1. 19 Jan, 2021 1 commit
    • Nick R. Papior's avatar
      enh: minor changes to finalize for the release · e18f308e
      Nick R. Papior authored
      This adds a link to the latest article describing siesta
      developments and allows users to see this.
      I also note that users should clearly state which version
      of siesta they are using to clarify this for readers.
      Added MODULOP which makes mod operations on negative indices
      much easier.
  2. 14 Jan, 2021 1 commit
  3. 13 Jan, 2021 3 commits
  4. 12 Jan, 2021 1 commit
  5. 08 Jan, 2021 1 commit
  6. 15 Dec, 2020 1 commit
  7. 01 Dec, 2020 1 commit
  8. 27 Nov, 2020 1 commit
  9. 26 Nov, 2020 2 commits
  10. 24 Nov, 2020 2 commits
    • Nick R. Papior's avatar
      M(!40) LDAU->DFTU · 5b654f8e
      Nick R. Papior authored
      Name change for consistency and not
      confusing end-users.
    • Nick R. Papior's avatar
      mnt: changed LDAU to DFTU · dccd9f17
      Nick R. Papior authored
      Name convention change to DFTU to clarify it may
      be used for both LDA and GGA.
      No additional code changes are done.
      Note that old LDAU flags are still used while
      the DFTU flags have precedence.
      Fixes #67.
  11. 21 Nov, 2020 1 commit
  12. 17 Nov, 2020 1 commit
  13. 12 Nov, 2020 1 commit
    • Nick R. Papior's avatar
      bug: fixed un-even blocks for MPI-IO in NetCDF · 4a7b06d2
      Nick R. Papior authored
      This got discovered by Albert and will only show
      it self for systems where there are one processor
      having more blocks than other processors of the sparsity pattern.
      Apparently my tests were only testing the "easy" case. :(
  14. 11 Nov, 2020 3 commits
    • Nick R. Papior's avatar
      make: updated makefiles · 13619023
      Nick R. Papior authored
    • Nick R. Papior's avatar
      Merge branch 'rel-4.0' · c81f10b3
      Nick R. Papior authored
      Fixed !58 and removed some memory leaks.
    • Nick R. Papior's avatar
      Changed TS dq tolerance, now per SCF step. · 4782fc72
      Nick R. Papior authored
      A report was made about the SCF not converging.
      Clearly the user did not read the message since it clearly
      stated that there was abnormal charge.
      I have now made the checks be in the SCF loops which is probably
      easier to understand. This may end up in situations where
      the SCF never stops due to dQ being too high.
      There is no good way around this, I think.
      Perhaps this scheme requires another iteration.
  15. 09 Nov, 2020 1 commit
    • Nick R. Papior's avatar
      bug: removed lots of memory leaks for k-point sampling · 3ffa072a
      Nick R. Papior authored
      Memory-leaks removed here:
      - atomlist.f
        lots of arrays were not deallocated upon exit
      - cdiag.f
        rwork was not handled correctly, fixed
      - listsc.f
        lists are now de-allocated
      - radial.f
        sub-optimal handling of the arrays, and not allowing
        them to be reset.
      Fixes 58. But still some memory elements that are not
      deallocated upon exit.
  16. 05 Nov, 2020 1 commit
  17. 02 Nov, 2020 1 commit
  18. 26 Oct, 2020 7 commits
    • Nick R. Papior's avatar
      Merge !20 fixed spin spiral code · 281b1271
      Nick R. Papior authored
      This merge fixes the spin spiral code with several changes.
      1. It now works in parallel
      2. Bugs for mesh-sub-divisions was corrected
      3. Changed from real-arrays to complex-arrays in the
         diag2kspiral code (easier to debug and read)
      4. Added documentation about the Spin.Spiral code and
         that the code is not production stable.
    • Nick R. Papior's avatar
      maint: redid test for new spiral phases · e02dbb14
      Nick R. Papior authored
    • Nick R. Papior's avatar
      maint: changed sign and diag-spiral code · 77f411a6
      Nick R. Papior authored
      Moved from pure real arrays to complex arrays.
      This is much easier to read and clarifies some things.
      This uncovered that a sign-difference between the original
      spiral code and the current Siesta base.
