Parallelized contraction of tensor trains or matrix product operators dataset

Description

The dataset represents the runtime and accuracy obtained by the fit algorithm used to compress Tensor Trains (TT) or Matrix Product Operators (MPO). site_accN.out contains the accuracies obtained with MPOs in site canonical form using N parallel processes. inv_accN.out contains the accuracies obtained with MPOs in inverse canonical form using N parallel processes. site_suN.out contains the runtimes obtained with MPOs in site canonical form using N parallel processes. inv_suN.out contains the runtimes obtained with MPOs in inverse canonical form using N parallel processes. The data is stored as ["N", "Chi", "Sweeps", "Form", "Err"] and ["N", "Chi", "Sweeps", "Form", "Time"] Where "N" is the number of processes, "Chi" is the maximum bond dimension, "Sweeps" is the number of sweeps, "Form" is the canonical form used (site or inverse), "Err" is the error compared to the naive contraction, and "Time" is the runtime. rand_su.out contains the runtimes of the MPO contraction using 1 process with in both canonical forms, with and without the randomized mode. The data is stored as ["Chi", "Form", "Time"] Where "Chi" is the maximum bond dimension, "Form" is the canonical form used, specified if the randomization has been used ("invdet" for inverse canonical form without randomization, "invrand" for inverse canonical form with randomization, "sitedet" for site canonical form without randomization, "siterand" for site canonical form with randomization), "Time" is the runtime.

Institutions

• Ludwig-Maximilians-Universität München

  Bavaria, Munich

Categories

Licence