Suvo Banik1,Srilok Srinivasan2,Sukriti Manna1,Troy Loeffler2,Henry Chan2,Pierre Darancet2,Subramanian Sankaranarayanan2
University of Illinois at Chicago1,Argonne National Laboratory2
Suvo Banik1,Srilok Srinivasan2,Sukriti Manna1,Troy Loeffler2,Henry Chan2,Pierre Darancet2,Subramanian Sankaranarayanan2
University of Illinois at Chicago1,Argonne National Laboratory2
The most common and popular methods for structure search and optimization are based on evolutionary design. This can often be cumbersome, limited to a few tens of parameters, and fails for large structural configurations or design problems with high degrees of freedom. Reinforcement learning approaches mostly operate in discrete action space such as in the Go game but the applications of that to inverse problems are limited since most inverse problems deal with continuous action space. There are many inverse structural search problems ranging from crystal structure search in material sciences to topology design in Quantum information, where it is highly desirable to optimize structure/configuration to target desired properties or functionalities. We develop CASTING as a scalable framework platform for molecular, crystal structure, defects, topology, device, and component design. We demonstrate the ability of our modified cMCTS algorithm on traditional trial functions. The predicational capability of the CASTING on crystals of 0D, 2D, 3D, polymorphs of carbon and complex system such as perovskite Nickelates. We also compare the cost and prediction performance of our CASTING framework with traditional Evolutionary algorithm-based approaches.