1: %=============================================================================%
2: \chapter{Preliminaries} \label{ch:defs}
3:
4: In this chapter we provide formal definitions of the quantum interactive proof
5: system and quantum refereed game models of computation as well as several
6: collections of complexity classes based upon these models.
7: We then summarize what is currently known of these models and state the
8: contributions of this thesis.
9:
10: %=============================================================================%
11: \section{Formalizations of Quantum Interaction}
12: \label{sec:defs:formalizations}
13:
14: Quantum interactive proof systems were introduced by Watrous in 1999
15: \cite{Watrous03} and it is that formalization of the model that we reproduce
16: here.
17: Although quantum refereed games had not yet been considered prior to the work of
18: the present thesis, the formalization of that model is a straightforward
19: extension of the quantum interactive proof system model.
20:
21: Quantum interactions consist of a \emph{verifier} and one or more \emph{provers}.
22: For any function $r : \mathbb{N} \to \mathbb{N} \setminus \set{0}$, an
23: \emph{$r$-round prover} $P$ is a mapping on input strings $x \in \set{0,1}^*$
24: where
25: $$P(x) = (P_1,\dots,P_{r(|x|)})$$
26: is an $r(|x|)$-tuple of quantum circuits, each of which acts upon the same
27: number of qubits.
28: No restrictions are placed on the complexity of the prover's circuits, which
29: captures the notion that the prover has unlimited computational power---each of
30: the prover's circuits can be viewed as an arbitrary unitary operation on its
31: input qubits.
32:
33: Similarly, an \emph{$r$-round verifier} $V$ is a mapping on input strings
34: $x \in \set{0,1}^*$ where
35: $$V(x) = (V_0,\dots,V_{r(|x|)})$$
36: is an $(r(|x|)+1)$-tuple of quantum circuits, each of which acts upon the same
37: number of qubits.
38: Unlike a prover, however, we require that the verifier's circuits be generated
39: by a polynomial-time Turing machine on input $x$.
40: This uniformity constraint captures the notion that the verifier's computational
41: power is limited and implicitly restricts the quantity $r(|x|)$ so that
42: $r \in \poly$ as one might expect.
43: We often abbreviate $r(|x|)$ to $r$ for easier readability.
44:
45: %=============================================================================%
46: \subsection{Quantum Interactive Proof Systems}
47: \label{subsec:defs:formalizations:QIPs}
48:
49: A \emph{quantum interactive proof system} has a verifier $V$ and a prover $P$.
50: The qubits upon which each of the circuits in the prover's $r$-tuple acts are
51: partitioned into two sets: one set of qubits is private to the prover and the
52: other is shared with the verifier.
53: These shared qubits act as a quantum channel between the verifier and the
54: prover.
55: The Hilbert spaces corresponding to the private and shared qubits of the prover
56: are denoted $\hilb{P}$ and $\hilb{M}$ respectively.
57:
58: Similarly, the qubits upon which each of the circuits in the verifier's $(r+1)$-
59: tuple acts are partitioned into two sets: one set of qubits is private to the
60: verifier and the other is shared with the prover.
61: The Hilbert spaces corresponding to the private and shared qubits of the
62: verifier are denoted $\hilb{V}$ and $\hilb{M}$ respectively.
63:
64: For any input string $x \in \set{0,1}^*$ we create a composite circuit
65: $(V,P)(x)$ by concatenating the circuits
66: $$V_0, P_1, V_1, \dots, V_{r-1}, P_r, V_r$$
67: in sequence, each circuit acting only upon the sets of qubits stipulated
68: previously.
69: Such a circuit is illustrated in Figure \ref{fig:qip} for the case $r=2$.
70: \begin{figure}
71: \begin{center}
72: \input{QIP.latex}
73: \end{center}
74: \caption{Quantum circuit for a two-round quantum interactive proof}
75: \label{fig:qip}
76: \end{figure}
77: The Hilbert space upon which $(V,P)(x)$ acts is denoted
78: $$\hilb{S} = \hilb{P} \otimes \hilb{M} \otimes \hilb{V}.$$
79: The quantum interactive proof system is implemented by applying the circuit
80: $(V,P)(x)$ to the initial pure state $\ket{0_\hilb{S}} \in \hilb{S}$.
