|
Roadmap on Machine Learning in Electronic Structure
Kulik, Heather and Hammerschmidt, Thomas and Schmidt, Jonathan and Botti, Silvana and Marques, Miguel A. L. and Boley, Mario and Scheffler, Matthias and Todorovic, Milica and Rinke, Patrick and Oses, Corey and Smolyanyuk, Andriy and Curtarolo, Stefano and Tkatchenko, Alexandre and Bartok, Albert and Manzhos, Sergei and Ihara, Manabu and Carrington, Tucker and Behler, Jorg and Isayev, Olexandr and Veit, Max and Grisafi, Andrea and Nigam, Jigyasa and Ceriotti, Michele and Schutt, Kristoff T and Westermayr, Julia and Gastegger, Michael and Maurer, Reinhard and Kalita, Bhupalee and Burke, Kieron and Nagai, Ryo and Akashi, Ryosuke and Sugino, Osamu and Hermann, Jan and Noe, Frank and Pilati, Sebastiano and Draxl, Claudia and Kuban, Martin and Rigamonti, Santiago and Scheidgen, Markus and Esters, Marco and Hicks, David and Toher, Cormac and Balachandran, Prasanna and Tamblyn, Isaac and Whitelam, Stephen and Bellinger, Colin and Ghiringhelli, Luca M. Electronic Structure, 4, 2 (2022) In recent years, machine learning has begun to fundamentally transform computational materials science and electronic structure theory, complementing, extending, and in some cases replacing traditional methods developed over the past century. This comprehensive roadmap article brings together contributions from leading experts across the field to discuss the current state and future challenges of integrating machine learning into electronic structure calculations. Topics covered span a broad range, including the prediction of materials properties, the construction of machine-learned interatomic potentials and force fields, the development of exchange-correlation functionals for density-functional theory, solutions to the quantum many-body problem, and the creation of shared data infrastructure. The article highlights that initial efforts to integrate ML into computational workflows show substantial promise for accelerating materials discovery, and anticipates that tighter integration between computational scientists and experimentalists will enable autonomous discovery of new functional materials. |
