Repository navigation

Feature: LR-TDDFT analytical forces (gradients) by Z-vector method and excited-state geometry relaxation - #8069

Merged

mohanchen merged 134 commits into

deepmodeling:developfrom

maki49:LR-Grad-relax

Oct 18, 2026

Merged

Feature: LR-TDDFT analytical forces (gradients) by Z-vector method and excited-state geometry relaxation#8069
mohanchen merged 134 commits into
deepmodeling:developfrom
maki49:LR-Grad-relax

Commits on Oct 7, 2026
- LR-TDDFT analytical gradients: port onto refactored develop
  maki49
  committed
  991ebc0
- fix(lr-grad): replace LibRI RI::LR with a local two-density-matrix EXX force kernel
  maki49
  committed
  a51848b
- fix(lr-grad): correct nspin in the LR grid-force ModuleGint calls, close HSolverLR timer
  maki49
  committed
  531542e
- fix(lr-grad): gather the RHS before the LAPACK Z-vector solve
  maki49
  committed
  3bbb90d
- fix: singlet gate for gxc
  maki49
  committed
  d5d5d10
- fix: Hxc dm-trans term (symmetrization and factor)
  maki49
  committed
  495b7c0
- Fix some factor bugs in ceZc, K_ii terms (visible in DZP)
  maki49
  committed
  72970b7
- Fix: transpose dm_trans in Z-vector eq. RHS
  maki49
  committed
  cd937ea
- Fix pot_hxc_gs: correct TDRPA@LDA (linear vxc=fxc[gs](T+Z) instead of vxc[T+Z])
  maki49
  committed
  5e1bb03
- Feature: implement gxc
  maki49
  committed
  2bdaa6a
- Perf: share gxc kernels to save memory
  maki49
  committed
  2a4f778
- Perf: put drho out to cal_v_eff for gxc, to save memory
  maki49
  committed
  a54e18b
- fix nspin=1
  maki49
  committed
  51df46b
- small fixes
  maki49
  committed
  e6dbb01
- Perf: speed up the LR kernel-to-potential integrands by ~45%
  
  maki49
  and
  claude
  committed
  68199dc
- feat: lr-grad openshell implementation
  maki49
  committed
  ca76c00
- Feat: enable LRC kernel
  maki49
  committed
  a5c73b0
- Remove the LR-Grad EXX debug scaffolding
  
  maki49
  and
  claude
  committed
  d91a131
- fix(lr): make the KS-orbital window copy and paraX_ setup re-entrant
  
  maki49
  and
  claude
  committed
  135c52d
- refactor(lr): keep the KS solver as a member and alias its geometry objects
  
  maki49
  and
  claude
  committed
  35254a9
- feat(lr): re-run the ground state from runner(istep)
  
  maki49
  and
  claude
  committed
  dbd737a
- feat(lr): drive geometry relaxation from an excited-state surface
  
  maki49
  and
  claude
  committed
  ae72baf
- fix(lr): scope the relax-target inputs, and clear out the re-entrancy leftovers
  
  maki49
  and
  claude
  committed
  b2a246c
- fix(lr): the excited-state force adds the LR part, and log the energy
  
  maki49
  and
  claude
  committed
  2e8c043
- fix(lr): build an Exx_LRI on the ks-lr path for a hybrid ground state
  
  maki49
  and
  claude
  committed
  6a3d3f3
- fix: use an spin-index mixing in close-shell grad: affect triplet gradients with degenerate KS basis up ≠ down
  maki49
  committed
  399ec7e
- fix: expand Z-window to nbands for complete orbital rotation space
  maki49
  committed
  f1b237c
- fix: small fixes about nupdown and degeneracy information
  maki49
  committed
  4094aea
- fix openshell bug: parallel index in HamiltULR and add-on op-lr-diag
  maki49
  committed
  aaf4032
- fix: adapt LR-gradient code to develop interface changes after rebase
  
  maki49
  and
  claude
  committed
  31df937
- fix compile warning
  maki49
  committed
  68e56c6
- fix: clamp sigma for HSE06's wpbeh kernel; fix complex CXC index bug
  maki49
  committed
  d33af90
- chore: remove diagnostic env-var switches from the LR-Grad investigation
  maki49
  committed
  157822d
- feat(lr-grad): algebra for the degenerate-subspace gradient matrix
  
  maki49
  and
  claude
  committed
  9d0c84a
- refactor(lr-grad): let the excited-state gradient take X from the caller
  
  maki49
  and
  claude
  committed
  138b2ca
- feat(lr-grad): compute the gradient matrix of a degenerate multiplet
  
  maki49
  and
  claude
  committed
  d7dc54a
- feat(lr-grad): store the degenerate gradient matrix, print its eigenvalues, add the average relax mode
  
  maki49
  and
  claude
  committed
  82d2640
- feat: state following by max overlap
  
  maki49
  and
  claude
  committed
  818dc96
- feat(lr-grad): find the Jahn-Teller direction from the degenerate gradient matrix
  
