Research Initiation Award: The Rapid Evaluation and Global Minimization of Potential Energy Functions in Molecular and Protein Conformations

薛 9409285这个项目的目标是产生更有效的顺序和并行算法和软件的快速评估和全球最小化的非凸势能函数，出现在分子和蛋白质构象问题。 本研究将联合收割机结合快速多极子演算法与自动微分演算法，以产生一个n粒子系统的函数评估与梯度向量评估的O（n）运算演算法。 由于所有现有的算法需要至少O（n2）的操作来评估梯度向量，这项研究将改善几乎所有现有的分子和蛋白质构象的函数和梯度的重复评估是必需的。 这种成功依赖于两个编程范式：离散连续编程范式，使用物理知识的问题，以指导连续最小化搜索，和并行两级模拟退火编程范式，提供负载平衡动态和移动从一个本地极小到另一个努力找到一个全球最小。

Xue 9409285 The objective of this project is to produce more effective sequential and parallel algorithms and software for the rapid evaluation and global minimization of nonconvex potential energy functions that arise in molecular and protein conformation problems. The proposed research will combine the fast multipole algorithms and automatic differentiation to produce O (n) operation algorithms for both the function evaluation and the gradient vector evaluation for a system of n particles. Since all existing algorithms require at least O (n2) operations to evaluate the gradient vector, this research will improve almost all existing algorithms for molecular and protein conformations where repeated evaluation of the function and gradient is required. This success relies on two programming paradigms: a discrete-continuous programming paradigm that uses physical knowledge on the problem to guide continuous minimization searchers, and a parallel two-level simulated annealing programming paradigm which provides load balancing dynamically and moves from one local minimizer to another in an effort to locate a global minimizer.