81: Hence, the pure state of the system after $(V,P)(x)$ is applied is precisely
82: $$V_r P_r V_{r-1} \cdots V_1 P_1 V_0 \ket{0_\hilb{S}} \in \hilb{S}.$$
83: Acceptance is dictated by a standard measurement of the output qubit of
84: $\hilb{S}$, which is assumed to belong to the verifier.
85: In particular, $(V,P)(x)$ accepts $x$ with probability
86: $$\Norm{ \Pi_\mathrm{accept} V_r P_r V_{r-1} \cdots V_1 P_1 V_0
87: \ket{0_\hilb{S}} }^2$$
88: and rejects $x$ with probability
89: $$\Norm{ \Pi_\mathrm{reject} V_r P_r V_{r-1} \cdots V_1 P_1 V_0
90: \ket{0_\hilb{S}} }^2$$
91: where $\Pi_\mathrm{accept}, \Pi_\mathrm{reject} \in \mapset{Pos}{\hilb{S}}$ are
92: as defined in Section \ref{subsec:intro:quantum:measurement}.
93:
94: We now define a collection of complexity classes based upon quantum interactive
95: proof systems.
96: For any polynomial-time computable functions $c,s : \mathbb{N} \to [0,1]$, the
97: class $\cls{QIP}(c,s)$\label{QIP} consists of all languages
98: $L \subseteq \set{0,1}^*$ for which there exists an $r$-round verifier $V$ that
99: satisfies the following completeness and soundness conditions:
100: \begin{description}
101:
102: \item[Completeness.]
103: There exists an $r$-round prover $P$ such that, for all $x \in L$, $(V,P)(x)$
104: rejects $x$ with probability at most $c(|x|)$.
105: In other words, there exist unitary matrices
106: $P_1,\dots,P_r \in \mapset{U}{\hilb{P} \otimes \hilb{M}}$ such that
107: $$\Norm{ \Pi_\mathrm{reject} V_r P_r V_{r-1} \cdots V_1 P_1 V_0
108: \ket{0_\hilb{S}} }^2 \leq c(|x|).$$
109:
110: \item[Soundness.]
111: \begin{sloppypar}
112: For all $r$-round provers $P$ and all $x \not \in L$, $(V,P)(x)$ accepts $x$
113: with probability at most $s(|x|)$.
114: In other words, for all unitary matrices
115: $P_1,\dots,P_r \in \mapset{U}{\hilb{P} \otimes \hilb{M}}$, we have
116: $$\Norm{ \Pi_\mathrm{accept} V_r P_r V_{r-1} \cdots V_1 P_1 V_0
117: \ket{0_\hilb{S}} }^2 \leq s(|x|).$$
118: \end{sloppypar}
119:
120: \end{description}
121: As with classical interactive proof systems, the functions $c$ and $s$ are
122: called the \emph{completeness error} and \emph{soundness error} respectively.
123: We often abbreviate $c(|x|)$ to $c$ and $s(|x|)$ to $s$ for easier readability.
124:
125: %=============================================================================%
126: \subsection{Quantum Refereed Games}
127:
128: A \emph{quantum refereed game} has a verifier $V$ and two provers $Y$ and $N$.
129: As with quantum interactive proof systems, the qubits upon which each of the
130: circuits in the provers' $r$-tuples acts are partitioned into two sets: one set
131: of qubits is private to that prover and the other is shared with the verifier.
132: These shared qubits act as a quantum channel between the verifier and that
133: prover.
134:
135: For clarity, $Y$ is called a \emph{yes-prover} and $N$ is called a
136: \emph{no-prover}.
137: This distinction is purely a notational convenience: the Hilbert spaces
138: corresponding to the private and shared qubits of a yes-prover are denoted
139: $\hilb{Y}$ and $\hilb{M}_Y$ respectively, whereas the Hilbert spaces
140: corresponding to the private and shared qubits of a no-prover are denoted
141: $\hilb{N}$ and $\hilb{M}_N$ respectively.
142:
143: In a quantum refereed game, the qubits upon which each of the circuits in the
144: verifier's $(r+1)$-tuple acts are partitioned into three sets: one set of
145: qubits, with corresponding Hilbert space $\hilb{V}$, is private to the verifier
146: and the two remaining sets have corresponding Hilbert spaces $\hilb{M}_Y$ and
147: $\hilb{M}_N$ and are shared with the yes- and no-provers respectively.
