[an error occurred while processing this directive]

NERSC 3 Greenbook

Next: The Energy Research Computational Up: Computational and Technology Research Previous: Center for Computational Science

Computational Chemistry for Nuclear Waste Characterization and Processing: Relativistic Quantum Chemistry of Actinides

Robert J. Harrison and Ricky A. Kendall, Pacific Northwest National Laboratory

Research Objectives

We aim to develop and apply the methods of relativistic quantum chemistry to assist in the understanding and prediction of the chemistry of actinide and lanthanide compounds.

Computational Approach

The work involves determination of the electronic structure of molecules including relativistic effects necessary for heavy elements. Most calculations are very challenging and well suited to the CRAY-T3E. There are five major categories of activities:

• Benchmarking of methods: Detailed and systematic comparison of various theoretical approaches with each other and with experiment. Few such studies are available for rigorous relativistic methods and fewer still for systems containing actinides.
• Application work: We are studying the speciation of aqueous uranium (VI) carbonates. A detailed understanding of the actinide-carbonate-water system is essential to modeling the fate and transport of actinides in the environment.
• Method and computer program development: Existing programs are being parallelized and extended to enable calculations on larger molecules at higher levels of accuracy.
• Computer Science: Extensions of Global Arrays, parallel I/O, new linear algebra, meta-computing, and prototyping of new parallel programming tools.
• Collaboration tools: We use Internet video conferencing to augment weekly telephone voice conferences and are evaluating CORE-2000 and other collaboration tools.

Accomplishments

All components of the project are well underway. For benchmark purposes we have parallelized existing all-electron Dirac-Hartree-Fock and four-component second-order perturbation theory which treat relativistic effects very rigorously. Calculations at these levels of theory can take several hours on 64 nodes of the T3E. These results are being used to validate calculations using relativistic effective core potentials on larger molecules. Determination of the vibrational frequencies of the molecule in the figure took about 30 hours on 256 nodes of the T3E using the NWChem DFT module. The data from calculations on a sequence of related molecules is being correlated with experiment data. A spin-orbit configuration interaction code has also been parallelized and has just commenced production use on the T3E.

Significance

Most radioactive waste involves actinides, and their large atomic number implies that relativistic effects have important chemical consequences. Our implementation of relativistic quantum chemical methods on massively parallel computers will provide capabilities for modeling heavy-element compounds similar to those currently available for light-element compounds. The theoretical and computational methodology so developed will supplement current, very expensive experimental studies of the actinides and lanthanides. This will allow limited experimental data to be extrapolated to many other regimes of interest.

The program objectives will be attained through a multi-laboratory, multi-university and multi-disciplinary collaboration. These techniques will be applied to important molecular systems and processes, including the interaction of actinides with: 1) organic complexing agents present in tank wastes; 2) natural aqueous systems (carbonates), in order to better understand fate and transport in the environment; and 3) new materials, such as phosphates and amides, for the design of in situ remediation technologies and separation systems.

DFT/RECP calculations have been performed on this polymeric uranium carbonate (c.f., Figure 38), and results agree well with experimental observations.

  

Figure 38: Polymeric uranium carbonate system

Future Requirements

Within the scope of this project and the accuracy requirements of future scientific simulations the default minimum specifications for an MPP system are:

• 512 Mbytes memory per node,
• per-processor disk I/O bandwidth of 8 Mbytes/sec,
• 700 Mflops peak performance per-processor,
• 1 micro-sec communication latency,
• 500 Mbytes/sec communication bandwidth.

To reach all the simulation goals the project will need between 1.5 and 2 cpu months of this MPP system.

NERSC 3 Greenbook

Next: The Energy Research Computational Up: Computational and Technology Research Previous: Center for Computational Science