Line data Source code
1 : !--------------------------------------------------------------------------------------------------!
2 : ! CP2K: A general program to perform molecular dynamics simulations !
3 : ! Copyright 2000-2024 CP2K developers group <https://cp2k.org> !
4 : ! !
5 : ! SPDX-License-Identifier: GPL-2.0-or-later !
6 : !--------------------------------------------------------------------------------------------------!
7 :
8 : ! **************************************************************************************************
9 : !> \brief Second order perturbation correction to XAS_TDP spectra (i.e. shift)
10 : !> \author A. Bussy (01.2020)
11 : ! **************************************************************************************************
12 :
13 : MODULE xas_tdp_correction
14 : USE admm_types, ONLY: admm_type
15 : USE admm_utils, ONLY: admm_correct_for_eigenvalues
16 : USE bibliography, ONLY: Bussy2021b,&
17 : Shigeta2001,&
18 : cite_reference
19 : USE cp_array_utils, ONLY: cp_1d_i_p_type,&
20 : cp_1d_r_p_type
21 : USE cp_blacs_env, ONLY: cp_blacs_env_type
22 : USE cp_cfm_types, ONLY: cp_cfm_create,&
23 : cp_cfm_get_submatrix,&
24 : cp_cfm_release,&
25 : cp_cfm_type,&
26 : cp_fm_to_cfm
27 : USE cp_control_types, ONLY: dft_control_type
28 : USE cp_dbcsr_api, ONLY: &
29 : dbcsr_copy, dbcsr_create, dbcsr_distribution_get, dbcsr_distribution_new, &
30 : dbcsr_distribution_release, dbcsr_distribution_type, dbcsr_get_info, dbcsr_p_type, &
31 : dbcsr_release, dbcsr_type
32 : USE cp_dbcsr_operations, ONLY: copy_fm_to_dbcsr,&
33 : cp_dbcsr_sm_fm_multiply,&
34 : dbcsr_deallocate_matrix_set
35 : USE cp_fm_diag, ONLY: choose_eigv_solver
36 : USE cp_fm_struct, ONLY: cp_fm_struct_create,&
37 : cp_fm_struct_p_type,&
38 : cp_fm_struct_release,&
39 : cp_fm_struct_type
40 : USE cp_fm_types, ONLY: cp_fm_create,&
41 : cp_fm_get_diag,&
42 : cp_fm_get_submatrix,&
43 : cp_fm_release,&
44 : cp_fm_to_fm,&
45 : cp_fm_to_fm_submat,&
46 : cp_fm_type
47 : USE cp_log_handling, ONLY: cp_logger_get_default_io_unit
48 : USE dbt_api, ONLY: &
49 : dbt_contract, dbt_copy, dbt_copy_matrix_to_tensor, dbt_create, dbt_default_distvec, &
50 : dbt_destroy, dbt_distribution_destroy, dbt_distribution_new, dbt_distribution_type, &
51 : dbt_finalize, dbt_get_block, dbt_get_info, dbt_iterator_blocks_left, &
52 : dbt_iterator_next_block, dbt_iterator_start, dbt_iterator_stop, dbt_iterator_type, &
53 : dbt_pgrid_create, dbt_pgrid_destroy, dbt_pgrid_type, dbt_put_block, dbt_type
54 : USE hfx_admm_utils, ONLY: create_admm_xc_section
55 : USE input_section_types, ONLY: section_vals_create,&
56 : section_vals_get_subs_vals,&
57 : section_vals_release,&
58 : section_vals_retain,&
59 : section_vals_set_subs_vals,&
60 : section_vals_type
61 : USE kinds, ONLY: dp
62 : USE machine, ONLY: m_flush
63 : USE mathlib, ONLY: complex_diag
64 : USE message_passing, ONLY: mp_para_env_type
65 : USE parallel_gemm_api, ONLY: parallel_gemm
66 : USE physcon, ONLY: evolt
67 : USE qs_environment_types, ONLY: get_qs_env,&
68 : qs_environment_type
69 : USE qs_ks_methods, ONLY: qs_ks_build_kohn_sham_matrix
70 : USE qs_mo_types, ONLY: deallocate_mo_set,&
71 : duplicate_mo_set,&
72 : get_mo_set,&
73 : mo_set_type,&
74 : reassign_allocated_mos
75 : USE util, ONLY: get_limit
76 : USE xas_tdp_kernel, ONLY: contract2_AO_to_doMO,&
77 : ri_all_blocks_mm
78 : USE xas_tdp_types, ONLY: donor_state_type,&
79 : xas_tdp_control_type,&
80 : xas_tdp_env_type
81 :
82 : !$ USE OMP_LIB, ONLY: omp_get_max_threads, omp_get_thread_num
83 : #include "./base/base_uses.f90"
84 :
85 : IMPLICIT NONE
86 : PRIVATE
87 :
88 : CHARACTER(len=*), PARAMETER, PRIVATE :: moduleN = 'xas_tdp_correction'
89 :
90 : PUBLIC :: gw2x_shift, get_soc_splitting
91 :
92 : CONTAINS
93 :
94 : ! **************************************************************************************************
95 : !> \brief Computes the ionization potential using the GW2X method of Shigeta et. al. The result cam
96 : !> be used for XAS correction (shift) or XPS directly.
97 : !> \param donor_state ...
98 : !> \param xas_tdp_env ...
99 : !> \param xas_tdp_control ...
100 : !> \param qs_env ...
