Computing RPA Adsorption Enthalpies by Machine Learning Thermodynamic Perturbation Theory

Area of Science:

• Computational Chemistry
• Materials Science
• Machine Learning

Background:

• Correlated quantum-chemical methods, like random phase approximation (RPA), offer higher accuracy than standard density functional theory for condensed matter.
• The computational expense of these advanced methods limits their use in finite-temperature molecular dynamics simulations.
• Predicting finite-temperature properties of materials is crucial for understanding their behavior under realistic conditions.

Purpose of the Study:

• To develop a computationally efficient method for estimating finite-temperature properties using high-accuracy correlated quantum-chemical approximations.
• To enable the application of expensive quantum-chemical methods to larger systems and longer timescales relevant to condensed matter simulations.
• To reduce the computational burden associated with finite-temperature molecular dynamics in materials science.

Main Methods:

• Coupling machine learning techniques with thermodynamic perturbation theory.
• Training a machine learning model using a small dataset of energies computed with correlated approximations (e.g., RPA).
• Estimating finite-temperature properties, such as enthalpies of adsorption in zeolites.

Main Results:

• The proposed method successfully estimates finite-temperature properties with reliable accuracy.
• Accurate results were achieved by training the machine learning model with as few as 10 random phase approximation energies.
• Demonstrated applicability to calculating enthalpies of adsorption in zeolites.

Conclusions:

• The integration of machine learning and thermodynamic perturbation theory significantly reduces the computational cost of using advanced quantum-chemical methods.
• This approach facilitates the broader application of computationally demanding methods for predicting finite-temperature properties of condensed matter systems.
• Paves the way for more accurate and efficient materials simulations under realistic thermal conditions.