitive, where the automorphism group $\aut(Y)$ acts
750: transitively on the vertex set.) The adjacency operator
751: (\ref{eq:qnadj}) of the weighted path is exactly the $x$ angular
752: momentum operator $J_x$ in the spin-$\frac{n}{2}$ irrep.\ of
753: $\text{{\sl SU}}(2)$. One can envisage the quantum state
754: propagating from the first vertex to the last vertex in $P^Q_n$ as
755: a localised state at $z=+\frac{n}{2}$ on north pole of the Bloch
756: sphere propagating across the sphere (and hence becoming
757: completely delocalised around the equator in the process) to the
758: south pole $z=-\frac{n}{2}$.
759: \end{remark}
760:
761: Our second example concerns the Cayley graph for an arbitrary
762: finite abelian or nonabelian group $G$. To introduce it we need a
763: definition.
764:
765: \begin{definition}
766: Given a subgroup $H$ of a finite group $G$ and a generating subset
767: $S$ (recall $s\in S$ implies $s^{-1}\in S$ and $e\not\in S$) of
768: $G$, which we call the edge set, the \emph{Schreier graph}
769: $X=X(G/H, S)$ has as vertices the cosets $gH$, $g\in G$. Two
770: vertices $gH$ and $s^{-1}gH$ are joined by an edge for all $s\in
771: S$.
772: \end{definition}
773:
774: \begin{remark}
775: Note that unless $H$ is a normal subgroup $H\trianglelefteq G$
776: then the Schreier graph $X(G/H, S)$ will contain loops. In the
777: case that $H$ is normal then $X(G/H, S)$ is exactly the Cayley
778: graph of the group $G/H$ with generating set $S$. Note also that
779: our definition of graph covering spaces includes graphs with loops
780: (i.e.\ when $w(u,u)>0$) so that the following results are
781: well-defined.
782: \end{remark}
783:
784: \begin{lemma}
785: The adjacency matrix $A(X)$ for a Cayley graph $X(G,S)$ can be
786: written in the following way
787: \begin{equation}
788: A(X) = \sum_{s\in S}\sum_{x\in G} |x\rangle\langle s^{-1}x|.
789: \end{equation}
790: Similarly, the adjacency matrix $A(X(G/H,S))$ for the Schreier
791: graph $X(G/H,S)$ is given by
792: \begin{equation}
793: A(X(G/H,S)) = \sum_{s\in S}\sum_{gH\in H} |gH\rangle\langle
794: s^{-1}gH|.
795: \end{equation}
796: \end{lemma}
797:
798:
799: We aim to show that if $Y$ is a Cayley graph $Y=X(G,S)$ for a
800: finite group $G$ with generating set $S$ then it is a covering
801: space for the Schreier graphs $X(G/H, S)$ for all $H\le G$. In
802: order to show this we define the projection map $\pi:Y\rightarrow
803: X(G/H, S)$ by $\pi(g) = gH$. The corresponding pull-back operator
804: is then given by
805: \begin{equation}
806: P = \sum_{\substack{ gH\in G/H
807: \\ g'\in gH }} \frac{1}{\sqrt{|H|}}|gH\rangle \langle g'|.
808: \end{equation}
809:
810: In order to establish our claim we need to show that $PA(Y)$ and
811: $A(X(G/H,S))P$ are equal and apply lemma~\ref{lem:godsil1}:
812: \begin{equation}
813: \begin{split}
814: A(X(G/H,S))P &= \frac{1}{\sqrt{|H|}}\sum_{s\in S}\sum_{gH\in
815: G/H}\sum_{\substack{kH\in G/H\\ k'\in kH }} |gH\rangle\langle
816: s^{-1}gH|kH\rangle\langle k'| \\
817: &= \frac{1}{\sqrt{|H|}}\sum_{s\in S}\sum_{\substack{kH\in G/H\\
818: k'\in kH }}
819: \delta_{s^{-1}gH,kH}|gH\rangle\langle k'| \\
820: &= \frac{1}{\sqrt{|H|}}\sum_{s\in S}\sum_{\substack{kH\in G/H\\
821: k'\in kH }} |kH\rangle\langle s^{-1}k'|\\
822: &= PA(Y).
823: \end{split}
824: \end{equation}
825: The transition from the second-last line to the last follows by a
826: change of variable $kH\mapsto skH$ and $k'\mapsto sk'$, and the
827: definition of $P$ and $A(Y)$. This discussion, together with
828: lemma~\ref{lem:coverwalk}, constitutes the following corollary.
829:
830: \begin{corollary}\label{cor:schreierwalk}
831: A quantum walk on a Cayley graph $X(G,S)$ which begins on the
832: fibre-constant state $|\phi_H\rangle = \sqrt{|H|/|G|}\sum_{gH\in
833: G/ H} c_{gH}|\pi^{-1}(gH)\rangle$ is isomorphic to a quantum walk
834: on induced the Schreier graph $X(G/H,S)$ for the subgroup $H$
835: which begins on the induced state $|\phi'\rangle
836: =\sqrt{|H|/|G|}\sum_{gH\in G/ H} c_{gH}|gH\rangle$.
837: \end{corollary}
838:
839:
840:
841: \section{Efficient Gate Sequences for Continuous-Time Quantum
842: Walks}\label{sec:efficient}
843:
844: In this section we explain how to construct efficient gate
845: sequences simulating continuous-time quantum walks on certain
846: classes of graphs $Y$. The principle feature intuitively exploited
847: in these gate sequences is that the graph $Y$ covers a tower
848: $Y\xrightarrow{\pi_{N}}X_{N}\xrightarrow{\pi_{N-1}}X_{N-1}
849: \xrightarrow{\pi_{N-2}}\cdots\xrightarrow{\pi_{1}}X_1$ of graphs
850: $X_j$. An interesting feature of these gate sequences is that
851: their length doesn't depend on the time $t$ the walk is simulated
852: for.
