1: % ************************************************************
2: % Intrinsic Quantum Computation
3: % was Memory and Synchronization in Finitary Quantum Processes
4: % jpc: 7/14/06, 7/22/06, 8/1/06, 8/6/06, 11/19/06, 1/30/07
5: % kw: 21/8/06, 08/11/06
6:
7: \documentclass[pre,twocolumn,showpacs,superscriptaddress,preprintnumbers,floatfix]{revtex4}
8:
9: % Packages
10: \usepackage{dcolumn}
11: \usepackage{url}
12: \usepackage{amsmath}
13: \usepackage{amssymb}
14: \usepackage{graphicx}
15: \usepackage{bm} % bold math
16: \usepackage{bbm}
17: \usepackage{verbatim}
18: \usepackage{stmaryrd}
19: %\usepackage{theorem}
20: \usepackage{amsthm}
21:
22: \def\bra#1{\mathinner{\langle{#1}|}}
23: \def\ket#1{\mathinner{|{#1}\rangle}}
24: \def\braket#1{\mathinner{\langle{#1}\rangle}}
25:
26: \theoremstyle{break} \newtheorem{Cor}{Corollary}
27: \theoremstyle{plain} \newtheorem*{ProCor}{Proof}
28: \theoremstyle{break} \newtheorem{The}{Theorem}
29: \theoremstyle{plain} \newtheorem*{ProThe}{Proof}
30: \theoremstyle{break} \newtheorem{Prop}{Proposition}
31: \theoremstyle{plain} \newtheorem*{ProProp}{Proof}
32: \theoremstyle{break} \newtheorem{Conj}{Conjecture}
33: \theoremstyle{plain} \newtheorem{Rem}{Remark}
34: \theoremstyle{break} \newtheorem*{Def}{Definition}
35: \theoremstyle{plain} \newtheorem*{Not}{Notation}
36:
37: \newcommand{\Prob} {\mathrm{Pr}}
38: \newcommand{\Lang} {\mathcal{L}}
39: \newcommand{\supp} {\mathrm{supp}\;}
40: \newcommand{\sub} {\mathrm{sub}}
41: \newcommand{\prcl} {\mathcal{P}}
42: \newcommand{\prfl} {\mathcal{F}}
43: \newcommand{\stochL} {\mathcal{S}}
44: \newcommand{\ql} {\mathcal{Q}}
45: \newcommand{\qpl} {\mathcal{P}}
46: \newcommand{\field}[1]{\mathbb{#1}}
47: \newcommand{\rank} {\mathrm{rank}}
48: \newcommand{\svector}[1]{\bra{ {#1} }}
49: \newcommand{\asympt} {\rho^s }
50: \newcommand{\mperiod} {\mathnormal{p}}
51: \newcommand{\Ldiv} {\mathcal{D}}
52: \newcommand{\subw} {\mathrm{sub}}
53:
54: \newcommand{\past}{\stackrel{\leftarrow}{S}}
55: \newcommand{\pastt}{\stackrel{\leftarrow}{S_t}}
56: \newcommand{\pasttp}{\stackrel{\leftarrow}{S_{t^\Prime}}}
57: \newcommand{\future}{\stackrel{\rightarrow}{S}}
58: \newcommand{\all}{\stackrel{\leftrightarrow}{S}}
59: \newcommand{\pastL}{\stackrel{\leftarrow}{S_L}}
60: \newcommand{\futureL}{\stackrel{\rightarrow}{S_L}}
61: \def\gmu {g_\mu}
62: \def\hmu {h_\mu}
63: \def\Cmu {C_\mu}
64: \def\Abet {\mathcal{A}}
65: \def\EE {{\bf E} }
66: \def\SI {{\bf S} }
67: \def\TP {{\bf G}}
68: \def\TI {{\bf T} }
69: \def\l2 {{\rm log}_2}
70:
71: \addtolength{\abovedisplayskip}{-.05in}
72: \addtolength{\belowdisplayskip}{-.05in}
73: \addtolength{\dbltextfloatsep}{-.10in}
74: \addtolength{\abovecaptionskip}{-.10in}
75: \addtolength{\belowcaptionskip}{-.10in}
76: \parskip 0pt
77:
78: \begin{document}
79:
80: \title{Intrinsic Quantum Computation}
81:
82: \author{James P. Crutchfield}
83: \email{chaos@cse.ucdavis.edu}
84: \affiliation{Center for Computational Science \& Engineering and Physics Department,
85: University of California Davis, One Shields Avenue, Davis, CA 95616}
86:
87: \author{Karoline Wiesner}
88: \email{karoline@cse.ucdavis.edu}
89: \affiliation{Center for Computational Science \& Engineering and Physics Department,
90: University of California Davis, One Shields Avenue, Davis, CA 95616}
91:
92: \date{\today}
93:
94: \bibliographystyle{unsrt}
95:
96: % ************************* ABSTRACT *************************
97: \begin{abstract}
98: We introduce ways to measure information storage in quantum systems, using
99: a recently introduced computation-theoretic model that accounts for
100: measurement effects. The first, the quantum excess entropy, quantifies the
101: shared information between a quantum process's past and its future. The
102: second, the quantum transient information, determines the difficulty with
103: which an observer comes to know the internal state of a quantum process
104: through measurements. We contrast these with von Neumann entropy and quantum
105: entropy rate and provide a closed-form expression for the latter for the
106: class of deterministic quantum processes.
