Algorithms for Global Optimization, Enhanced Sampling and Reaction Pathway Determination

约翰·斯特劳布得到了理论与计算化学计划的资助，继续他在开发全局优化、增强采样和确定反应途径的新算法方面的工作。将探讨算法开发的两个领域。第一个领域涉及与Gibbs-Boltzmann统计相反的广义统计的使用，广义统计可用于描述在标准统计中具有强局域化的系统的配置的高度离域分布。这种技术可以提高发现低能量结构的可能性。第二个研究领域涉及一种被称为量子力学退火的优化技术，在这种技术中，普朗克常数的值被大大夸大，导致系统在空间中“扩散”，并通过穿越能量障碍更有效地寻求最低能量配置，否则很难爬过能量障碍。该技术以一种新颖的方式利用虚时路径积分量子模拟。应用包括：1)模型蛋白折叠；2)最优反应路径的确定；3)增强了分子构型的采样以计算平均热力学性质。计算机模拟提供了许多分子水平的见解，以了解发生在溶液和表面上的化学过程。这项技术非常强大，但受到与当前超级计算机的速度和执行此类模拟的可用时间相关的统计抽样问题的限制。斯特劳布正在开发一种执行分子模拟的新算法，这应该是统计抽样的重大改进。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采样统计所需的计算资源。

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.