Full-Trajectory Learning of Open Quantum Systems: Dynamical Emulation and the Limits of Hamiltonian Identifiability
The work proposes a unified Quantum Neural Network (QNN) to learn unknown quantum Hamiltonians and their dissipation parameters by analyzing the full temporal evolution of density matrices under Lindblad dynamics. The framework uses trajectory loss optimization, enhanced by chirped excitation and randomized initial states, to reconstruct the Hamiltonian coefficients. This results in a data-driven method for creating physical quantum emulators.