107: \end{abstract}
108:
109: \pacs{
110: % 61.72.Dd, % Experimental determination of defects by diff/scattering
111: % 61.10.Nz, % Single crystal/powder diffraction
112: % 61.43.-j, % Disordered Solids
113: % 81.30.Hd % Constant-Composition Solid State Transformations
114: % 02.50.Ey % Stochastic processes
115: % 02.50.Ga % Markov processes
116: % 05.20.-y % Classical statistical mechanics
117: % 05.45.Tp % Time series analysis
118: % 65.40.Gr % Thermodynamics of solids: Entropy, other quantities
119: % 89.75.Kd % Complex Systems: Patterns
120: % 02.50.-r % Probability theory, stochastic processes, and statistics
121: 03.67.-a % Quantum information
122: 89.70.+c % Information science
123: 05.45.-a % Nonlinear dynamics and nonlinear dynamical systems
124: 03.67.Lx % Quantum computation
125: }
126: %
127: \preprint{Santa Fe Institute Working Paper 06-11-045}
128: \preprint{arxiv.org/quant-ph/0611202}
129:
130: \maketitle
131:
132: % ****************************************************************
133:
134: % \tableofcontents
135:
136: % ************************* INTRODUCTION *************************
137:
138: \vspace{-.25in}
139:
140: Poincar\'e discovered that classical mechanical systems can appear to
141: be random \cite{Poin92}. Kolmogorov, adapting Shannon's theory of
142: communication \cite{Shan48a}, showed that their degree of randomness
143: can be measured as a rate of information production \cite{Kolm59}.
144: Shannon, in fact, adopted the word ``entropy'' to describe information
145: transmitted through a communication channel based on a suggestion by von
146: Neumann, who had recently used entropy to describe the distribution of
147: states in quantum systems \cite[Ch. 5]{Neum32a}. Information
148: has a long history in quantifying degrees of disorder
149: in both classical and quantum mechanical systems.
150:
151: In a seemingly unrelated effort, Feynman proposed to develop quantum
152: computers \cite{feynman:82} with the goal of (greatly) accelerating
153: simulation of quantum systems. Their potential power, though, was brought
154: to the fore most recently by the discovery of algorithms that would run
155: markedly faster on quantum computers than on classical computers.
156: Experimental efforts to find a suitable physical substrate for a quantum
157: computer have been well underway for almost a decade now \cite{Zole05a}.
158:
159: In parallel, the study of the quantum behavior of classically chaotic
160: systems gained much interest \cite{reic04a}, most recently including
161: the role of measurement. It turns out that measurement interaction
162: leads to genuinely chaotic behavior in quantum systems, even far from
163: the semi-classical limit \cite{habib:06}.
164:
165: How are computing, information creation, and dynamics related? A contemporary
166: view of these three historical threads is that they are not so
167: disparate. We show here that a synthesis leads to methods to
168: analyze how quantum processes store and manipulate information---what
169: we refer to as \emph{intrinsic quantum computation}.
