Adaptive Multiscale Molecular Dynamics of Macromolecular Fluids

Until now, adaptive atomistic–coarse-grain (A/CG) molecular dynamics simulations have had very limited applicability because the on-the-fly $\mathrm{CG}\to \mathrm{A}$ transformation is problematic for all but those molecules whose CG representation consists of a single particle. Here, we solve this problem by combining a transitional healing region with a rotational dynamics of rigid atomistic fragments in the CG region. Error control is obtained by analysis of the $\mathrm{A}\leftrightarrow \mathrm{CG}$ energy flow. We illustrate the method with adaptive multiscale simulations of liquid hexane and of a dilute polymer solution in a theta solvent.

Figure 1

Rotational dynamics of rigid atomistic fragments on top of the dynamics of CG molecules. A bond interaction between atoms induces a torque $t_1$ and a rotation of the atoms (A). CG bond stretching is removed from the rotational dynamics (B). CG bending couples to the rotational dynamics (C).

Figure 2

Total energies (offset) of MMD simulations of liquid hexane, using different widths of the HR, ranging from $d_{\mathrm{HR}}=1$ to 6 Å. Without the rotational dynamics, the total energy is not conserved (see irregular line using the right axis).

Figure 3

Snapshots from the MMD simulation of the swelling of a polyethylene chain in a theta solvent (hexane), taken at $t=0$, $t=100$, and $t=1000\mathrm{ps}$. Solvent molecules completely outside the AR are omitted for clarity. The large transparent sphere indicates the AR, which is centered at the middle of the chain (small bright sphere).

Figure 4

Top panel: Increasing radius of gyration of the polymer during the first ns. Bottom panel: number of carbon atoms of polyethylene (lower gray line) and hexane solvent (upper gray line) within the AR plus HR. Linear fits to the first and second 500 ps are shown with dashed lines. The slight decrease in the total number of carbon atoms (black line) is due to the lower carbon atom density of the solvent.