148:
149: For any input string $x \in \set{0,1}^*$ we create a composite circuit
150: $(V,Y,N)(x)$ by concatenating the circuits
151: $$V_0, N_1, Y_1, V_1, \dots, V_{r-1}, N_r, Y_r, V_r$$
152: in sequence, each circuit acting only upon the sets of qubits stipulated
153: previously.
154: Such a circuit is illustrated in Figure \ref{fig:qrg} for the case $r=2$.
155: \begin{figure}
156: \begin{center}
157: \input{QRG.latex}
158: \end{center}
159: \caption{Quantum circuit for a two-round quantum refereed game}
160: \label{fig:qrg}
161: \end{figure}
162: The Hilbert space upon which $(V,Y,N)(x)$ acts is denoted
163: $$\hilb{S} = \hilb{Y} \otimes \hilb{M}_Y \otimes \hilb{V} \otimes \hilb{M}_N
164: \otimes \hilb{N}.$$
165: Although $\hilb{S}$ also denotes the Hilbert space for quantum interactive proof
166: systems, any ambiguity is always resolved by context.
167: The quantum refereed game is implemented by applying the circuit
168: $(V,Y,N)(x)$ to the initial pure state $\ket{0_\hilb{S}} \in \hilb{S}$.
169: Hence, the pure state of the system after $(V,Y,N)(x)$ is applied is precisely
170: $$V_r Y_r N_r V_{r-1} \cdots V_1 Y_1 N_1 V_0 \ket{0_\hilb{S}} \in \hilb{S}.$$
171: As with quantum interactive proof systems, acceptance is dictated by a standard
172: measurement of the output qubit of $\hilb{S}$, which is assumed to belong to the
173: verifier.
174: In particular, $(V,Y,N)(x)$ accepts $x$ with probability
175: $$\Norm{ \Pi_\mathrm{accept} V_r Y_r N_r V_{r-1}
176: \cdots V_1 Y_1 N_1 V_0 \ket{0_\hilb{S}} }^2$$
177: and rejects $x$ with probability
178: $$\Norm{ \Pi_\mathrm{reject} V_r Y_r N_r V_{r-1}
179: \cdots V_1 Y_1 N_1 V_0 \ket{0_\hilb{S}} }^2.$$
180:
181: We now define a collection of complexity classes based upon quantum refereed
182: games.
183: For any polynomial-time computable functions $c,s : \mathbb{N} \to [0,1]$, the
184: class $\cls{QRG}(c,s)$\label{QRG} consists of all languages
185: $L \subseteq \set{0,1}^*$ for which there exists an $r$-round verifier $V$ that
186: satisfies the following completeness and soundness conditions:
187: \begin{description}
188:
189: \item[Completeness.]
190: There exists an $r$-round yes-prover $Y$ such that, for all
191: $r$-round no-provers $N$ and all $x \in L$, $(V,Y,N)(x)$ rejects $x$ with
192: probability at most $c(|x|)$.
193: In other words, there exist unitary matrices
194: $Y_1,\dots,Y_r \in \mapset{U}{\hilb{Y} \otimes \hilb{M}_Y}$ such that
195: $$\Norm{ \Pi_\mathrm{reject} V_r Y_r N_r V_{r-1}
196: \cdots V_1 Y_1 N_1 V_0 \ket{0_\hilb{S}} }^2 \leq c(|x|)$$
197: for all unitary matrices
198: $N_1,\dots,N_r \in \mapset{U}{\hilb{M}_N \otimes \hilb{N}}$.
199:
200: \item[Soundness.]
201: There exists an $r$-round no-prover $N$ such that, for all
202: $r$-round yes-provers $Y$ and all $x \not \in L$, $(V,Y,N)(x)$ accepts $x$
203: with probability at most $s(|x|)$.
204: In other words, there exist unitary matrices
205: $N_1,\dots,N_r \in \mapset{U}{\hilb{M}_N \otimes \hilb{N}}$ such that
206: $$\Norm{ \Pi_\mathrm{accept} V_r Y_r N_r V_{r-1}
207: \cdots V_1 Y_1 N_1 V_0 \ket{0_\hilb{S}} }^2 \leq s(|x|)$$
208: for all unitary matrices
209: $Y_1,\dots,Y_r \in \mapset{U}{\hilb{Y} \otimes \hilb{M}_Y}$.
210:
211: \end{description}
212: As with quantum interactive proof systems, the functions $c$ and $s$ are the
213: completeness error and soundness error respectively.
214:
215: %=============================================================================%
216: \subsection{Short Quantum Games}
217:
218: For most of this thesis we restrict our attention to a specific class of quantum
219: refereed games that we call \emph{short quantum games}.
