Active learning phase boundaries of a quantum many-body system

Abstract

We describe how to use techniques from the field of Machine Learning to direct a variational energy minimization scheme to search for phase boundaries of a quantum many-body system. The modeled physical system presents states of finite momentum condensate, also known as FFLO states, as well as a uniform superfluid phase-all of which is interesting in its own right; however, a full description of the multitude of phase boundaries is expensive from a computational standpoint. In this work, we treat the output of the energy minimization as a labeled sythetic data set to train a support vector classifier to separate states of finite momentum condensate from superfluid and normal states. We can then use the trained support vector classifier to refocus the minimizer to intensify its calculations near the boundary separating each of the three regions. Doing so will preclude using the minimizer to perform expensive calculations deep within the normal or superfluid regions, resulting in more efficient use of compute time. The application of the procedure we describe is straightforward and should be applicable in any computational search of phase boundaries.