Symmetry Breaking Slows Convergence of the ADAPT Variational Quantum Eigensolver

Luke W. Bertels, Harper R. Grimsley, Sophia E. Economou, Edwin Barnes, Nicholas J. Mayhall

Research output: Contribution to journalArticlepeer-review

5 Scopus citations

Abstract

Because quantum simulation of molecular systems is expected to provide the strongest advantage over classical computing methods for systems exhibiting strong electron correlation, it is critical that the performance of VQEs be assessed for strongly correlated systems. For classical simulation, strong correlation often results in symmetry breaking of the Hartree-Fock reference, leading to Löwdin's well-known "symmetry dilemma", whereby accuracy in the energy can be increased by breaking spin or spatial symmetries. Here, we explore the impact of symmetry breaking on the performance of ADAPT-VQE using two strongly correlated systems: (i) the "fermionized"anisotropic Heisenberg model, where the anisotropy parameter controls the correlation in the system, and (ii) symmetrically stretched linear H4, where correlation increases with increasing H-H separation. In both of these cases, increasing the level of correlation of the system leads to spontaneous symmetry breaking (parity and S^2, respectively) of the mean-field solutions. We analyze the role that symmetry breaking in the reference states and orbital mappings of the fermionic Hamiltonians have in the compactness and performance of ADAPT-VQE. We observe that improving the energy of the reference states by breaking symmetry has a deleterious effect on ADAPT-VQE by increasing the length of the ansatz necessary for energy convergence and exacerbating the problem of "gradient troughs".

Original languageEnglish
Pages (from-to)6656-6669
Number of pages14
JournalJournal of Chemical Theory and Computation
Volume18
Issue number11
DOIs
StatePublished - Nov 8 2022
Externally publishedYes

Fingerprint

Dive into the research topics of 'Symmetry Breaking Slows Convergence of the ADAPT Variational Quantum Eigensolver'. Together they form a unique fingerprint.

Cite this