This Title All WIREs
How to cite this WIREs title:
WIREs Comput Mol Sci
Impact Factor: 16.778

Algorithm improvements for molecular dynamics simulations

Full article on Wiley Online Library:   HTML PDF

Can't access this content? Tell your librarian.

Abstract High‐performance implementations of molecular dynamics (MD) simulations play an important role in the study of macromolecules. Recent advances in both hardware and simulation software have extended the accessible time scales significantly, but the more complex algorithms used in many codes today occasionally make it difficult to understand the program flow and data structures without at least some knowledge about the underlying ideas used to improve performance. In this review, we discuss some of the currently most important areas of algorithm improvement to accelerate MD, including floating‐point maths, techniques to accelerate nonbonded interactions, and methods to allow multiple or extended time steps. There is also a strong trend of increased parallelization on different levels, including both distributed memory domain decomposition, stream processing algorithms running, e.g., on graphics processing units hardware, and last but not least techniques to decouple simulations to enable massive parallelism on next‐generation supercomputers or distributed computing. We describe some of the impacts these algorithms are having in current performance, and also how we believe they can be combined in the future. © 2011 John Wiley & Sons, Ltd. WIREs Comput Mol Sci 2011 1 93–108 DOI: 10.1002/wcms.3 This article is categorized under: Structure and Mechanism > Molecular Structures Computer and Information Science > Computer Algorithms and Programming Molecular and Statistical Mechanics > Molecular Dynamics and Monte-Carlo Methods

Removing hydrogen angle vibrations in an NH3 group using virtual interaction sites. Two dummy masses (in yellow) are required to preserve the rotational degree of freedom of the NH3 group. The virtual site construction and force redistribution itself conserves energy perfectly, since it uses analytic derivative chain rules.

[ Normal View | Magnified View ]

(a) Subdivision of atoms on streaming architectures into blocks of 32 entries. The early NVIDIA approach evaluated the complete interaction matrix43 (left), although modern hardware makes it possible to use Newton's third law is used to exclude half the interactions (middle). In the Eastman and Pande47 approach, atoms are sorted and only tiles with distances within the cutoff are included, leading to a sparse interaction matrix (right). For large systems, many more tiles will be excluded, increasing performance. (b) Alternatively, the decomposition can be based on voxels as suggested by Anderson et al.,46 applying either all‐versus‐all interactions to voxel pairs, or using them to construct neighbor lists.

[ Normal View | Magnified View ]

Cutoff sphere partitioning shown in two‐dimensional. The actual simulation cell is shaded, and large numbers refer to the ‘shift index’ vector. Periodic boundaries means particle i will interact with copies of some neighbors (primed indices). The list with shift (0,0) will contain neighbors 1, 2, and 3, the list with shift (0,−1) particle 4, the list with index (1, −1) particle 5, and finally the neighbor list with shift (1,0) particle 6. Neighbor lists with other shifts are empty for particle i.

[ Normal View | Magnified View ]

A Markov state model (MSM) approach to protein folding. Kinetic clustering of states from initial simulations can be used to create ‘macrostates,’ after which the folding problem becomes a question of approximate first‐order transitions between macrostates. The figure shows representative such macrostates (14 out of 2000) from an MSM of global folding dynamics of NTL9 using 1.5 ms of aggregated simulation. States with low equilibrium free energy are drawn larger, and arrows represent superpositions of pathways between states, with the size proportional to the flux between states based on transition path theory. The two lowest‐free energy states in this model correspond to one ‘unfolded’ (red) and one ‘native’ state (green), but there is more than one possible folding/unfolding pathway between them. Figure kindly provided by Voelz et al.90

[ Normal View | Magnified View ]

Example of actual load balancing between cores in two‐dimensional (last dimension not shown). Some domains are smaller and others bigger to compensate for workload differences. For this case, the protein is significantly more expensive to calculate than the surrounding lipids, and the water around the membrane cheapest (not shown). Although the figure shows a cubic unit cell, the same principle applies to general triclinic cells.

[ Normal View | Magnified View ]

The communication volume of three spatial decomposition methods in two‐dimensional where the cutoff distance (rc) is smaller than the domain size.

[ Normal View | Magnified View ]

Browse by Topic

Molecular and Statistical Mechanics > Molecular Dynamics and Monte-Carlo Methods
Computer and Information Science > Computer Algorithms and Programming
Structure and Mechanism > Molecular Structures

Access to this WIREs title is by subscription only.

Recommend to Your
Librarian Now!

The latest WIREs articles in your inbox

Sign Up for Article Alerts