Computational Material Science Applied to the Screening of Materials

Context: The properties of heterogeneous catalysts with sizes from “single-atom” to

nanoscale range have been a longstanding interest of both experimental and

computational researchers. Although atomistic understanding of the molecular-level

properties of such catalysts has improved considerably in the past two decades, the

mechanisms by which the catalyst structure is coupled to the chemical environment,

and the impact of this coupling on catalytic properties, remain incompletely understood,

thereby limiting opportunities for discovery of new and exciting materials. To extend the

range of applications for real-life catalysis, we will develop and extend theoretical

methods to predict how the structure of multielemental catalytic nanoparticles works.

General Strategy: The proposed computational strategy will use periodic DFT

calculations of the energies of multielemental surface structures of differing

compositions and local structures to provide raw data for input into machine learning

algorithms based on crystal graph convolution neural network formalisms [4]. These

formalisms will regress the DFT data to yield compact potential energy expressions for

nanostructures with varying distributions of surface features and local compositions.

These expressions will then be exploited by global optimization and in predicting the

structures of nanoscale materials based on empirical potential energy functions, to

predict ensemble average nanocatalyst geometries at given reactor temperatures.

International Collaboration: This postdoc project will be supervised by Juarez L. F. Da

Silva in close collaboration with Jeffrey Greely (University of Purdue, CISTAR, USA).

Activities: The project is composed by several activities, and the most important ones

are listed: (1) Selection of the materials with potential for those applications according

to the chemical reactions; (2) Density functional theory calculations for surfaces and

finite-size particles to be used as training data for neural-network algorithm. (3)

combination with the global optimization implementation employed at the QTNano

group. (4) simulations to address the structure changes under different environments

and its effects on the particle catalytic activities. (5) as possible, comparison with

experimental results.

Expected Results: Atomistic understanding of the structural changes in finite-size

nanoparticles at high-temperate conditions and their effects in chemical reactions.