Thursday, June 13, 20192:00 PM - 3:00 PMCNLS Conference Room (TA-3, Bldg 1690)|
Accelerating materials design and ab initio simulation via statistical inference and machine learning
Machine learning and statistical inference are emerging as powerful approaches to accelerate scientific discovery and computation. Regardingthe former, I will discuss recent advances in the systemic design of colloidal interactions for self assembly of complex structural morphologies. Simultaneous tuning of bulk thermodynamic properties (i.e., pressure) with the microstructure is also possible. In regards toaccelerated computation, I will discuss recent attempts to accelerate ab initio simulation of chemically reactive mixtures. Using a nested MonteCarlo approach, it is possible to exactly sample, at equilibrium, from a high level of theory (i.e., density functional theory) using long composite trial moves generated by an inexpensive reference potential. Utilizing only pair potential reference interactions, speedups up to a factor of six seem feasible--more complex interactions are probablyrequired to break the "order of magnitude barrier".
Host: David Métivier