Algorithms for Global Optimization, Enhanced Sampling and Reaction Pathway Determination

John Straub得到了理论和计算化学计划的资助，继续他的工作，开发用于全局优化、增强采样和确定反应路径的新算法。将探索算法开发的两个领域。第一个领域涉及广义统计的使用，而不是Gibbs-Boltzmann统计，后者可以用来描述在标准统计中强烈局域的系统的高度离域分布。这项技术导致发现低能量构型的可能性更高。第二个研究领域涉及一种被称为量子力学退火法的最优化技术，在这种技术中，普朗克常数的值被极大地夸大，导致系统在空间中展开，并通过隧道穿越原本难以逾越的能量屏障，更有效地寻找最低的能量配置。该技术以一种新颖的方式使用了虚时间路径积分量子模拟。应用包括：1)蛋白质折叠模型；2)最佳反应路径确定；以及3)用于计算平均热力学性质的分子构型的增强采样。计算机模拟已经为溶液中和表面上发生的化学过程提供了许多分子水平的见解。这项技术非常强大，但受到与当前超级计算机的速度和可用于执行此类模拟的时间相关的统计抽样问题的限制。Straub正在开发一种执行分子模拟的新算法，这应该是统计抽样方面的一个重大改进。如果成功，这种新算法可以显著减少在分子模拟中获得有意义的抽样统计所需的计算资源量。

John Straub is supported by a grant from the Theoretical and Computational Chemistry Program to continue his work on developing new algorithms for global optimization, enhanced sampling and the determination of reaction pathways. Two areas of algorithmic development will be explored. The first area involves the use of generalized statistics, as opposed to Gibbs-Boltzmann statistics, which can be used to describe highly delocalized distributions of configurations for systems that are strongly localized in the standard statistics. This technique leads to a higher likelihood of finding low energy configurations. The second area of research involves an optimization technique known as quantum mechanical annealing, in which the value of Planck's constant is greatly exaggerated causing the system to `spread out` in space and more effectively seek the lowest energy configurations by tunneling through energy barriers that would otherwise be hard to climb over. This technique employs imaginary-time path integral quantum simulations in a novel way. Applications include: 1) model protein folding; 2) optimal reaction path determination; and 3) enhanced sampling of molecular configurations for the computation of average thermodynamic properties. Computer simulations have provided a number of molecular level insights into chemical processes which occur in solution and on surfaces. The technique is very powerful, but is limited by statistical sampling problems associated with the speed of current supercomputers and the time available to perform such simulations. Straub is developing a new algorithm for performing molecular simulations which should be a major improvement in statistical sampling. If successful, this new algorithm could provide a significant decrease in the amount of computational resources required to obtain meaningful sampling statistics in molecular simulations.