Project Description

New materials with engineered properties are central to addressing pressing societal challenges, such as energy and climate. Predictive computational tools to describe materials from first principles can significantly accelerate this process by reducing the number of costly experimental tests. Atomistic simulations can capture the chemistry, mechanics, thermodynamics, and transport of materials, but they are restricted to sub-micron scales and short timescales, even in modern supercomputers. Thus, bridging between these atomic scales and continuum models is paramount. Traditional approaches to multiscale modeling use physical intuition to select the collective degrees of freedom of retaining at the coarse scales (e.g., dislocation density or the fraction of intermediates in a chemical reaction) and the laws that govern their evolution and use atomistic simulations to parameterize these models. This effort will develop and apply a radically new approach. We will use modern tools of machine learning to discover optimal collective variables and the laws that govern their evolution. For the first step, we will explore dimensionality reduction techniques (both linear and non-linear), and for the second, tools to model time-series. In both cases, the models will incorporate the physics of the problem, for example, symmetries and conservation laws. 

Start Date

Position available immediately

Postdoc Qualifications

PhD in Materials Science, Mechanical Engineering, Physics, Computer Science or related fields. 

Co-advisors

Ilias Bilionis , Mechanical Engineering.
Ale Strachan , Materials Engineering. https://www.strachanlab.org.

Bibliography