A Physics-Informed Neuro-Fuzzy Framework for Quantum Error Attribution

Abstract

As quantum processors scale beyond 100 qubits, distinguishing software bugs from stochastic hardware noise becomes a critical diagnostic challenge. We present a neuro-fuzzy framework that addresses this attribution problem by combining Adaptive Neuro-Fuzzy Inference Systems (ANFIS) with physics-grounded feature engineering. We introduce the Bhattacharyya veto, a hard constraint calibrated to the empirically characterized noise floor of the device, which classifies as buggy any output distribution whose divergence from the ideal exceeds what the device's noise produces under normal operation. The framework is validated on two IBM Heron r2 processors. On ibm_fez, across 105 circuits spanning 17 algorithm families, it attains 89.5% accuracy (Wilson 95% CI [82.2, 94.0]) with no abstentions; every residual error is a bug whose ideal computational-basis distribution is identical to the intended circuit's, or whose divergence lies below the device's noise floor. On a held-out suite of 30 deeper circuits (ISA depth to 100) executed on ibm_kingston, with all thresholds frozen, it attains 96.7% effective and 93.3% strict accuracy. We resolve key ambiguities - such as distinguishing correct Grover amplification from bug-induced collapse - and identify fundamental limits of single-basis diagnostics, including a Z-basis blind spot where phase-only errors remain statistically invisible. The extracted rule base is analyzed rather than asserted: the learned rules reproduce the physics of the veto, of the Grover boundary, and of depth-dependent noise without any of them being hard-coded. This work establishes an interpretable diagnostic layer that prevents error mitigation from being applied to logically flawed circuits.