170:
171: Computation-theoretic comparisons of classical (stochastic) and
172: measured quantum systems showed that a quantum system's
173: behavior depends sensitively on how it is measured. The differences
174: were summarized in a hierarchy of computational model classes for
175: quantum processes \cite{wiesner:06b}. Here we adopt an
176: information-theoretic approach that, on the one hand, is more quantitative
177: than and, on the other, is complementary to the structural view emphasized
178: by the computation-theoretic analysis. To start, recall the
179: \emph{finite-state quantum generators} defined there. They consist of a finite
180: set of \emph{internal states} $Q = \{q_i: i = 0, \ldots, |Q|-1 \}$. The
181: \emph{state vector} over the internal states is an element of a
182: $|Q|$-dimensional Hilbert
183: space: $\bra{\psi} \in \mathcal{H}$. At each time step a
184: quantum generator outputs a symbol $s \in \Abet$ and updates its
185: state vector.
186:
187: The temporal dynamics is governed by a set of $|Q|$-dimensional
188: \emph{transition matrices} $\{T(s) = U \cdot P(s), s \in \Abet \}$,
189: whose components are elements of the complex unit disk and where each
190: is a product of a unitary matrix $U$ and a projection operator $P(s)$.
191: $U$ governs the evolution of
192: the state vector. $\mathbf{P} =\{ P(s): s \in \Abet \}$ is a set of
193: \emph{projection operators} that determine how the state vector is measured.
194: We base our analysis on the class of projective measurements, applicable to
195: closed quantum systems\footnote{The generalization to open quantum systems
196: using any (including non-orthogonal) \emph{positive operator valued measures}
197: (POVM) requires a description
198: of the state after measurement in addition to the measurement statistics.
199: Repeated measurement, however, is the core of a quantum process as the term
200: is used here. We therefore leave the discussion of quantum processes observed
201: with a general POVM to a separate study.}. In the measurement setting used here the
202: output symbol $s$ is identified with the measurement outcome and
203: labels the system's eigenvalues. We represent the event of no measurement
204: with the symbol $\lambda$; $P(\lambda)$ can be thought of as the identity
205: matrix. Thus, starting with state vector $\svector{\psi_0}$, a single
206: time-step yields $\langle \psi(s) \vert = \langle\psi_0\vert U \cdot P(s)$.
207:
208: To describe how an observer chooses to measure a quantum system we introduce
209: the notion of a \emph{measurement protocol}. A
210: \emph{measurement act}---applying
211: $P(m), m \in \Abet \bigcup \lambda$---returns a value $s \in \Abet$
212: or nothing ($m = \lambda$). A \emph{measurement protocol}, then, is a choice
213: of a sequence of measurement acts $m_t \in \Abet \bigcup \lambda$.
214: If an observer asks if the measurement outcomes $s_1 s_2 s_3$ occur,
215: the answer depends on the measurement protocol, since the observer
216: could choose protocol
217: $s_1 \lambda \lambda s_2 \lambda s_3$, $s_1 s_2 s_3$, or others.
218: Fixing a measurement protocol, then, the state vector after observing the
219: measurement series $m^N = m_1 \ldots m_N$ is
220: $
221: \langle \psi(m^N) \vert
222: = \langle\psi_0\vert
223: U \cdot P(m_1) \cdot U \cdot P(m_2) \cdots U \cdot P(m_N)
224: $.
225:
226: A {\em quantum process} is the joint distribution
227: ${\rm Pr} ( \ldots S_{-2} S_{-1} S_0 S_1 \ldots)$ over the infinite chain of
228: measurement random variables $S_t$. Defined in this way, it is the quantum
229: analog of what Shannon
230: referred to as an information source \cite{cover}. Starting a generator
231: in $\langle\psi_0\vert$ the probability of output $s$ is given by
232: the state vector without renormalization:
233: $\Prob(s) = \left\Vert \psi(s) \right\Vert^2$. By extension,
234: the \emph{word distribution}, the probability of
235: outcomes $s^L$ from a sequence of $L$ measurements, is
236: $\Prob(s^L) = \left\Vert \psi(s^L) \right\Vert^2$.
237:
238: We can use the observed behavior, as reflected in the word distribution,
239: to come to a number of conclusions about how a quantum process generates
240: randomness and stores and transforms historical information. The
241: {\em Shannon entropy} of length-$L$ sequences is defined
242: \begin{align}
243: \label{eq:HL}
244: H(L) &\equiv - \sum_{ s^L \in {\cal A}^L } \Prob (s^L) \l2 \Prob (s^L) ~.
