Time-dependent density-functional theory with Turbomole
Analyzing the time development of a wave function in real time gives nice insight into chemical reactions and electron transfer processes in systems on the molecular scale. The working horse of theoretical chemistry still remains density-functional theory (DFT) because of the good information to computational cost ratio. To get results a lot of time steps have to be involved, each as quick as possible. The faster the code, the bigger the system that can be analyzed, the deeper the insight into chemical systems where quantum mechanical effects matter. A lot of fields ranging from surface science to catalysis, to pharmaceutical industry directly profit from that.
The static, time free implementation of the density-functional theory as implemented in the Turbomole program package is basis of the time-dependent algorithm. Therefore the code was compiled and adapted to run on the new multicore compute cluster LiMa at the Regionales Rechenzentrum of Erlangen. A couple of realistic test calculations have been set up and the scaling behavior of the MPI implementation and the single core performance of each process have been analyzed. The performance was compared to current GPU implementations of the density-functional theory. Strategies to prevent IO bottlenecks have been discussed.
- KONWIHR funding: two months during Multicore-Software-Initiative 2009/2010
- Thorsten Wölfle, Theoretische Chemie, Uni-Erlangen