Adaptive pruning-based optimization of parameterized quantum circuits

Variational hybrid quantum-classical algorithms are powerful tools to maximize the use of Noisy Intermediate Scale Quantum devices. While past studies have developed and expressive ansatze, their near-term applications been limited by difficulty optimizing in vast parameter space. In this work, we propose a heuristic optimization strategy for such ansatze used variational quantum algorithms, which call "Parameter-Efficient Circuit Training" (PECT). Instead all ansatz parameters at once, PECT launches sequence each iteration algorithm activates optimizes subset total set. To update between iterations, adapt dynamic sparse reparameterization scheme Mostafa et al. (arXiv:1902.05967). We demonstrate Eigensolver, benchmark unitary coupled-cluster including UCCSD k-UpCCGSD, as well low-depth circuit (LDCA), estimate ground state energies molecular systems. additionally layerwise variant optimize hardware-efficient Sycamore processor energy densities one-dimensional Fermi-Hubbard model. From our numerical data, find that can enable optimizations certain were previously difficult converge more generally improve performance reducing runtime and/or depth circuits encode solution candidate(s).