245: \end{align}
246: It measures the average surprise in observing the ``event'' $s^L$.
247: Ref. \cite{crutchfield:03} showed that a stochastic process's informational
248: properties can be derived systematically by taking derivatives and then
249: integrals of $H(L)$, as a function of $L$.
250: For example, the {\em source entropy rate} $\hmu$ is the rate of increase with
251: respect to $L$ of the Shannon entropy in the large-$L$ limit:
252: \begin{equation}
253: \hmu \equiv \lim_{L \rightarrow \infty} \left[ H(L) - H(L-1) \right] \; ,
254: \label{ent.def}
255: \end{equation}
256: where the units are \emph{bits/measurement} \cite{cover}. The entropy rate
257: $h_\mu$ quantifies the irreducible randomness in measurement sequences
258: produced by a process: the randomness that remains after the correlations and
259: structures in longer and longer sequences are taken into account.
260: The latter, in turn, are measured by two complementary quantities.
261: The amount $I(\past;\future)$ of mutual information \cite{cover} shared
262: between a process's past $\past$ and its future $\future$ is given by the
263: \emph{excess entropy} $\EE$ \cite{crutchfield:03}. It is the subextensive
264: part of $H(L)$:
265: \begin{equation}
266: \EE = \lim_{L \rightarrow \infty} [ H(L) - \hmu L ]\;.
267: \label{EEfromEntropyGrowth}
268: \end{equation}
269: Note that the units here are \emph{bits}.
270: Ref. \cite{crutchfield:03} also showed that the amount of information an
271: observer must extract from measurements in order to know the internal state
272: is given by the \emph{transient information}:
273: \begin{equation}
274: \TI \equiv \sum_{L=0}^\infty \left[ \EE + \hmu L - H(L) \right] \;,
275: \label{T.def}
276: \end{equation}
277: where the units are {\em bits $\times$ measurements}.
278:
279: If one can determine the word distribution $\Prob (s^L)$, then, in principle
280: at least, one can calculate a quantum process's informational properties:
281: $h_\mu$, $\EE$, and $\TI$. Fortunately, there are several classes of quantum
282: process for which one can give closed-form expressions. For example, we will
283: provide a way of computing $h_{\mu}$ exactly, using the finite-state machine
284: representation of a quantum process, similar to what is known for classical
285: stochastic processes \cite{Crut97a}. We then give examples of various
286: quantum processes at the end, measuring their intrinsic computation.
287:
288: In quantum theory one distinguishes between complete and incomplete
289: measurements. A \emph{complete measurement} projects onto a one-dimensional
290: subspace of $\mathcal{H}$. A \emph{complete quantum generator} (CQG) is
291: simply a quantum generator observed with complete measurements.
292:
293: Another, as it turns out, more general class of quantum processes are those
294: that can be described by a \emph{deterministic quantum generator} (DQG),
295: where each matrix $T(s)$ has at most one nonzero entry per row.
296: The importance of \emph{determinism} comes from the fact that it guarantees
297: that an internal-state sequence $q_t q_{t+1} q_{t+2} \ldots$ is in 1-to-1
298: correspondence with a measurement sequence $s_t s_{t+1} s_{t+2}$.
299: \footnote{Determinism here refers, as it does in automata theory\cite{hopcroft},
300: to the stated property; it does \emph{not} imply ``non-stochastic''.}
301:
302: It simplifies matters if the word distribution is independent of the initial
303: state vector $\bra{\psi_0}$. This is achieved by switching to the density matrix
304: formalism \cite{wiesner:06b}. A \emph{stationary state distribution} $\rho^s$
305: can then be found for DQGs and
306: $\Prob(s^L) = \mathrm{Tr} \left[ T^{\dagger}(s^L)\rho^s T(s^L) \right]$
307: is start-state independent.
308:
309: We showed that every DQG has an equivalent deterministic classical generator that produces
310: the same stochastic process \cite{wiesner:06b}. Specifically,
311: given a DQG $M = \{U,P(s)\}$, the \emph{equivalent} classical generator
312: $M^{\prime} = \{T,P(s)\}$ has unistochastic transition matrix
313: $T_{ij} = | U_{ij} |^2$.
