Variational quantum eigensolvers for sparse Hamiltonians

TL;DR

This paper extends the variational quantum eigensolver (VQE) to efficiently handle sparse Hamiltonians, broadening its applicability beyond Pauli representations by decomposing sparse Hamiltonians into self-inverse, Hermitian terms.

Contribution

The authors develop a method to decompose general sparse Hamiltonians into a manageable number of self-inverse, Hermitian terms, enabling VQE to work with sparse matrix representations.

Findings

Decomposition of fermionic Hamiltonians into one-sparse, Hermitian terms.

Extension of VQE to general sparse Hamiltonians.

Sample complexity scales as (\u03b5) for expectation value estimation.

Abstract

Hybrid quantum-classical variational algorithms such as the variational quantum eigensolver (VQE) and the quantum approximate optimization algorithm (QAOA) are promising applications for noisy, intermediate-scale quantum (NISQ) computers. Both VQE and QAOA variationally extremize the expectation value of a Hamiltonian. All work to date on VQE and QAOA has been limited to Pauli representations of Hamiltonians. However, many cases exist in which a sparse representation of the Hamiltonian is known but there is no efficient Pauli representation. We extend VQE to general sparse Hamiltonians. We provide a decomposition of a fermionic second-quantized Hamiltonian into a number of one-sparse, self-inverse, Hermitian terms linear in the number of ladder operator monomials in the second-quantized representation. We provide a decomposition of a general $d$-sparse Hamiltonian into $O(d^2)$ such terms. In both cases a single sample of any term can be obtained using two ansatz state preparations and at most six oracle queries. The number of samples required to estimate the expectation value to precision $\epsilon$ scales as $\epsilon^{-2}$ as for Pauli-based VQE. This widens the domain of applicability of VQE to systems whose Hamiltonian and other observables are most efficiently described in terms of sparse matrices.