Stiff-spring approximation revisited: inertial effects in non-equilibrium trajectories

TL;DR

This paper revisits the stiff spring approximation in steered molecular dynamics, revealing inertial effects at high spring constants that bias free energy estimates, and proposes methods to mitigate these errors for more accurate simulations.

Contribution

It demonstrates how inertial effects at high spring constants bias work distributions and introduces strategies to optimize spring constants and use peak work values for accurate PMF calculations.

Findings

Inertial effects cause skewed and broader work distributions.

Bias in PMF estimates increases with simulation time.

Using peak work values reduces bias and improves accuracy.

Abstract

Use of harmonic guiding potentials is the most common method for implementing steered molecular dynamics (SMD) simulations, performed to obtain potentials of mean force (PMFs) of molecular systems using non-equilibrium work (NEW) theorems. Harmonic guiding potentials are also the natural choice in single molecule force spectroscopy experiments. The stiff spring approximation (SSA) of Schulten and coworkers enables to use the work performed along SMD trajectories to obtain the PMF. We discuss and demonstrate how a high spring constant, k, required for the validity of the SSA can violate another requirement of this theory, i.e., the validity of Brownian dynamics of the system. Violation of the Brownian condition results in the introduction of kinetic energy contributions to the external work, performed during SMD simulations. These inertial effects result in skewed work distributions, rather than the Gaussian distributions predicted by SSA. The inertial effects also result in broader work distributions, which worsen the effect of the skewness when calculating work averages. Remarkably, our results strongly suggest that the skew and width of work distributions are independent of the average drift velocity and physical asymmetries. The skew and broadening of work distributions result in biased estimation of the PMF. The bias manifests itself in the form of a systematic error that increases with simulation time. We discuss the proper upper limit for k, such that the inertial effects are avoided. This limit, used together with the relation for the lower limit of k, enables to conduct accurate steering while satisfying the Brownian dynamics. Furthermore, we argue and demonstrate that using the peak-value (rather than the statistical mean) of the work distributions vastly reduces the bias in the calculated PMFs and improves the accuracy.