853:
854: Consider the cycle $C_{2^n}$ on $2^n$ vertices, which is the
855: Cayley graph $C_{2^n} = X(\mathbb{Z}/2^n\mathbb{Z}, \{\pm 1\})$ of
856: the cyclic group. We begin by presenting a quantum circuit which
857: simulates the quantum walk on the cycle $C_{2^n}$ which uses only
858: $O(\poly(n))$ gates. A special feature of this gate sequence is
859: that it can simulate the walk for any given time $t$, including
860: exponential times $O(t) = 2^{\poly(n)}$.
861:
862: Recall \cite{lubotzky:1995a} that the eigenvalues and eigenstates
863: of the adjacency matrix $A(C_{m})$ are given in terms of sums of
864: the characters $\mu^j = e^{i\frac{2\pi}{m}j}$, $j=0, \ldots, m-1$,
865: of $\mathbb{Z}/m\mathbb{Z}$. Specifically,
866: \begin{equation}
867: A(C_{m}) = \sum_{j=0}^{m-1} (\mu^j+\mu^{-j})|W(j)\rangle \langle
868: W(j)|,
869: \end{equation}
870: where
871: \begin{equation}
872: |W(j)\rangle = \frac{1}{\sqrt{m}}\sum_{k=0}^{m-1} \mu^{jk}
873: |k\rangle.
874: \end{equation}
875: Note that the vectors $|W(j)\rangle$ are precisely those given by
876: applying the quantum fourier transformation to $|j\rangle$:
877: \begin{equation}
878: |W(j)\rangle = \text{QFT}|j\rangle =
879: \frac{1}{\sqrt{N}}\sum_{k,l=0}^{m-1} \mu^{kl} |k\rangle\langle
880: l|j\rangle.
881: \end{equation}
882:
883: Our objective is to simulate, with a quantum circuit, the
884: evolution
885: \begin{equation}
886: U[C_{2^n}](t) = \sum_{j=0}^{m-1} e^{-i 2\cos\left(\frac{2\pi}{m}
887: j\right) t} |W(j)\rangle\langle W(j)|,
888: \end{equation}
889: where $m=2^n$. We note that this unitary can be written as
890: \begin{equation}
891: U[C_{2^n}](t)=\text{QFT} \Phi(t) \text{QFT}^\dag,
892: \end{equation}
893: where
894: \begin{equation}
895: \Phi(t) = \sum_{j=0}^{m-1} e^{-i 2\cos\left(\frac{2\pi}{m}
896: j\right) t} |j\rangle\langle j|.
897: \end{equation}
898: The phase operation $\Phi(t)$ can be performed efficiently using,
899: for instance, the phase kickback trick \cite{cleve:1997a}. The
900: phase kickback trick shows that if the computation
901: \begin{equation}
902: |j\rangle \mapsto |j\rangle|f(j)\rangle,
903: \end{equation}
904: where $|j\rangle$ is a computational basis state of $n$ qubits and
905: $f(j)$ is an $n$ bit approximation to $(2\cos\left(\frac{2\pi}{m}
906: j\right) t) \mod 2\pi$ (calculable efficiently classically), is
907: implementable efficiently, then the phase changing operation
908: \begin{equation}
909: |j\rangle \mapsto e^{-i 2\cos\left(\frac{2\pi}{m} j\right) t}
910: |j\rangle,
911: \end{equation}
912: is implementable using $O(\poly(n))$ quantum gates
913: \cite{cleve:1997a}. (It is easy to obtain arbitrary accuracy by
914: appending more qubits to the register for $|f(x)\rangle$.)
915:
916: Hence, because there is an efficient (i.e., using $O(\poly(n))$
917: elementary gates) quantum circuit for the quantum fourier
918: transform \cite{nielsen:2000a}, there is an efficient gate
919: sequence approximating the propagator $U[C_{2^n}](t)$. Moreover,
920: the number of gates required to simulate the propagator does not
921: depend on the time $t$.
922:
923: We now note some features of our gate sequence for $U[C_m](t)$.
924: Firstly, we point out that the circuits for $U[C_m](t)$ can be
925: easily generalised to simulate quantum walks on a many other
926: families of graphs. This is because any circulant matrix is
927: diagonalised by the discrete fourier transformation, and so has
928: the same eigenvectors as $C_m$. This observation allows us to
929: generalise our simulation sequence to any circulant matrix whose
930: eigenvalue set $\{\lambda_j\}$ is efficiently calculable using
931: classical or quantum means. In particular, this means that all
932: Cayley graphs of $\mathbb{Z}/m\mathbb{Z}$ and the Paley
933: graphs\footnote{Given a finite field $\mathbb{F}_q$ with $q$
934: elements, the Paley graph $P(\mathbb{F}_q)$ is the graph with
935: vertex set $V=\mathbb{F}_q$ where two vertices are joined when
936: their difference is a square in the field. This is an undirected
937: graph when $q$ is congruent $1 (\text{mod} 4)$ \cite{chung:1997a}.
938: Note that Paley graphs have edge density approximately
939: $\frac{1}{2}$, and so are not efficiently simulatable using the
940: recipe of \cite{aharonov:2003a}.} are efficiently simulatable
941: using relatives of the quantum circuit we have constructed.
942: Additionally, any \emph{cartesian product} of graphs which are
943: efficiently simulatable via the gate sequence above will be
944: efficiently simulatable. This is because the adjacency matrix for
945: the cartesian product of two graphs $X$ and $Y$ is $A(X\times Y) =
946: A(X)\otimes I + I\otimes A(Y)$. In particular, a gate sequence for
947: the $m$-torus $T^m_{2^n}$ on $2^n$ vertices follows from
948: \begin{equation}
949: e^{-iA(T^m) t} = \bigotimes_{j=0}^{m-1} e^{-iA(C_{2^n})t}.