101 : ! **************************************************************************************************
102 30 : SUBROUTINE GW2X_shift(donor_state, xas_tdp_env, xas_tdp_control, qs_env)
103 :
104 : TYPE(donor_state_type), POINTER :: donor_state
105 : TYPE(xas_tdp_env_type), POINTER :: xas_tdp_env
106 : TYPE(xas_tdp_control_type), POINTER :: xas_tdp_control
107 : TYPE(qs_environment_type), POINTER :: qs_env
108 :
109 : CHARACTER(len=*), PARAMETER :: routineN = 'GW2X_shift'
110 :
111 : INTEGER :: ex_idx, exat, first_domo(2), handle, i, ido_mo, iloc, ilocat, ispin, jspin, &
112 : locat, nao, natom, ndo_mo, nhomo(2), nlumo(2), nonloc, nspins, start_sgf
113 30 : INTEGER, DIMENSION(:), POINTER :: nsgf_blk
114 : LOGICAL :: pseudo_canonical
115 : REAL(dp) :: og_hfx_frac
116 30 : REAL(dp), ALLOCATABLE, DIMENSION(:, :) :: contract_coeffs_backup
117 : TYPE(admm_type), POINTER :: admm_env
118 30 : TYPE(cp_1d_r_p_type), ALLOCATABLE, DIMENSION(:) :: homo_evals, lumo_evals
119 : TYPE(cp_blacs_env_type), POINTER :: blacs_env
120 : TYPE(cp_fm_struct_p_type), ALLOCATABLE, &
121 30 : DIMENSION(:) :: all_struct, homo_struct, lumo_struct
122 : TYPE(cp_fm_struct_type), POINTER :: hoho_struct, lulu_struct
123 : TYPE(cp_fm_type) :: hoho_fock, hoho_work, homo_work, &
124 : lulu_fock, lulu_work, lumo_work
125 30 : TYPE(cp_fm_type), ALLOCATABLE, DIMENSION(:) :: all_coeffs, homo_coeffs, lumo_coeffs
126 : TYPE(cp_fm_type), POINTER :: mo_coeff
127 30 : TYPE(dbcsr_p_type), DIMENSION(:), POINTER :: dbcsr_work, fock_matrix, matrix_ks
128 30 : TYPE(dbt_type), ALLOCATABLE, DIMENSION(:) :: ja_X, oI_Y
129 30 : TYPE(dbt_type), ALLOCATABLE, DIMENSION(:, :) :: mo_template
130 : TYPE(dft_control_type), POINTER :: dft_control
131 30 : TYPE(mo_set_type), DIMENSION(:), POINTER :: mos
132 : TYPE(mp_para_env_type), POINTER :: para_env
133 : TYPE(section_vals_type), POINTER :: xc_fun_empty, xc_fun_original, xc_section
134 :
135 30 : NULLIFY (xc_fun_empty, xc_fun_original, xc_section, mos, dft_control, dbcsr_work, &
136 30 : fock_matrix, matrix_ks, para_env, mo_coeff, blacs_env, nsgf_blk)
137 :
138 30 : CALL cite_reference(Shigeta2001)
139 30 : CALL cite_reference(Bussy2021b)
140 :
141 30 : CALL timeset(routineN, handle)
142 :
143 : !The GW2X correction we want to compute goes like this, where omega is the corrected epsilon_I:
144 : !omega = eps_I + 0.5 * sum_ajk |<Ia||jk>|^2/(omega + eps_a - eps_j - eps_k)
145 : ! + 0.5 * sum_jab |<Ij||ab>|^2/(omega + eps_j - eps_a - eps_b)
146 : ! j,k denote occupied spin-orbitals and a,b denote virtual spin orbitals
147 :
148 : !The strategy is the following (we assume restricted closed-shell):
149 : !1) Get the LUMOs from xas_tdp_env
150 : !2) Get the HOMOs from qs_env
151 : !3) Compute or fetch the generalize Fock matric
152 : !4) Diagonalize it in the subspace of HOMOs and LUMOs (or just take diagonal matrix elements)
153 : !5) Build the full HOMO-LUMO basis that we will use and compute eigenvalues
154 : !6) Iterate over GW2X steps to compute the self energy
155 :
156 : !We implement 2 approaches => diagonal elements of Fock matrix with original MOs and
157 : !pseudo-canonical MOs
158 30 : pseudo_canonical = xas_tdp_control%pseudo_canonical
159 :
160 : !Get donor state info
161 30 : ndo_mo = donor_state%ndo_mo
162 30 : nspins = 1; IF (xas_tdp_control%do_uks .OR. xas_tdp_control%do_roks) nspins = 2
163 :
164 : !1) Get the LUMO coefficients from the xas_tdp_env, that have been precomputed
165 :
166 : CALL get_qs_env(qs_env, matrix_ks=matrix_ks, mos=mos, para_env=para_env, &
167 30 : blacs_env=blacs_env, natom=natom)
168 :
169 158 : ALLOCATE (lumo_struct(nspins), lumo_coeffs(nspins))
170 :
171 64 : DO ispin = 1, nspins
172 34 : CALL get_mo_set(mos(ispin), homo=nhomo(ispin), nao=nao)
173 34 : nlumo(ispin) = nao - nhomo(ispin)
174 :
175 : CALL cp_fm_struct_create(lumo_struct(ispin)%struct, para_env=para_env, context=blacs_env, &
176 34 : ncol_global=nlumo(ispin), nrow_global=nao)
177 :
178 34 : CALL cp_fm_create(lumo_coeffs(ispin), lumo_struct(ispin)%struct)
179 98 : CALL cp_fm_to_fm(xas_tdp_env%lumo_evecs(ispin), lumo_coeffs(ispin))
180 : END DO
181 :
182 : !2) get the HOMO coeffs. Reminder: keep all non-localized MOs + those localized on core atom
