Mass-zero constrained molecular dynamics for electrostatic interactions

Area of Science:

• Computational physics and chemistry
• Molecular dynamics simulations
• Scientific computing

Background:

• Evaluating electrostatic forces in large molecular dynamics systems is computationally intensive.
• Current particle-mesh Ewald methods use fast Fourier transforms (FFTs), which face scalability issues with increasing processor counts.
• Real-space methods offer an alternative but require efficient Poisson equation solvers.

Purpose of the Study:

• To develop a novel, scalable real-space approach for electrostatic force evaluation in molecular dynamics.
• To address the limitations of FFT-based methods in supercomputing environments.
• To achieve better computational scaling for very large molecular systems.

Main Methods:

• Introduced a novel real-space method utilizing an extended Lagrangian formulation.
• Treated grid point field values as auxiliary variables with zero inertia.
• Enforced the discretized Poisson equation as a dynamical constraint, leading to an efficiently solvable linear system.
• Employed state-of-the-art iterative solvers, demonstrating linear scaling with multigrid approaches.

Main Results:

• The novel method achieves linear scaling, outperforming semi-logarithmic scaling of particle-mesh Ewald methods.
• The approach demonstrated faster convergence in iterative solver cycles compared to other real-space methods.
• Simulations of molten NaCl validated the method's accuracy in reproducing structural and transport properties.
• Linear scaling was demonstrated on a non-trivial benchmark system.

Conclusions:

• The developed real-space method offers an efficient and scalable alternative for electrostatic force calculations in large molecular dynamics simulations.
• The extended Lagrangian approach with dynamical constraints provides a robust framework for solving the Poisson equation.
• This work contributes to optimizing supercomputing resource utilization for complex scientific simulations.