arXiv:2610.05524
Optimization Geometry of QAOA and Variational Quantum Algorithms
Vojtěch Novák, Ivan Zelinka, Silvie Illésová, Swagatam Das, Martin Beseda
Abstract
Variational quantum algorithms turn choices of Hamiltonian, ansatz, and parameterization into a classical nonconvex optimization problem. We study how this objective function can be visualized and characterized in ways that help explain optimizer behavior. We distinguish two properties of the objective: the number of local minima encountered along sampled directions and the differences in quality among local-search endpoints. We then ask how a local optimizer, represented by BFGS, compares with adaptive differential evolution, represented by jSO. Rather than comparing jSO with a single local run, we allow BFGS multiple starts within the same function-evaluation budget. This gives local search repeated opportunities to explore different basins and provides a stronger baseline for asking when a global evolutionary solver is useful. We study these questions using VQE and QAOA, focusing on how frustra- tion, circuit depth, parameter tying, mixed locality, and nonlinear repa- rameterization change the Hamiltonian expectation-value objective seen by the classical optimizer. Increasing independent circuit depth raises the sampled local-minimum count without making global search more effective. Parameter tying, by contrast, produces both more repeated local structure and much larger differences in quality among local-search endpoints, and in this regime adaptive differential evolution outperforms function-evaluation-matched multistart BFGS. The comparison shows that the number of local minima alone does not determine whether global search is advantageous: the important distinction is whether different basins lead to similarly good solutions or to substantially different objective values. These results provide a practical workflow for connecting model construction, objective- function geometry, empirical diagnostics, and optimizer choice.
AI Summary
The paper asks when global optimization helps VQAs beyond multistart local search, distinguishing sampled directional multimodality from variation in the quality of local-search endpoints. It combines coordinate slices, optimizer-informed planes, basin visualizations, random-line minimum counts, and random-start BFGS endpoint statistics across QAOA and VQE constructions. The diagnostics examine how Hamiltonians, depth, parameter tying, and nonlinear reparameterization reshape the objective. The main optimizer comparison uses exact statevector simulations of four random 3-regular MaxCut graphs with $N=10$. It compares standard QAOA at $p=3$, independent $p=6$, and an 18-layer tied construction retaining six parameters. Multistart BFGS and adaptive differential evolution (jSO) each receive 15,000 function evaluations per run, with eight matched seeds. Increasing independent depth raises the median sampled minima count per line from 12 to 14.75, while the median local-endpoint gap remains similar: 2.01 versus 1.79 spectral-range percentage points. jSO performs worse by 4.63 points. Parameter tying instead yields 21.75 minima per line, a 16.74-point median endpoint gap, and 82.81% of endpoints more than five points above the best sampled endpoint; jSO outperforms multistart BFGS by 1.88 points. These examples associate global-search benefit with unequal endpoint quality rather than minimum counts alone. The stated limitations include small instances, fixed budgets, and best-observed rather than certified optima. Line minima and selected slices do not characterize the full landscape, and basin accessibility depends on the local optimizer. Finite-shot sampling, hardware noise, and barren-plateau regimes are not tested; optimizer rankings need not transfer to those settings.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.