On the performance of molecular dynamics applications on current high-end systems

The effective exploitation of current high performance computing (HPC) platforms in molecular simulation relies on the ability of the present generation of parallel molecular dynamics code to make effective utilisation of these platforms and their components, including CPUs and memory. In this paper, we investigate the efficiency and scaling of a series of popular molecular dynamics codes on the UK's national HPC resources, an IBM p690+ cluster and an SGI Altix 3700. Focusing primarily on the AMBER, DL_POLY and NAMD simulation codes, we demonstrate the major performance and scalability advantages that arise through a distributed, rather than a replicated data approach.     

Developing community codes for materials modeling

For this article, we call scientific software a community code if it is freely available, written by a team of developers who welcome user input, and has attracted users beyond the developers. There are obviously many such materials modeling codes. The authors have been part of such efforts for many years in the field of atomistic simulation, specifically for two community codes, the LAMMPS and GULP packages for molecular dynamics and lattice dynamics respectively. Here we highlight lessons we have learned about how to create such codes and the pros and cons of being part of a community effort. Many of our experiences are similar, but we also have some differences of opinion (like modeling vs modelling). Our hope is that readers will find these lessons useful as they design, implement, and distribute their own materials modelling software for others to use.     

Molecular dynamics simulation of a DNA containing a single strand break

Molecular dynamics simulations were performed for a dodecamer DNA containing a single strand break (SSB), which has been represented by a 3'-OH deoxyribose and 5'-OH phosphate in the middle of the strand. Molecular force field parameters of the 5'-OH phosphate region were determined from an ab initio calculation at the HF/6-31G level using the program package GAMESS. The DNA was placed in a periodic boundary box with water molecules and Na+ counter-ions to produce a neutralised system. After minimisation, the system was heated to 300 K, equilibrated and a production run at constant NTP was executed for 1 ns using AMBER 4.1. Snapshots of the SSB-containing DNA and a detailed analysis of the equilibrated average structure revealed surprisingly small conformational changes compared to normal DNA. However, dynamic properties calculated using the essential dynamics method showed some features that may be important for the recognition of this damage by repair enzymes.     

Diffusion mechanisms at metallic grain boundaries

Summary As the sizes of electronic devices continue to shrink, understanding key atomic phenomena vital to macroscopic processes become increasingly important. Mass transport along grain boundaries (GBs) is such a key process. We have studied diffusion mechanisms at metallic GBs of Ag and Al with the embedded-atom method, with molecular statics (MS) and molecular dynamics (MD), as well as with massively parallel computers at Oak Ridge National Laboratory and Sandia National Laboratory. Formation and migration energies of interstitials and vacancies at the optimal symmetric tilt GBs of different angles in Ag were first obtained using MS at 0 K. Extensive MD simulations were then carried out for selected Ag GBs using massively parallel computers at different finite temperatures up to the melting point. In this way, we were able to determine the dominant diffusion mechanism within different temperature regimes by comparison of the activation energies from MS results for the identified diffusion processes with the MD simulation results. For the first time, the results of this study on Ag GBs had provided realistic explanation and simulation-based evidence for the discrepancy between the activation energies from the recent low-temperature experiments by Ma and Balluffi and those at high temperatures reported in the literature. Preliminary MD results on Al GB diffusion are in excellent agreement with experiment and on-going work on Al and Al-Cu systems aimed to further understand electromigration phenomena will be briefly discussed.     

Hybrid molecular-continuum fluid models: implementation within a general coupling framework

Over the past three years we have been developing a new approach for the modelling and simulation of complex fluids. This approach is based on a multiscale hybrid scheme, in which two or more contiguous subdomains are dynamically coupled together. One subdomain is described by molecular dynamics while the other is described by continuum fluid dynamics; such coupled models are of considerable importance for the study of fluid dynamics problems in which only a restricted aspect requires a fully molecular representation. Our model is representative of the generic set of coupled models whose algorithmic structure presents interesting opportunities for deployment on a range of architectures including computational grids. Here we describe the implementation of our HybridMD code within a coupling framework that facilitates flexible deployment on such architectures.     

Long-term dynamic stability of discrete dislocations in graphene at finite temperature

We present an assessment of the finite-temperature dynamical stability of discrete dislocations in graphene. In order to ascertain stability, we insert discrete dislocation quadrupole configurations into molecular dynamics calculations as initial conditions. In calculations we use Sandia National Laboratories Large-scale Atomic/Molecular Massively Parallel Simulator (LAMMPS) and the Adaptive Intermolecular Reactive Empirical Bond-Order (AIREBO) potential. The analysis shows that the core structures predicted by discrete dislocation theory are dynamically stable up to temperatures of 2,500 K, though they tend to relax somewhat in the course of molecular dynamics. In addition, we find that discrete dislocation theory accurately predicts energies, though it exhibits a slight overly-stiff bias.     

Large scale molecular dynamics simulation of native and mutant dihydropteroate synthase-sulphanilamide complexes suggests the molecular basis for dihydropteroate synthase drug resistance

Antibiotic resistance is hampering the efficacy of drugs in the treatment of several pathological infections. Dihydropteroate synthase (DHPS) has been targeted by sulphonamide inhibitors for the past 60 years and has developed different amino acid mutations to survive sulpha drug action. We couple homology modelling techniques and massively parallel molecular dynamics simulations to study both the drug-bound and apo forms of native and mutant DHPS. Simulations of the complex between sulphanilamide and Streptomyces pneumoniae, DHPS shows how sulphanilamide is able to position itself close to 6-hydroxymethyl-7, 8-dihydropteridine-phosphate in a suitable position for the enzymatic transformation whereas in the mutant complex the sulpha drug is expelled from the catalytic site. Our simulations, therefore, provide insight into the molecular basis for drug resistance with S. pneumoniae DHPS.     

Scalable concurrent computing

This paper focusses on the challenge of building and programming scalable concurrent computers. The paper describes the inadequacy of current models of computing for programming massively parallel computers and discusses three universal models of concurrent computing — developed respectively by programming, architecture and algorithm perspectives. These models provide a powerful representation for parallel computing and are shown to be quite close. Issues in building systems architectures which efficiently represent and utilize parallel hardware resources are then discussed. Finally, we argue that by using a flexible universal programming model, an environment supporting heterogeneous programming languages can be developed.     

Empirical force field for the simulation of a class of chromophores in a photosynthetic center

We present the results of the parametrization of the AMBER molecular mechanical force field for simulating Bacteriochlorophylls (BChls) and Bacteriopheophytins (BPhs). Ab initio Density Functional Theory (DFT) calculations on fragments and the entire chromophores in their gas-phase have been used to obtain a new set of force field parameters which are able to reproduce static properties and vibrational frequencies of the chromophores. Molecular dynamics (MD) simulations of crystalline BPh and BChl have been performed to test our intramolecular parameterization as well as the non-bonded parameters.     

高性能计算机的关键技术和发展趋势

介绍高性能计算机的关键技术和发展趋势。简要回顾高性能计算机的发展历史和当前形势,重点讨论大规模并行处理（MPP）所面临的挑战,包括可扩展性、友善性和可用性。介绍神威高性能计算机及其应用情况,并对如何发展我国高性能计算机提出一些初浅的看法。     

Simulations of the elastic response of single-walled carbon nanotubes

Using the Tersoff-Brenner potential we have performed molecular dynamics simulations of nanotubes under axial strain, analyzing both compression and stretching forces. These large-scale simulations were carried out on a MasPar massively parallel computer. The elastic response is investigated and expressions for various elastic constants are derived from the simulations. Typical failure modes are also shown and discussed.     

Atomistic simulations of fatigue crack growth and the associated fatigue crack tip stress evolution in magnesium single crystals

Using Large-scale Atomic Molecular Massively Parallel Simulator (LAMMPS), a classical molecular dynamics code, atomistic simulations were performed to investigate the fatigue crack growth rate and the evolution of the associated atomic stress fields near the crack tip during fatigue crack growth in magnesium single crystals. The interatomic bonds of atoms were described using the EAM potential. The specimens with initial edge cracks were subjected to uniaxial Mode I cyclic loading. For the sake of revealing the influence of the initial cracks’ crystal orientations, three different orientations were considered. The fatigue growth rate can be expressed by da/dN = cCTOD, where the values of constant c are determined by the atomistic simulations. Notably, the values of the constant c are much larger for magnesium single crystals than for FCC single crystals and vary widely from one orientation to another. The simulation results show that the evolution of atomic stress fields was highly dependent on the crystal orientations due to anisotropy and magnesium single crystals’ HCP structure. Interestingly, the von Mises stress or normal stress around the crack tip controlled the fatigue crack growth behaviors.     

Feature Activated Molecular Dynamics: Parallelization and Application to Systems with Globally Varying Mechanical Fields

A recently developed hybrid molecular dynamics method (Feature Activated Molecular Dynamics, or FAMD), which was originally designed to extend the scope of certain types of molecular dynamics simulations, is extended here in two ways. First, the method is modified to execute on parallel computer architectures using the MPI communication interface. The parallel FAMD algorithm is demonstrated to be computationally efficient and to substantially increase the length scales accessible with molecular dynamics. The performance of the parallel algorithm is demonstrated using a crystalline system containing 1× 10⁶ atoms, in which 1000 supersaturated self-interstitials are introduced and allowed to aggregate for about 4 ns. In the second part of this paper, the FAMD method is applied to problems in which spatio-temporally varying stress fields are present throughout the simulation cell. In particular, we consider the evolution of a spherical void in a hydrostatically stressed silicon crystal and show that the method can capture the extremely rapid void cavitation dynamics following material failure. Once again, the FAMD approach is demonstrated to provide substantial computational advantages over standard molecular dynamics.     

