Gaussian process model kernels for noisy optimization in variational quantum algorithms

TL;DR

This paper introduces trigonometric kernels for Gaussian Process Models to improve noisy optimization in Variational Quantum Algorithms, demonstrating their effectiveness in quantum chemistry and combinatorial problems.

Contribution

The paper proposes novel trigonometric kernels for GPMs tailored to VQAs and systematically compares their performance with existing kernels.

Findings

Trigonometric kernels outperform others in most tested cases.

RotoGP optimizer benefits from these kernels to mitigate noise.

Improved convergence in quantum chemistry and combinatorial optimization problems.

Abstract

Variational Quantum Algorithms (VQAs) aim at solving classical or quantum optimization problems by optimizing parametrized trial states on a quantum device, based on the outcomes of noisy projective measurements. The associated optimization process benefits from an accurate modeling of the cost function landscape using Gaussian Process Models (GPMs), whose performance is critically affected by the choice of their kernel. Here we introduce trigonometric kernels, inspired by the observation that typical VQA cost functions display oscillatory behaviour with only few frequencies. Appropriate scores to benchmark the reliability of a GPM are defined, and a systematic comparison between different kernels is carried out on prototypical problems from quantum chemistry and combinatorial optimization. We further introduce RotoGP, a sequential line-search optimizer equipped with a GPM, and test how different kernels can help mitigate noise and improve optimization convergence. Overall, we observe that the trigonometric kernels show the best performance in most of the cases under study.