Title:Forecasting Quantum Observables: A Compressed Sensing Approach with Performance Guarantees

Abstract:Noise in near-term quantum devices limits the timescales over which measurements can be reliably obtained. Existing data-driven extrapolation methods extend the dynamics of quantum observables from measurements, but they cannot guarantee that the recovered model reflects the true dynamics. In this paper, we introduce atomic norm minimization as a framework to certify that a model learned by any algorithm accurately captures the underlying dynamics of quantum observables. This certification is valid when the system is governed by a small number of well-separated Bohr frequencies. We validate the framework across multiple algorithms on numerical experiments with spin-chain Hamiltonians involving 8-20 qubits. Comparing with exact diagonalization, certification yields an average forecasting error below 0.1 (within the observable's range of $[-1, 1]$) in 98% of cases and below 0.05 in 89-97% of cases, depending on the forecasting algorithm. Even in the presence of realistic shot noise, certified models remain robust, with success rates decreasing only to 88-95% for the 0.1 error threshold.