314:
315: As a consequence, when a quantum process can be represented by a complete
316: or a deterministic generator, closed-form expressions exist for several of
317: the information quantities. For example, adapting Ref. \cite{Crut97a} to
318: DQGs we obtain the \emph{quantum entropy rate}:
319: \begin{equation}
320: h_\mu = - |Q|^{-1} \sum_{i=0}^{|Q|-1}\sum_{j=0}^{|Q|-1}
321: |U_{ij}|^2 \log_2 |U_{ij}|^2 ~,
322: \end{equation}
323: using the fact that $\rho^s_{ii}=|Q|^{-1}$ for DQGs. This should be compared
324: to the entropy rates for nondeterministic classical and quantum processes
325: which, in general, have no closed-form expression.
326: For DQGs we introduce the
327: \emph{internal-state quantum entropy}:
328: \begin{equation}
329: S_q = - \sum_{i=0}^{|Q|-1} \rho^s_{ii} \log_2 \rho^s_{ii} ~,
330: \end{equation}
331: which measures the average uncertainty in knowing the internal state. Again,
332: $\rho_{ii}=|Q|^{-1}$ for DQGs allows us to simplify: $S_q = \log_2 |Q|$.
333:
334: $S_q$, as defined here, is nothing other than
335: the \emph{von Neumann entropy} $S(\rho)$ of the density matrix $\rho$
336: \cite{nielsen}.
337: Note, that our use of the density matrix is as a time average, not an ensemble
338: average. This should be further compared to the
339: von Neumann entropy
340: $S(\otimes_{i=0}^{L-1} \rho_i)$ over a sequence of density matrices produced
341: by a stationary quantum source \cite{wehrl:78}. Using this, a
342: \emph{density-matrix quantum entropy rate} can be defined as the limit of
343: $S(\otimes_{i=0}^{L-1} \rho_i)/L$ for large $L$. Note that these alternative
344: definitions of entropy and entropy rate suffer from two problems. First,
345: they refer to internal variables that are not directly measurable.
346: Second, they do not take the effects of measurement into account.
347: Importantly, $\hmu$, $\EE$, and $\TI$ do not suffer from these problems.
348: Let's explore their consequences for characterizing intrinsic computation
349: in several simple quantum dynamical systems. (An exhaustive analysis of
350: all (deterministic and non-deterministic) few-qubit quantum processes will
351: appear elsewhere.)
352:
353: The \emph{iterated beam-splitter} (Fig.~\ref{fig:IteratedBeamSplitter})
354: is a quantum system that, despite its simplicity, makes a direct connection
355: to familiar experiments. Photons are sent through a beam-splitter,
356: producing two possible paths, which are redirected by mirrors and recombined
357: at the beam-splitter after passing around the feedback loop determined by the
358: mirrors. Nondestructive detectors are located along the upper and lower paths.
359:
360: \begin{figure}
361: \begin{center}
362: \resizebox{!}{1.50in}{\includegraphics{IteratedBeamSplitter.eps}}
363: \end{center}
364: \caption{Iterated beam splitter: Thick solid lines are mirrors. Photon
365: detectors, marked as D, are placed in upper and lower photon paths
366: (solid gray lines).
367: }
368: \label{fig:IteratedBeamSplitter}
369: \end{figure}
370:
371: The iterated beam splitter is a quantum dynamical system with
372: a two-dimensional state space
373: (which path is taken around the loop) with eigenstates ``above''
374: $\svector{\phi_A}$ and ``below''$\svector{\phi_B}$. Its dynamics
375: are given by a unitary operator for the beam-splitter, the Hadamard matrix
376: $
377: U_H = \tfrac{1}{\sqrt{2}}
378: \left(
379: \begin{smallmatrix}
380: 1 & 1 \\
381: 1 & -1
382: \end{smallmatrix}
383: \right) $,
384: and measurement operators representing the detectors:
385: $P(0) = \ket{0}\bra{0}$ and $P(1) = \ket{1}\bra{1}$,
386: in the experiment's eigenbasis. Measurement symbol $0$
387: stands for ``above'' and symbol $1$ stands for ``below''.