183 : ! For this to work, it is assumed that the LOCALIZE keyword is used
184 188 : ALLOCATE (homo_struct(nspins), homo_coeffs(nspins))
185 :
186 64 : DO ispin = 1, nspins
187 34 : nonloc = nhomo(ispin) - xas_tdp_control%n_search
188 34 : exat = donor_state%at_index
189 80 : ex_idx = MINLOC(ABS(xas_tdp_env%ex_atom_indices - exat), 1)
190 156 : locat = COUNT(xas_tdp_env%mos_of_ex_atoms(:, ex_idx, ispin) == 1)
191 :
192 : CALL cp_fm_struct_create(homo_struct(ispin)%struct, para_env=para_env, context=blacs_env, &
193 34 : ncol_global=locat + nonloc, nrow_global=nao)
194 34 : CALL cp_fm_create(homo_coeffs(ispin), homo_struct(ispin)%struct)
195 :
196 34 : CALL get_mo_set(mos(ispin), mo_coeff=mo_coeff)
197 : CALL cp_fm_to_fm_submat(mo_coeff, homo_coeffs(ispin), nrow=nao, ncol=nonloc, s_firstrow=1, &
198 34 : s_firstcol=xas_tdp_control%n_search + 1, t_firstrow=1, t_firstcol=locat + 1)
199 :
200 : !this bit is taken from xas_tdp_methods
201 34 : ilocat = 1
202 156 : DO iloc = 1, xas_tdp_control%n_search
203 122 : IF (xas_tdp_env%mos_of_ex_atoms(iloc, ex_idx, ispin) == -1) CYCLE
204 : CALL cp_fm_to_fm_submat(mo_coeff, homo_coeffs(ispin), nrow=nao, ncol=1, s_firstrow=1, &
205 82 : s_firstcol=iloc, t_firstrow=1, t_firstcol=ilocat)
206 : !keep track of donor MO index
207 82 : IF (iloc == donor_state%mo_indices(1, ispin)) first_domo(ispin) = ilocat !first donor MO
208 :
209 156 : ilocat = ilocat + 1
210 : END DO
211 64 : nhomo(ispin) = locat + nonloc
212 : END DO
213 :
214 : !3) Computing the generalized Fock Matrix, if not there already
215 30 : IF (ASSOCIATED(xas_tdp_env%fock_matrix)) THEN
216 12 : fock_matrix => xas_tdp_env%fock_matrix
217 : ELSE
218 18 : BLOCK
219 18 : TYPE(mo_set_type), DIMENSION(:), ALLOCATABLE :: backup_mos
220 :
221 56 : ALLOCATE (xas_tdp_env%fock_matrix(nspins))
222 18 : fock_matrix => xas_tdp_env%fock_matrix
223 :
224 : ! remove the xc_functionals and set HF fraction to 1
225 18 : xc_section => section_vals_get_subs_vals(qs_env%input, "DFT%XC")
226 18 : xc_fun_original => section_vals_get_subs_vals(xc_section, "XC_FUNCTIONAL")
227 18 : CALL section_vals_retain(xc_fun_original)
228 18 : CALL section_vals_create(xc_fun_empty, xc_fun_original%section)
229 18 : CALL section_vals_set_subs_vals(xc_section, "XC_FUNCTIONAL", xc_fun_empty)
230 18 : CALL section_vals_release(xc_fun_empty)
231 18 : og_hfx_frac = qs_env%x_data(1, 1)%general_parameter%fraction
232 54 : qs_env%x_data(:, :)%general_parameter%fraction = 1.0_dp
233 :
234 : !In case of ADMM, we need to re-create the admm XC section for the new hfx_fraction
235 : !We also need to make a backup of the MOs as theiy are modified
236 18 : CALL get_qs_env(qs_env, dft_control=dft_control, admm_env=admm_env)
237 18 : IF (dft_control%do_admm) THEN
238 2 : IF (ASSOCIATED(admm_env%xc_section_primary)) CALL section_vals_release(admm_env%xc_section_primary)
239 2 : IF (ASSOCIATED(admm_env%xc_section_aux)) CALL section_vals_release(admm_env%xc_section_aux)
240 2 : CALL create_admm_xc_section(qs_env%x_data, xc_section, admm_env)
241 :
242 10 : ALLOCATE (backup_mos(SIZE(mos)))
243 4 : DO i = 1, SIZE(mos)
244 4 : CALL duplicate_mo_set(backup_mos(i), mos(i))
245 : END DO
246 : END IF
247 :
248 56 : ALLOCATE (dbcsr_work(nspins))
249 38 : DO ispin = 1, nspins
250 20 : ALLOCATE (dbcsr_work(ispin)%matrix)
251 38 : CALL dbcsr_copy(dbcsr_work(ispin)%matrix, matrix_ks(ispin)%matrix)
252 : END DO
253 :
254 : !both spins treated internally
255 18 : CALL qs_ks_build_kohn_sham_matrix(qs_env, calculate_forces=.FALSE., just_energy=.FALSE.)
256 :
257 38 : DO ispin = 1, nspins
258 20 : ALLOCATE (fock_matrix(ispin)%matrix)
259 20 : CALL dbcsr_copy(fock_matrix(ispin)%matrix, matrix_ks(ispin)%matrix, name="FOCK MATRIX")
260 20 : CALL dbcsr_release(matrix_ks(ispin)%matrix)
261 38 : CALL dbcsr_copy(matrix_ks(ispin)%matrix, dbcsr_work(ispin)%matrix)
262 : END DO
263 18 : CALL dbcsr_deallocate_matrix_set(dbcsr_work)
264 :
265 : !In case of ADMM, we want to correct for eigenvalues
266 18 : IF (dft_control%do_admm) THEN
267 4 : DO ispin = 1, nspins
268 4 : CALL admm_correct_for_eigenvalues(ispin, admm_env, fock_matrix(ispin)%matrix)
269 : END DO
