This paper presents a novel approach for performing finite-temperature coupled cluster simulations of periodic materials using machine learning techniques. The high computational cost of coupled cluster theory, especially when considering finite-temperature effects, is addressed by combining efficient periodic coupled cluster implementations with machine learning models trained on a limited number of high-level calculations. This allows for the accurate prediction of thermodynamic observables, such as the enthalpy of adsorption, which is demonstrated through the calculation of CO2 adsorption in a protonated chabazite zeolite. The results are validated by comparison with experimental data and show excellent agreement.

Basile Herzog, Alejandro Gallo, Felix Hummel, Michael Badawi, Tomáš Bučko, Sébastien Lebegue, Andreas Grüneis, Dario Rocca