Repository navigation

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

Open

maki49 wants to merge 79 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

Commits on Oct 7, 2026

Commits on Oct 8, 2026