  maki49
  and
  claude
  committed
  fa31179
- feat(lr-grad): lr_relax_degen_mode = jt, descending the Jahn-Teller branch
  
  maki49
  and
  claude
  committed
  b14820c
- fix: adapt LR-gradient code to develop interface changes after rebase (part 2)
  
  maki49
  and
  claude
  committed
  c1183ba
- test(lr): add cal_force to the gamma-only LR-TDDFT integrate tests
  
  maki49
  and
  claude
  committed
  b4ef783
- fix: guard MPI-only and EXX-only symbols for the no-ELPA/no-EXX CI builds
  
  maki49
  and
  claude
  committed
  e4cf077
- fix: more MPI-only guards, and wire the Grad sources into Makefile.Objects
  
  maki49
  and
  claude
  committed
  12776d4
- fix: make the gradient path build again under -DENABLE_MPI=OFF/Intel Makefile CI, and fix a parallel force bug
  
  maki49
  and
  claude
  committed
  9a47836
- fix: reduce global dependency budget flagged by agent governance check
  
  maki49
  and
  claude
  committed
  d144b9b
- feat(lr-grad): distributed ScaLAPACK and ELPA solvers for the Z-vector equation
  
  maki49
  and
  claude
  committed
  58cd935
- feat(lr-grad): add INPUT parameter lr_grad_solver to choose the Z-vector solver
  
  maki49
  and
  claude
  committed
  d41fbf8
- docs(lr): move hand-written input-main.md text into the C++ descriptions, regenerate
  
  maki49
  and
  claude
  committed
  a981272
- feat(lr-grad): add lr_grad_solver = scalapack_chol, a ScaLAPACK Cholesky Z-vector solve
  
  maki49
  and
  claude
  committed
  4791751
- fix(lr): clamp sigma in xc_gga_vxc for HSE06, matching fxc/kxc
  maki49
  committed
  dffff57
- fix(lr-grad): build the LR unit tests against the 6-argument gather_2d_to_full
  
  maki49
  and
  claude
  committed
  fdd43e7
- refactor(lr-grad): flatten Grad/ into module_lr, cap filenames at 15 chars
  
  maki49
  and
  claude
  committed
  c04ee92
- chore(lr-grad): drop unused includes left over from removed diagnostics
  
  maki49
  and
  claude
  committed
  af63607
- fix: match Makefile.Objects' LR-Grad object name to zeqlin_solv.cpp
  
  maki49
  and
  claude
  committed
  ea71a3d
- fix para: ks-lr nb sync with groudstate; nrxx=0 guard
  maki49
  committed
  739cfd5
- fix(lr-grad): replace #pragma once with include guards
  
  maki49
  and
  claude
  committed
  1912461
- fix para: dav_subspace bcast type and spectra spin index
  maki49
  committed
  fb90da4
- fix(lr): derive occupied windows from charged spin populations
  maki49
  committed
  c79ffec
- fix(lr): update two nocc tests left stale by the post-SCF resolution move
  