270 : END IF
271 :
272 : !restore xc and HF fraction
273 18 : CALL section_vals_set_subs_vals(xc_section, "XC_FUNCTIONAL", xc_fun_original)
274 18 : CALL section_vals_release(xc_fun_original)
275 54 : qs_env%x_data(:, :)%general_parameter%fraction = og_hfx_frac
276 :
277 20 : IF (dft_control%do_admm) THEN
278 2 : IF (ASSOCIATED(admm_env%xc_section_primary)) CALL section_vals_release(admm_env%xc_section_primary)
279 2 : IF (ASSOCIATED(admm_env%xc_section_aux)) CALL section_vals_release(admm_env%xc_section_aux)
280 2 : CALL create_admm_xc_section(qs_env%x_data, xc_section, admm_env)
281 :
282 4 : DO i = 1, SIZE(mos)
283 2 : CALL reassign_allocated_mos(mos(i), backup_mos(i))
284 4 : CALL deallocate_mo_set(backup_mos(i))
285 : END DO
286 2 : DEALLOCATE (backup_mos)
287 : END IF
288 : END BLOCK
289 : END IF
290 :
291 : !4,5) Build pseudo-canonical MOs if needed + get related Fock matrix elements
292 188 : ALLOCATE (all_struct(nspins), all_coeffs(nspins))
293 158 : ALLOCATE (homo_evals(nspins), lumo_evals(nspins))
294 30 : CALL dbcsr_get_info(matrix_ks(1)%matrix, row_blk_size=nsgf_blk)
295 120 : ALLOCATE (contract_coeffs_backup(nsgf_blk(exat), nspins*ndo_mo))
296 :
297 64 : DO ispin = 1, nspins
298 : CALL cp_fm_struct_create(hoho_struct, para_env=para_env, context=blacs_env, &
299 34 : ncol_global=nhomo(ispin), nrow_global=nhomo(ispin))
300 : CALL cp_fm_struct_create(lulu_struct, para_env=para_env, context=blacs_env, &
301 34 : ncol_global=nlumo(ispin), nrow_global=nlumo(ispin))
302 :
303 34 : CALL cp_fm_create(hoho_work, hoho_struct)
304 34 : CALL cp_fm_create(lulu_work, lulu_struct)
305 34 : CALL cp_fm_create(homo_work, homo_struct(ispin)%struct)
306 34 : CALL cp_fm_create(lumo_work, lumo_struct(ispin)%struct)
307 :
308 34 : IF (pseudo_canonical) THEN
309 : !That is where we rotate the MOs to make them pseudo canonical
310 : !The eigenvalues we get from the diagonalization
311 :
312 : !The Fock matrix in the HOMO subspace
313 32 : CALL cp_fm_create(hoho_fock, hoho_struct)
314 32 : NULLIFY (homo_evals(ispin)%array)
315 96 : ALLOCATE (homo_evals(ispin)%array(nhomo(ispin)))
316 : CALL cp_dbcsr_sm_fm_multiply(fock_matrix(ispin)%matrix, homo_coeffs(ispin), &
317 32 : homo_work, ncol=nhomo(ispin))
318 : CALL parallel_gemm('T', 'N', nhomo(ispin), nhomo(ispin), nao, 1.0_dp, homo_coeffs(ispin), &
319 32 : homo_work, 0.0_dp, hoho_fock)
320 :
321 : !diagonalize and get pseudo-canonical MOs
322 32 : CALL choose_eigv_solver(hoho_fock, hoho_work, homo_evals(ispin)%array)
323 : CALL parallel_gemm('N', 'N', nao, nhomo(ispin), nhomo(ispin), 1.0_dp, homo_coeffs(ispin), &
324 32 : hoho_work, 0.0_dp, homo_work)
325 32 : CALL cp_fm_to_fm(homo_work, homo_coeffs(ispin))
326 :
327 : !overwrite the donor_state's contract coeffs with those
328 : contract_coeffs_backup(:, (ispin - 1)*ndo_mo + 1:ispin*ndo_mo) = &
329 1052 : donor_state%contract_coeffs(:, (ispin - 1)*ndo_mo + 1:ispin*ndo_mo)
330 46 : start_sgf = SUM(nsgf_blk(1:exat - 1)) + 1
331 : CALL cp_fm_get_submatrix(homo_coeffs(ispin), &
332 : donor_state%contract_coeffs(:, (ispin - 1)*ndo_mo + 1:ispin*ndo_mo), &
333 : start_row=start_sgf, start_col=first_domo(ispin), &
334 32 : n_rows=nsgf_blk(exat), n_cols=ndo_mo)
335 :
336 : !do the same for the pseudo-LUMOs
337 32 : CALL cp_fm_create(lulu_fock, lulu_struct)
338 32 : NULLIFY (lumo_evals(ispin)%array)
339 96 : ALLOCATE (lumo_evals(ispin)%array(nlumo(ispin)))
340 : CALL cp_dbcsr_sm_fm_multiply(fock_matrix(ispin)%matrix, lumo_coeffs(ispin), &
341 32 : lumo_work, ncol=nlumo(ispin))
342 : CALL parallel_gemm('T', 'N', nlumo(ispin), nlumo(ispin), nao, 1.0_dp, lumo_coeffs(ispin), &
343 32 : lumo_work, 0.0_dp, lulu_fock)
344 :
345 : !diagonalize and get pseudo-canonical MOs
346 32 : CALL choose_eigv_solver(lulu_fock, lulu_work, lumo_evals(ispin)%array)
347 : CALL parallel_gemm('N', 'N', nao, nlumo(ispin), nlumo(ispin), 1.0_dp, lumo_coeffs(ispin), &
348 32 : lulu_work, 0.0_dp, lumo_work)
349 32 : CALL cp_fm_to_fm(lumo_work, lumo_coeffs(ispin))
350 :
351 32 : CALL cp_fm_release(lulu_fock)
352 96 : CALL cp_fm_release(hoho_fock)
353 :
354 : ELSE !using the generalized Fock matrix diagonal elements
355 :
356 : !Compute their Fock matrix diagonal
357 6 : ALLOCATE (homo_evals(ispin)%array(nhomo(ispin)))
358 : CALL cp_dbcsr_sm_fm_multiply(fock_matrix(ispin)%matrix, homo_coeffs(ispin), &
359 2 : homo_work, ncol=nhomo(ispin))
360 : CALL parallel_gemm('T', 'N', nhomo(ispin), nhomo(ispin), nao, 1.0_dp, homo_coeffs(ispin), &
361 2 : homo_work, 0.0_dp, hoho_work)
362 2 : CALL cp_fm_get_diag(hoho_work, homo_evals(ispin)%array)
363 :
364 6 : ALLOCATE (lumo_evals(ispin)%array(nlumo(ispin)))
365 : CALL cp_dbcsr_sm_fm_multiply(fock_matrix(ispin)%matrix, lumo_coeffs(ispin), &
366 2 : lumo_work, ncol=nlumo(ispin))
367 : CALL parallel_gemm('T', 'N', nlumo(ispin), nlumo(ispin), nao, 1.0_dp, lumo_coeffs(ispin), &
368 2 : lumo_work, 0.0_dp, lulu_work)
369 2 : CALL cp_fm_get_diag(lulu_work, lumo_evals(ispin)%array)
370 :
371 : END IF
372 34 : CALL cp_fm_release(homo_work)
373 34 : CALL cp_fm_release(hoho_work)
374 34 : CALL cp_fm_struct_release(hoho_struct)
375 34 : CALL cp_fm_release(lumo_work)
376 34 : CALL cp_fm_release(lulu_work)
377 34 : CALL cp_fm_struct_release(lulu_struct)
378 :
379 : !Put back homo and lumo coeffs together, to fit tensor structure
380 : CALL cp_fm_struct_create(all_struct(ispin)%struct, para_env=para_env, context=blacs_env, &
381 34 : ncol_global=nhomo(ispin) + nlumo(ispin), nrow_global=nao)
382 34 : CALL cp_fm_create(all_coeffs(ispin), all_struct(ispin)%struct)
383 : CALL cp_fm_to_fm(homo_coeffs(ispin), all_coeffs(ispin), ncol=nhomo(ispin), &
384 34 : source_start=1, target_start=1)
385 : CALL cp_fm_to_fm(lumo_coeffs(ispin), all_coeffs(ispin), ncol=nlumo(ispin), &
386 98 : source_start=1, target_start=nhomo(ispin) + 1)
387 :
388 : END DO !ispin
389 :
390 : !get semi-contracted tensor (AOs to MOs, keep RI uncontracted)
391 : CALL contract_AOs_to_MOs(ja_X, oI_Y, mo_template, all_coeffs, nhomo, nlumo, &
392 30 : donor_state, xas_tdp_env, xas_tdp_control, qs_env)
393 :
394 : !intermediate clean-up
395 64 : DO ispin = 1, nspins
396 34 : CALL cp_fm_release(all_coeffs(ispin))
397 34 : CALL cp_fm_release(homo_coeffs(ispin))
398 34 : CALL cp_fm_release(lumo_coeffs(ispin))
399 34 : CALL cp_fm_struct_release(all_struct(ispin)%struct)
400 34 : CALL cp_fm_struct_release(lumo_struct(ispin)%struct)
401 64 : CALL cp_fm_struct_release(homo_struct(ispin)%struct)
402 : END DO
403 :
404 : !6) GW2X iterations
405 :
406 30 : IF (nspins == 1) THEN
407 : !restricted-closed shell: only alpha spin
408 : CALL GW2X_rcs_iterations(first_domo(1), ja_X(1), oI_Y, mo_template(1, 1), homo_evals(1)%array, &
409 26 : lumo_evals(1)%array, donor_state, xas_tdp_control, qs_env)
410 : ELSE
411 : !open-shell, need both spins
412 : CALL GW2X_os_iterations(first_domo, ja_X, oI_Y, mo_template, homo_evals, lumo_evals, &
413 4 : donor_state, xas_tdp_control, qs_env)
414 : END IF
415 :
416 : !restore proper contract_coeffs
417 30 : IF (pseudo_canonical) THEN
418 1048 : donor_state%contract_coeffs(:, :) = contract_coeffs_backup(:, :)
419 : END IF
420 :
421 : !Final clean-up
422 72 : DO ido_mo = 1, nspins*ndo_mo
423 72 : CALL dbt_destroy(oI_Y(ido_mo))
424 : END DO
425 64 : DO ispin = 1, nspins
426 34 : CALL dbt_destroy(ja_X(ispin))
427 34 : DEALLOCATE (homo_evals(ispin)%array)
428 34 : DEALLOCATE (lumo_evals(ispin)%array)
429 106 : DO jspin = 1, nspins
430 76 : CALL dbt_destroy(mo_template(ispin, jspin))
431 : END DO
432 : END DO
433 72 : DEALLOCATE (oI_Y, homo_evals, lumo_evals)
434 :
435 30 : CALL timestop(handle)
436 :
437 166 : END SUBROUTINE GW2X_shift
438 :
439 : ! **************************************************************************************************
440 : !> \brief Preforms the GW2X iterations in the restricted-closed shell formalism according to the
441 : !> Newton-Raphson method
442 : !> \param first_domo index of the first core donor MO to consider
443 : !> \param ja_X semi-contracted tensor with j: occupied MO, a: virtual MO, X: RI basis element
444 : !> \param oI_Y semi-contracted tensors with o: all MOs, I donor core MO, Y: RI basis element
445 : !> \param mo_template tensor template for fully MO contracted tensor
446 : !> \param homo_evals ...
447 : !> \param lumo_evals ...
448 : !> \param donor_state ...
449 : !> \param xas_tdp_control ...
450 : !> \param qs_env ...
