FNOs Predict All Local Observables From Subset Measurements

Researchers have developed a new framework using Fourier neural operators (FNOs) capable of predicting all local observables from a subset of quantum measurements without information about the Hamiltonian, a capability exceeding that of conventional simulation methods. Addressing a major challenge in quantum dynamics, where computational cost grows exponentially with system size, this approach achieves polynomial scaling by employing a spatially local basis. The team demonstrates FNO’s versatility through reconstructing Floquet Hamiltonians, predicting expectation values, and learning quantum information spreading, achieving significant speedups compared to exact numerical methods. Once trained, the model transfers reliably across driving frequencies, system parameters, and time discretizations, and FNOs can also extrapolate beyond the limits of current experimental observation, enabling access to observables and operator-spreading dynamics that might be beyond the coherence…