Emulating the First Principles of Matter: A Probabilistic Roadmap

TL;DR

This paper reviews first principles methods for modeling matter, emphasizing probabilistic surrogate models that incorporate uncertainty quantification, and discusses recent advances integrating physics-based modeling with data science for faster, more accurate predictions.

Contribution

It introduces probabilistic surrogate models for first principles calculations, combining Gaussian processes with physical symmetries to improve efficiency and accuracy.

Findings

Probabilistic models quantify uncertainty in emulation.

Gaussian process kernels encode physical symmetries.

Integration of physics and data science enhances computational efficiency.

Abstract

This chapter provides a tutorial overview of first principles methods to describe the properties of matter at the ground state or equilibrium. It begins with a brief introduction to quantum and statistical mechanics for predicting the electronic structure and diverse static properties of of many-particle systems useful for practical applications. Pedagogical examples are given to illustrate the basic concepts and simple applications of quantum Monte Carlo and density functional theory -- two representative methods commonly used in the literature of first principles modeling. In addition, this chapter highlights the practical needs for the integration of physics-based modeling and data-science approaches to reduce the computational cost and expand the scope of applicability. A special emphasis is placed on recent developments of statistical surrogate models to emulate first principles calculation from a probabilistic point of view. The probabilistic approach provides an internal assessment of the approximation accuracy of emulation that quantifies the uncertainty in predictions. Various recent advances toward this direction establish a new marriage between Gaussian processes and first principles calculation, with physical properties, such as translational, rotational, and permutation symmetry, naturally encoded in new kernel functions. Finally, it concludes with some prospects on future advances in the field toward faster yet more accurate computation leveraging a synergetic combination {of} novel theoretical concepts and efficient numerical algorithms.