Parameter optimization in differential geometry based solvation models.

Area of Science:

• Computational chemistry
• Theoretical chemistry
• Physical chemistry

Background:

• Differential geometry (DG) based solvation models offer a novel approach to implicit solvent modeling, avoiding unphysical boundaries.
• Previous DG models showed promise in non-polar solvation but faced challenges in full solvation due to complex parameterization and equation stability.
• The accurate and self-consistent coupling of polar and non-polar interactions remains a challenge in computational solvent modeling.

Purpose of the Study:

• To introduce novel parameter learning algorithms for DG-based solvation models.
• To stabilize the numerical solutions of the underlying nonlinear equations for optimal model parametrization.
• To demonstrate the capability of the improved DG model for accurate solvation free energy predictions in a unified framework.

Main Methods:

• Development of new parameter learning algorithms utilizing perturbation and convex optimization theories.
• Stabilization of numerical solutions for strongly coupled nonlinear Laplace-Beltrami and Poisson-Boltzmann equations.
• Application and validation of the optimized DG-based solvation model across a diverse set of molecules.

Main Results:

• The new algorithms successfully stabilize the DG-based solvation model, enabling optimal parameter learning.
• The unified DG model accurately predicts solvation free energies for both polar and non-polar molecules.
• Extensive numerical experiments confirm the high accuracy of the DG model compared to existing methods.

Conclusions:

• The optimized DG-based solvation model provides a robust and accurate method for predicting solvation free energies.
• The developed parameter learning strategies overcome previous limitations in DG model application.
• This work advances implicit solvent modeling, offering a unified approach for diverse chemical environments.