Arxiv:2105.01193V1

Improved approximation algorithms for bounded-degree local Hamiltonians Anurag Anshu1, David Gosset2, Karen J. Morenz Korol3, and Mehdi Soleimanifar4 1 Department of EECS & Challenge Institute for Quantum Computation, University of California, Berkeley, USA and Simons Institute for the Theory of Computing, Berkeley, California, USA. 2 Department of Combinatorics and Optimization and Institute for Quantum Computing, University of Waterloo, Canada 3 Department of Chemistry, University of Toronto, Canada and 4 Center for Theoretical Physics, Massachusetts Institute of Technology, USA We consider the task of approximating the ground state energy of two-local quantum Hamiltonians on bounded-degree graphs. Most existing algorithms optimize the energy over the set of product states. Here we describe a family of shallow quantum circuits that can be used to improve the approximation ratio achieved by a given product state. The algorithm takes as input an n-qubit 2 product state |vi with mean energy e0 = hv|H|vi and variance Var = hv|(H − e0) |vi, and outputs a 2 state with an energy that is lower than e0 by an amount proportional to Var /n. In a typical case, we have Var = Ω(n) and the energy improvement is proportional to the number of edges in the graph. When applied to an initial random product state, we recover and generalize the performance guarantees of known algorithms for bounded-occurrence classical constraint satisfaction problems. We extend our results to k-local Hamiltonians and entangled initial states. Quantum computers are capable of efficiently comput- Hamiltonian ing the dynamics of quantum many-body systems [1], and it is anticipated that they can be useful for scien- H = hij (1) i,j E tific applications in physics, materials science and quan- { X}∈ tum chemistry. The extent of the quantum advantage with n = V qubits and nearest-neighbor interactions for other important simulation tasks, such as comput- h that act| | nontrivially only on qubits i, j at ver- ing low temperature properties of quantum systems, is ij tices connected by an edge. We assume{ without} loss still unknown. In this paper we consider the task of ap- of generality that h 1. We are interested in the proximating the ground state energy of local Hamiltoni- ij problem of approximatingk k ≤ the ground energy or small- ans. Here it is natural to expect some improvement over est eigenvalue λ (H) of the Hamiltonian. It will be classical machines which cannot even store the state of min convenient to instead approximate the largest eigenvalue such systems efficiently. Indeed, classical methods such λ (H); this convention matches the one used in clas- as the mean-field or Hartree-Fock approximations do not max sical optimization and is without loss of generality, since capture the entanglement structure present in the true λ (H)= λ ( H). In the worst case, the problem ground state. min − max − of estimating the largest eigenvalue λmax(H)ofEq.(1) to Motivated by small quantum computers that may be within an additive error scaling inverse polynomially with available in the near future, there has been increased in- n is believed to be intractable for quantum or classical terest in devising algorithms that consume few quantum computers [9]. Here we consider the approximation task resources and can be implemented across a wide range where the goal is to compute an estimate e λ (H) ≤ max of hardware platforms. In this vein, heuristic algorithms such that the approximation ratio r e/λmax(H) is as for ground state preparation have been proposed based large as possible. We will also be interested≡ in efficient on variationally minimizing the energy over the output quantum algorithms that prepare states ψ with good arXiv:2105.01193v1 [quant-ph] 3 May 2021 states of shallow (low-depth) quantum circuits [2–4]. Al- approximation ratios. | i though variational algorithms have been rigorously an- Besides describing local interactions encountered in alyzed for specific problems and some limitations are physics, Hamiltonians of the form Eq. (1) can encode no- known [5–8], no general treatment of their efficacy exists. table cost-functions considered in computer science and Characterizing the advantage offered by shallow quantum thus provide a physically motivated extension of the clas- circuits and variational quantum algorithms stands as a sical approximation algorithm setting [10]. For exam- pressing challenge. ple, one may consider an Ising Hamiltonian for which In this paper, we derive rigorous bounds on the per- hij = (I + ZiZj)/2, where Z is the Pauli operator. This formance of shallow quantum circuits in estimating the Hamiltonian is classical—that is, diagonal in the compu- ground state