Phase Transitions from the perspective of Machine Learning

LSU Department of Physics & Astronomy

Over the past few years, machine learning algorithms have gradually been recognized as powerful tools for making decisions by learning patterns from data. Increasing amount of resources from tech companies have been invested in machine learning. At present, rather sophisticated packages are readily available. Time is ripe for leveraging the investments from business companies back to science and engineering. In this talk, I will demonstrate how to unleash the potential of machine learning to gain insight from diverse set of data from molecular dynamics to quantum Monte Carlo. In contrast with conventional methods, machine learning approach does not rely directly on a priory knowledge of the systems. It finds the ‘order parameter’ from the data by itself. The order parameter can then be used to identify phase transitions. We demonstrate this new approach with two examples: 1) We find the melting point of aluminum from molecular dynamics data, and 2) We estimate the crossover coupling strength between the Kondo phase and local moment phase of a quantum impurity problem from quantum Monte Carlo data.