Improving Moderate Density Polymer Simulation Initialization Routines
Faculty Mentor Information
Dr. Eric Jankowski, Boise State University; and Stephanie McCallum, Boise State University
Presentation Date
7-15-2026
Abstract
Initialization of systems of polymers is a ubiquitous, tedious, necessary step to run molecular simulations. Standard methods of initialization typically pack straight polymer chains in a box at a low density, using PACKMOL or lattices, followed by simulations that shrink the box to the target density. These simulations can be long and numerically unstable, which may necessitate multiple attempts or further relaxations.We present a two-step initialization routine: We first generate polymer chains at the desired density, but with quickly-generated, possibly-overlapping configuration. We next relax this configuration using dissipative particle dynamics. We sample roughly 25,000 model parameterizations for systems of 100 100-mers on both consumer grade laptop hardware and graphics processing units -on a high performance computing cluster to optimize initialization. We find a parameter set that initializes100,100-mers at number density 0.85 at 0.0002 seconds per particle on an Nvidia V100 GPU, which is over 20 times faster than a PACKMOL initialization of the same system size and final density on the same hardware, which averages 0.0044 seconds per particle.
Improving Moderate Density Polymer Simulation Initialization Routines
Initialization of systems of polymers is a ubiquitous, tedious, necessary step to run molecular simulations. Standard methods of initialization typically pack straight polymer chains in a box at a low density, using PACKMOL or lattices, followed by simulations that shrink the box to the target density. These simulations can be long and numerically unstable, which may necessitate multiple attempts or further relaxations.We present a two-step initialization routine: We first generate polymer chains at the desired density, but with quickly-generated, possibly-overlapping configuration. We next relax this configuration using dissipative particle dynamics. We sample roughly 25,000 model parameterizations for systems of 100 100-mers on both consumer grade laptop hardware and graphics processing units -on a high performance computing cluster to optimize initialization. We find a parameter set that initializes100,100-mers at number density 0.85 at 0.0002 seconds per particle on an Nvidia V100 GPU, which is over 20 times faster than a PACKMOL initialization of the same system size and final density on the same hardware, which averages 0.0044 seconds per particle.