Repository navigation
Tags
Tags
6.5.0 Code updated. This version feature a number of improvements to speedup execution. * A revised mpi_transpose (key in MVP operations) which wrap up communications step. Tests have indicated about 2x gain. * A reduced Lanczos error threshold. Unlike ED method here it is not necessary to push error convergence so much. * A new wavefunction prediction scheme for OBC and PBC infinite DMRG method. It works fine with Neigen=1 NOT for Neigen>1. This is due to the fact that Arpack currently accepts only one guess. The last point will be solved using PRIMME with Jacobi-Davidson blocks methods, which is now integrated in SciFortran > 4.28.0. However this will require some non-trivial work.
6.4.0 new feature and possible-bug fix. The code so far was targeting a single state (THE groundstate). So, when degeneracies are present, this approach disregards part of the degenerate manifold leading to possible physics inconsistencies. The updated version automatically check for degenerate solutions, provided the user is smart enought to set the LANC_NSTATES to a number > 1, within a threhsold GS_DEG_THRESHOLD (set to 1d-10). This solved the problem with apparent AFM ordering at large U in Hubbard systems (checked for the BHZ-HubKan model). This update involved simple changes throughout the code.
6.3.1 Update following new version SciFortran 4.26.0 sligthly breaking back-compatibility. SciFortran now includes support for generic sparse_matrices. Here we use a special local version of sparse_matrices with the same naming convention. This required to fine tune SciFortran use.
6.3.0 Updated Quantum Numbers input.
Merge branch 'dev_QN_shift'
The code now uses `QN_Density` and `QN_Sfhit` as main inputs to set QN.
The target QN is obtained for each lenght as:
Q(L) = int(QN_Density*L) + QN_Shift
In this way the user has full access to sectors with different QN.
There is no more distinction in `local` or `global` QN or anything like that.
We use density instead, e.g. magnetization density m for spin, total spin-occupation density n_\up and n_\down for electrons.
(we still miss orbital resolved spin-occupations n_{a\sigma} for Hubbard systems).
See the README.md for a more thorough description AND/OR check the tests.
TODO: add an examples directory
6.2.5 Added several new measurements in DMRG_MEASURE: * Polarization/Topological marker Z (localization marker as from Resta & Sorella). * Structure factors from correlations C_AB(i,j) using discrete Fourier transforms C(q) = \sum_ij exp(iq(ri-rj))<A^+_iB_j> * Calculation of the structure factor using A_q = \sum_i exp(iqr_i)A_i * Various interfaces to measure. > Automatic test from CODEX using regression test: Heisenberg and Hubbard with OBC/PBC and correlation raw/connected, dble/cmplx e post-processing da checkpoint.
PreviousNext