Solvation free energies from neural thermodynamic integration

Area of Science:

• Computational chemistry
• Statistical mechanics
• Machine learning in molecular modeling

Background:

• Calculating free-energy differences is crucial for understanding molecular processes.
• Traditional methods can be computationally expensive and limited in accuracy.
• Neural network potentials offer a promising avenue for more efficient molecular simulations.

Purpose of the Study:

• To introduce a novel method for computing free-energy differences.
• To leverage neural network potentials within thermodynamic integration.
• To accurately model molecular systems at atomistic resolution.

Main Methods:

• Utilizing thermodynamic integration with a neural network potential.
• Interpolating between target Hamiltonians at the sample distribution level.
• Optimizing neural network potentials to match equilibrium potentials at intermediate steps.
• Simultaneously coupling Lennard-Jones and electrostatic interactions.
• Modeling rigid-body rotation of molecules for molecular systems.

Main Results:

• Accurate free-energy difference calculations for benchmark systems.
• Successful application to a Lennard-Jones particle in a Lennard-Jones fluid.
• Precise simulation of water and methane solute insertion in water solvent.
• Demonstrated accuracy using a three-body neural-network potential.

Conclusions:

• The proposed method provides an accurate and efficient approach for free-energy calculations.
• Neural network potentials integrated with thermodynamic integration are effective for molecular simulations.
• This method advances the capability of atomistic simulations for complex chemical systems.