piwik-script

Intern
    SFB 1170

    Z03

    Code development and optimization: harnessing the power of massively parallel computer architectures

    Summary

    Cutting edge research in computational physics is based on innovative algorithmic ideas. Translating these ideas into efficient computer codes that harness the power of modern computer architectures is a highly non-trivial task. The aim of this project is to provide high-level support for all the groups within the Collaborative Research Center that are confronted with this challenging task. In this proposal, we will concentrate on a subset of algorithms, including tensor networks (with emphasis on the density matrix renormalization group, DMRG), different quantum Monte Carlo (QMC) methods for impurity and lattice systems, and non-equilibrium Green function schemes for transport. Each algorithm will be critically reviewed with emphasis on optimization for massively parallel machines. Novel algorithms, in particular QMC and tensor network methods, will be developed and tested. These efforts will produce a library of optimal codes that will greatly profit the Collaborative Research Center.

    Our present hardware is adapted to massively parallel tasks with a small requirement of random access memory per core. With further financial support by the Collaborative Research Center, we plan to purchase fat nodes and thereby gain access to a state-of-the-art local computer cluster, at least for the first funding period.

    Link to the webpage of the Z03 project