University of Heidelberg
P-RICDΣ: A highly scalable parallel platform for calculating ionization potentials and electron affinities in large quantum systems via Green’s Function method
Monday, June 27, 2:00 p.m.
307 Eberly Hall
Overview: To treat the electronic structure in large quantum systems via state-of-the-art second-order electron propagator method (ADC(2)) we developed a highly scalable parallel computer program called P−RICDΣ . The parallel efficiency has been achieved by utilizing the hybrid MPI/OpenMP programming model and exploiting the sparsity of the two-electron integral matrix by using Cholesky decomposition techniques. By using the P−RICDΣ program and a massively parallel computer we are now able to tackle grand challenge-class problems (in the size-range of a few hundredths of atoms and thousands of basis functions) which were unattainable before within the framework of conventional techniques.
The talk is organized as follows:
- Brief introduction to the basics of the Cholesky decomposition technique and its various modifications;
- Introduction to the second-order Green’s function method; working equations;
- Technical aspects of the efficient MPI/OpenMP parallelization of working equations
- Timings and scaling behavior of the P−RICDΣ program
- Future prospect and perspectives: utilizing a specialized floating-point mathematical coprocessors (GPGPU) in single precision mode .
- Victor P. Vysotskiy and Lorenz S. Cederbaum, Journal of Chemical Physics 132, 044110 (2010)
- Victor P. Vysotskiy and Lorenz S. Cederbaum, Journal of Chemical Theory and Computation 7, 320 (2011)
Sponsored by: Center for Molecular and Materials Simulations at the University of