Multireference Generalization of the Weighted Thermodynamic Perturbation Method.

Area of Science:

• Computational Chemistry
• Molecular Dynamics
• Free Energy Calculations

Background:

• Estimating free energy surfaces for complex systems requires computationally expensive high-level potentials.
• Existing methods like weighted thermodynamic perturbation (wTP) use a single low-level potential, limiting flexibility.
• Machine learning potentials (MLPs) offer speed but require accurate training against high-level data.

Purpose of the Study:

• To introduce the generalized weighted thermodynamic perturbation (gwTP) method for improved free energy surface estimation.
• To enable the combination of multiple low-level reference potentials for enhanced sampling and accuracy.
• To demonstrate the application of gwTP with machine learning potentials for computationally demanding simulations.

Main Methods:

• Developed the generalized weighted thermodynamic perturbation (gwTP) method, extending the wTP approach.
• Utilized multiple inexpensive low-level reference potentials for umbrella sampling.
• Applied gwTP to estimate the free energy surface of nonenzymatic RNA 2'-O-transesterification using MLPs trained on QM/MM data.

Main Results:

• The gwTP method successfully combined sampling from diverse low-level potentials (MNDO/d, DFTB2/MIO, DPRc).
• Extended sampling with reference potentials identified unequilibrated simulation regions.
• gwTP yielded more accurate ab initio free energy predictions compared to single-reference wTP or direct high-level sampling.

Conclusions:

• The gwTP method provides a flexible and powerful framework for free energy calculations.
• It significantly enhances sampling efficiency and accuracy, especially when combined with MLPs.
• This approach allows for more reliable predictions of reaction mechanisms and molecular properties.