388:
389: The detectors after
390: the first beam splitter detect the photon in the upper or lower path
391: with equal probability. Once the photon is measured, though, it is in
392: that detector's path with probability $1$. And so it enters the beam
393: splitter again in the same state as when it first entered. Thus,
394: the measurement outcome after the second beam splitter will have
395: the same uncertainty as after the first: the detectors still report
396: ``above'' or ``below'' with equal probability. The resulting sequence
397: of detector outcomes after many circuits of the feedback loop is simply
398: a random sequence. Call this \emph{measurement protocol I}.
399:
400: Now alter the experiment slightly by activating the detectors only after
401: every other circuit of the feedback loop. In this set-up, call it
402: \emph{protocol II}, the photon enters the first beam splitter, passes an
403: inactive detector and interferes with itself when it returns to the beam
404: splitter. This,
405: as we will confirm, leads to destructive interference of one path after
406: the beam splitter. The photon is thus in the same path after the second
407: visit to the beam splitter as it was on the first. The now-active detector
408: therefore reports with probability $1$ that the photon is in the upper path,
409: if the photon was initially in the upper path. If it was initially in the
410: lower path, then the detector reports that it is in the upper path with
411: probability $0$. The resulting sequence of path detections is a very
412: predictable sequence, compared to the random sequence from protocol I.
413: Note that both protocols are complete measurements.
414:
415: We now construct a complete quantum generator for the iterated-beam splitter.
416: The output alphabet consists of two symbols denoting detection ``above'' or
417: ``below'': $\Abet = \{0,1\}$. There are two internal states ``above'' and
418: ``below'', each associated with one of the two system eigenstates:
419: $Q = \{A, B\}$. The transition matrices are $T(0) = U_H P(0)$ and
420: $T(1) = U_H P(1)$. We assume that the quantum process has been operating
421: for some time and so take
422: $\asympt = |Q|^{-1} \sum_{i\in Q}\ket{\phi_i}\bra{\phi_i}$.
423:
424: One can readily verify that this representation of the iterated beam
425: splitter is a DQG. And so we can determine its classical equivalent
426: generator has transition matrices
427: $
428: T(0) = \frac{1}{2}
429: \left(
430: \begin{smallmatrix}
431: 1 & 0 \\
432: 1 & 0
433: \end{smallmatrix}
434: \right)
435: ~\mathrm{and}~
436: T(1) = \frac{1}{2}
437: \left(
438: \begin{smallmatrix}
439: 0 & 1 \\
440: 0 & 1
441: \end{smallmatrix}
442: \right) $.
443: The sequences it generates for protocol I are described by the
444: uniform distribution at all lengths: $\Prob(s^L) = 2^{-L}$.
445:
446: Note, however, that the probability distribution of the sequences for the
447: classical generator under protocol II is still the uniform distribution
448: for all lengths $L$. This could not be more different from the
449: behavior of the (quantum) iterated beam splitter under protocol II.
450: The classical generator is simply unable to capture the interference
451: effects present in this case. A second classical generator must be
452: constructed from the quantum generator's transition matrices for protocol
453: II. One finds
454: $
455: T(0) = \tfrac{1}{2}
456: \left(
457: \begin{smallmatrix}
458: 1 & 0 \\
459: 0 & 0
460: \end{smallmatrix}
461: \right)
462: ~\mathrm{and}~
463: T(1) = \tfrac{1}{2}
464: \left(
465: \begin{smallmatrix}
466: 0 & 0 \\
467: 0 & 1
468: \end{smallmatrix}
469: \right) $.
470: Starting the photon in the upper path again, for protocol II one finds
471: $\Prob(00\ldots) = \Prob(11\ldots) = 1/2$ and all other words have
472: zero probability.
473:
474: Table \ref{tab:info} gives the informational quantities for the iterated
475: beam splitter under the protocols. Under protocol I it is maximally random
476: ($\hmu = 1$), as expected. It also, according to $\EE = 0$, does not
477: store any historical information and the observer comes to know this
478: immediately ($\TI = 0$). In stark contrast, under protocol II, the iterated
479: beam splitter is quite predictable ($\hmu = 0$) and stores one bit of
480: information ($\EE = 1$)---whether the measurement sequence is $000\ldots$
481: or $111\ldots$. Learning which requires extracting one bit ($\TI = 1$)
482: of information from the measurements.
