Algorithms for enhanced sampling and global optimization

波士顿大学的约翰·斯特劳布（John Straub）得到了理论与计算化学计划的资助，继续他在增强采样和全局优化算法方面的研究。这种算法在生物分子、低温团簇和过冷液体的模拟中有广泛的应用。本研究的目标是开发能够有效采样复杂多维势能面的搜索算法。理想情况下，这种技术将1)分离出热力学上重要的低洼能量盆地，2)有效地对这些盆地进行采样，并以一种产生精确热力学平均值的方式在它们之间移动。提出了两种新的算法。第一种方法涉及一种新颖的蒙特卡罗方法，该方法允许以满足详细平衡的方式在低能量最小值的邻域之间进行长距离移动。第二种算法涉及全局优化的量子力学退火方法的持续发展。这涉及到费曼路径积分表示的重新表述，该表示使用有效势来允许超越标准量子离域的增强采样。提出了许多应用，包括：1)模拟溶剂化肽；2)低温分子团簇热力学平均值的计算；3)肌红蛋白中配体再结合机制的评价。由于计算机的可用性和速度的急剧提高，以及用于模拟动力和热力学现象的有效算法的发展，凝聚态系统的模拟已经向前发展。生物分子、过冷和玻璃成型系统的特性具有特别的基础和实际意义。对于这样的系统，隔离最低自由能构象或计算热力学平均值的任务，会因为崎岖的能量格局而受挫。斯特劳布正在开发的新技术应该在生物系统的模拟中得到广泛的应用。

John Straub of Boston University is supported by a grant from the Theoretical and Computational Chemistry Program to continue his research in algorithms for enhanced sampling and global optimization. Such algorithms have extensive applications in simulations involving biomolecules, low temperature clusters, and supercooled liquids. The goal of this research is to develop search algorithms that can sample complex multidimensional potential energy surfaces efficiently. Ideally such a technique would 1) isolate the thermodynamically important low lying energy basins, and 2) effectively sample those basins and move between them in a way that produces exact thermodynamic averages. Two new algorithms are proposed. The first involves a novel Monte Carlo method that allows long range moves between neighborhoods of low energy minima in a way that satisfies detailed balance. The second algorithm involves the continued development of quantum mechanical annealing methods for global optimization. This involves a reformulation of the Feynman path integral representation that uses an effective potential to allow enhanced sampling beyond that of the standard quantum delocalization. A number of applications are proposed including: 1) the simulation of solvated peptides; 2) the computation of thermodynamic averages for low temperature molecular clusters; and 3) the evaluation of the mechanism for ligand rebinding in myoglobin.The simulation of condensed matter systems has been moved forward by dramatic increases in the availability and speed of computers and the development of effective algorithms for the simulation of dynamical and thermodynamic phenomena. Of particular fundamental and practical interest are the properties of biomolecular, supercooled and glass forming systems. For such systems, the task of isolating conformations of lowest free energy or computing thermodynamic averages is frustrated by the rugged energy landscape. The new techniques being developed by Straub should find widespread applications in the simulation of biological systems.