220: A short quantum game has a two-round verifier $V$, a one-round yes-prover $Y$,
221: and a one-round no-prover $N$.
222: In these games, the composite circuit $(V,Y,N)'(x)$ is created by concatenating
223: the circuits
224: $$V_0, Y_1, V_1, N_1, V_2$$
225: in sequence.
226: In other words, short quantum games are one-round quantum games in which the
227: verifier is permitted to process the yes-prover's response before sending a
228: message to the no-prover.
229: Figure \ref{fig:sqg} illustrates the quantum circuit for a short quantum game.
230:
231: \begin{figure}
232: \begin{center}
233: \input{SQG.latex}
234: \end{center}
235: \caption{Quantum circuit for a short quantum game} \label{fig:sqg}
236: \end{figure}
237:
238: For polynomial-time computable $c,s : \mathbb{N} \to [0,1]$, we define the
239: collection $\cls{SQG}(c,s)$\label{SQG} of complexity classes by restricting the
240: definition of $\cls{QRG}(c,s)$ to short quantum games.
241: We also define the collection $\cls{SQG}_*(c,s)$\label{SQGstar} of complexity
242: classes by further restricting the definition of $\cls{SQG}(c,s)$ to short
243: quantum games in which the verifier cannot send a message to the yes-prover
244: (in other words, the verifier's first circuit is empty, so that $V_0=I$).
245:
246: %=============================================================================%
247: \section{Remarks and Contributions}
248: \label{sec:defs:remarks}
249:
250: This section contains a summary of what is currently known of quantum
251: interactive proof systems and quantum refereed games.
252: The contributions of this thesis are stated in Section
253: \ref{subsec:defs:contributions}.
254:
255: %=============================================================================%
256: \subsection{Known Results}
257: \label{subsec:defs:remarks:results}
258:
259: Several inclusions follow immediately from the definitions in Section
260: \ref{sec:defs:formalizations}.
261: For example, because a quantum interactive proof system is just a quantum
262: refereed game without a no-prover, it is clear that
263: $$\cls{QIP}(c,s) \subseteq \cls{QRG}(c,s).$$
264: Furthermore, we have
265: $$\cls{SQG}_*(c,s) \subseteq \cls{SQG}(c,s) \subseteq \cls{QRG}(c,s),$$
266: as any shorter quantum refereed game can be simulated by a longer quantum
267: refereed game.
268:
269: Also, it is clear that any classical interaction can be simulated by a
270: quantum interaction in which the verifier simply measures every message he
271: receives from any prover (thus collapsing each message to a completely classical
272: state) and otherwise behaves in a classical manner.
273: In other words, we have $\cls{IP}(c,s) \subseteq \cls{QIP}(c,s)$,
274: $\cls{RG}(c,s) \subseteq \cls{QRG}(c,s)$, and
275: $\cls{RG}_1(c,s) \subseteq \cls{SQG}(c,s)$.
276: It is instructive to note that the relation
277: $\cls{RG}_1(c,s) \subseteq \cls{SQG}_*(c,s)$ is not immediately seen to hold, as
278: $\cls{SQG}_*(c,s)$ does not permit games in which the verifier sends a message
279: to the yes-prover.
280:
281: The method of sequential repetition used in Section
282: \ref{subsec:intro:complexity:results} to demonstrate the robustness of classical
283: interaction can be applied without complication in the quantum setting.
284: That is, for every $\varepsilon \in 2^{-\poly}$ and every reasonable
285: $c,s : \mathbb{N} \to [0,1]$ we have
286: $$
287: \cls{QIP}(c,s) \subseteq \cls{QIP}(\varepsilon,\varepsilon)
288: \qquad \textrm{and} \qquad
289: \cls{QRG}(c,s) \subseteq \cls{QRG}(\varepsilon,\varepsilon).
290: $$
291: Similar to the classical case, quantum interactive proof systems with reasonable
292: error can be assumed to have zero completeness error \cite{KitaevW00}, yielding
293: the inclusion
294: $$\cls{QIP}(c,s) \subseteq \cls{QIP}(0,\varepsilon).$$
295: In light of this robustness, we define the following shorthand notations:
296: \begin{itemize}
297:
298: \item
299: $\cls{QIP}$ is the complexity class of all languages
300: $L \subseteq \set{0,1}^*$ such that
301: $L \in \cls{QIP}(0,\varepsilon)$ for every
302: $\varepsilon \in 2^{-\poly}$.