483:
484: Note that the internal-state (von Neumann) entropy $S_q = 1$ under both
485: protocols: there are two equally likely states in both cases. It simply
486: reflects the single qubit in the iterated beam splitter, not whether that
487: qubit is useful in supporting intrinsic computation.
488:
489: \begin{table}[tbp]
490: \begin{tabular}{|c||c|c||c|c|}
491: \hline
492: Quantum & \multicolumn{2}{c||}{Iterated Beam} & \multicolumn{2}{c|}{Spin-1} \\
493: Dynamical System & \multicolumn{2}{c||}{Splitter} & \multicolumn{2}{c|}{Particle } \\
494: \hline
495: Protocol & ~~I~~ & II & I & II \\
496: \hline
497: \hline
498: $h_\mu$ [\emph{bits/measurement}] & 1 & 0 & 0.666 & 0.666 \\
499: $S_q$ [\emph{bits}] & 1 & 1 & 1.585 & 1.585 \\
500: $\EE$ [\emph{bits}] & 0 & 1 & 0.252 & 0.902 \\
501: $\TI$ [\emph{bits$\times$measurement}] & 0 & 1 & 0.252 & 3.03 \\
502: \hline
503: \end{tabular}
504: \caption{Information storage and generation for example quantum processes:
505: entropy rate $h_\mu$, internal-state entropy $S_q$, excess entropy $\EE$,
506: and transient information $\TI$.
507: }
508: \label{tab:info}
509: \end{table}
510:
511: Now consider a second, more complex example: A spin-$1$ particle subject to
512: a magnetic field that rotates its spin. The state evolution can be described
513: by the unitary matrix
514: $
515: U = \left(
516: \begin{smallmatrix}
517: 1/\sqrt{2} & 1/\sqrt{2} & 0
518: \\ 0 & 0 & -1
519: \\ -1/\sqrt{2} & 1/\sqrt{2} & 0
520: \end{smallmatrix}
521: \right)
522: $,
523: which is a rotation in $\mathbb{R}^3$ around the y-axis by angle
524: $\tfrac{\pi}{4}$ followed by a rotation around the x-axis by $\tfrac{\pi}{2}$.
525:
526: Using a suitable representation of the spin operators $J_i$ \cite[p.199]{peres},
527: such as:
528: $
529: J_x = \left(
530: \begin{smallmatrix}
531: 0 & 0 & 0 \\
532: 0 & 0 & i \\
533: 0 & -i & 0
534: \end{smallmatrix}
535: \right)
536: $,
537: $
538: J_y = \left(
539: \begin{smallmatrix}
540: 0 & 0 & i \\
541: 0 & 0 & 0 \\
542: -i & 0 & 0
543: \end{smallmatrix}
544: \right)
545: $, and
546: $
547: J_z = \left(
548: \begin{smallmatrix}
549: 0 & i & 0 \\
550: -i & 0 & 0 \\
551: 0 & 0 & 0
552: \end{smallmatrix}
553: \right) ,
554: $
555: the relation $P_i = 1 - J_i^2$
556: defines a one-to-one correspondence between the projector $P_i$ and the
557: square of the spin component along the $i$-axis. The resulting measurement
558: poses the yes-no question, Is the square of the spin component along
559: the $i$-axis zero?
560:
561: Define two protocols, this time differing in the projection operators. First,
562: consider measuring $J_y^2$; call this \emph{protocol I}. Then $U$ and the
563: projection operators
564: $P(0) = \ket{100}\bra{100} + \ket{001}\bra{001}$ and $P(1) = \ket{010}\bra{010}$
565: define a quantum generator.
566:
567: The stochastic language produced by this process is the so-called
568: \emph{Golden-Mean Process} language \cite{crutchfield:03}. It is
569: defined by the set of \emph{irreducible forbidden words}
570: $\mathcal{F} = \{00\}$. That is, all measurement sequences occur,
571: except for those with consecutive $0$s. For
572: the spin-$1$ particle this means that the spin component along the
573: y-axis never equals 0 twice in a row. We call this
574: \emph{short-range correlation} since there is a correlation between a
575: measurement outcome at time $t$ and the immediately preceding one
576: at time $t-1$. If the outcome is $0$, the next outcome will be
577: $1$ with certainty. If the outcome is $1$, the next measurement is
578: maximally uncertain: outcomes $0$ and $1$ occur with equal probability.
