Chapter 24 - Improving semiempirical quantum mechanical methods with machine learning

Abstract

As has been seen in the corresponding introductory Chapter 3, semiempirical quantum mechanical (SQM) methods are among the fastest quantum chemical approaches, but they are generally less accurate than other approaches. Machine learning (ML) offers itself as an excellent tool for improving SQM methods in various ways. ML can be used to improve the predictions made with an SQM method in a Δ-learning way, or ML can be used to improve an SQM Hamiltonian by learning better parameters as well as complimenting existing parameters and parametric functions. This chapter discusses the general strategies to improve SQM methods with ML and shows examples of the resulting ML-SQM methods. We also demonstrate the improved accuracy of ML-SQM methods compared to the traditional SQM methods for selected applications. In the case study section, we give a hands-on example on the use of a general-purpose ML-SQM method AIQM1.