Efficient simulation of semiflexible polymers

TL;DR

This paper introduces an efficient algorithm for simulating semiflexible polymers like dsDNA by modeling fluctuation modes, significantly increasing simulation speed while maintaining accuracy, and enabling practical long-time dynamics studies.

Contribution

The authors develop a mode-based simulation algorithm for semiflexible polymers that vastly improves computational efficiency over traditional bead-based models.

Findings

Achieves 5-6 orders of magnitude speed-up over inextensible WLC model.

Allows for larger time steps with 5% accuracy in key observables.

Successfully simulates dsDNA tumbling in shear flow.

Abstract

Using a recently developed bead-spring model for semiflexible polymers that takes into account their natural extensibility, we report an efficient algorithm to simulate the dynamics for polymers like double-stranded DNA (dsDNA) in the absence of hydrodynamic interactions. The dsDNA is modelled with one bead-spring element per basepair, and the polymer dynamics is described by the Langevin equation. The key to efficiency is that we describe the equations of motion for the polymer in terms of the amplitudes of the polymer's fluctuation modes, as opposed to the use of the physical positions of the beads. We show that, within an accuracy tolerance level of $5\%$ of several key observables, the model allows for single Langevin time steps of $\approx1.6$, 8, 16 and 16 ps for a dsDNA model-chain consisting of 64, 128, 256 and 512 basepairs (i.e., chains of 0.55, 1.11, 2.24 and 4.48 persistence lengths) respectively. Correspondingly, in one hour, a standard desktop computer can simulate 0.23, 0.56, 0.56 and 0.26 ms of these dsDNA chains respectively. We compare our results to those obtained from other methods, in particular, the (inextensible discretised) WLC model. Importantly, we demonstrate that at the same level of discretisation, i.e., when each discretisation element is one basepair long, our algorithm gains about 5-6 orders of magnitude in the size of time steps over the inextensible WLC model. Further, we show that our model can be mapped one-on-one to a discretised version of the extensible WLC model; implying that the speed-up we achieve in our model must hold equally well for the latter. We also demonstrate the use of the method by simulating efficiently the tumbling behaviour of a dsDNA segment in a shear flow.