Dy-Part: A Dynamic, Noise-Aware Scheduler for Optimizing Hybrid Quantum-Classical Algorithms

Hybrid quantum-classical algorithms are a leading approach for applying noisy intermediate-scale quantum (NISQ) devices to practical problems. A key challenge in this paradigm is determining how to partition large computational problems into smaller sub-tasks that can be executed on size-limited quantum hardware. Current methods often rely on static, predetermined partitioning strategies that primarily consider the problem's structure, often failing to adapt to the quantum processor's dynamic noise characteristics. In this work, we introduce and analyze a dynamic scheduling framework, named Dy-Part, designed to address this challenge. Dy-Part automates the partitioning decision by using a heuristic cost model that balances two competing factors: the expected infidelity of the quantum computation, which increases with circuit size, and the classical post-processing overhead, which increases as the problem is broken into more pieces. To find the optimal partition size that minimizes this cost, our framework employs an efficient ternary search, whose output guides a fast, greedy graph partitioning heuristic. We demonstrate and validate this approach using the Variational Quantum Eigensolver (VQE) to solve the Max-Cut problem. Our results, averaged over 30 random graph instances, show that while a static strategy is effective in low-noise regimes, Dy-Part provides a more robust solution as noise increases. For instance, on 12-node graphs with a high gate error rate (\((\epsilon_{\text{gate}} > 10^{-2})\)), Dy-Part's dynamic strategy yields a mean approximation ratio more than double that of the static baseline. These results show that a dynamic, noise-aware scheduling approach can provide a robust method for configuring hybrid workflows, offering a practical pathway to maximizing the performance of near-term quantum computers.