  maki49
  and
  claude
  committed
  9a2b75e
- fix(lr): balance force timers in degenerate gradient paths
  maki49
  committed
  a0700ca
- fix(lr): handle empty grid ranks and align gamma CI coverage
  maki49
  committed
  27cadf1
- fix CI precision
  maki49
  committed
  b571f36
- perf(lr): reuse transition density across output spins
  maki49
  committed
  7f2ee94
- fix: CI test lr force thr 1e-6
  maki49
  committed
  bf1387c
- perf: vectorize Gamma and multi-k LR exchange projection
  maki49
  committed
  4fdd372
- refactor(dm): make internal real-space density updates non-const
  maki49
  committed
  4607c35
- refactor(lr): pass orbital layout explicitly to density transposition
  maki49
  committed
  ea6e740
- refactor(lr): move gradient evaluation out of ESolver
  maki49
  committed
  463a022
- refactor(lr): lowercase new filenames and split gradient controllers
  maki49
  committed
  36e9095
- refactor(lr): remove mutable operator work buffers
  maki49
  committed
  7659d84
- refactor(lr): keep new filename stems within fifteen characters
  maki49
  committed
  1f2d4b2
- refactor(input): shorten LR degeneracy parameter names
  maki49
  committed
  5f0765d
- fix(lr): restore CI builds after gradient refactor
  maki49
  committed
  d6f3b3b
Commits on Oct 8, 2026
- fix(lr): address blocking PR review findings
  maki49
  committed
  9e8ae7f
- fix(lr): make gradient ownership and solver failures explicit
  maki49
  committed
  6c59371
- refactor(lr): separate production force and validation helpers
  maki49
  committed
  232363d
- fix(lr): apply BSE singlet weighting once and cover nspin one
  maki49
  committed
  6ff2094
- refactor(lr): reuse gradient inputs in EDM W and Z interfaces
  maki49
  committed
  e28cff8
Commits on Oct 9, 2026
- fix(lr): correct amplitude offsets and stationary JT tracking
  maki49
  committed
  1012d7f
- fix(lr): use response XC kernel for quadratic gradient terms
  maki49
  committed
  73ee3d0
- fix(lr): isolate electronic EXX state while sharing geometry tensors
  maki49
  committed
  8bec3f7
- fix(lr): reject unsupported NLCC forces and relaxation
  maki49
  committed
  58cafa6
- fix(lr): preserve signed sigma for non-HSE kernels
  maki49
  committed
  ca201a4
- fix(lr): track roots with distributed cross-geometry overlaps
  maki49
  committed
  413b786
- fix(lr): track degenerate subspaces and preserve the actual JT branch
  maki49
  committed
  a055dd8
- test(lr): cover degenerate relaxation and enable CI XC gradients
  maki49
  committed
  40bcc10
- Merge origin/develop into LR-Grad-relax
  maki49
  committed
  b143939
Commits on Oct 10, 2026
- Merge origin/develop into LR-Grad-relax
  maki49
  committed
  aaf7da8
- fix(lr): report resolved KS band counts in initialization log
  maki49
  committed
  cc578ff
- Merge branch 'develop' into LR-Grad-relax
  maki49
  authored
  a70cc80
- fix(lr): check runtime LibXC KXC availability for gradients
  maki49
  committed
  c62bc3a
- Merge remote-tracking branch 'my/LR-Grad-relax' into LR-Grad-relax
  maki49
  committed
  9f39177
- Merge origin/develop and adapt LR cube output API
  maki49
  committed
  a546107
- refactor(lr): clarify state tracking names and orthonormality warning
  maki49
  committed
  ae1f3f2
- refactor(lr): split ESolver initialization and orbital setup
  maki49
  committed
  8a2c449
- Merge origin/develop and adapt relocated LR initialization
  maki49
  committed
  8b86aa7
- fix(test): adapt overlap fixtures to UnitCell vector maps
  maki49
  committed
  fd32f16
Commits on Oct 11, 2026
Commits on Oct 13, 2026
Commits on Oct 14, 2026
- Merge remote-tracking branch 'origin/develop' into LR-Grad-relax
  maki49
  committed
  ab811c9
- Fix LR review issues in state projection and force validation
  maki49
  committed
  324cacf
- Avoid constructing discarded Casida chains for Z operators
  maki49
  committed
  4e42a4b
- Make LR XC kernel preparation explicit
  maki49
  committed
  34ee454
- Remove the ELPA-backed LR Z-vector solver
  maki49
  committed
  5b27615
- Document and warn about truncated LR force windows
  maki49
  committed
  1ccd41e
- Document the finite-tolerance approximation in JT relaxation
  maki49
  committed
  0bf7c32
- Compare signed LR force components in integration tests
  maki49
  committed
  3399002
- Update serial INPUT tests after removing the ELPA Z solver
  maki49
  committed
  d962f16
Commits on Oct 15, 2026
- Merge develop and separate LR implementation from interfaces
  maki49
  committed
  398a482
- Pass LR force settings explicitly and use numerical connectors
  maki49
  committed
  17d5a66
- Use column-major connectors for LR Hermitian eigensolvers
  maki49
  committed
  07fb8e4
- Use numerical connectors throughout the LR additions
  maki49
  committed
  e741632
- Pass LR runtime dependencies explicitly and clarify captures
  maki49
  committed
  9bc3be3
- Expand compact PR additions and make LR test setup consistent
  maki49
  committed
  0a212fd
Commits on Oct 17, 2026
- Merge origin/develop into LR-Grad-relax
  maki49
  committed
  3570618
- Rename EXX workspace files to reflect RI ownership
  maki49
  committed
  5bc7213
- Make LR reductions and elementwise calculations explicit
  maki49
  committed
  abe308b
- Merge branch 'develop' into LR-Grad-relax
  mohanchen
  authored
  ac4bf0d
Commits on Oct 18, 2026
- Fix LR review issues and isolate LR density regressions
  maki49
  committed
  3ec4712
- Make LR interfaces explicit and reject unsupported force modes
  maki49
  committed
  7b9c4eb
- Split remaining LR force matrix declarations
  maki49
  committed
  2e13b88
- Reuse the common sparse writer behind thin LR adapters
  maki49
  committed
  2ec57f5
- Restore ground-state XC selection during LR force reconstruction
  maki49
  committed
  4e8e8dc
- Allow empty local blocks in LR matrix transformations
  maki49
  committed
  95d9b1b
- Support empty LR ownership through eigensolvers and force response
  maki49
  committed
  5db76ae
- Preserve LR exchange linearity and certify Z-CG residuals
  maki49
  committed
  87328f1
- Allow disabling LR optical postprocessing and accelerate CI cases
  maki49
  committed
  05e8146
- Fix missing reduction header in Z solver test
  maki49
  committed
  c852f67

Commits

Commits on Oct 7, 2026

Commits on Oct 8, 2026

Commits on Oct 9, 2026

Commits on Oct 10, 2026

Commits on Oct 11, 2026

Commits on Oct 13, 2026

Commits on Oct 14, 2026

Commits on Oct 15, 2026

Commits on Oct 17, 2026

Commits on Oct 18, 2026