      If one compares the diagk code and the original spin-spiral
      code there was a mismatch between the phases. This should
      now be fixed and meant reversing signs in vmatsp and rhoofdsp
      as well.
      The code is still very much un-tested since we don't have
      a decent test for this.
    • Nick R. Papior's avatar
      maint: rerunned test and updated documentation and TRsym for spiral · 49e1b38e
      Nick R. Papior authored
      1. Reran test
      2. Updated documentation a bit
      3. Ensured TRSym is false for spiral, regardless of flags
      Fixed sub-mesh division calculations in spiral
      Now the code correctly handles mesh sub divisions
      by taking into account the displacements for the smaller
      Removed a global "spiral" variable in dhscf which was never used.
      rhoofdsp could be further optimized since the density was
      still accessed in a triangular way. This means that it is very
      similar to the rhoofd routine. It now uses less memory, has
      a better memory access pattern (moved nspin to first index)
      and I removed superfluous if statements by unrolling loops.
      In vmatsp some of the same things are done; fixed
      sub mesh division calculations; moved spin index to first.
    • Nick R. Papior's avatar
      enh: updated vmatsp to F90 and changed to follow vmat more closely · 621b7d21
      Nick R. Papior authored
      Still not threaded, but now is more comparable.
      Also fixed parallel problem
      It was related to two things:
      1. In parallel the Dscf array was not distributed correctly
      2. offset for phase was incorrect.
    • Nick R. Papior's avatar
    • Nick R. Papior's avatar
      bug: fixed spin-spiral calculations · f8c08aef
      Nick R. Papior authored
      I find it hard to believe they ever worked (besides in serial). :(
      Now I have fixed indices and array dimensions.
      This part of the code was also exposed to the bug in #44.
      I have added a test, but I have no idea what it should give.
      We need some feedback on this part of the code.
  19. 15 Oct, 2020 1 commit
  20. 14 Oct, 2020 1 commit
  21. 13 Oct, 2020 1 commit
  22. 09 Oct, 2020 5 commits
    • Alberto Garcia's avatar
    • Alberto Garcia's avatar
      Better processing of the valence charge for ps generation · 4602d5ea
      Alberto Garcia authored
      The header information in psf files records information about the
      valence configuration for pseudopotential generation. Some generators
      (such as ATOM) cannot pseudize multiple shells for the same angular
      momentum.  When semicore states are present, ATOM uses an ionic
      configuration, with a reduced valence charge. The legacy header,
      having one field per 'l', is still appropriate. Other programs (such
      as ONCVPSP) can deal with multiple shells per 'l' in the valence, but
      the psf header cannot contain the full information, and the total
      valence charge at the time of pseudopotential generation has to be
      handled in a special way, beyond adding up the contributions of the
      different sections of the header.
      An extra field in psf files that records the total valence charge for
      pseudopotential generation was present already ('gen_zval', for the
      needs of the VCA tools), but, due to a bug, Siesta was not processing
      properly this field, even if set.
      The 'gen_zval' value can be significant. Its difference from the
      effective valence of the pseudo-ion is used by default as an 'extra
      charge' for PAO generation, and it also determines the charge used for
      KB projector generation.
      The above was an issue when using psf files generated from PSML files
      with semicore states, but with a non-ionic configuration. In this
      case, Siesta set incorrectly the 'gen_zval' value, forcing the use of
      a non-zero effective charge for PAO generation. This led typically to
      shorter orbitals.
      Siesta now processes correctly the 'gen_zval' field (if present). Note
      that in this version the user still has to provide a PAO.Basis block
      with the correct suite of shells if semicore states are included in
      the valence. And, of course, the psf file should correctly reflect the
      proper valence charge, if appropriate.
    • Nick R. Papior's avatar
    • Nick R. Papior's avatar
      Merge branch 'rel-4.0' into rel-4.1 · ef8d8f23
      Nick R. Papior authored
      Moved ChangeLog.md to root folder.
      Also amended and cleaned the entries
      for easier search.
    • Nick R. Papior's avatar
      maint: moved ChangeLog.md to root directory · fb35f0bc
      Nick R. Papior authored
      This makes it faster to find and pushes people
      to read it.
      Also updated missing entries.
  23. 08 Oct, 2020 2 commits