CASTING

CASTING is a Python framework for crystal-structure prediction and inverse materials design using a tree-search algorithm in continuous action space. It integrates ab initio energy evaluators (VASP, LAMMPS) with a Monte-Carlo-Tree-Search-inspired optimizer to explore high-dimensional potential energy surfaces and discover stable and metastable phases.

CASTING extends classical inverse-design strategies — which typically rely on genetic algorithms or basin hopping — by treating structure discovery as a sequential decision-making problem. The continuous-action formulation enables efficient exploration of complex configuration landscapes and has been used to discover metastable phases of boron and superhard carbon polymorphs.

Status: Maintained · Published in npj Computational Materials (2023)

Role: Core contributor

Applications

• Crystal structure prediction across the periodic table
• Discovery of metastable phases of boron and superhard carbon polymorphs
• High-dimensional potential-energy-surface exploration

Links

• Paper: Banik et al., npj Computational Materials 9, 177 (2023)

Citation

@article{banik2023casting,
  title   = {A continuous action space tree search for inverse design ({CASTING})
             framework for materials discovery},
  author  = {Banik, Suvo and Loeffler, Troy D. and Manna, Sukriti and
             Srinivasan, Srilok and Darancet, Pierre and Chan, Henry and
             Hexemer, Alexander and Sankaranarayanan, Subramanian K. R. S.},
  journal = {npj Computational Materials},
  volume  = {9},
  number  = {1},
  pages   = {177},
  year    = {2023},
  doi     = {10.1038/s41524-023-01128-y}
}