Ab Initio Molecular Electronic Structure Computation on Parallel Computers

在理论与计算化学项目和先进科学计算学部新技术项目的支持下，King教授和Furlani博士将在分布式内存MIMD计算机上实现大规模的从头算电子结构计算。算法将被设计为实现高效的处理器间通信、负载平衡、消息传递与计算的重叠以及机器可移植性。将为非常大的轨道基集开发一个直接的自一致域程序，例如，N=4000，在具有100个或更多处理器的机器上，具有5兆字节的本地内存，没有磁盘存储。积分求值和矩阵对角化计算负荷以及矩阵存储存储器将被分配到处理器上。应用将包括具有生物学意义的大分子和松散结合的分子簇。为了将精确的组态相互作用计算结果外推到完整的轨道基集极限，将研究小原子、分子和可溶模型系统中的短程动态电子相关。这个项目的结果将使具有并行结构的超级计算机对化学家更有用，使其有可能在更大的化学有趣的系统上进行理论计算。

With support from the Theoretical and Computational Chemistry Program, and of the New Technologies Program of the Division of Advanced Scientific Computing, Professor King and Dr. Furlani will implement large scale ab-initio electronic structure computations on distributed memory MIMD computers. Algorithms will be designed to achieve efficient interprocessor communications, load balance, overlapping of message passing with computation, and machine portability. A direct self-consistent-field program will be developed for very large orbital basis sets, e.g., N=4000, on machines with a hundred or more processors, with 5 MBytes of local memory and no disk storage. The integral evaluation and matrix diagonalization computational load as well as the matrix storage memory will be distributed over the processors. Applications will include large molecules of biological interest and loosely bound molecular clusters. Short range dynamic electron correlation in small atoms, molecules, and soluble model systems will be investigated for the purpose of extrapolating the results of accurate configuration interaction calculations to the complete orbital basis set limit. %%% The results of this project will make supercomputers with parallel architectures much more useful to chemists, making it possible to do theoretical calculations on much larger chemically interesting systems.