energy of local Hamiltonians. For simplicity, tational basis—and computing its maximum eigenvalue we state our results for a system of qubits with two-local is equivalent to finding the Max-Cut of the graph G, a interactions. In the Supplemental Material, we discuss well-studied classical optimization problem. More gener- extensions of our results to k-local Hamiltonians. ally, two-local quantum Hamiltonians may involve non- To begin, let G = (V, E) be a graph, and consider a commuting terms such as Heisenberg interactions hij = 1/4(I X X Y Y Z Z ) (with Pauli X, Y and Z oper- − i j − i j − i j 2 ators); the resulting optimization problem can be viewed v1 v2 ... vn where each vi is a single- as a quantum analogue of Max-Cut [11]. Quantum ap- qubit| i ⊗ state. | i⊗ ⊗ | i | i proximation algorithms aim to estimate the largest eigen- Theorem 1. Given a product state v , we can efficiently value of such Hamiltonians and have been studied in sev- | i eral previous works. This includes the Heisenberg inter- compute a depth-(d + 1) quantum circuit U such that the state ψ = U v satisfies actions mentioned above [11, 12] and more general set- | i | i tings in which the interaction terms h are restricted to ij Var (H)2 be positive semidefinite [13–15], or traceless [16, 17]. ψ H ψ v H v +Ω v . (2) h | | i ≥ h | | i d2 E Despite considerable interest, the ultimate limits of ef- | | ficient algorithms for quantum approximation algorithms are poorly understood. Approximation ratios approach- This result applies broadly to quantum optimization ing 1 are only known to be achievable for certain special problems, but does not provide any improvement when families of graphs, including lattices or bounded-degree specialized to the classical setting. To see this, note that planar graphs using tensor product of O(1)-qubit states condition (i) is not satisfied in the purely classical case [18] or high degree graphs using tensor products of single- where v is a computational basis state and H is diagonal | i qubit states [13, 18, 19]. In certain cases, one may ascer- in the computational basis. Indeed, we have Varv(H)=0 tain limitations on efficient achievable approximation ra- whenever v is an eigenstate of H. On the other hand, tios from the classical Probabilistically Checkable Proof condition (i)| i is fairly mild in the quantum setting, which (PCP) theorem [20–22], though stronger and more gen- can be seen from the following expression for the vari- eral limitations may follow from the quantum PCP con- ance: jecture if some version of it can be proven [23]. Var (H)= ( v h h v v h v v h v ) . A quantum approximation algorithm typically outputs v h | ij kl| i − h | ij | i · h | kl| i i,j k,l = an estimate of the form v H v where v is a quantum { }∩{X }6 ∅ h | | i | i state computed by the algorithm. A central challenge is Since G is d-regular, the number of terms in the sum to understand the structure of quantum states v that | i is O(d E ). So condition (i) is satisfied if the sum is achieve high approximation ratios in the general case. proportional| | to the number of terms appearing in it. Most existing algorithms are based on tensor products of one- or few-qubit states, while Ref. [12] also considers Simple examples demonstrate that neither of the two states prepared by shallow quantum circuits. In this work conditions alone is enough to even guarantee the exis- tence of a state with approximation ratio better than v we describe conditions under which the performance of | i such algorithms can be improved. We restrict our at- for large regular graphs. Condition (ii) alone is not suf- tention to local Hamiltonians on bounded-degree graphs ficient because it is possible for a product state to have and consider an improvement strategy based on shallow maximal energy λmax(H) (i.e., this occurs for all classi- quantum circuits. cal Hamiltonians). To see that condition (i) is not suf- ficient, one can consider the Max-Cut Hamiltonian on Improvement of product states To this end, suppose (say) an even cycle graph, and let v be an equal su- we are given an n-qubit state v and a Hamiltonian | i | i perposition of two eigenstates of H, one with maximal Eq. (1) defined on a graph G = (V, E) with maximum de- energy E and one with energy E Θ( E ). The re- gree d 2. It will be convenient to assume (without loss | | | |− | | 1/2 sulting state has approximation ratio 1 O( E − ) and of generality)≥ that G is d-regular—we can ensure this by −p | | variance Varv(H) = Ω( E ). Thus condition (i) is satis- possibly adding some local terms hij which are equal to fied, but the approximation| | ratio cannot be improved by zero. We imagine that v may be the output of some ap- an additive constant.

Load more