Berlin 2015 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
TT: Fachverband Tiefe Temperaturen
TT 72: Low-Dimensional Systems: Poster Session
TT 72.19: Poster
Wednesday, March 18, 2015, 15:00–18:00, Poster B
Hardware-independent Parallelization of Matrix Product State Codes Using SciPAL — •Thomas Köhler1, Johannes Hagemann2, Salvatore R. Manmana1, and Stephan C. Kramer3 — 1Institut f. Theoretische Physik, Universität Göttingen — 2Institut f. Röntgenphysik, Universität Göttingen — 3Max-Planck-Institut f. biophysikalische Chemie, Göttingen
The physical properties of low-dimensional, strongly-correlated quantum systems, as described by Hubbard-like or Heisenberg models, can efficiently be accessed by matrix product state (MPS) and density-matrix renormalization group (DMRG) methods. The key to an efficient MPS code is the optimization of the evaluation of concatenated matrix-matrix products for building the multi-particle states and an efficient compression of single-particle states by truncated singular value decompositions. Based on SciPAL's [1] expression templates for BLAS operations we have designed a domain-specific embedded language for computing the properties of MPS. Because SciPAL's expression templates are hardware-independent we can provide an objective comparison of the performance of CPU- or GPU-based computations and hybrid forms thereof.
[1] SciPAL: Expression Templates and Composition Closure Objects for High Performance Computational Physics with CUDA and OpenMP, S. C. Kramer and J. Hagemann, ACM TOPC (to appear).