Electrostatic point charge fitting as an inverse problem: Revealing the underlying ill-conditioning.

Area of Science:

• Computational Chemistry
• Molecular Modeling
• Biophysics

Background:

• Atom-centered point charge (PC) models are crucial for atomistic biomolecular simulations.
• Parameterization via least-squares (LS) fitting to electrostatic potentials suffers from numerical instabilities.
• The ill-conditioned nature of the LS problem hinders accurate electrostatic modeling.

Purpose of the Study:

• To investigate the origins of numerical instabilities in PC model parameterization.
• To develop a more stable and accurate method for calculating molecular electrostatics.
• To propose an alternative to current PC models in force fields.

Main Methods:

• Treated PC fitting as an inverse problem.
• Developed an analytical model using Lebedev quadrature for spherical charge arrangements.
• Contrasted the analytical model with traditional atom-centered PC models.
• Analyzed the Hessian matrix of the LS sum to understand ill-conditioning.

Main Results:

• Identified ill-conditioning due to decaying curvatures associated with LS sum Hessian eigenvectors.
• Demonstrated that the Lebedev grid model aligns eigenvectors with multipole moments, unlike atom-centered models.
• Showed that the proposed analytical model alleviates ill-conditioning without external restraints.
• The Lebedev grid PC model can reproduce multipole moments up to a specified rank.

Conclusions:

• The analytical Lebedev grid PC model provides a stable and accurate alternative for molecular electrostatics.
• This model can potentially replace explicit multipole terms in force fields.
• Understanding the inverse problem nature of PC fitting is key to improving simulation accuracy.