Efficient and accurate implicit solvent models are important tools for boosting the time- and length-scale achievable using molecular simulations. A recent implicit solvent model, denoted IS-SPA, presents an appealing platform to achieve this goal. The model utilizes atomistically parameterized solvent radial distribution functions and the Kirkwood superposition approximation to approximate the solvent density around a solute configuration. The mean solvent force resulting from this distribution is sampled using a novel Monte Carlo sampling procedure. Here, we demonstrate that this algorithm is readily portable to graphic processing units using straightforward CUDA code. We start be determining practical cutoffs for the model to allow the use of standard neighboring listing special decomposition schemes. Subsequently, a three kernel algorithm is presented that allows for enhanced memory coalescence and ultimately improves efficiency over more naïve approaches. Finally, the code is used to investigate the self-assembly of alanine dipeptide at experimentally relevant concentrations.

Developing a CUDA Algorithm for implicit solvation using the superposition approximation (IS-SPA)

ACS Spring Meeting 2021

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.