📝 SummarXiv

Abstract

Achieving chemical accuracy in quantum simulations is often constrained by the measurement bottleneck: estimating operators requires a large number of shots, which remains costly even on fault-tolerant devices and is further exacerbated on today's noisy hardware by finite circuit fidelity and error-mitigation overhead. Addressing this challenge involves a multiobjective optimization problem that balances total shot count, the number of distinct measurement circuits, and hardware-specific compilation constraints. Existing methods typically rely on heuristic graph-coloring strategies to group commuting or qubit-wise commuting Hamiltonian terms, or on greedy allocation schemes for distributing measurements. Such approaches explore only a limited portion of the combinatorial solution space, potentially missing superior solutions. We introduce an algorithm that adapts Generative Flow Networks (GFlowNets) for coloring graph representations of Hamiltonians, enabling the generation of diverse, high-quality groupings. Our approach samples colored graphs in proportion to a user-defined reward, naturally capturing multiobjective trade-offs and discovering multiple competitive solutions. On benchmark molecular Hamiltonians, our method outperforms sorted-insertion baselines by reducing measurement costs. We further analyze the role of composite rewards, incorporating both the number of circuits and measurement costs, leading to additional improvements. This generative policy framework not only reduces measurement costs but also offers flexibility for potential hardware-aware adaptations through its reward function.