Lab Home | Phone | Search
Center for Nonlinear Studies  Center for Nonlinear Studies
 Home 
 People 
 Current 
 Postdocs 
 Visitors 
 Students 
 Research 
 Publications 
 Conferences 
 Workshops 
 Sponsorship 
 Talks 
 Seminars 
 Postdoc Seminars Archive 
 Quantum Lunch 
 Quantum Lunch Archive 
 P/T Colloquia 
 Archive 
 Ulam Scholar 
 
 Postdoc Nominations 
 Students 
 Student Program 
 Visitors 
 Description 
 Past Visitors 
 Services 
 General 
 
 History of CNLS 
 
 Maps, Directions 
 CNLS Office 
 T-Division 
 LANL 
 
Tuesday, April 30, 2024
2:00 PM - 3:00 PM
CNLS Conference Room (TA-3, Bldg 1690)

Seminar

Pyiron: Simulation Workflows for Data-Driven Materials Design

Jan Janssen
Group Leader, Max Planck Institute for Sustainable Materials, Dusseldorf, Germany

Sampling the chemical complexity of the periodic table at different temperatures and concentration ranges to identify alloying elements with favorable material properties, requires thousands up to millions of sample points. This is not only experimentally prohibitive but also remains a challenge for theoretical predictions and simulations. Workflow frameworks like pyiron [1,2] address this challenge by providing building blocks for the rapid prototyping of the simulation protocol and the up-scaling on high performance computers. As a demonstration, we present a workflow which combines the VASP DFT simulation code, the FitSNAP fitting code for interatomic potentials, and the LAMMPS molecular dynamics code to develop an interatomic potential and validate it. With this workflow, we find that the cut-off radius of a machine-learned interatomic potential is less of a physically-defined parameter but rather a numerical parameter which has to be adjusted corresponding to the complexity of the machine-learned interatomic potential to achieve optimal computational efficiency. Furthermore, this data-driven approach to fitting machine-learned interatomic potentials highlights how low-precision density functional theory (DFT) calculation can be leveraged for constructing machine-learned interatomic potentials that capture the properties of high-precision DFT. In summary, by developing automated workflows for the fitting of machine learned interatomic potentials and applying them for data-driven parameter studies, we accelerate the prediction of material properties like the phase diagram to enable a quantitative comparison to experiment.

[1]: http://pyiron.org
[2]: J. Janssen, et al., Comp. Mat. Sci. 161 (2019)

Host: Danny Perez (T-1)