Skip to content

feat: add BSE module - #7718

Merged
mohanchen merged 9 commits into
deepmodeling:developfrom
Fisherd99:develop
Aug 13, 2026
Merged

feat: add BSE module#7718
mohanchen merged 9 commits into
deepmodeling:developfrom
Fisherd99:develop

Conversation

@Fisherd99

@Fisherd99 Fisherd99 commented Jul 31, 2026 •
edited
Loading

Copy link
Copy Markdown
  1. feat: add BSE module. Relevant tutorial will update in https://fisherd99.github.io/abacus-librpa-bse-tutorial.html.
    ATTENTION: BSE module depends on LibRI#11 Implement CVC algorithm and enhance LR class abacusmodeling/LibRI#11
  2. feat: add ExcitonPlotter class.
  3. add input parameter rpa_outdir and rpa_out_vel for Output files are not in the correct directories #7661

Reminder

  • I have read AGENTS.md and docs/developers_guide/agent_governance.md.
  • I have linked an issue or explained why this PR does not need one.
  • I have added adequate unit tests and/or case tests, or explained why not.
    Not yet. The BSE module depends on LibRPA output files for reading. I don't know whether we can upload these files.
  • I have listed the exact verification commands run and their results.
  • I have described user-visible behavior changes, including INPUT parameter changes.
  • I have explained core-module impact for ESolver, HSolver, ElecState, Hamilt, Operator, Psi, or other source/ changes.
  • I have requested any needed governance exception below.

Linked Issue

Fix #7661

Unit Tests and/or Case Tests for my changes

  • Commands run:
  • Result summary:
  • Checks not run, with reason:

What's changed?

  • Example: brief summary of the user-visible or developer-facing change.

Governance Notes

  • INPUT/docs changes:
  • Core module impact:
  • Exceptions requested:

@mohanchen mohanchen added Feature Discussed The features will be discussed first but will not be implemented soon Refactor Refactor ABACUS codes labels Jul 31, 2026
mohanchen
mohanchen reviewed Jul 31, 2026
View reviewed changes
Comment thread source/source_esolver/esolver_lr_lcao_bse.h
@mohanchen mohanchen added the Tests/Examples Issues/PR related to unit tests and integrate tests label Jul 31, 2026
@Growl1234

Growl1234 commented Jul 31, 2026 •
edited
Loading

Copy link
Copy Markdown

Just a note: the current code is not well encapsulated:

  1. Generic LCAO builds now pull in BSE and ELPA dependencies unconditionally, which also causes the current CI failure.
  2. ELPA-related macro handling, such as HAVE_SKEWSYMMETRIC, should not leak through BSE headers.

@Fisherd99
Fisherd99 force-pushed the develop branch 7 times, most recently from 124ce0e to 1962112 Compare August 12, 2026 20:41
@Fisherd99
1. feat: add BSE module
d95c4ae 
2. feat: add ExcitonPlotter
3. add input parameter `rpa_outdir` and `rpa_out_vel`
@Fisherd99
- Enhanced RPA_LRI.hpp to improve MPI file writing for eigenvector …
3880c60 
…outputs, ensuring robust handling of parallel I/O.

- Refactor exciton plotting utilities.
- Delete unnecessary 'PARAM‘ and 'GlobalV'.
@Fisherd99
revise the description of rpa_outdir
1c83156
@Fisherd99
unitcell to basecell
42241c4
@Fisherd99
update FindLibRI
470c199 
fix without MPI compile bug

fix elpa interface
@Fisherd99
update Pgrid interface
8ce6477
@Fisherd99
refactor: separate BSE module and input items
f32902d
@Fisherd99
fix elpa_skew_eigenvector interface
2ba8ddc
@mohanchen mohanchen added Features Needed The features are indeed needed, and developers should have sophisticated knowledge Input&Output Suitable for coders without knowing too many DFT details and removed Feature Discussed The features will be discussed first but will not be implemented soon labels Aug 13, 2026
mohanchen
mohanchen approved these changes Aug 13, 2026
View reviewed changes

@mohanchen mohanchen left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is an important update, thanks for your contributions! However, keep in mind that refactors are needed in near future.

@mohanchen
mohanchen merged commit 1e4c1c6 into deepmodeling:develop Aug 13, 2026
20 of 21 checks passed
dyzheng pushed a commit to dyzheng/abacus-develop that referenced this pull request Aug 18, 2026
dyzheng
Merge remote-tracking branch 'origin/develop' into fix/dmk-dmr-fourie…
4a0cf9c 
…r-sign

Conflict resolutions (build fixes required by the merged tree):
- module_operator_lcao/test/CMakeLists.txt: ORB_atomic_lm.cpp ->
  orb_atomic_lm.cpp, matching develop's filename normalization (deepmodeling#7797)
  in the dm_trace test target sources.
- module_lr/utils/exciton_plotter.cpp: add missing 'template' keyword
  for the dependent-template call eval_wfc_bloch<T> (upstream BSE
  module, deepmodeling#7718) that fails to compile with the standard C++11 rule.
maki49 added a commit to maki49/abacus-develop that referenced this pull request Sep 4, 2026
@maki49 @claude
fix(lr): the excited-state force adds the LR part, and log the energy
2793d95 
The LR part was being subtracted. It should be added: what cal_force(int)
returns already follows the ABACUS force convention F = -dE/dR, despite
the "Gradients of each excited state" heading it prints under. The
finite-difference reference it was validated against, abacus-fd
lr-custom, goes through run_diff_custom_lr, where energies[0] is the
+dx/2 point and the result is (energies[1] - energies[0])/dx, i.e.
-dE/dx -- so that table compares force to force. (run_diff_all_kslr in
the same file uses the opposite ordering; do not read the convention off
that one.)

With the wrong sign a relaxation still converged, because F_gs - F_Omega
does reach zero somewhere -- just not at a stationary point of
E_gs + Omega. On BH/lda it stopped at 1.2864 A while the energy minimum
was near 1.25 A.

What made that visible is the other half of this commit: one line per
ionic step giving E_gs, Omega, E_exc and the two force magnitudes.
Before it the log carried only a force, so a force that was
self-consistently zero looked like convergence and nothing could be
cross-checked against the energy.

After the fix, BH/lda relaxes the first singlet to 1.25628 A, which is
the lowest E_exc of every geometry sampled and where the two force
halves cancel (0.29326 vs 0.29359). The ground state relaxes to
1.27430 A, so the excited bond is shorter by 0.018 A -- the direction
and size seen for BH A1Pi (1.2195 vs 1.2324 A).

Also restores the aims guard on set_atomic_trace. d4fe3fe ("Support
different basis number from aims") deliberately replaced the
unconditional call with a guarded one, because with aims_nbasis the
per-atom orbital counts behind iat2iwt do not match nbasis. 1e4c1c6
(BSE, deepmodeling#7718) then re-added an unconditional call above it, silently
undoing that. The previous commit here kept the wrong one of the two.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GUvgXxW4C5E1sRzUXqegEu
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@mohanchen mohanchen mohanchen approved these changes

Assignees

No one assigned

Labels

Features Needed The features are indeed needed, and developers should have sophisticated knowledge Input&Output Suitable for coders without knowing too many DFT details Refactor Refactor ABACUS codes Tests/Examples Issues/PR related to unit tests and integrate tests

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

Output files are not in the correct directories

3 participants

@Fisherd99 @Growl1234 @mohanchen