Machine learning of Gaussian basis sets for use in computational chemistry J. Grant Hill and Robert A. Shaw University of Sheffield, UK Basis sets for use in molecular electronic structure calculations have a large influence on both the accuracy and efficiency of results, yet the development of new basis sets is a painstakingly slow process carried out by a small number of groups across the globe. Recent work within our group to produce novel basis sets and to make the process more accessible to other scientists will be presented. The Python package BasisOpt is a tool for the automated optimisation of basis sets that can be easily adapted to almost any electronic structure program and quantum chemical method. Applications of the package will be presented, including the optimisation of auxiliary basis sets for use in density fitting, optimisation of basis set exponents for molecules (rather than the usual optimisation for atomic energies), and the automated reduction of a large basis into the most accurate basis possible for a given number of functions. Analysis of data from molecule-optimised basis sets using BasisOpt leads to the use of machine learning (ML) to predict new sets that are tailor-made for different chemical applications. Results of ML basis sets will be presented and compared to those from both existing basis sets and new basis sets that were optimised using BasisOpt. It will be demonstrated that this approach is effective in producing accurate and efficient basis sets without requiring any basis set optimisation expertise.
DC03
© The Author(s), 2023
Made with FlippingBook Learn more on our blog