Conversation
Port the average/roks dual-semantics EnsembleRKS reference (fixed 2/1/0 occupations, Dz = 0) and align Dz0SCF with the shared reference_energy semantics so NTTDA can treat both reference types uniformly.
Port the NTTDA response/gradient stack (deltaS = -1/0 channels, XC ledger kernels, ensemble orbital-response backend) and wire NTTDA.nuc_grad_method to nest.grad.nttda.Gradients. The linear solver uses pyscf.tdscf._lr_eig: the nest._lr_eig fork does not converge on the ensemble orbital Hessian.
Add the ensemble FD/response, scalar-ledger, gradient-layer, and known-value NTTDA energy suites.
Drop the EnsembleRKS reference path and drive the NTTDA gradient from Dz0SCF (average-occupation) and ROKS only. Dz0SCF now exposes the average-occupation marker plus a charge-RKS companion for the spin-unpolarized response, rebuilt on every call so a reused mean-field object (reset/new geometry/xc change) never reuses a stale reference. Port the ensemble suites to Dz0SCF and add a cache-invalidation regression test.
Remove the stale spin-lowering-only transition methods that shadow the upstream all-channel implementation after merging main. Restore nest._lr_eig so oscillator strengths retain upstream convergence behavior. The existing all-channel oscillator-strength regression passes without changing its tolerance.
Keep spin-channel amplitudes and coefficients explicit while sharing orbital partitions, reference responses, J/K derivative batching, Fz exchange and adjoint assembly. Consolidate identical GGA and meta-GGA quadrature flows with explicit density dimensions and kernel dispatch. Remove 1216 net lines from the gradient package; retain existing numerical checks and tolerances.
Retain the branch's PySCF response solver: the shared NEST solver fails the existing small-system Dz0SCF convergence checks. Explicitly converge oscillator-strength amplitudes to 1e-9 with lindep=1e-18 rather than relying on the default 1e-5 residual; reference values and assertion tolerances are unchanged. Seven targeted tests and eight subtests pass.
Build channel projections once per gradient and reuse the shared transition-potential projection. Share adjoint assembly and Krylov mechanics while retaining reference-specific Hessians and preconditioners. Use a singleton density-feature axis for LDA to share the GGA/MGGA quadrature loops, retaining distinct pair kernels. Consolidate HF/DFT Fock derivatives and remove unreferenced channel projection wrappers. Net reduction: 604 implementation lines; existing numerical tests and tolerances are unchanged.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Adds NTTDA analytic nuclear gradients for ΔS = −1 and 0 with both conventional ROKS and average-occupation ROKS (AOCSCF) references. Finite-difference gradients are available for all three spin channels.
td.Gradients().kernel(state=n)and the gradient scanner return derivatives of the selected total energy.Implementation and compatibility
nest._lr_eigwith its existing subspace sizing.td._scf.e_totandtd.e_totenergy interfaces. Includes selected-state scanner handling, symmetry-aware atom selection, maintained examples, and documentation.Validation
Run in the
nestConda environment with PySCF 2.14.0 andOMP_NUM_THREADS=2:test_nttda.py,test_nttda_aocscf.py,test_nttda_oscillator_strength.py): 17 passed, 36 subtests passed.test_nttda_grad.py,test_nttda_aocscf_fd.py,test_nttda_aocscf_response.py,test_nttda_gradient_layers.py,test_nttda_scalar_ledger.py): 18 passed, 105 subtests passed.python -m ruff check .andgit diff --check main...HEAD: passed.nobetasettings.The full repository suite has not been rerun on the final commit; the results above cover the affected NTTDA and gradient tests.
Limitations
Analytic ΔS = +1 gradients remain unsupported. Analytic DFT gradients omit numerical-grid response; finite-difference validation uses fixed grids by default. The NTTDA gradient driver does not support density fitting, X2C, solvent response, or nonlocal correlation.