Identification of low-performance operating conditions and candidate robust operating regions in a SERF atomic magnetometer

Abstract To address the difficulty of identifying low-performance states and determining robust operating regions in a spin-exchange relaxation-free (SERF) atomic magnetometer under multiparameter coupling, operating-condition identification and operating-region analysis are performed using multiparameter experimental data. Low-performance conditions are first identified in a joint output-state space comprising signal amplitude, noise level, and the derived signal-to-noise ratio. The results show that low-performance states are characterized primarily by a significant increase in noise without a significant decrease in signal amplitude, indicating a noise-amplification degradation mode. Physics-informed features associated with optical detuning and power matching are then constructed, and key parameter features are analyzed using random-forest permutation importance and bootstrap resampling. A SafeScore metric incorporating the mean signal-to-noise ratio, performance variability, and the proportion of low-performance samples is subsequently introduced to identify candidate robust operating regions within the experimental input-parameter space. Weight-sensitivity analysis and repeated holdout validation show that the identified candidate regions exhibit good ranking consistency and some out-of-sample performance discrimination. These results provide a basis for multiparameter operating-state analysis and experimental parameter optimization of SERF atomic magnetometers.