950: \end{equation}
951:
952:
953: The second feature we would like to point out is that the covering
954: space properties of $C_{2^n}$, in particular that $C_{2^n}$ is a
955: covering space for $C_{2^{n/2}}$, allow us to ``rediscover'',
956: using geometric constructions, an efficient quantum circuit for
957: the quantum fourier transform. Roughly speaking, an eigenvector of
958: the base graph $C_{2^{n/2}}$ can be lifted to $2^{n/2}$
959: eigenvectors of $C_{2^n}$ via a simple unitary recipe. This is a
960: general feature of the laplacian on a covering space, and extends
961: even to the continuous case \cite{rosenberg:1997a}. This idea is
962: explored further in \cite{osborne:2004a}.
963:
964: We believe that the existence of covering spaces structures in
965: graphs may be utilised more generally to supply efficient quantum
966: circuits for quantum walks on graphs. We sketch one result which
967: is indicative of this idea.
968:
969: We restrict our attention to regular graphs. Let $Y$ be a regular
970: graph on $2^n$ vertices which is a covering space for a tower of
971: graphs $Y\xrightarrow{\pi_{N}}X_{N}\xrightarrow{\pi_{N-1}}X_{N-1}
972: \xrightarrow{\pi_{N-2}}\cdots\xrightarrow{\pi_{1}}X_1$ with the
973: property that $|V(X_{j+1})|\approx \sqrt{|V(X_j)|}$. In this case
974: $N$ must be $O(\poly(\log(|V(Y|)))$.
975:
976: We assume that $m=|\pi^{-1}_N(u)|$ is the same for all $u\in
977: V(X_N)$. Consider the \emph{transition matrix} $A(u,v)$ for an
978: edge $(u,v)\in E(X_N)$, which is the $2m\times 2m$ matrix whose
979: $(x,y)$ entry, where $x,y\in\pi^{-1}_N(u)\cup\pi^{-1}_N(v)$, is
980: given by $A(Y)_{x,y}$.
981:
982: Suppose we can label the vertices in $\pi^{-1}_N(u)$ so that
983: \begin{equation}
984: A(u,v) = \left(\begin{matrix}0& 1\\ 1 &
985: 0\end{matrix}\right)\otimes I_m,\quad \text{or} \quad A(u,v) =
986: \left(\begin{matrix}0& 1\\ 1 & 0\end{matrix}\right)\otimes
987: \frac{1}{\sqrt{m}}J_m,
988: \end{equation}
989: where $I_m$ is the $m\times m$ identity matrix and $J_m$ is the
990: $m\times m$ all $1$'s matrix. We call such a transition matrix
991: \emph{trivial}, and the we say that the edge $(u,v)$ has
992: \emph{trivial pull-back}.
993:
994:
995: Suppose the edges of $A(X_N)$ have trivial pull-back for all but
996: $O(\poly(\log(|V(Y)|)))$ edges. In the case we can write the
997: adjacency matrix for $A(Y)$ as
998: \begin{equation}\label{eq:adjdecomp}
999: A(Y) = A(X_N)\otimes I_m + \mathcal{D}, \quad \text{or} \quad A(Y)
1000: = A(X_N)\otimes \frac{1}{\sqrt{l}}J_m + \mathcal{D},
1001: \end{equation}
1002: where $\mathcal{D}$ is a symmetric $|V(Y)|\times |V(Y)|$ matrix
1003: whose $(x,y)$ entry, $x,y\in\pi^{-1}_N(u)\cup\pi^{-1}_N(v)$, is
1004: nonzero only when there is an edge $(u,v)\in E(X_N)$ whose
1005: transition matrix is nontrivial. When the edges of $A(X_N)$ have
1006: trivial pull-back for all but $O(\poly(\log(|V(Y)|)))$ edges then
1007: $\mathcal{D}$ is sparse, and has $O(\poly(\log(|V(Y)|)))$ nonzero
1008: entries in each row (in the language of \cite{aharonov:2003a},
1009: $\mathcal{D}$ is \emph{row sparse}).
1010:
1011: We want to find a quantum circuit which simulates
1012: $U[Y](t)=e^{-iA(Y) t}$ (note that because $Y$ is regular the
1013: action of the adjacency matrix and laplacian are equivalent). To
1014: do this we apply the Trotter formula (see \cite{nielsen:2000a} for
1015: details and further discussion)
1016: \begin{equation}
1017: \lim_{n\rightarrow \infty} (e^{-iAt/n}e^{-iBt/n})^n = e^{-i
1018: (A+B)t}.
1019: \end{equation}
1020: By taking $O(n)=\poly(\uinvnorm A \uinvnorm, \uinvnorm B\uinvnorm,
1021: \log(|V(Y)|))$ we gain a good approximation to the time evolution
1022: of $A+B$ for time $O(t) = \poly(n)$ \cite{nielsen:2000a} .
1023:
1024: Applying the Trotter formula to (\ref{eq:adjdecomp}) we find that
1025: \begin{equation}
1026: U[Y](t) \approx ((e^{-iA(X_N)\otimes I_mt/n})
1027: e^{-i\mathcal{D}t/n})^n, \quad \text{or} \quad U[Y](t) \approx
1028: ((e^{-iA(X_N)\otimes J_mt/n}) e^{-i\mathcal{D}t/n})^n.
1029: \end{equation}
1030: Because $\mathcal{D}$ is \emph{row sparse}, the simulation
1031: algorithm of \cite{aharonov:2003a} can be applied to simulate
1032: $e^{-i\mathcal{D}t/n}$ efficiently. (In order to guarantee the
1033: applicability of the Trotter formula we assume that $\uinvnorm
1034: A(Y) \uinvnorm$ grows polylogarithmically with $|V(Y)|$.)
1035:
1036: We recursively reapply this construction to $A(X_N)$ (assuming, at
1037: each step, that the pull-back of the edges $X_j$ is trivial for
1038: all but a small number of edges.) until we have expressed all
1039: instances of $A(Y)$ with $A(X_1)$. This construction furnishes an
1040: efficient (i.e.\ using $O(\poly(\log(|V(Y)|)))$ elementary quantum
1041: gates) quantum circuit which simulates the propagator $U[Y](t)$
1042: accurately for times $t$ which are $O(\poly(\log(|V(Y)|)))$.
1043:
1044: The calculations in this section should be seen as representative
1045: of a general theory analogous to that initiated for quantum
1046: fourier transforms in \cite{moore:2003a}. That is, given a tower
1047: of covering spaces
1048: $Y\xrightarrow{\pi_{N}}X_{N}\xrightarrow{\pi_{N-1}}X_{N-1}
1049: \xrightarrow{\pi_{N-2}}\cdots\xrightarrow{\pi_{1}}X_1$, what are
1050: the conditions the $X_j$ must satisfy in order to give rise to an
1051: efficient gate sequence? We believe that such a theory is worth
1052: developing because it will potentially provide a class of quantum
1053: circuits whose behaviour may be interesting from an algorithmic
1054: point of view. Interestingly, except for Cayley graphs, such
1055: quantum circuits will be unrelated to discrete fourier transforms.
1056:
1057: \section{The Hidden Cover Problem}\label{sec:hiddencover}
1058: The discussion in the previous section indicates that
1059: continuous-time quantum walks on certain graphs admit efficient
1060: gate decompositions whose size doesn't depend on the length of
1061: time the walk is simulated for. This feature can be exploited to
1062: give polynomial time (in $\log(|V(Y)|)$) quantum algorithms which
1063: can measure the hamiltonian $A(Y)$. As a simple corollary of this
1064: we obtain an alternative observable for the hidden subgroup
1065: problem which is not immediately equivalent to that employed by
1066: Shor's algorithm (and related quantum algorithms). Motivated by
1067: these new (efficiently implementable) observables we propose
1068: another generalisation of the hidden subgroup problem: \emph{the
1069: hidden covering space problem} {\sf HiddenCover}.
1070:
1071: The (coherent sampling) hidden subgroup problem for a group $G$
1072: (see, for example, \cite{nielsen:2000a} and references therein,
1073: for a discussion of the hidden subgroup problem) consists of a
1074: black box which outputs at random quantum states
1075: $|\psi_{gH}\rangle$ which are equal superpositions of elements of
1076: cosets of a hidden subgroup $H\le G$. (The Hilbert space
1077: $\mathcal{H}$ in this case is taken to be the group algebra
1078: $\mathbb{C}[G]$.) The problem is to determine $H$ using as few of
1079: the coset states $|\psi_{gH}\rangle$ and as little quantum
1080: computational time as possible.
1081:
1082: The solution of the specific case $G=\mathbb{Z}/pq\mathbb{Z}$,
1083: where $p$ and $q$ are prime and $H=\mathbb{Z}/p\mathbb{Z}$ or
1084: $\mathbb{Z}/q\mathbb{Z}$, is well-known --- this is Shor's
1085: factoring algorithm.
1086:
1087: In the following we refer to a quantum state $|\phi\rangle$ in
1088: $\mathbb{C}[\mathbb{Z}/n\mathbb{Z}]$, $n=pq$, as a
1089: \emph{constant-coset} state on cosets of a subgroup
1090: $\mathbb{Z}/q\mathbb{Z}\le \mathbb{Z}/pq\mathbb{Z}$ if it can be
1091: written
1092: \begin{equation}
1093: |\phi\rangle = \sum_{j=0}^{p-1} c_j|\alpha_j\rangle,
1094: \end{equation}
1095: where $|\alpha_j\rangle = 1/\sqrt{q}\sum_{l=0}^q |j+lp\rangle$, $j
1096: = 1,\ldots, p-1$.
1097:
1098: The expansion of a constant coset state $|\phi\rangle$ in the
1099: basis $|W(j)\rangle$ can be found via (this is essentially the
1100: discrete Poisson summation formula \cite{terras:1999a}):
1101: \begin{equation}\label{eq:expcoeffcoset}
1102: \begin{split}
1103: \langle W(k)|\alpha_j\rangle &= \frac{1}{q\sqrt{p}}\sum_{l=0}^q
1104: e^{-\frac{2\pi i}{pq}(j+lp)k} \\
1105: &= \frac{e^{-\frac{2\pi i}{pq}jk}}{q\sqrt{p}}\sum_{l=0}^q
1106: e^{-\frac{2\pi i}{q} lk} \\
1107: &= \frac{e^{-\frac{2\pi i}{pq}jk}}{\sqrt{p}}\delta_{k,\lambda q},
1108: \quad \lambda = 0,\ldots,p-1.
1109: \end{split}
1110: \end{equation}
1111: The expansion coefficients $\langle W(k)|\alpha_j\rangle$ are
1112: nonzero only when $k$ is a multiple of $q$.
1113:
1114: In the abelian hidden subgroup problem we have a black box which
1115: outputs at random the constant-coset states $|\alpha_j\rangle$.
1116: The standard solution of the hidden subgroup problem proceeds by
1117: applying the quantum fourier transform to the $|\alpha_j\rangle$
1118: and then measuring in the computational basis. This yields an
1119: approximation $\widetilde{r/q}$ to the number $r/q$, for random
1120: integer $r$. After enough samples the identity of the hidden
1121: subgroup can be inferred by applying the continued-fractions
1122: algorithm to $\widetilde{r/q}$.
1123:
1124: We now supply an alternative procedure to the standard quantum
1125: fourier transform which uses a quantum walk on the Cayley graph
1126: $X(\mathbb{Z}/pq\mathbb{Z},\{\pm 1\})$. We don't claim that this
1127: is any different to the standard quantum fourier transform
1128: algorithm for the HSP on $\mathbb{Z}/pq\mathbb{Z}$. The point is
1129: that the generalisation of this procedure to other graphs will
1130: \emph{not} be equivalent to the quantum fourier transform method.
1131: At the moment, however, we can only perform the algorithm for
1132: cyclic groups.
1133:
1134: The eigenvalues of the hamiltonian (i.e.\ the Laplacian
1135: $\triangle$) for the quantum walk on the cycle
1136: $X(\mathbb{Z}/pq\mathbb{Z}, \{\pm 1\})$ are given by
1137: \begin{equation}
1138: \lambda_j = \cos\left(\frac{2\pi}{pq}j\right).
1139: \end{equation}
1140: Consequently, if the hamiltonian $\triangle [X]$ is measured
1141: exactly on a constant-coset state $|\phi\rangle$ then, by the
1142: discussion surrounding (\ref{eq:expcoeffcoset}), the only
1143: eigenvalues that can be measured are those of the form $\lambda_{j
1144: q}=\cos\left(\frac{2\pi}{p}j\right)$, for some random $0\le j\le
1145: q-1$.
1146:
1147: We can effectively measure $\triangle[X]$ using $U[X](t)$ ---
1148: which, as discussed in \S\ref{sec:efficient}, can be implemented
1149: efficiently, using the discretisation of von Neumann's
1150: prescription for measuring a hermitian operator given by
1151: \cite{childs:2002b}. Implementing this measurement yields an
1152: approximation $\widetilde{\lambda}_j$ to an eigenvalue of
1153: $\triangle[X]$ for random $j$. Because $\cos(x)$ is continuous,
1154: the function
1155: \begin{equation}
1156: \widetilde{j/q} = \cos^{-1}(\widetilde{\lambda}_j)
1157: \end{equation}
1158: is a good approximation to the ratio ${j/q}$, for random $0\le
1159: j\le q-1$. Applying the continued fractions algorithm yields $q$.
1160:
1161: The previous result is suggestive of generalisations in the
1162: following way.
1163:
1164: Imagine we have a graph $Y$ which is a covering space
1165: $\pi:Y\rightarrow X$ for $X$. Imagine, further, we have a black
1166: box which outputs at random fibre-constant states
1167: $|\psi_{\pi}\rangle$. Recall that any fibre-constant state can be
1168: written in terms of the pull-backs of the eigenstates of $X$:
1169: \begin{equation}
1170: |\psi_{\pi}\rangle = \sum_{j=0}^{|V(X)|-1} c_j P^\dag
1171: |E_j(X)\rangle,
1172: \end{equation}
1173: where $|E_j(X)\rangle$ are the eigenstates of $X$.
1174:
1175: \emph{This means that if the hamiltonian $\triangle[Y]$ is
1176: measured on $|\psi_{\pi}\rangle$ then it can only report
1177: eigenvalues of $X$, rather than eigenvalues from the full possible
1178: spectrum of $Y$. If the spectrum of $\lambda(X)$ can be
1179: distinguished from the spectra $\lambda(Z_l)$ of the other graphs
1180: $Z_l$ that $Y$ is a covering space for then this measurement can
1181: identify the hidden graph $X$. As we showed previously, in the
1182: case where $Y=X(\mathbb{Z}/pq\mathbb{Z}, \{\pm 1\})$, and
1183: $X=X(\mathbb{Z}/q\mathbb{Z}, \{\pm 1\})$, then this is equivalent
1184: to solving the abelian hidden subgroup problem.}
1185:
1186:
1187: Thus, generalising boldly, we believe that quantum computers ought
1188: to be able to efficiently solve the following problem.
1189:
1190: \begin{framed}
1191: {\large \bf The hidden covering space problem {\sf HiddenCover}}
1192: \\
1193: \noindent {\bf Input:}
1194: \begin{enumerate}
1195: \item A class $\mathcal{C}$ of mathematical objects.%
1196: \item A quantisation scheme which associates a unitary matrix
1197: $U(Y)$, with each $Y\in \mathcal{C}$, acting on an associated
1198: Hilbert space $\mathcal{H}(Y)$. %
1199: \item An object $Y$ from $\mathcal{C}$ with the promise that $Y$
1200: is a covering space $\pi_l:Y\rightarrow X_l$ only for a (known)
1201: set of objects $X_l$, $l=0, \ldots, n$. %
1202: \item A black box which randomly emits fibre-constant states
1203: $|\psi_{\pi_l}\rangle$ for some (unknown) projection $\pi_l$, for
1204: some $l=0, \ldots, n$.
1205: \end{enumerate}
1206: \noindent {\bf Task:} Determine the projection $\pi_l$, and hence
1207: identify the base space $X_l$.
1208: \end{framed}
1209:
1210: We are willing to conjecture that {\sf HiddenCover} is efficiently
1211: solvable on a quantum computer for certain classes of simple
1212: graphs using continuous-time quantum walks.
1213:
1214: A natural question arises: what happens when we apply the
1215: algorithms sketched above to the Cayley graph for the dihedral
1216: group $D_n = \langle s,t \,|\, s^n = t^2 = e\rangle$? In this
1217: case, we are after a hidden transposition $\langle e,
1218: s^jt\rangle$. Unfortunately, it can be verified that the Schreier
1219: graphs for the $n$ hidden transpositions are isospectral, so that,
1220: with one fibre-constant state $|\psi_{\pi}\rangle$ it is
1221: impossible to tell which transposition $\langle e, s^jt\rangle$ is
1222: hidden.
1223:
1224:
1225: \section{Conclusions and Future Directions}
1226: In this paper we have explored two related ideas, both of which
1227: are connected with covering space structures. The first is that a
1228: quantum evolution for an object $Y$ which is a covering space for
1229: another object $X$ can be equivalent to an evolution on the
1230: smaller object $X$. The second is that covering space structures
1231: can be exploited to give efficient gate sequences for their
1232: quantisations. The first idea can be exploited to give
1233: hitting-time speedups for such evolutions as quantum walks.
1234: Additionally, the first idea leads to the notion that hidden
1235: covering spaces can be identified spectrally. The second leads to
1236: the notion that quantum computers can do this efficiently.
1237:
1238: We shall conclude with a list of future directions.
1239: \begin{enumerate}
1240: \item For the dihedral group consider walking on a graph $Y$ which
1241: is not a simple cartesian product of $\log|D_n|$ copies of the
1242: Cayley graph of $D_n$ (this is what happens when you measure the
1243: propagator $\log|D_n|$ times). Possible graphs to try might be
1244: certain graph products of the Cayley graph $X(D_n, S)$ for the dihedral group with generating set $S$. %
1245: \item What about other mathematical objects? There are some
1246: promising candidates, such as algebraic number fields, smooth
1247: manifolds, and knots which have natural structures amenable to
1248: quantisation. %
1249: \item What sorts of computational problems are expressible as
1250: variants of {\sf HiddenCover}? Are there any interesting
1251: computational problems?
1252: \end{enumerate}
1253:
1254: Finally, a word on discrete-time quantum walks. Discrete-time
1255: quantum walks represent another quantisation scheme for simple
1256: graphs where the topology of the graph is encoded in the unitary
1257: quantisation. It is natural to ask how the ideas of this paper
1258: extend to the discrete-time quantum walks? It is interesting to
1259: remark that a coined quantum walk on a graph $X$ is in fact a
1260: discrete-time quantum walk on a certain directed graph $Y$ (namely
1261: $Y$ is the line digraph of $X$ --- see, for example,
1262: \cite{gusfield:1998a} for an application of line digraph in the
1263: design of algorithms) which is homomorphic to $X$. It can be
1264: observed that $Y$ covers $X$ (the spectrum of $Y$ is the spectrum
1265: of $X$ plus a zero eigenvalues with the appropriate multiplicity
1266: \cite{rosenfeld:2001a}). Now, the construction mentioned above is
1267: only one possible quantisation of graphs. Are there other natural
1268: quantisations? Perhaps one can pick out a canonical quantisation
1269: by demanding that it respect covering space structures?
1270:
1271: Given a matrix $M$ (over any field), a directed graph $X$ is said
1272: to be the \emph{graph of} $M$ if the $uv$-th entry of $M$ is
1273: nonzero if and only if there is a directed arc $(u,v)$ for every
1274: pair of vertices $u,v$. A characterisation of graphs of unitary
1275: matrices is still missing (see, for example, the open problems
1276: section of \cite{zyczkowski:2003a}). In general, what is the
1277: relation between graphs of unitary matrices and covering spaces?
1278: The study of this problem may be useful in understanding the
1279: combinatorial properties of certain combinatorial designs like
1280: weighing matrices.
1281:
1282: \subsection*{Acknowledgments} We are very grateful to Andreas Winter
1283: for many helpful and inspiring discussions as well as for many
1284: helpful comments on this paper. We are also very grateful to
1285: Andrew Childs for carefully reading this paper and many helpful
1286: suggestions. TJO is grateful to the EU for support for this
1287: research under the IST project RESQ. SS was supported by a
1288: University of Bristol Research Scholarship.
1289:
1290: \newcommand{\etalchar}[1]{$^{#1}$}
1291: \providecommand{\bysame}{\leavevmode\hbox
1292: to3em{\hrulefill}\thinspace}
1293: \begin{thebibliography}{AvDK{\etalchar{+}}04}
1294: \providecommand{\MR}{\relax\ifhmode\unskip\space\fi MR }
1295: % \MRhref is called by the amsart/book/proc definition of \MR.
1296: \providecommand{\MRhref}[2]{%
1297: \href{http://www.ams.org/mathscinet-getitem?mr=#1}{#2}
1298: } \providecommand{\href}[2]{#2} \expandafter\ifx\csname
1299: url\endcsname\relax
1300: \def\url#1{\texttt{#1}}\fi
1301: \expandafter\ifx\csname
1302: urlprefix\endcsname\relax\def\urlprefix{URL}\fi
1303: \providecommand{\eprint}[2][]{\url{#2}}
1304:
1305: \bibitem[AKR04]{ambainis:2004a}
1306: Andris Ambainis, Julia Kempe, and Alexander Rivosh, \emph{Coins
1307: {M}ake
1308: {Q}uantum {W}alks {F}aster}, 2004, \eprint{quant-ph/0402107}
1309:
1310: \bibitem[Amb03]{ambainis:2003a}
1311: Andris Ambainis, \emph{Quantum walk algorithm for element
1312: distinctness}, 2003,
1313: \eprint{quant-ph/0311001}
1314:
1315: \bibitem[ATS03]{aharonov:2003a}
1316: Dorit Aharonov and Amnon Ta-Shma, \emph{Adiabatic quantum state
1317: generation and
1318: statistical zero knowledge}, Proceedings of the 35th Annual ACM Symposium on
1319: Theory of Computing held in San Diego, CA, June 9--11, 2003 (New York),
1320: Association for Computing Machinery (ACM), 2003, pp.~20--29,
1321: \eprint{quant-ph/0301023}
1322:
1323: \bibitem[AvDK{\etalchar{+}}04]{aharonov:2004a}
1324: D.~Aharonov, W.~van Dam, J.~Kempe, Z.~Landau, S.~Lloyd, and
1325: O.~Regev, \emph{On
1326: the universality of adiabatic quantum computation}, 2004
1327:
1328: \bibitem[Big93]{biggs:1993a}
1329: Norman Biggs, \emph{Algebraic graph theory}, 2nd ed., Cambridge
1330: Mathematical
1331: Library, Cambridge University Press, Cambridge, 1993; \MR{95h:05105}
1332:
1333: \bibitem[BV97]{bernstein:1997a}
1334: Ethan Bernstein and Umesh Vazirani, \emph{Quantum complexity
1335: theory}, SIAM J.
1336: Comput. \textbf{26} (1997), no.~5, 1411--1473; \MR{99a:68053}
1337:
1338: \bibitem[CCD{\etalchar{+}}03]{childs:2003a}
1339: Andrew~M. Childs, Richard Cleve, Enrico Deotto, Edward Farhi, Sam
1340: Gutmann, and
1341: Daniel~A. Spielman, \emph{Exponential algorithmic speedup by a quantum walk},
1342: Proceedings of the 35th Annual ACM Symposium on Theory of Computing held in
1343: San Diego, CA, June 9--11, 2003 (New York), Association for Computing
1344: Machinery (ACM), 2003, pp.~59--68
1345:
1346: \bibitem[CDF{\etalchar{+}}02]{childs:2002b}
1347: Andrew~M. Childs, Enrico Deotto, Edward Farhi, Jeffrey Goldstone,
1348: Sam Gutmann,
1349: and Andrew~J. Landahl, \emph{Quantum search by measurement}, Phys. Rev. A
1350: \textbf{66} (2002), no.~3, 032314, \eprint{quant-ph/0204013}
1351:
1352: \bibitem[CDS95]{cvetkovic:1995a}
1353: Drago{\v{s}}~M. Cvetkovi{\'c}, Michael Doob, and Horst Sachs,
1354: \emph{Spectra of
1355: graphs}, 3rd ed., Johann Ambrosius Barth, Heidelberg, 1995; \MR{96b:05108}
1356:
1357: \bibitem[CE03]{childs:2003b}
1358: Andrew~M. Childs and Jason~M. Eisenberg, \emph{Quantum algorithms
1359: for subset
1360: finding}, 2003, \eprint{quant-ph/0311038}
1361:
1362: \bibitem[CEMM98]{cleve:1997a}
1363: R.~Cleve, A.~Ekert, C.~Macchiavello, and M.~Mosca, \emph{Quantum
1364: algorithms
1365: revisited}, R. Soc. Lond. Proc. Ser. A Math. Phys. Eng. Sci. \textbf{454}
1366: (1998), no.~1969, 339--354, Quantum coherence and decoherence (Santa Barbara,
1367: CA, 1996), \eprint{quant-ph/9708016}; \MR{99m:81030}
1368:
1369: \bibitem[CFG02]{childs:2002a}
1370: Andrew~M. Childs, Edward Farhi, and Sam Gutmann, \emph{An example
1371: of the
1372: difference between quantum and classical random walks}, Quantum Inf. Process.
1373: \textbf{1} (2002), no.~1-2, 35--43, \eprint{quant-ph/0103020}; \MR{1 964 317}
1374:
1375: \bibitem[CG03]{childs:2003c}
1376: Andrew~M. Childs and Jeffrey Goldstone, \emph{Spatial seach by
1377: quantum walk},
1378: 2003, \eprint{quant-ph/0306054}
1379:
1380: \bibitem[Chu97]{chung:1997a}
1381: Fan R.~K. Chung, \emph{Spectral graph theory}, CBMS Regional
1382: Conference Series
1383: in Mathematics, vol.~92, Published for the Conference Board of the
1384: Mathematical Sciences, Washington, DC, 1997; \MR{97k:58183}
1385:
1386: \bibitem[CY99]{chung:1999a}
1387: Fan Chung and S.-T. Yau, \emph{Coverings, heat kernels and
1388: spanning trees},
1389: Electron. J. Combin. \textbf{6} (1999), no.~1, Research Paper 12, 21 pp.\
1390: (electronic); \MR{2000g:35079}
1391:
1392: \bibitem[GKWS98]{gusfield:1998a}
1393: Dan Gusfield, Richard Karp, Lusheng Wang, and Paul Stelling,
1394: \emph{Graph
1395: traversals, genes and matroids: an efficient case of the travelling salesman
1396: problem}, Discrete Appl. Math. \textbf{88} (1998), no.~1-3, 167--180;
1397: \MR{99k:90133}
1398:
1399: \bibitem[GM80]{godsil:1980a}
1400: C.~D. Godsil and B.~D. McKay, \emph{Feasibility conditions for the
1401: existence of
1402: walk-regular graphs}, Linear Algebra Appl. \textbf{30} (1980), 51--61;
1403: \MR{83d:05068}
1404:
1405: \bibitem[Hat02]{hatcher:2002a}
1406: Allen Hatcher, \emph{Algebraic topology}, Cambridge University
1407: Press,
1408: Cambridge, 2002; \MR{2002k:55001}
1409:
1410: \bibitem[Kem03a]{kempe:2003b}
1411: Julia Kempe, \emph{Quantum random walks: an introductory
1412: overview}, Contemp.
1413: Phys. \textbf{44} (2003), no.~4, 307--327, \eprint{quant-ph/0303081}
1414:
1415: \bibitem[Kem03b]{kempe:2003a}
1416: Julia Kempe, \emph{Quantum {R}andom {W}alks {H}it {E}xponentially
1417: {F}aster},
1418: RANDOM-APPROX (Sanjeev Arora, Klaus Jansen, Jos{\'e} D.~P. Rolim, and Amit
1419: Sahai, eds.), Lecture Notes in Computer Science, vol. 2764, Springer, 2003,
1420: pp.~354--369, \eprint{quant-ph/0205083}
1421:
1422: \bibitem[Lub95]{lubotzky:1995a}
1423: Alexander Lubotzky, \emph{Cayley graphs: eigenvalues, expanders
1424: and random
1425: walks}, Surveys in combinatorics, 1995 (Stirling), London Math. Soc. Lecture
1426: Note Ser., vol. 218, Cambridge Univ. Press, Cambridge, 1995, pp.~155--189;
1427: \MR{96k:05081}
1428:
1429: \bibitem[ML98]{maclane:1998a}
1430: Saunders Mac~Lane, \emph{Categories for the working
1431: mathematician}, 2nd ed.,
1432: Graduate Texts in Mathematics, vol.~5, Springer-Verlag, New York, 1998;
1433: \MR{2001j:18001}
1434:
1435: \bibitem[MR02]{moore:2002a}
1436: Cristopher Moore and Alexander Russell, \emph{Quantum walks on the
1437: hypercube},
1438: RANDOM (Jos{\'e} D.~P. Rolim and Salil~P. Vadhan, eds.), Lecture Notes in
1439: Computer Science, vol. 2483, Springer, 2002, pp.~164--178,
1440: \eprint{quant-ph/0104137}
1441:
1442: \bibitem[MRR03]{moore:2003a}
1443: Cristopher Moore, Daniel Rockmore, and Alexander Russell,
1444: \emph{Generic
1445: {Q}uantum {F}ourier {T}ransforms}, To appear in SODA 2004,
1446: \eprint{quant-ph/0304064}
1447:
1448: \bibitem[MSS03]{magniez:2003a}
1449: Fr{\'e}d{\'e}ric Magniez, Miklos Santha, and Mario Szegedy,
1450: \emph{An
1451: ${O}(n^{1.3})$ {Q}uantum {A}lgorithm for the {T}riangle {P}roblem}, 2003,
1452: \eprint{quant-ph/0310134}
1453:
1454: \bibitem[NC00]{nielsen:2000a}
1455: Michael~A. Nielsen and Isaac~L. Chuang, \emph{Quantum computation
1456: and quantum
1457: information}, Cambridge University Press, Cambridge, 2000; \MR{1 796 805}
1458:
1459: \bibitem[Osb04]{osborne:2004a}
1460: Tobias~J. Osborne, \emph{Efficient {Q}uantum {G}ate {S}equences
1461: from {G}raph
1462: {C}overing {S}paces}, 2004
1463:
1464: \bibitem[Pre98]{preskillnotes}
1465: John Preskill, {P}hysics 229: {A}dvanced {M}athematical {M}ethods
1466: of {P}hysics
1467: --- {Q}uan\-tum {C}omputation and {I}nformation. {C}alifornia {I}nstitute of
1468: {T}echnology, \texttt{http://www.theory.caltech/edu/people/}
1469: \texttt{preskill/ph229/}, 1998
1470:
1471: \bibitem[RB01]{briegel:2001b}
1472: Robert Raussendorf and Hans~J. Briegel, \emph{A {O}ne-{W}ay
1473: {Q}uantum
1474: {C}omputer}, Phys. Rev. Lett. \textbf{86} (2001), no.~22, 5188--5191,
1475: \eprint{quant-ph/0010033}
1476:
1477: \bibitem[Ros97]{rosenberg:1997a}
1478: Steven Rosenberg, \emph{The {L}aplacian on a {R}iemannian
1479: manifold}, London
1480: Mathematical Society Student Texts, vol.~31, Cambridge University Press,
1481: Cambridge, 1997, An introduction to analysis on manifolds; \MR{98k:58206}
1482:
1483: \bibitem[Ros01]{rosenfeld:2001a}
1484: Vladimir~Raphael Rosenfeld, \emph{Some spectral properties of the
1485: arc-graph},
1486: Match (2001), no.~43, 41--48; \MR{2002c:05115}
1487:
1488: \bibitem[Sho94]{shor:1994a}
1489: Peter~W. Shor, \emph{Algorithms for quantum computation: discrete
1490: logarithms
1491: and factoring}, 35th Annual Symposium on Foundations of Computer Science
1492: (Santa Fe, NM, 1994), IEEE Comput. Soc. Press, Los Alamitos, CA, 1994,
1493: pp.~124--134, \eprint{quant-ph/9508027}; \MR{1 489 242}
1494:
1495: \bibitem[SKW03]{shenvi:2003a}
1496: Neil Shenvi, Julia Kempe, and K.~Birgitta Whaley, \emph{Quantum
1497: random-walk
1498: search algorithm}, Phys. Rev. A \textbf{67} (2003), no.~5, 052307,
1499: \eprint{quant-ph/0210064}
1500:
1501: \bibitem[ST96]{stark:1996a}
1502: H.~M. Stark and A.~A. Terras, \emph{Zeta functions of finite
1503: graphs and
1504: coverings}, Adv. Math. \textbf{121} (1996), no.~1, 124--165; \MR{98b:11094}
1505:
1506: \bibitem[ST00]{stark:2000a}
1507: H.~M. Stark and A.~A. Terras, \emph{Zeta functions of finite
1508: graphs and
1509: coverings. {II}}, Adv. Math. \textbf{154} (2000), no.~1, 132--195;
1510: \MR{2002f:11123}
1511:
1512: \bibitem[ST01]{stark:1999a}
1513: H.~M. Stark and A.~A. Terras, \emph{Artin {$L$}-functions of graph
1514: coverings},
1515: Dynamical, spectral, and arithmetic zeta functions (San Antonio, TX, 1999),
1516: Contemp. Math., vol. 290, Amer. Math. Soc., Providence, RI, 2001,
1517: pp.~181--195; \MR{2003c:11113}
1518:
1519: \bibitem[Sze04]{szegedy:2004b}
1520: Mario Szegedy, \emph{Spectra of {Q}uantized {W}alks and a
1521: $\sqrt{\delta\epsilon}$-{R}ule}, 2004, \eprint{quant-ph/0401053}
1522:
1523: \bibitem[Ter99]{terras:1999a}
1524: Audrey Terras, \emph{Fourier analysis on finite groups and
1525: applications},
1526: London Mathematical Society Student Texts, vol.~43, Cambridge University
1527: Press, Cambridge, 1999; \MR{2000d:11003}
1528:
1529: \bibitem[Ter02]{terras:2002a}
1530: Audrey Terras, \emph{Finite quantum chaos}, Amer. Math. Monthly
1531: \textbf{109}
1532: (2002), no.~2, 121--139; \MR{1 903 150}
1533:
1534: \bibitem[{\.Z}KSS03]{zyczkowski:2003a}
1535: Karol {\.Z}yczkowski, Marek Ku{\'s}, Wojciech S{\l}omczy{\'n}ski,
1536: and
1537: Hans-J{\"u}rgen Sommers, \emph{Random unistochastic matrices}, J. Phys. A
1538: \textbf{36} (2003), no.~12, 3425--3450; \MR{1 986 428}
1539:
1540: \end{thebibliography}
1541:
1542:
1543: \end{document}
1544: