Cálculos de Química Quântica em GPUs: proposta de um algoritmo paralelo para a pseudodiagonalização de matrizes simétricas usando a plataforma NVIDIA/CUDA.

• Júlio Maia UFPB
• Lucídio Cabral UFPB
• Gerd Rocha UFPB

Abstract

Eigenvalue and eigenvectors notions are essential for the calculations in Quantum Chemistry, since molecules and atoms energy levels are calculated based on the eigenvectors of some matrices. The diagonalisation procedure is a crucial step of the Hartree-Fock-Roothaan method, since it provides the molecular orbitals energies. However, finding an accurate set of eigenvalues and eigenvectors is not always mandatory. This paper focuses on a parallel implementation for an approximated eigenvector calculation algorithm on GPUs using the NVIDIA/CUDA platform.

References

Beierlein, F., & Clark, T. (2005). Computer Simulations of Enzyme Reaction Mechanisms: Simulation of Protein Spectra. In S. Wagner, W. Hanke, A. Bode, & F. Durst (Eds.), High Performance Computing in Science and Engineering, Munich 2004 (pp. 245–259). Springer Berlin Heidelberg.

Hwu, W. W. (2011). GPU Computing Gems Emerald Edition (1 edition.). Burlington, MA: Morgan Kaufmann.

Maia, J. D. C., Urquiza Carvalho, G. A., Mangueira, C. P., Santana, S. R., Cabral, L. A. F., & Rocha, G. B. (2012). GPU Linear Algebra Libraries and GPGPU Programming for Accelerating MOPAC Semiempirical Quantum Chemistry Calculations. Journal of Chemical Theory and Computation, 8(9), 3072–3081.

Rocha, G. B., & Simas, A. M. (2007). Métodos Semi-empíricos de Estrutura Eletrônica em Química Quântica. In N. H. Morgon & K. Coutinho (Eds.), Métodos de Química Teórica e Modelagem Molecular (1a ed., pp. 29 – 71). São Paulo: Livraria da Física.

Stewart, J. J. P., Császár, P., & Pulay, P. (1982). Fast semiempirical calculations. Journal of Computational Chemistry, 3(2), 227–228.