Scalable quantum error mitigation with physically informed graph neural networks

Communications in Mathematical Physics · Springer Nature · 2026-10-13 · DOI: 10.1007/s42484-026-00462-w

Аннотация

Quantum error mitigation (QEM) provides a practical route for estimating reliable observables on noisy intermediate-scale quantum (NISQ) devices. Conventional methods, including zero-noise extrapolation (ZNE) and Clifford data regression (CDR), rely on noise scaling or target-specific calibration and may incur substantial overhead as circuit size and complexity increase. We propose a graph-enhanced mitigation (GEM) framework that incorporates physical hardware information into an attributed graph representation. Local calibration parameters, including $$T_1$$, $$T_2$$, and readout errors, are encoded as node features, while coupling-related quantities such as two-qubit gate errors are represented as edge features. Topology-constrained graph message passing captures hardware-dependent interactions across the physical coupling structure. A gated local-global residual correction combines qubit-specific features with circuit-level context, and parameter sharing allows the same model to operate across different system sizes. We investigate GEM using 10-, 16-, and 22-qubit random circuits executed on the Tianyan-176 superconducting quantum processor. The Origin Wukong-180 superconducting quantum processor was used for preliminary workflow testing. The results show that GEM is competitive with CDR at 10 qubits, achieves lower mean absolute error in zero-shot transfer from 10 to 16 qubits, and remains comparable with CDR at 22 qubits. We further examine its robustness across independent hardware runs and its ability to support reusable mitigation across circuits and system sizes through a shared representation of local hardware structure, coupler-level information, and circuit context. Ablation studies confirm that informative edge attributes and graph message passing contribute directly to mitigation performance. These results establish GEM as a scalable and physically informed framework for observable correction on superconducting quantum processors.

Читать у издателя →

Похожие статьи