Simulating Many-Body Systems with a Projective Quantum Eigensolver

TL;DR

This paper introduces the projective quantum eigensolver (PQE), a hybrid quantum-classical algorithm for optimizing wave functions on near-term quantum hardware, demonstrating comparable or superior performance to existing methods on molecular systems.

Contribution

The paper presents PQE and its variant SPQE, novel algorithms that optimize quantum states using residuals instead of gradients, suitable for noisy quantum devices and capable of handling strong correlation effects.

Findings

PQE converges to energies similar to variational methods with fewer resources.

SPQE achieves comparable accuracy to adaptive variational algorithms with the same number of parameters.

SPQE outperforms some classical methods like DMRG on strongly correlated systems.

Abstract

We present a new hybrid quantum-classical algorithm for optimizing unitary coupled-cluster (UCC) wave functions deemed the projective quantum eigensolver (PQE), amenable to near-term noisy quantum hardware. Contrary to variational quantum algorithms, PQE optimizes a trial state using residuals (projections of the Schr\"{o}dinger equation) rather than energy gradients. We show that the residuals may be evaluated by simply measuring two energy expectation values per element. We also introduce a selected variant of PQE (SPQE) that uses an adaptive ansatz built from arbitrary-order particle-hole operators, offering an alternative to gradient-based selection procedures. PQE and SPQE are tested on a set of molecular systems covering both the weak and strong correlation regimes, including hydrogen clusters with 4-10 atoms and the BeH$_2$ molecule. When employing a fixed ansatz, we find that PQE can converge disentangled (factorized) UCC wave functions to essentially identical energies as variational optimization while requiring fewer computational resources. A comparison of SPQE and adaptive variational quantum algorithms shows that - for ans\"{a}tze containing the same number of parameters - the two methods yield results of comparable accuracy. Finally, we show that SPQE performs similar to, and in some cases, better than selected configuration interaction and the density matrix renormalization group on 1-3 dimensional strongly correlated H$_{10}$ systems in terms of energy accuracy for a given number of variational parameters.