A Bayesian Optimization Algorithm for Functionals: Application to Materials Modeling

Tuesday, February 10, 2015 - 2:00pm - 3:00pm
Lind 305
Paul Patrone
In computational materials science, low-dimensional coarse-grained (CG) simulations are often used as surrogates for their more expensive fully-atomistic (i.e. high-dimensional) counterparts. Mathematically, CG simulations can be viewed as mappings from a function f(x), which describes interparticle forces, to a real number that describes some material property. Typically, the f(x) that yields best agreement between the CG and atomistic models is at best known to within some probability. In this talk, I present an iterative Bayesian-type algorithm that can be used to rapidly improve our knowledge of optimal f(x) under certain conditions on the functional map.