579:
580: Second, consider measuring the observable $J_x^2$; call this
581: \emph{protocol II}. Then $U$ and $P(0) = \ket{100}\bra{100}$ and
582: $P(1) = \ket{010}\bra{010} + \ket{001}\bra{001}$ define a quantum finite-state
583: generator. The stochastic language generated is the \emph{Even Process}
584: language \cite{crutchfield:03}. It is characterized by an infinite set
585: of irreducible forbidden words $\mathcal{F} = \{01^{2k-1}0\}, k=1,2,3,...$.
586: That is, if the spin component equals $0$ along the $x$-axis, it will be zero
587: an \emph{arbitrary large}, even number of consecutive measurements before
588: being observed to be nonzero.
589:
590: Table \ref{tab:info} gives the intrinsic-computation analysis
591: of the spin-$1$ system. Comparing it to the iterated beam
592: splitter, it's clear that the spin-$1$ system is richer---a
593: process that does not neatly fall into one or the other extreme of
594: exactly predictable and utterly unpredictable. In fact, under
595: protocols I and II it appears equally unpredictable: $\hmu \approx
596: 0.333$. As before the von Neumann entropy is also the same
597: under both protocols. The amount of information that the spin-$1$
598: system communicates from the past to the future is nonzero,
599: with the amount under protocol I ($\EE \approx 0.252$) being
600: less than under protocol II ($\EE \approx 0.902$). These accord
601: with our observation that in the latter case there is a kind
602: of infinite-range temporal correlation. Not too much information
603: ($\TI \approx 0.252$) must be extracted by the observer under protocol
604: I in order to see the relatively little memory stored in this
605: process. Interestingly, however, under protocol II this is
606: markedly larger ($\TI \approx 3.03$), indicating again that
607: the observer must extract more information to see just how
608: this process is monitoring ``evenness''.
609:
610: We close by looking at a repeatedly measured quantum system in the context
611: of recent developments in quantum computation. Quantum control theory---a
612: paradigm of repeated classical-state measurement and feedback control---has
613: only recently been implemented as a means to drive quantum systems
614: toward a desired state or dynamic \cite{geremia:04}. Current
615: implementations are finite-dimensional and aim at the control of a known
616: quantum state. Therefore, our formalism can be used to characterize the
617: intrinsic computation of these quantum systems. In fact, it can be used on
618: any quantum system observed over time, whether its exact state is known or
619: not. If it is known and the applied measurement protocol generates a
620: deterministic quantum process the above measures of intrinsic computation
621: can be calculated exactly using its quantum generator representation. Generally,
622: though, the intrinsic quantum computation of an unknown quantum state can be
623: measured experimentally by recording a time series of measurement outcomes
624: and computing the probability distribution. In fact, from the probability
625: distribution of a discrete time series of measurements one can compute the
626: intrinsic computation of both open and closed quantum systems.
627:
628: We introduced several new information-theoretic quantities that
629: reflect a quantum process's intrinsic computation: its information
630: production rate, how much memory it apparently stores in generating
631: information, and how hard it is for an observer to synchronize.
632: We discussed how several of these informational quantities are related
633: to existing notions of entropy in quantum theory. The contrast
634: allowed us to demonstrate how much more they tell one about the
635: intrinsic computation supported by quantum processes and to
636: highlight the crucial role of measurement. For example, simply
637: knowing that a quantum process is built out of some number of qubits
638: only gives an upper bound on the possible information processing.
639: It does not reflect how the quantum system actually uses the qubits
640: to process information nor how much of its information
641: processing can be observed. For these, one needs the quantum
642: $\hmu$, $\EE$, and $\TI$.
643:
644:
645: % **********************************************************************
646:
647: UCD and the Santa Fe Institute supported this work via the Network
648: Dynamics Program funded by Intel Corporation. The Wenner-Gren Foundations,
649: Stockholm, Sweden, provided KW's postdoctoral fellowship.
650:
651: % **************************** REFERENCES ****************************
652:
653: \vspace{-.25in}
654:
655: \bibliography{ref}
656:
657: \end{document}
658: