Development of Parallel Search Algorithms

并行搜索算法的开发

• 批准号: EP/F01130X/1
• 负责人: Adrian Wander
• 金额: $ 18.16万
• 依托单位: Science and Technology Facilities Council
• 依托单位国家: 英国
• 项目类别: Research Grant
• 财政年份: 2007
• 资助国家: 英国
• 起止时间: 2007 至 无数据
• 项目状态: 已结题
• 来源: https://gtr.ukri.org/projects?ref=EP%2FF01130X%2F1
• 关键词: Development; Parallel; Search; Algorithms

Many computational tools in the materials and chemistry discipline make use of geometry optimisation techniques, often minimisation of the energy with respect to the atomic positions to find equilibrium structures, or saddle points on the potential energy surface to locate transition states for a chemical reaction. Traditionally these stationary points are found by performing energy and force calculation at an initial, guessed geometry and then approaching the stationary point by successive, sequentially calculations on new geometries, hopefully converging to the required stationary point.The availibility of large scale parallel computers, with tens of 1000s of processors offers both challenges and opportunities to the computational chemistry and materials science disciplines. The chief challenge is that the effective exploitation of such a machine requires software to have an extremely high parallel efficiency and for many modern techniques this is very difficult to achieve when computing the energy and forces for a given geometry. One way to increase the level of parallelism is to consider multiple geometries simultaneously and we propose to explore the possibility by implementing and testing a number of parallel geometry optimisation algorithms. These will include genetic algorithms and stochastic searches as well the exploration of multiple starting points. The evaluation of second derivatives by finite difference of first derivatives (forces), while expensive, may become possible when large scale parallelism is available.We will develop the new code in such a way that it can be used within a variety of computational science packages, including the materials chemistry codes CASTEP and CRYSTAL, the classical forcefield MD code DL_POLY, the ab initio quantum chemistry package GAMESS-UK and ChemShell which performs mixed Quantum/Classical calculations (for example, combining GAMESS-UK and DL_POLY). The resulting codes will accellerate the solution of many materials modelling and computational chemistry problems on large-scale facilities such as the UK's next national facility (HeCTOR) and similar facilities worldwide.

材料和化学学科中的许多计算工具利用几何优化技术，通常最小化相对于原子位置的能量以找到平衡结构，或势能面上的鞍点以定位化学反应的过渡态。传统上，这些稳定点是通过在一个初始的，猜测的几何形状进行能量和力的计算，然后通过连续的，顺序的计算新的几何形状接近稳定点，希望收敛到所需的稳定点。大规模并行计算机的可用性，与几十个1000的处理器提供了计算化学和材料科学学科的挑战和机遇。主要的挑战是，这种机器的有效利用需要软件具有极高的并行效率，对于许多现代技术，在计算给定几何形状的能量和力时，这是很难实现的。提高并行度的一种方法是同时考虑多个几何形状，我们建议通过实施和测试一些并行几何优化算法来探索这种可能性。这些将包括遗传算法和随机搜索以及多个起点的探索。通过一阶导数（力）的有限差分来计算二阶导数虽然昂贵，但当大规模并行可用时，可能成为可能。我们将开发新的代码，使其可以在各种计算科学软件包中使用，包括材料化学代码CASTEP和CRYSTAL，经典力场MD代码DL_POLY，从头算量子化学软件包GAMESS-UK和ChemShell，可执行混合量子/经典计算（例如，结合GAMESS-UK和DL_POLY）。由此产生的代码将加速解决许多材料建模和计算化学问题的大型设施，如英国的下一个国家设施（HeCTOR）和世界各地的类似设施。