Typically, first-principle calculations are computationally expensive. This problem is usually aggravated when modelling materials due to the large number of atoms involved in the models. Therefore, the design of robust and fast optimizers has always been a hot topic in this field. The desired algorithms must minimize the number of <i>ab initio </i>function calls as much as possible without compromising the accuracy of the simulated properties. Our approach is based on a machine learning surrogate model which allows to substantially reduce the number of function calls in the search of the optimal solution (see Fig. 1). Here, we present the evidence of the aforementioned acceleration for two of the most commonly encountered optimizations problems in materials science: energy minimization and transition-state search, such as structural relaxations and Nudged Elastic Band (NEB) calculations.

Accelerating <i>ab initio</i> simulation using surrogate machine learning models

2019 Spring National Meeting

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.

Accelerating ab initio simulation using surrogate machine learning models