University of Cambridge > Talks.cam > Machine learning in Physics, Chemistry and Materials discussion group (MLDG) > Global optimisation of atomistic structure with evolutionary algorithms and reinforcement learning

Global optimisation of atomistic structure with evolutionary algorithms and reinforcement learning

Add to your list(s) Download to your calendar using vCal

If you have a question about this talk, please contact Bingqing Cheng .

Atomistic simulations of the physico-chemical processes at inorganic surfaces often require knowledge of the energetically optimal state of the surfaces. In this talk, examples are given of intricate surface reconstructions and surprising shapes assumed by metal nano-particles supported on oxide surfaces. The focus of the talk will be on how to identify such optimal structure given a costly total energy method, typically based on density functional theory (DFT). A number of approaches will be presented. 1) A purely evolutionary approach in which new structural candidates are created by random cross-over and mutation operations, 2) a machine learned-enhanced evolutionary approach in which an on-the-fly learned surrogate energy landscape directs the candidate production, and finally 3) a pure reinforcement learning approach in which image recognition via a convolutionary neural network is used to build up rational knowledge about the energy landscape, that eventually leads to construction of the globally optimal structure.

[1] Evolutionary approach (EA): Phys. Rev. Lett. 108, 126101 (2012) [2] ML assisted EA: Phys. Rev. Lett. 124, 086102 (2020) and https://gofee.au.dk [3] Reinforcement learning: Phys. Rev. B, 102, 075427 (2020).

This talk is part of the Machine learning in Physics, Chemistry and Materials discussion group (MLDG) series.

Tell a friend about this talk:

This talk is included in these lists:

Note that ex-directory lists are not shown.

 

© 2006-2024 Talks.cam, University of Cambridge. Contact Us | Help and Documentation | Privacy and Publicity