A parallel contact detection algorithm for transient solid dynamics simulations using PRONTO3D

An efficient, scalable, parallel algorithm for treating material surface contacts in solid mechanics finite element programs has been implemented in a modular way for multiple-instruction, multiple-data (MIMD) parallel computers. The serial contact detection algorithm that was developed previously for the transient dynamics finite element code PRONTO3D has been extended for use in parallel computation by utilizing a dynamic (adaptive) load balancing algorithm. This approach is scalable to thousands of computational nodes¹     

Massively parallel computer simulations of fullerenes and Si-clusters

A general overview of cluster and bulk molecular dynamics algorithms for parallel computers is given. Specific results are presented for the MasPar MP-1 (DECmpp 12000) series of computers, which possess a single-instruction multiple-data (SIMD) architecture with processors arranged in a two-dimensional mesh. Many-body potentials used were selected to describe CC- and Si-clusters and ground-states were determined through simulated annealing. The difficulty of attaining the absolute minimum in this way for larger clusters is discussed and some possible remedies are suggested.     

Capturing nanoscale effects by peridynamics

Molecular dynamic simulations inevitably demand large computational resources for structures of liner measures even as small as a few tens or hundreds of nanometers. Thus, a computationally efficient method to simulate larger structures and, at the same time, retain the properties and the mechanical response at the atomic scale is in demand. One such approach is peridynamics, which is a nonlocal extension of continuum mechanics. In this study, we investigate the possibility to efficiently reproduce results from molecular dynamic (MD) simulations by calibration of two parameters inherent in peridynamics: the length scale parameter and the interparticle bond strength. The free-ware LAMMPS supports both numerical approaches, and thus LAMMPS has been used as the common framework. Beams of single-crystal fcc copper of various sizes and under tension along the crystallographic [100]- and [110]-directions act as the modeling example. The force–displacement curves and the elastic–plastic transitions have been compared between the approaches. The conclusion is that proper calibration of the peridynamic two parameters to MD simulations results in proper reproduction of the molecular dynamic results. This in turn allows for geometrical upscaling or simulation of geometrically more complicated structures, without loss of features derived from the atomic scale but to a much lower computational cost.     

Molecular dynamics simulations of cluster collisions

Molecular dynamics simulations are very powerful to get more insight into the dynamics of collision processes. Simulations of such processes have been successfully performed using empirical interaction potentials. More recently the combination of the Density-Functional-Theory within the Local Density Approximation (DFT-LDA) with Molecular Dynamics (MD) has been realized. A simplified LCAO-DFT-LDA scheme, which enables to consider the electronic states in the calculation of the forces on the atoms, was developed and applied for simulations of molecule-cluster and cluster-cluster collisions.

In the present paper the authors demonstrate the extension of such calculations under the consideration of the statistics in cluster collision processes by using parallel computer techniques.     

Molecular dynamics simulations of end-tethered single-stranded DNA probes on a silica surface

The structure of DNA molecules tethered to surfaces may significantly affect the efficiency of hybridization on the DNA microarray. Understanding the structure of single-stranded DNA (ssDNA) tethered to surfaces is critical for applying the molecular recognition function of DNA microarrays. Although a number of experimental methods have been applied to determine the structure of the DNA probe on surfaces, they can not provide enough information on the dynamical behavior of the ssDNAs on surfaces. Herein, we investigated the dynamics and interaction of seven DNA probes tethered on a silica surface by a molecular dynamics simulation. From the simulation results, we examined the structure and dynamics of the ssDNAs, by calculating the root-mean-square derivations, the tilt angles, the radius of gyration, and the distances of the neighboring ssDNAs. The data obtained from our simulation suggests the packing density has a significant effect on the overall structure and molecular orientation change of surface-tethered ssDNAs, which is complementary to the recent experimental reports. Our simulation provided a structural insight, which is helpful to better understand the behavior of ssDNA on surfaces and optimize the design of DNA microarray.     

Parallel finite element fluid analysis on an element-by-element basis

A finite element fluid analysis code, which is based on an element-by-element scheme and the matrix-storage free formulation, is developed and implemented to the massively parallel computer; KSR1. Since the element-by-element scheme coupled with the CG-type iterative solver is suitable for parallel processing, the matrix-storage free formulation will enable the large-scale computation within a reasonable time.After the verification of the code by some numerical examples, the cavity flow and cyclinder flow, the parallel efficiency is discussed. In the analysis of cavity flow, the speed-up using 16 CPUs is 15.35, which corresponds to the parallel efficiency of 95.9%.     

6H-SiC辐照点缺陷诱发化学无序的分子动力学分析

为配合6H-SiC中化学无序对其导电性能影响的研究, 本研究运用经典分子动力学方法, 采用LAMMPS软件对6H-SiC的线性级联碰撞过程进行了模拟, 给出了在不同能量、不同种类PKA(Primary Knock-on Atom)的情况下, 6H-SiC单次线性级联碰撞和多次线性级联碰撞过程中主要点缺陷的演化过程, 并统计了多次级联碰撞后化学无序的演化和六种点缺陷各自最终所占的比例。结果表明, 级联碰撞产生的Si-Si键比C-C键更易形成且更加稳定, Si-Si键主要由Si_C反位缺陷形成, C-C键主要由C间隙原子聚集形成, PKA的种类及初始能量影响点缺陷的产额和化学无序的程度, 但不影响六种点缺陷各自的占比。