A Complete Characterization of Unitary Quantum Space

A Complete Characterization of Unitary Quantum Space∗†‡ Bill Fefferman1 and Cedric Yen-Yu Lin2 1 2 Electrical Engineering and Computer Science, University of California, Berkeley and Joint Center for Quantum Information and Computer Science (QuICS), University of Maryland, and NIST, Gaithersburg, MD, USA Joint Center for Quantum Information and Computer Science (QuICS), University of Maryland, College Park, MD, USA Abstract Motivated by understanding the power of quantum computation with restricted number of qubits, we give two complete characterizations of unitary quantum space bounded computation. First we show that approximating an element of the inverse of a well-conditioned efficiently encoded 2k(n) × 2k(n) matrix is complete for the class of problems solvable by quantum circuits acting on O(k(n)) qubits with all measurements at the end of the computation. Similarly, estimating the minimum eigenvalue of an efficiently encoded Hermitian 2k(n) × 2k(n) matrix is also complete for this class. In the logspace case, our results improve on previous results of Ta-Shma [30] by giving new space-efficient quantum algorithms that avoid intermediate measurements, as well as showing matching hardness results. Additionally, as a consequence we show that PreciseQMA, the version of QMA with exponentially small completeness-soundess gap, is equal to PSPACE. Thus, the problem of estimating the minimum eigenvalue of a local Hamiltonian to inverse exponential precision is PSPACE-complete, which we show holds even in the frustration-free case. Finally, we can use this characterization to give a provable setting in which the ability to prepare the ground state of a local Hamiltonian is more powerful than the ability to prepare PEPS states. Interestingly, by suitably changing the parameterization of either of these problems we can completely characterize the power of quantum computation with simultaneously bounded time and space. 1998 ACM Subject Classification F.1.3 Complexity Measures and Classes, F.2 Analysis of Algorithms and Problem Complexity Keywords and phrases Quantum complexity, space complexity, complete problems, QMA Digital Object Identifier 10.4230/LIPIcs.ITCS.2018.4 1 Introduction How powerful is quantum computation with a restricted number of qubits? In this work we will study unitary quantum space-bounded classes - those problems solvable using a given amount of (quantum and classical) space, with all quantum measurements performed at the ∗ This work was supported by the Department of Defense. A full version of the paper is available at https://arxiv.org/abs/1604.01384 ‡ This work is a contribution of the National Institute of Standards and Technology and is not subject to U.S. copyright. † © Bill Fefferman and Cedric Yen-Yu Lin; licensed under Creative Commons License CC-BY 9th Innovations in Theoretical Computer Science Conference (ITCS 2018). Editor: Anna R. Karlin; Article No. 4; pp. 4:1–4:21 Leibniz International Proceedings in Informatics Schloss Dagstuhl – Leibniz-Zentrum für Informatik, Dagstuhl Publishing, Germany 4:2 A Complete Characterization of Unitary Quantum Space end of the computation. We give two sets of complete problems for these classes; to the best of our knowledge, these are the first natural complete problems proposed for quantum space-bounded classes. The first problem we consider, the k(n)-Well-conditioned Matrix Inversion problem, is a well-conditioned version of the ubiquitous matrix inversion problem. The second problem we consider, the k(n)-Minimum Eigenvalue problem, asks us to compute the minimum eigenvalue of a Hermitian matrix to high precision – in the context of quantum complexity, this is a natural generalization of the familiar local Hamiltonian problem [23]. Interestingly enough, the first (resp. second) problem is the space-bounded variant of a BQP-complete [18] (resp. QMA-complete [23]) problem; their complexities coincide in the space-bounded regime. For the sake of readability, we defer precise definitions of these problems and statements of our results until Sections 3 and 4. We now proceed to give some justification for the importance of our results. In the following discussion, BQU SPACE[k(n)] refers to the class of problems solvable with bounded error by a quantum algorithm running in O(k(n)); the subscript U indicates that the algorithm is unitary, i.e. employs no intermediate measurements. 1.1 Background and Motivation The Matrix Inversion problem is of central importance in computational complexity theory. Matrix inversion is known to be complete for DET, the class of functions as hard as computing the determinant of an integer matrix, which can be solved in classical O(log2 (n)) space [5, 12]. It is a major open problem to determine if Matrix Inversion can be solved in classical logarithmic space, which would imply L = NL = DET. Recently, Ta-Shma [30], building on work of Harrow, Hassidim, and Lloyd [18], showed that a well-conditioned n × n matrix can be inverted (up to 1/ poly(n) error) by a quantum O(log n) space algorithm using intermediate measurements. Similarly, Ta-Shma also gives an algorithm for computing eigenvalues of a Hermitian matrix with similar space. These algorithms achieve a quadratic advantage in space over the best known classical algorithms, which require Ω(log2 n) space. This is the maximum quantum advantage possible, since Watrous has shown BQSPACE[k(n)] ⊆ SPACE[O(k(n)2 )] [35, 36] even for quantum algorithms with intermediate measurements. Our completeness result for matrix inversion, along with observing our algorithm for matrix inversion (Theorem 14) actually gives a high-precision approximation, gives the following corollary in the logspace case (see Remark 2.3). I Corollary 1. The problem of approximating, to constant precision, an entry of the inverse of an n × n positive semidefinite matrix with condition number at most poly(n) is BQU Lcomplete under L-reductions, where BQU L is the set of problems solvable in unitary quantum logspace. This problem remains in BQU L even if 1/ poly(n) precision is required. Similarly, restricting Thereom 4 to the logspace case gives the following corollary. I Corollary 2. The problem of approximating, to 1/ poly(n) precision, the minimum eigenvalue of an n × n positive semidefinite matrix is BQU L-complete under L-reductions. These corollaries improve upon Ta-Shma’s results [30] in two ways. First, our algorithms solve these problems without needing intermediate measurements. Unlike in time complexity, where the “Principle of safe storage” gives a time-efficient procedure to defer intermediate measurements, these methods may incur an exponential blow-up in space. B. Fefferman and C. Y. -Y. Lin 4:3 One might wonder why we care so much about avoiding intermediate measurements. The main reason is that removing intermediate measurements from the computation allows us to give matching hardness results, showing the optimality of our algorit (...truncated)