451 : ! **************************************************************************************************
452 26 : SUBROUTINE GW2X_rcs_iterations(first_domo, ja_X, oI_Y, mo_template, homo_evals, lumo_evals, &
453 : donor_state, xas_tdp_control, qs_env)
454 :
455 : INTEGER, INTENT(IN) :: first_domo
456 : TYPE(dbt_type), INTENT(inout) :: ja_X
457 : TYPE(dbt_type), DIMENSION(:), INTENT(inout) :: oI_Y
458 : TYPE(dbt_type), INTENT(inout) :: mo_template
459 : REAL(dp), DIMENSION(:), INTENT(IN) :: homo_evals, lumo_evals
460 : TYPE(donor_state_type), POINTER :: donor_state
461 : TYPE(xas_tdp_control_type), POINTER :: xas_tdp_control
462 : TYPE(qs_environment_type), POINTER :: qs_env
463 :
464 : CHARACTER(len=*), PARAMETER :: routineN = 'GW2X_rcs_iterations'
465 :
466 : INTEGER :: batch_size, bounds_1d(2), bounds_2d(2, 2), handle, i, ibatch, ido_mo, iloop, &
467 : max_iter, nbatch_occ, nbatch_virt, nblk_occ, nblk_virt, nblks(3), ndo_mo, nhomo, nlumo, &
468 : occ_bo(2), output_unit, tmp_sum, virt_bo(2)
469 26 : INTEGER, ALLOCATABLE, DIMENSION(:) :: mo_blk_size
470 : REAL(dp) :: c_os, c_ss, dg, diff, ds1, ds2, eps_I, &
471 : eps_iter, g, omega_k, parts(4), s1, s2
472 858 : TYPE(dbt_type) :: aj_Ib, aj_Ib_diff, aj_X, ja_Ik, &
473 234 : ja_Ik_diff
474 : TYPE(mp_para_env_type), POINTER :: para_env
475 :
476 26 : CALL timeset(routineN, handle)
477 :
478 26 : eps_iter = xas_tdp_control%gw2x_eps
479 26 : max_iter = xas_tdp_control%max_gw2x_iter
480 26 : c_os = xas_tdp_control%c_os
481 26 : c_ss = xas_tdp_control%c_ss
482 26 : batch_size = xas_tdp_control%batch_size
483 :
484 26 : ndo_mo = donor_state%ndo_mo
485 26 : output_unit = cp_logger_get_default_io_unit()
486 :
487 26 : nhomo = SIZE(homo_evals)
488 26 : nlumo = SIZE(lumo_evals)
489 :
490 26 : CALL get_qs_env(qs_env, para_env=para_env)
491 :
492 : !We use the Newton-Raphson method to find the zero of the function:
493 : !g(omega) = eps_I - omega + mp2 terms, dg(omega) = -1 + d/d_omega (mp2 terms)
494 : !We simply compute at each iteration: omega_k+1 = omega_k - g(omega_k)/dg(omega_k)
495 :
496 : !need transposed tensor of (ja|X) for optimal contraction scheme (s.t. (aj|X) block is on same
497 : !processor as (ja|X))
498 26 : CALL dbt_create(ja_X, aj_X)
499 26 : CALL dbt_copy(ja_X, aj_X, order=[2, 1, 3])
500 :
501 : !split the MO blocks into batches for memory friendly batched contraction
502 : !huge dense tensors never need to be stored
503 26 : CALL dbt_get_info(ja_X, nblks_total=nblks)
504 78 : ALLOCATE (mo_blk_size(nblks(1)))
505 26 : CALL dbt_get_info(ja_X, blk_size_1=mo_blk_size)
506 :
507 26 : tmp_sum = 0
508 52 : DO i = 1, nblks(1)
509 52 : tmp_sum = tmp_sum + mo_blk_size(i)
510 52 : IF (tmp_sum == nhomo) THEN
511 26 : nblk_occ = i
512 26 : nblk_virt = nblks(1) - i
513 26 : EXIT
514 : END IF
515 : END DO
516 26 : nbatch_occ = MAX(1, nblk_occ/batch_size)
517 26 : nbatch_virt = MAX(1, nblk_virt/batch_size)
518 :
519 : !Loop over donor_states
520 60 : DO ido_mo = 1, ndo_mo
521 34 : IF (output_unit > 0) THEN
522 : WRITE (UNIT=output_unit, FMT="(/,T5,A,I2,A,I4,A,/,T5,A)") &
523 17 : "- GW2X correction for donor MO with spin ", 1, &
524 17 : " and MO index ", donor_state%mo_indices(ido_mo, 1), ":", &
525 34 : " iteration convergence (eV)"
526 17 : CALL m_flush(output_unit)
527 : END IF
528 :
529 : !starting values
530 34 : eps_I = homo_evals(first_domo + ido_mo - 1)
531 34 : omega_k = eps_I
532 34 : iloop = 0
533 34 : diff = 2.0_dp*eps_iter
534 :
535 168 : DO WHILE (ABS(diff) > eps_iter)
536 134 : iloop = iloop + 1
537 :
538 : !Compute the mp2 terms and their first derivative
539 134 : parts = 0.0_dp
540 :
541 : !We do batched contraction for (ja|Ik) and (ja|Ib) to never have to carry the full tensor
542 276 : DO ibatch = 1, nbatch_occ
543 :
544 142 : occ_bo = get_limit(nblk_occ, nbatch_occ, ibatch - 1)
545 710 : bounds_1d = [SUM(mo_blk_size(1:occ_bo(1) - 1)) + 1, SUM(mo_blk_size(1:occ_bo(2)))]
546 :
547 142 : CALL dbt_create(mo_template, ja_Ik)
548 : CALL dbt_contract(alpha=1.0_dp, tensor_1=ja_X, tensor_2=oI_Y(ido_mo), &
549 : beta=0.0_dp, tensor_3=ja_Ik, contract_1=[3], &
550 : notcontract_1=[1, 2], contract_2=[2], notcontract_2=[1], &
551 142 : map_1=[1, 2], map_2=[3], bounds_3=bounds_1d)
552 :
553 : !opposite-spin contribution
554 : CALL calc_os_oov_contrib(parts(1), parts(2), ja_Ik, homo_evals, lumo_evals, homo_evals, &
555 142 : omega_k, c_os, nhomo)
556 :
557 426 : bounds_2d(:, 2) = bounds_1d
558 142 : bounds_2d(1, 1) = nhomo + 1
559 142 : bounds_2d(2, 1) = nhomo + nlumo
560 :
561 : !same-spin contribution. Contraction only neede if c_ss != 0
562 : !directly compute the difference (ja|Ik) - (ka|Ij)
563 142 : IF (ABS(c_ss) > EPSILON(1.0_dp)) THEN
564 :
565 142 : CALL dbt_create(ja_Ik, ja_Ik_diff, map1_2d=[1], map2_2d=[2, 3])
566 142 : CALL dbt_copy(ja_Ik, ja_Ik_diff, move_data=.TRUE.)