303:
304: \item
305: $\cls{QRG}$ is the complexity class of all languages
306: $L \subseteq \set{0,1}^*$ such that
307: $L \in \cls{QRG}(\varepsilon,\varepsilon)$ for every
308: $\varepsilon \in 2^{-\poly}$.
309:
310: \item
311: $\cls{SQG}$ is the complexity class of all languages
312: $L \subseteq \set{0,1}^*$ such that
313: $L \in \cls{SQG}(\varepsilon,\varepsilon)$ for every
314: $\varepsilon \in 2^{-\poly}$.
315:
316: \item
317: $\cls{SQG}_*$ is the complexity class of all languages
318: $L \subseteq \set{0,1}^*$ such that
319: $L \in \cls{SQG}_*(\varepsilon,\varepsilon)$ for every
320: $\varepsilon \in 2^{-\poly}$.
321:
322: \end{itemize}
323:
324: The well-known characterization $\cls{IP} = \cls{PSPACE}$ mentioned in Section
325: \ref{subsec:intro:complexity:results} implies
326: $$\cls{PSPACE} \subseteq \cls{QIP}.$$
327: Watrous gave a three-message quantum interactive proof system
328: demonstrating this containment \cite{Watrous03}.
329: The existence of a three-message quantum interactive proof system for
330: $\cls{PSPACE}$ contrasts with the classical case, wherein it is widely believed
331: that the verifier and prover must exchange a nonconstant number of messages in
332: order to decide $\cls{PSPACE}$
333: (see Section \ref{subsec:end:open:parallelization}).
334:
335: This result was later extended to show that any quantum interactive proof system
336: can be simulated by a three-message quantum interactive proof system and that
337: these simulations are robust with respect to error \cite{KitaevW00}.
338: This extension gives rise to a natural complete promise problem for $\cls{QIP}$
339: known as \textsc{close-images}, which can in turn be reduced to an
340: exponential-size instance of the semidefinite programming problem
341: \cite{KitaevW00}.
342: As semidefinite programs can be solved in deterministic polynomial time, it
343: follows from the $\cls{QIP}$-completeness of \textsc{close-images} that
344: $$\cls{QIP} \subseteq \cls{EXP}.$$
345: In a later talk on quantum coin-flipping, Kitaev gave a semidefinite program
346: that directly simulates many-message quantum interactive proof systems
347: \cite{Kitaev02}, thus yielding a more direct proof of the containment of
348: $\cls{QIP}$ in $\cls{EXP}$.
349:
350: Although quantum refereed games were not considered prior to this work, it is
351: nonetheless appropriate to mention several facts in this subsection.
352: First, it is clear that the characterizations $\cls{RG} = \cls{EXP}$ and
353: $\cls{RG}_1 = \cls{PSPACE}$ imply
354: $$
355: \cls{EXP} \subseteq \cls{QRG}
356: \qquad \textrm{and} \qquad
357: \cls{PSPACE} \subseteq \cls{SQG}.
358: $$
359:
360: Also, Kitaev's variant \cite{Kitaev02} of the semidefinite program found in
361: Reference \cite{KitaevW00} is easily extended to yield
362: $\cls{QRG} \subseteq \cls{NEXP}$ \cite{Gutoski05}.
363: Due to the symmetric nature of quantum refereed games, it is clear that
364: $\cls{QRG}$ is closed under complement, from which it follows that $\cls{QRG}$
365: is also contained in $\cls{coNEXP}$.
366: In other words, we have
367: $$\cls{QRG} \subseteq \cls{NEXP} \cap \cls{coNEXP}.$$
368: In Section \ref{subsec:end:open:QRGeqEXP} we we discuss possible implications of
369: the curious fact that $\cls{QRG}$ contains $\cls{EXP}$ and is in turn contained
370: in $\cls{NEXP} \cap \cls{coNEXP}$.
371: Chapter \ref{ch:end} also offers a diagram of relationships among the complexity
372: classes considered in this thesis.
