Coupled cluster finite temperature simulations of periodic materials via machine learning

Density functional theory is the workhorse of materials simulations. Unfortunately, the quality of results often varies depending on the specific choice of the exchange-correlation functional, and this significantly limits the predictive power of this approach. Coupled cluster theory, including single, double and perturbative triple particle-hole excitation operators, is widely considered as the `gold standard` of quantum chemistry as it can achieve chemical accuracy for non-strongly correlated applications. Because of the high computational cost, the application of coupled cluster theory in materials simulations is rare, and this is particularly true if finite-temperature properties are of interest for which molecular dynamics simulations have to be performed. By combining recent progress in machine learning models with low data requirements for energy surfaces and in the implementation of coupled cluster theory for periodic materials, we show that chemically accurate simulations of materials are practical and could soon become significantly widespread. As an example of this numerical approach, we consider the calculation of the enthalpy of adsorption of CO2 in a porous material.