DL_POLY version 4: a major shift in length- and time-scale limitations in Molecular Dynamics simulations of heterogeneous phenomena

用原子和分子分辨率模拟固体、液体、膜或纳米粒子等物质是过去30年来的一个主要发展领域。计算方法和硬件的发展意味着，这样的分子模拟现在可以以10年前无法想象的真实感和准确性进行。然而，在长度和时间尺度上仍然有基本的限制，可以用分子模拟来探索。这些阻碍了科学家认识到分子模拟的全部力量。高性能计算的改进已经到了我们可以开始消除这些限制的阶段。特别是，大规模并行处理器(MPP)的出现，其中100个或1000个处理器被同时使用，这意味着可以使用分子模拟来模拟具有更真实长度尺度的系统。到目前为止，这种进展主要应用于分子均匀分布在整个材料中的材料：最常见的材料(金属、陶瓷、塑料、液体)都是这种形式。然而，在许多非常重要的材料中，分子分布不均匀/要么是因为它们涉及不同相的混合物(例如在水和空气之间的界面上)，要么是因为非常不同性质的颗粒混合在一起(如许多催化剂、纳米颗粒、聚合物复合材料和骨等天然材料)。在这种不均匀的材料中，重要性质显现的长度和时间尺度通常都很长，因此确实需要在最新的MPP计算机上使用快速高效的代码来准确建模。赫克托的出现为在分子水平上对非均质材料建模提供了一个独特的机会，从而获得了对这些材料的性质以及如何控制它们的性质的迄今无法获得的见解。不幸的是，现有的能够对各种重要的非均质材料进行模拟的分子模拟软件中，没有一种能够充分扩展到赫克托能力的机器。因此，如果要抓住赫克托提供的机会，就迫切需要对现有的分子模拟包进行重大修改。这个项目的目的是开发这个新的代码，并优化在赫克托上的使用。该项目将从现有的程序包DL_Poly开始，该程序包在英国编写，并在以前的高端计算机设施中广泛使用，以模拟同质材料。将对DL_POLY进行几个主要修改，以使程序能够：(I)高效地对具有或演化为高度非均匀结构的系统进行建模；(Ii)通过一系列非常快但非常不频繁的原子跳跃来对实时演化非常缓慢的系统进行建模(这是大多数固态系统随时间变化的基础)；以及(Iii)利用Hector的体系结构来获得最大性能。新计划的威力将在以下方面得到展示：分布在空气-水界面上的纳米粒子的性质、分子晶体和框架材料的成核、离子通过特殊玻璃的流动性以及蛋白质在控制蛋壳形成中的作用。

The simulation of matter/such as solids, liquids, membranes or nanoparticles/with atomic and molecular resolution has been a major growth area over the last 30 years. Developments in the computational methods and hardware have meant that such molecular simulations can now be performed with a level of realism and accuracy that was unimaginable even 10 years ago. However, there are still fundamental limits on the length- and time-scales that can be probed with molecular simulation. These are preventing scientists from realising the full power of molecular simulation.Improvements in high performance computing have got to the stage where we can begin to remove these limitations. In particular, the advent massively parallel processors (MPP), where 100s or 1000s of processors are used simultaneously, have meant that systems with much more realistic lengthscales can be modelled using molecular simulations. To date such progress has been applied mainly to materials in which the molecules are spread evenly throughout the material: most common materials (metals, ceramics, plastics, liquids) are of this form. However, there are many very important materials where the molecules are spread unevenly/either because they involve a mixture of different phases (such as at the interface between water and air) or particles of very different character are mixed together (as occurs with many catalysts, nanoparticles, polymer composites and natural materials such as bone). In such inhomogeneous materials, both the length- and time-scale on which important properties are manifest are often very long, and so really demand fast efficient code on the latest MPP computers to model accurately. The advent of HECToR opens a unique opportunity to model inhomogeneous materials at a molecular level, and thereby gain hitherto inaccessible insights into the nature of these materials and how their properties can be controlled. Unfortunately, of the existing molecular simulation software that is capable of modelling the full range of important inhomogeneous materials, none will scale adequately to a machine of the power of HECToR. Thus, major revisions of an existing molecular simulation package are needed urgently if the opportunities offered by HECToR are to be grasped.The purpose of this project is to develop this new code and optimise for use on HECToR. The project will start with an existing package, DL_POLY, written in the UK and used extensively on previous high-end computer facilities to model homogeneous materials. Several major modifications will be made to DL_POLY that will enable the program: (i) to model efficiently systems which either possess, or evolve into, highly nonuniform structures; (ii) to model systems which evolve very slowly in real time through a series of very fast but very infrequent atomic jumps (this is the basis on which most solid state systems change over time); and (iii) to exploit the architecture of HECToR to gain maximum performance. The power of the new program will be demonstrated in calculations on the properties of nanoparticles distributed on an air-water interface, nucleation of molecular crystals and framework materials, the mobility of ions through specialised glasses, and the role of proteins in controlling eggshell formation.