567 :
568 : CALL dbt_contract(alpha=-1.0_dp, tensor_1=oI_Y(ido_mo), tensor_2=aj_X, &
569 : beta=1.0_dp, tensor_3=ja_Ik_diff, contract_1=[2], &
570 : notcontract_1=[1], contract_2=[3], notcontract_2=[1, 2], &
571 426 : map_1=[1], map_2=[2, 3], bounds_2=[1, nhomo], bounds_3=bounds_2d)
572 :
573 142 : CALL calc_ss_oov_contrib(parts(1), parts(2), ja_Ik_diff, homo_evals, lumo_evals, omega_k, c_ss)
574 :
575 142 : CALL dbt_destroy(ja_Ik_diff)
576 : END IF !c_ss != 0
577 :
578 276 : CALL dbt_destroy(ja_Ik)
579 : END DO
580 :
581 268 : DO ibatch = 1, nbatch_virt
582 :
583 134 : virt_bo = get_limit(nblk_virt, nbatch_virt, ibatch - 1)
584 : bounds_1d = [SUM(mo_blk_size(1:nblk_occ + virt_bo(1) - 1)) + 1, &
585 1252 : SUM(mo_blk_size(1:nblk_occ + virt_bo(2)))]
586 :
587 134 : CALL dbt_create(mo_template, aj_Ib)
588 : CALL dbt_contract(alpha=1.0_dp, tensor_1=aj_X, tensor_2=oI_Y(ido_mo), &
589 : beta=0.0_dp, tensor_3=aj_Ib, contract_1=[3], &
590 : notcontract_1=[1, 2], contract_2=[2], notcontract_2=[1], &
591 134 : map_1=[1, 2], map_2=[3], bounds_3=bounds_1d)
592 :
593 : !opposite-spin contribution
594 : CALL calc_os_ovv_contrib(parts(3), parts(4), aj_Ib, lumo_evals, homo_evals, lumo_evals, &
595 134 : omega_k, c_os, nhomo, nhomo)
596 :
597 : !same-spin contribution, only if c_ss is not 0
598 : !directly compute the difference (aj|Ib) - (bj|Ia)
599 134 : IF (ABS(c_ss) > EPSILON(1.0_dp)) THEN
600 134 : bounds_2d(1, 1) = 1
601 134 : bounds_2d(2, 1) = nhomo
602 402 : bounds_2d(:, 2) = bounds_1d
603 :
604 134 : CALL dbt_create(aj_Ib, aj_Ib_diff, map1_2d=[1], map2_2d=[2, 3])
605 134 : CALL dbt_copy(aj_Ib, aj_Ib_diff, move_data=.TRUE.)
606 :
607 : CALL dbt_contract(alpha=-1.0_dp, tensor_1=oI_Y(ido_mo), tensor_2=ja_X, &
608 : beta=1.0_dp, tensor_3=aj_Ib_diff, contract_1=[2], &
609 : notcontract_1=[1], contract_2=[3], notcontract_2=[1, 2], &
610 : map_1=[1], map_2=[2, 3], &
611 402 : bounds_2=[nhomo + 1, nhomo + nlumo], bounds_3=bounds_2d)
612 :
613 134 : CALL calc_ss_ovv_contrib(parts(3), parts(4), aj_Ib_diff, homo_evals, lumo_evals, omega_k, c_ss)
614 :
615 134 : CALL dbt_destroy(aj_Ib_diff)
616 : END IF ! c_ss not 0
617 :
618 268 : CALL dbt_destroy(aj_Ib)
619 : END DO
620 :
621 134 : CALL para_env%sum(parts)
622 134 : s1 = parts(1); ds1 = parts(2)
623 134 : s2 = parts(3); ds2 = parts(4)
624 :
625 : !evaluate g and its derivative
626 134 : g = eps_I - omega_k + s1 + s2
627 134 : dg = -1.0_dp + ds1 + ds2
628 :
629 : !compute the diff to the new step
630 134 : diff = -g/dg
631 :
632 : !and the new omega
633 134 : omega_k = omega_k + diff
634 134 : diff = diff*evolt
635 :
636 134 : IF (output_unit > 0) THEN
637 : WRITE (UNIT=output_unit, FMT="(T21,I18,F32.6)") &
638 67 : iloop, diff
639 67 : CALL m_flush(output_unit)
640 : END IF
641 :
642 168 : IF (iloop > max_iter) THEN
643 0 : CPWARN("GW2X iteration not converged.")
644 0 : EXIT
645 : END IF
646 : END DO !while loop on eps_iter
647 :
648 : !compute the shift and update donor_state
649 34 : donor_state%gw2x_evals(ido_mo, 1) = omega_k
650 :
651 60 : IF (output_unit > 0) THEN
652 : WRITE (UNIT=output_unit, FMT="(/T7,A,F11.6,/,T5,A,F11.6)") &
653 17 : "Final GW2X shift for this donor MO (eV):", &
654 34 : (donor_state%energy_evals(ido_mo, 1) - omega_k)*evolt
655 : END IF
656 :
657 : END DO !ido_mo
658 :
659 26 : CALL dbt_destroy(aj_X)
660 :
661 26 : CALL timestop(handle)
662 :
663 52 : END SUBROUTINE GW2X_rcs_iterations
664 :
665 : ! **************************************************************************************************
666 : !> \brief Preforms the GW2X iterations in the open-shell shell formalism according to the
667 : !> Newton-Raphson method
668 : !> \param first_domo index of the first core donor MO to consider, for each spin
669 : !> \param ja_X semi-contracted tensors with j: occupied MO, a: virtual MO, X: RI basis element
670 : !> \param oI_Y semi-contracted tensors with o: all MOs, I donor core MO, Y: RI basis element
671 : !> \param mo_template tensor template for fully MO contracted tensor, for each spin combination
672 : !> \param homo_evals ...
673 : !> \param lumo_evals ...
674 : !> \param donor_state ...
675 : !> \param xas_tdp_control ...
676 : !> \param qs_env ...
