Thermal reconstruction of chaotic quantum many-body systems

Thermal states are thermal with respect to a fixed Hamiltonian. How much information about this Hamiltonian can we ``bootstrap'' from the subsystems of a thermal state? We attack the problem by positioning it as a subspecies of the quantum marginal problem. In states that obey the quantum Markov property, the Petz recovery map captures the knowledge of the larger system inherent in a subsystem. We use the conditional mutual information to check the goodness of Petz recovery, analytically in a random-matrix-theory-inspired hopping model and numerically in an Ising-like spin chain model. We observe different behavior in chaotic versus integrable phases of the model: in the chaotic phase, the reconstruction works well at both very low and very high temperatures, with some intermediate critical temperature at which reconstruction works worst, whereas in the integrable phase reconstruction breaks down at low temperatures.