Supercomputer Chemistry At the University of Minnesota

Abstract

The Minnesota Supercomputer Institute (MSI) is a multidisciplinary research program of the University of Minnesota. Supercomputer research at MSI is carried out using the three supercomputers at the Minnesota Supercomputer Center, which is a short walk or campus-bus ride from the heart of the Minneapolis campus. The supercomputers include a two-pipe, 8-megaword Control Data Corporation CYBER 205 with a VSOS virtualmemory operating system, a one-processor, 16-megaword CRAY-2 with UNICOS UNIX operating system, and a four-processor, 256-megaword CRAY-2 also running UNICOS. We describe here a selection of the research projects carried out using these machines by researchers from the Departments of Chemistry, Chemical Engineering and Materials Science, and Medicinal Chemistry. CLUSTER CHEMISTRYThe goal of the cluster chemistry project is to understand and be able to ' predict chemical reactions and properties in the borderline area between molecules and bulk solids. This field of research encompasses molecules adsorbed on surfaces and local defects in solids as well as the study of clusters of atoms and the variation and convergence of their properties as their size increases. Our studies focus especially on adsorption (chemisorption) of small hydrocarbon molecules on metal and graphite surfaces, and on substitution defects in diamond and silicon. By carrying out quantum-chemical calculations on such systems, one can gain information about their structure and properties that would have been almost impossible to obtain from experiment.In quantum-chemical studies of molecules, the computational difficulties increase rapidly with the size of the molecular system under consideration. When infinite (bulk) systems are considered, the calculations are usually simplified by applying boundary conditions that utilize the periodic structure of a perfectly crystalline solid. For the systems considered here, however, the periodicity is violated by the very phenomenon of interest, that is, the crystal defect or the chemisorbed molecule. For these studies the bulk systems must be approximated by finite-size clusters, small enough to allow for as accurate a treatment of the electronic structure as the problem requires, yet sufficiently large to be representative for the bulk. To justify such an approach it is necessary first to study how rapidly the properties of these clusters converge with increasing size. Clusters of various sizes therefore constitute one important and natural target for this project. We need to study small and large clusters of carbon (diamond-and graphite-like), as well as transition-metal clusters in order to achieve that goal.The complex computational tasks encountered in electronic structure calculations include coupled integrodifferential equations in thousands of variables, evaluation and manipulation of millions of six-dimensional integrals, and eigenvalue problems for matrices of dimension 107 x 10~ or larger. This mission is extremely demanding on computing equipment, and the rapid development of elec...