 |
|
 |
||||||
|
||||||
 |
Home
| Mission
|
about SciDAC
|
Contact Us |
|||||
 |
 |
 |
||||
|
Alumni ProjectReliable Electronic Structure Calculations for Heavy Element Chemistry: Relativistic Pseudopotentionals for Very-Large-Core Lanthanide and Actinide SystemsWalter C. Ermler SummaryRealizing the full potential of terascale computing requires not only the use of next-generation hardware coupled with software specifically developed to make full use of such advanced platforms, but it also requires that the software employed incorporate the latest theoretical developments that will guarantee high accuracy and tractability for the systems of interest. In many instances, the tractability of a calculation depends on the number of electrons that must be treated explicitly, and this relationship holds even if the most highly advanced hardware platform is employed. Consequently, the purpose of this project is to develop the computational methodology needed to treat molecules that contain heavy elements (i.e., atoms that contain large numbers of electrons) at an accuracy comparable to what can be achieved for molecules containing first and second row atoms (e.g. carbon, which contains only six electrons) and to implement this technology in high performance software that is portable and scalable on massively parallel computer systems. The number of electrons treated explicitly may be reduced by realizing that not all electrons are involved in chemical bonding. Since only the valence (bonding) electrons must be treated explicitly, the rest of the electrons, which are labeled “core”, can be represented by a relativistic effective core potential (RECP). For example, americium (Am) contains 95 electrons, but the use of an RECP to represent the core electrons of Am reduces this number to 27, which represents over a 70% reduction in the number of electrons requiring explicit treatment for this atom alone. Despite such an impressive reduction, the valence space of Am actually contains only 9 electrons and, for certain structural studies, the valence space of Am+2 can be described using only one electron. The use of a 9- or 1-electron valence space RECP for Am or Am+2, however, would result in significant errors. While the RECP (as well as all other pseudopotential methods) relies on the principle of core/valence separability, nature actually contains a third type of electronic region, labeled “outer core” comprised of electrons that while affected by chemical bonding do not participate directly in the process. Since no pseudopotential method to date accurately describes the outer core, these electrons are routinely relegated to the valence region for explicit treatment, resulting in either unnecessary complexity or computational intractability, even if state-of-the-art hardware platforms are used along with software specifically optimized for such platforms. In this context, we have developed a relativistic pseudopotentional (RPP) for use in ab initio molecular electronic calculations. The RPP is a new form of RECP that is based on extending the usual two-space representation of atomic electrons (core and valence) to three spaces (core, outer core and valence), thus requiring only the smallest number of molecular electrons to be treated explicitly. The use of very large core RPPs in conjunction with advanced computing platforms will permit the highly accurate ab initio treatment of systems possessing orders of magnitude more electrons than are tractable using current codes and platforms. For example, an RECP would reduce the number of explicitly treated electrons in Am from 95 to 27. However, a RPP would reduce this number to 9. For electronic structure calculations scaling as N5 – N7 (which is normal for these types of systems), where N is the number of valence electrons, this leads to a reduction of 250 - 2000 in computational cost. In other words, a calculation that would require 2000 hours of computer time to complete with a RECP could be completed in approximately one hour with an RPP. Affiliated SciDAC researchers are R. M. Pitzer, B. E. Bursten and I. Shavitt of The Ohio State University. This group is expanding and enhancing the Columbus suite of electronic structure codes for efficient parallel computations of electronic wave functions for polyatomic systems containing heavy elements in which the core and outer core electrons are represented by RECPs and RPPs. We are developing new RECPs and RPPs for use by this group. We continue our contacts with D. A. Dixon, W. A. de Jong, T. L. Windus, and M. Gutowski of PNNL on relativistic electronic structure calculations and code development in the context of their NWChem system. Additionally, we collaborate with J. L. Tilson, Center for Computational Research (CCR), SUNY, Buffalo and A. F. Wagner and R. Shepard of ANL using our parallel spin-orbit CI modification of Columbus. We are continuing the computer implementation and calculation of RPPs for the individual elements, initially Am and Cl. We are utilizing a two-component basis set representation of small-core RECP-based atomic wavefunctions needed to generate an RPP, as opposed to the numerical, all-electron wavefunctions used in the generation of RECPs. We are also generating the angular momentum projection operators to be associated with the RPPs. Unlike their RECP counterparts, these operators are not developed in the context of the Wigner-Eckart theorem, thus allowing for the symmetry-breaking that takes place when outer core electrons polarize in response to bonding. This results in operators containing additional angular degrees of freedom, as opposed to the two angular-component-based operators of the RECPs. We are also continuing the coding of modules for molecular calculations utilizing the RPP, specifically for outer core polarization effects and for correlated calculations in the context of two-component basis sets. Upon completion of algorithm development and coding, implementation of the RPP formalism into the NWChem and Columbus suites will be carried out in collaboration with PNNL and OSU, respectively. Once this is accomplished, large-scale SOCI calculations on AmCl1+ will be carried out in collaboration with CCR. Additionally, code optimization and parallelization, as well as the efficient two-, three- and four-center extensions of the RPP molecular codes will be carried out in collaboration with the SciDAC Centers. For further information on this subject contact:Walter C. ErmlerDepartment of Chemistry The University of Memphis Memphis, TN 38152 Phone: 901-678-4422 wermler@memphis.edu
|
||||
Home | ASCR | Contact Us | DOE disclaimer |
|
|
