Estimating error in diffusion coefficients derived from molecular dynamics simulations.

Area of Science:

• Computational Chemistry
• Statistical Mechanics
• Biophysics

Background:

• Molecular dynamics (MD) simulations are computationally intensive, limiting accessible length and time scales.
• Accurate estimation of statistical uncertainty is vital for properties derived from MD simulations.
• The ergodic hypothesis is often assumed for property calculations from single MD simulations without validation.

Purpose of the Study:

• To demonstrate the unreliability of diffusion coefficient estimates from single MD simulations without validation.
• To highlight the increased severity of this issue for macromolecular diffusion.
• To advocate for the necessity of multiple independent simulations for robust results.

Main Methods:

• Analysis of self-diffusion coefficients in Lennard-Jones fluids using single MD simulations.
• Investigation of diffusion coefficient calculations for macromolecules.
• Examination of the validity of linear regression assumptions in standard analysis methods.
• Development of a rigorous approach to address statistical uncertainty.

Main Results:

• Single MD simulations without validation lead to poor estimates of self-diffusion coefficients.
• Macromolecular diffusion calculations are particularly susceptible to errors from single simulations.
• Standard methods for calculating diffusion coefficients, like linear fitting of mean squared displacement, violate regression assumptions.
• Multiple independent simulations are essential for accurate diffusion coefficients and their uncertainties.

Conclusions:

• The assumption of ergodicity in single MD simulations is often invalid and leads to inaccurate diffusion coefficients.
• Reliable estimation of diffusion coefficients and their statistical uncertainties requires multiple independent simulations.
• A rigorous methodology is needed to address the limitations of current MD simulation analysis for diffusion properties.