SOC (1e and 2e cases) by VladimirMakhnev · Pull Request #128 · Open-Quantum-Platform/openqp

VladimirMakhnev · 2026-05-18T04:32:53Z

Added: SOC calculations.

The calculation procedure:
SCF -- MRSF (singlet) -- MRSF (triplet) -- SOC
(see runfunc.py)

Keywords:

new runtype: "runtype = soc"
new input keyword: "soc_2e". Two modes supported:
"soc_2e = 0" -- only 1e SOC calculation
"soc_2e = 1" -- 1e and 2e SOC calculation

New input folder -- "SOC" with examples of thymine and CH3Br.

New module: "soc_mrsf.F90".

This reverts commit ea10080. revert commit with lx,ly functions

…case, 0 for 1e only.

mohsenkor · 2026-05-18T05:37:01Z

-        pass
-    else:
-        pass
+#    soc_type = mol.config['properties']['soc']


This part may be needed when we perform NAMD calculations considering SOC, so it may be better to make it compatible with this one.

What arrays are required for the NAMD calculations? What is prefereable output from SOC module for the NAMD?

Alireza-Lashkaripour · 2026-05-18T23:12:00Z

Thank you for your PR, here are the result of my review:
After run_test, I got two errors:
Execution Date: 2026-05-19 06:44:46
Git Branch Info: soc-mrsf
Git Commit Info: 3318034
Output dir: /bighome/alireza/clone/pr128/oqp/openqp_test_tmp_2026-05-19_06-43-24
Total tests: 201
Passed: 199
Failed: 0
Errors: 2

For the example/SOC files, please change the names of the inputs into more detailed format, like for example: C2H4_BHHLYP-MRSFTDDFT_MECI.inp
Also, we need to save the result for the SOC test in .json format for future tests.

Regarding Wiki, please give a more detail explanation for the keywords in SOC inputs, so we can put then in wiki as will as a detail explantion of inputs and log files for https://github.com/Open-Quantum-Platform/openqp/wiki/Example%20Input%20Files

ConstLike

LGTM! Thanks for your work!

Please address some minor corrections:

move new constants to source/physical_constants.F90
clean from GMS traces
Move all the cutoffs (CHANGE THE NAME TO CLEAR MEANING) to constants.F90?
keyword soc_2e -> [tdhf]?
add new SOC tests with PR.

ConstLike · 2026-05-20T11:56:41Z

+#    mol.data['OQP::soc_']
+


rm 4 lines.

ConstLike · 2026-05-20T11:58:36Z

+
+
+    !TEST
+!    write(*,'(a,i0)') 'DEBUG compute_soc2e_ints called, ng=', ng
+
+


ConstLike · 2026-05-20T11:58:55Z

+    if (allocated(gdat%fj   )) deallocate(gdat%fj   )
+
+  end subroutine soc2e_gdat_clean
+


ConstLike · 2026-05-20T11:59:09Z

+
+

ConstLike · 2026-05-20T11:59:36Z

+                ng, gdat%nroots, nmax, mmax, nimax, njmax, nkmax, nlmax, &
+                gdat%dij, gdat%dkl)
+
+!if (gdat%am(1)==0 .and. gdat%am(2)==0 .and. &


ConstLike · 2026-05-20T15:32:24Z

+  nstate  = ns + 3*nt
+  nstate4 = int(nstate, 4)
+
+  ! Fixed lwork: safe minimum for small matrices, avoids workspace query overhead


'lwork' is GMS style.

Chech comment

ConstLike · 2026-05-20T15:38:29Z

+  character(len=*), parameter :: module_name = "soc_mrsf_mod"
+
+
+  real(kind=dp), parameter :: ZEFF_321G(54) = [ &


Where do you use it?

ConstLike · 2026-05-20T15:40:55Z

+    mol.data.set_tdhf_multiplicity(1)
+    mol.singlet_energies = sp.excitation(ref_energy)
+
+    mol.data['OQP::td_singlet_energies'] = mol.data['OQP::td_energies']


maybe mol.data['OQP::td_singlet_energies'] = mol.data['OQP::td_energies'].copy()? check it.

ConstLike · 2026-05-20T16:24:40Z

+
+    ! --- Step 2b: 2e mean-field SOC correction ---
+    if (do_2e_soc) then
+      allocate(den_rohf(nbf, nbf), stat=ok)


den_rohf -> den_core?

ConstLike · 2026-05-20T16:26:30Z

+!           ---- Exchange 12 ----
+!           W(K,J) += -3*D(I,L)*VAL
+!           W(K,I) -= -3*D(J,L)*VAL
+            val3 = -3.0_dp


why -3? add comment.

ConstLike · 2026-05-20T16:50:36Z

Consider adding OpenMP multithreading support to the soc2e_driver.

karmachoi

Codex review focused on OpenMP/MPI readiness of the SOC integral paths:

P1 — 1e SOC integral path is not MPI/OpenMP-ready

compute_soc_ao in source/modules/soc_mrsf.F90 zeroes the full AO SOC buffers and then serially loops over all shell pairs while updating shared packed matrices. There is no MPI rank partitioning / allreduce, and adding OpenMP around this loop would race on lx_ao, ly_ao, and lz_ao through update_triang_matrix unless each thread uses private scratch/accumulators.

Please make the 1e SOC path follow the existing parallel integral pattern: split shell pairs across MPI ranks, use thread-private shpair_t/SOC block scratch and private AO accumulators for OpenMP, reduce the thread-local AO buffers, then MPI-allreduce the final Lx/Ly/Lz AO matrices.

P1 — 2e SOC / ERI path has MPI splitting but is not OpenMP-ready

soc2e_driver in source/integrals/grd2_rys.F90 partitions shell-pair work across MPI ranks, but the quartet loop remains serial and calls soc2e_rys_compute, which accumulates directly into the shared wao tensor inside compute_soc2e_ao. Sharing gdat or wao under an OpenMP loop would race.

Please make the 2e ERI/SOC path OpenMP-ready by using thread-private soc2e_int_data_t scratch plus per-thread wao accumulators, reducing those AO tensors before the existing MPI allreduce. This should cover the 2e ERI path rather than leaving it serial on each MPI-assigned chunk.

P2 — SOC output/debug path is not MPI-safe

soc_mrsf opens/appends the same log file from every rank, leaves debug_soc_prints = .true., and writes AO/debug/SOC tables without rank gating. Once 1e/2e integral work is distributed, this will produce duplicate or interleaved output and potentially duplicated tag publication.

Please gate printing and tag publication to rank 0, or use the repository’s existing parallel-aware printing/data-publication pattern.

Overall: the SOC implementation should not be merged as-is until both 1e and 2e integral paths are explicitly OpenMP- and MPI-ready.

Generated with Codex CLI and checked against PR #128 diff.

karmachoi

Codex CLI review for PR #128

Codex reviewed the PR diff in read-only mode and did not modify files or run project code.

Findings

Blocking: direct zheev call uses integer(4) LAPACK arguments. source/modules/soc_mrsf.F90:1006 and the call at source/modules/soc_mrsf.F90:1039 bypass OQP's blas_int abstraction used in source/mathlib/eigen.F90. On ILP64 builds this is an ABI mismatch and can corrupt the call or crash. Please use integer(blas_int)/oqp_linalg, or add a complex Hermitian wrapper consistent with eigen.
Blocking: SOC results are not returned in Python/JSON. compute_soc stores tagarray outputs, but pyoqp/oqp/molecule/molecule.py:228-233 still returns [] from get_soc(), and pyoqp/oqp/pyoqp.py:141 reports self.mol.soc, which is never populated from OQP::soc_*. Reference JSON/tests for runtype=soc will therefore show empty SOC even when Fortran produced SOC data.
High: 2e SOC debug logging is always enabled. source/modules/soc_mrsf.F90:91 sets debug_soc_prints = .true., causing shell tables and full AO SOC matrices to be printed for every 2e SOC run at source/modules/soc_mrsf.F90:166-193. This will make normal outputs very large and unstable for reference comparisons.
High: unconditional stdout debug prints remain in integral kernels. source/integrals/mod_1e_primitives.F90:2188-2195 writes derivative debug tables directly to stdout for d/d shell pairs. These bypass the log/verbosity system and will pollute CLI/test output.
Medium: SOC input help still says SOC is unimplemented. pyoqp/oqp/utils/input_checker.py:39 still tells users that "soc and neb are recognized but not implemented yet" even though soc was moved into supported runtypes. Please update the help text and docs/examples together.

Given the two blocking issues above, Codex would not merge this PR as-is.

# Conflicts: # pyoqp/oqp/utils/input_checker.py

VladimirMakhnev added 7 commits March 5, 2026 16:17

Initial Lx, Ly function commit

ea10080

Revert "Initial Lx, Ly function commit"

e1c5932

This reverts commit ea10080. revert commit with lx,ly functions

Added: full scalar relativistic correction, initial skeleton for SOC

b0229db

Full 1-e implementation of SOC. No switch 1e/2e case yet.

eb8a9cf

Examples commit

2bf0b1c

Full 2-e implementation of SOC. Add input parametr soc_2e = 1 for 2e …

dd01e55

…case, 0 for 1e only.

Added SOC state decomposition, updated EXAMPLES/SOC

3318034

VladimirMakhnev requested review from Alireza-Lashkaripour, ConstLike, karmachoi and mohsenkor May 18, 2026 04:32

mohsenkor reviewed May 18, 2026

View reviewed changes

VladimirMakhnev added 2 commits May 19, 2026 15:38

Merge branch 'main' into soc-mrsf

2d55d85

Added .JSON output for SOC case, updated examples for SOC

caf376d

ConstLike requested changes May 20, 2026

View reviewed changes

karmachoi reviewed May 25, 2026

View reviewed changes

Merge branch 'main' into soc-mrsf

b3db100

karmachoi reviewed May 25, 2026

View reviewed changes

cheolhochoi added 5 commits May 25, 2026 22:27

docs: document SOC and OpenqpView updates

2a9a824

Merge remote-tracking branch 'origin/main' into soc-mrsf

b0e2a55

# Conflicts: # pyoqp/oqp/utils/input_checker.py

docs: add geomeTRIC and advanced guess manuals

8b14a1e

docs: move README updates to separate PR

e8e10a9

docs: keep SOC documentation with SOC PR

aea3f99

karmachoi mentioned this pull request May 25, 2026

docs: update README and PyOQP manuals #138

Merged



		!TEST
		! write(*,'(a,i0)') 'DEBUG compute_soc2e_ints called, ng=', ng

		if (allocated(gdat%fj )) deallocate(gdat%fj )

		end subroutine soc2e_gdat_clean

		character(len=*), parameter :: module_name = "soc_mrsf_mod"


		real(kind=dp), parameter :: ZEFF_321G(54) = [ &

Conversation

VladimirMakhnev commented May 18, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Alireza-Lashkaripour commented May 18, 2026

Uh oh!

ConstLike left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ConstLike commented May 20, 2026

Uh oh!

karmachoi left a comment

Choose a reason for hiding this comment

Uh oh!

karmachoi left a comment

Choose a reason for hiding this comment

Codex CLI review for PR #128

Findings

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants