Advanced adaptive resolution methods for molecular simulation
Date issued
Authors
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
License
Abstract
Computer simulations have become a fundamental tool in molecular soft matter research. For example, they allow the investigation of microscopic details that are not accessible experimentally. However, even the most powerful supercomputers cannot simulate large systems on long time scales with arbitrarily high accuracy. Therefore, a large variety of computational methods have been developed, which range from highly accurate but numerically expensive quantum calculations to efficient but less detailed coarse-grained approaches.
A particular challenge is posed by systems whose macroscopic behavior is sensitively dependent on specific microscopic details. For these multiscale systems, simulation techniques are being developed that concurrently use two or more models with different computational complexity and accuracy in single simulations. In this way, one can use a demanding high-resolution model in a small but relevant subregion, while allowing overall large and long simulations by describing the rest of the system with an efficient, less detailed level. One such method is the adaptive resolution simulation scheme. In this approach, a predefined region of interest, for example a protein in a large box of solvent, is modeled atomistically while molecules far away from this region are described by a coarse-grained force field. A unique feature of the technique is that particles traveling between the two regions change their resolution on the fly, such that the high-resolution subsystem behaves as if embedded in an overall high-resolution environment, although at a significantly lower computational cost.
In this work, we first study the ability of adaptive resolution simulation methods to combine atomistic and coarse-grained models with very different thermodynamic properties. By coupling a highly structured liquid like water with an ideal gas of non-interacting particles, we demonstrate that the details of the coarse-grained region have a surprisingly small effect on the accuracy of the high resolution region. Next, we investigate the theoretical basis behind the described adaptive resolution approaches and show how a single unifying framework can be used to derive different kinds of adaptive resolution schemes. We also demonstrate that the relative entropy—a quantity characterizing differences in the configurational probability distributions of the models in the different regions—can be used as a guide to set up adaptive resolution simulations in an optimal manner. Furthermore, we devise a simulation algorithm that enables atomistic regions with arbitrary geometry and which can adapt during the simulation to follow, for example, the conformational change of a large biomolecule. Finally, we derive a new adaptive resolution scheme that allows a clean coupling of quantum mechanical path integral and classical atomistic models.
The results significantly advance the current state of adaptive resolution simulation methodologies both on the theoretical and the practical front. They shed light on the fundamentals behind such methods and enable more efficient computer simulations of relevant multiscale systems, such as complex biomolecules, membranes, DNA, or polymeric materials.