Title:	Large scale molecular simulation
DNr:	SNIC 2016/10-47
Project Type:	SNIC Large Compute
Principal Investigator:	Berk Hess <hess@kth.se>
Affiliation:	Kungliga Tekniska högskolan
Duration:	2016-07-01 – 2017-07-01
Classification:	20301 10402
Homepage:	http://tcbl10.scilifelab.se/~hess/
Keywords:

Abstract

The research in my group focuses on algorithms as well as applications for large scale molecular dynamics (MD) simulations. During the few decades that have passed since birth of molecular simulation, the time scales one would like to simulate for many applications have always exceeded what is computationally possible by one or several orders of magnitude. This is still the case, even though the computational power has increased exponentially with Moore's law. The reason for is is that as computers and software get faster, new scientific problems come within reach. Currently the increase in computational power mainly comes from the increase in CPU and GPU core count. This means that scientific codes need to extract more parallelism from scientific problems to be able to make full use of current and future supercomputers. This is the main focus of the work funded by my ERC starting grant entitled ``Million-Core Molecular Simulation''. For particular applications often more acceleration can be achieved by altering the sampling. An important class of MD applications we are studying is the sampling conformational transitions in bio-molecules. Here crossing of (free-)energy barriers can be increased exponentially by adjusting the weights of the sampling, by providing some, limited, information on the pathway. Bio-molecular systems are one of the most important applications of MD, but their computational scaling is limited due to the fixed problem size. Here improved sampling techniques are the only way to significantly reduce the time to solution. One of the few fields where simulations of larger systems is useful is wetting. When a liquid wets a substrate, molecular processes at the three-phase contact line drive the dynamics of wetting. Since these processes have at molecular length and time scales, they are difficult to access experimentally. The consensus in the community is that only MD simulations can reveal the physics at the contact line. This is especially the case for electro-wetting, i.e. wetting driven by electric fields, on which the next phase of this application will focus. All this work in done using the open-source GROMACS molecular simulation package and all algorithmic improvements will be made directly available to the community.