373:
374: %=============================================================================%
375: \subsection{Contributions of this Thesis}
376: \label{subsec:defs:contributions}
377:
378: We prove in this thesis the following two relationships among the complexity
379: classes defined in Section \ref{sec:defs:formalizations}:
380: \begin{itemize}
381:
382: \item
383: $\cls{QIP} \subseteq \cls{SQG}_*$ (Chapter \ref{ch:QIPinSQG})
384:
385: \item
386: $\cls{SQG} \subseteq \cls{EXP}$ (Chapter \ref{ch:SQGinEXP})
387:
388: \end{itemize}
389: The first result is proven in Reference \cite{GutoskiW05}, while the second is
390: proven in Reference \cite{Gutoski05}.
391: It is those proofs that we reproduce in Chapters \ref{ch:QIPinSQG} and
392: \ref{ch:SQGinEXP}.
393:
394: Two of the intermediate results that are proven in order to obtain the
395: containment $\cls{QIP} \subseteq \cls{SQG}_*$ might be of independent interest
396: and so we also list them here:
397: \begin{itemize}
398:
399: \item
400: For any two disjoint closed convex sets
401: $\mathcal{A}_0, \mathcal{A}_1 \subseteq \mapset{D}{\hilb{H}}$ of quantum states,
402: there exists a binary POVM such that, for any state
403: $\rho \in \mathcal{A}_0 \cup \mathcal{A}_1$,
404: the POVM will correctly determine whether $\rho \in \mathcal{A}_0$ or
405: $\rho \in \mathcal{A}_1$ with probability proportional to the minimal trace
406: distance between $\mathcal{A}_0$ and $\mathcal{A}_1$.
407:
408: \item
409: $\cls{SQG}(c,s) \subseteq \cls{SQG}(kc,s^k) \cap \cls{SQG}(c^k,ks)$ for
410: any $c,s : \mathbb{N} \to [0,1]$ and any $k \in \mathit{poly}$.
411: A similar containment holds with $\cls{SQG}_*$ in place of $\cls{SQG}$.
412:
413: \end{itemize}
414: The first of these intermediate results is a generalization of the well-known
415: fact that two quantum states can be distinguished with probability proportional
416: to their trace distance.
417: It can also be viewed as a quantitative version of the well-known separation
418: theorems in convex analysis.
419:
420: The second result indicates a partial robustness of short quantum games with
421: respect to error.
422: In particular, it states that the completeness (soundness) error can be made
423: exponentially small at the possible cost of soundness (completeness).
424: Error reduction results seem to be more elusive in quantum interaction than in
425: classical interaction and this result represents a first step toward that end
426: for short quantum games.
427:
428: The containment $\cls{SQG} \subseteq \cls{EXP}$ builds upon the
429: semidefinite program for $\cls{QIP}$ \cite{KitaevW00, Kitaev02}.
430: Hence, we offer a rigorous analysis of that semidefinite program in this thesis
431: as a precursor to our result.
432: This precursor leads to several extensions of the containment
433: $\cls{QIP} \subseteq \cls{EXP}$.
434: For example, we show that semidefinite programming can be used to simulate a
435: quantum interactive proof system in deterministic exponential time even if the
436: verifier's circuits are generated uniformly in \emph{exponential} time, so long
437: as they still act on only a polynomial number of qubits.
438: In particular, the verifier can exchange an exponential number of messages with
439: the prover and can use an exponential number of gates in his quantum circuits
440: without raising the power of the model beyond $\cls{EXP}$.
441:
442: The containment $\cls{QRG} \subseteq \cls{NEXP}$ is obtained by
443: nondeterministically guessing a yes-prover and solving the induced quantum
444: interactive proof system using the aforementioned semidefinite program
445: \cite{Gutoski05}.
446: Most of the extensions of that semidefinite program also apply to this
447: containment concerning quantum refereed games.
448: The only exception is that we still require that the verifier exchange only a
449: polynomial number of rounds of messages with the provers.
450: This strange restriction is brought on by the fact that a polynomial bound on
451: the number of qubits required by the provers in a quantum refereed game is known
452: to hold only if a polynomial number of messages are exchanged
453: (see Section \ref{subsec:4:bounds:proverqubits}).
454:
455: We prove $\cls{SQG} \subseteq \cls{EXP}$ by a repeated use of semidefinite
456: programming in concert with the ellipsoid method for convex feasibility.
457: Once again, many of the aforementioned extensions also apply to this
458: containment.
459: For example, our method can simulate a short quantum game in deterministic
460: exponential time even if the verifier is permitted to exchange an exponential
461: number of messages with the yes-prover, followed by an exponential number of
462: messages with the no-prover.
463:
464:
465:
466:
467:
468:
469:
470:
471: