Thom Dunning, Todd Martinez, and Robert Pennington of the University of Illinois, Michael Klein of the University of Pennsylvania, and Robert Harrison, Robert Hinde, and Gregory Peterson of the University of Tennessee are supported by the NSF Division of Chemistry under the Cyberinfrastructure and Research Facilities Program. This collaborative project will examine the application of new computing technologies--those most likely to be found in the petascale computers of tomorrow--to the chemical sciences, through a multidisciplinary team with expertise in chemistry, computer science and technology, and computer engineering. The initial focus will be on chemical applications that are known to make heavy use of existing supercomputing facilities. These span a broad range of computational chemistry and, thus, can be viewed as prototypes for a larger class of applications. The goal is to assess the performance of these applications on the technologies expected to be the basis for future petascale computers, identify the bottlenecks limiting the performance of the applications on the new technologies, and seek solutions to the problems.
This project has the potential to extend chemical simulations to the petascale level, which will result in breakthroughs in understanding of molecular structure, energetics, and dynamics. Many of the results on chemical applications will be applicable to related activities in other fields of science--materials science, condensed matter physics, and molecular biology. Workshops, tailored to expert code developers as well as to users of standard chemistry codes, will be offered along with a dedicated website for dissemination and interactive environments.
This project examined the application of new computing technologies to the chemical sciences, through a multidisciplinary team with expertise in chemistry, computer science and technology, and computer engineering. The focus was on chemical applications that are known to make heavy use of existing supercomputing facilities, such as those based on first principles. We developed an original approach to generate and run computationally intensive chemistry codes on the most advanced to date computer architectures using computer algebra systems. We used a meta-programming framework to generate complex software codes and implemented a number of advanced algorithms for electronic structure calculations, including both Hartree-Fock and generalized gradient approximation density functional theory methods. We have implemented analytic d-functions energy gradients for certain DFT functionals for geometry optimization and ab initio molecular dynamics. The work supported by this grant has formed the core of a new company (PetaChem, LLC) founded to improve, distribute, and augment the code developed in the course of this project. The latest version of the software product has been used by over 100 research teams in 19 countries. The program has been interfaced to a widely-used molecular simulation program AMBER to enable large scale quantum mechanical (QM) and mixed QM and molecular mechanical (MM) molecular dynamics simulations. Given the efficiency and high performance of our codes, we were able to demonstrate the first ab initio molecular dynamics calculations of a protein and the first ab initio calculations of the excited electronic state of a protein. We have used the new algorithms to carry out a systematic study of the performance of ab initio methods for protein structure. We used the developed software to carry out a systematic study of the predictive abilities of time dependent density functional theory (TDDFT) for the electronic absorption spectra of the photoactive yellow protein (PYP). In summary, the implemented codes now allow us to properly describe 3rd and 4th row elements of the periodic table. This enables the study of electronic structure of large biomolecules in explicit solvent and charge transfer effects at the active sites of the proteins with transition metals.