677 : ! **************************************************************************************************
678 4 : SUBROUTINE GW2X_os_iterations(first_domo, ja_X, oI_Y, mo_template, homo_evals, lumo_evals, &
679 : donor_state, xas_tdp_control, qs_env)
680 :
681 : INTEGER, INTENT(IN) :: first_domo(2)
682 : TYPE(dbt_type), DIMENSION(:), INTENT(inout) :: ja_X, oI_Y
683 : TYPE(dbt_type), DIMENSION(:, :), INTENT(inout) :: mo_template
684 : TYPE(cp_1d_r_p_type), DIMENSION(:), INTENT(in) :: homo_evals, lumo_evals
685 : TYPE(donor_state_type), POINTER :: donor_state
686 : TYPE(xas_tdp_control_type), POINTER :: xas_tdp_control
687 : TYPE(qs_environment_type), POINTER :: qs_env
688 :
689 : CHARACTER(len=*), PARAMETER :: routineN = 'GW2X_os_iterations'
690 :
691 : INTEGER :: batch_size, bounds_1d(2), bounds_2d(2, 2), handle, i, ibatch, ido_mo, iloop, &
692 : ispin, max_iter, nbatch_occ, nbatch_virt, nblk_occ, nblk_virt, nblks(3), ndo_mo, &
693 : nhomo(2), nlumo(2), nspins, occ_bo(2), other_spin, output_unit, tmp_sum, virt_bo(2)
694 4 : INTEGER, ALLOCATABLE, DIMENSION(:) :: mo_blk_size
695 : REAL(dp) :: c_os, c_ss, dg, diff, ds1, ds2, eps_I, &
696 : eps_iter, g, omega_k, parts(4), s1, s2
697 136 : TYPE(dbt_type) :: aj_Ib, aj_Ib_diff, ja_Ik, ja_Ik_diff
698 4 : TYPE(dbt_type), ALLOCATABLE, DIMENSION(:) :: aj_X
699 : TYPE(mp_para_env_type), POINTER :: para_env
700 :
701 4 : CALL timeset(routineN, handle)
702 :
703 4 : eps_iter = xas_tdp_control%gw2x_eps
704 4 : max_iter = xas_tdp_control%max_gw2x_iter
705 4 : c_os = xas_tdp_control%c_os
706 4 : c_ss = xas_tdp_control%c_ss
707 4 : batch_size = xas_tdp_control%batch_size
708 :
709 4 : nspins = 2
710 4 : ndo_mo = donor_state%ndo_mo
711 4 : output_unit = cp_logger_get_default_io_unit()
712 :
713 12 : DO ispin = 1, nspins
714 8 : nhomo(ispin) = SIZE(homo_evals(ispin)%array)
715 12 : nlumo(ispin) = SIZE(lumo_evals(ispin)%array)
716 : END DO
717 :
718 4 : CALL get_qs_env(qs_env, para_env=para_env)
719 :
720 : !We use the Newton-Raphson method to find the zero of the function:
721 : !g(omega) = eps_I - omega + mp2 terms, dg(omega) = -1 + d/d_omega (mp2 terms)
722 : !We simply compute at each iteration: omega_k+1 = omega_k - g(omega_k)/dg(omega_k)
723 :
724 48 : ALLOCATE (aj_X(2))
725 12 : DO ispin = 1, nspins
726 :
727 : !need transposed tensor of (ja|X) for optimal contraction scheme,
728 : !s.t. (aj|X) block is on same processor as (ja|X)) and differences can be taken
729 8 : CALL dbt_create(ja_X(ispin), aj_X(ispin))
730 12 : CALL dbt_copy(ja_X(ispin), aj_X(ispin), order=[2, 1, 3])
731 :
732 : END DO ! ispin
733 12 : DO ispin = 1, nspins
734 :
735 8 : other_spin = 3 - ispin
736 :
737 : !split the MO blocks into batches for memory friendly batched contraction
738 : !huge dense tensors never need to be stored. Split MOs for the current spin
739 8 : CALL dbt_get_info(ja_X(ispin), nblks_total=nblks)
740 24 : ALLOCATE (mo_blk_size(nblks(1)))
741 8 : CALL dbt_get_info(ja_X(ispin), blk_size_1=mo_blk_size)
742 :
743 8 : tmp_sum = 0
744 16 : DO i = 1, nblks(1)
745 16 : tmp_sum = tmp_sum + mo_blk_size(i)
746 16 : IF (tmp_sum == nhomo(ispin)) THEN
747 8 : nblk_occ = i
748 8 : nblk_virt = nblks(1) - i
749 8 : EXIT
750 : END IF
751 : END DO
752 8 : nbatch_occ = MAX(1, nblk_occ/batch_size)
753 8 : nbatch_virt = MAX(1, nblk_virt/batch_size)
754 :
755 : !Loop over donor_states of the current spin
756 16 : DO ido_mo = 1, ndo_mo
757 8 : IF (output_unit > 0) THEN
758 : WRITE (UNIT=output_unit, FMT="(/,T5,A,I2,A,I4,A,/,T5,A)") &
759 4 : "- GW2X correction for donor MO with spin ", ispin, &
760 4 : " and MO index ", donor_state%mo_indices(ido_mo, ispin), ":", &
761 8 : " iteration convergence (eV)"
762 4 : CALL m_flush(output_unit)
763 : END IF
764 :
765 : !starting values
766 8 : eps_I = homo_evals(ispin)%array(first_domo(ispin) + ido_mo - 1)
767 8 : omega_k = eps_I
768 8 : iloop = 0
769 8 : diff = 2.0_dp*eps_iter
770 :
771 40 : DO WHILE (ABS(diff) > eps_iter)
772 32 : iloop = iloop + 1
773 :
774 : !Compute the mp2 terms and their first derivative
775 32 : parts = 0.0_dp
776 :
777 : !We do batched contraction for (ja|Ik) and (ja|Ib) to never have to carry the full tensor
778 64 : DO ibatch = 1, nbatch_occ
779 :
780 : !opposite-spin contribution, i.e. (j_beta a_beta| I_alpha k_alpha) and vice-versa
781 : !do the batching along k because same spin as donor MO
782 32 : occ_bo = get_limit(nblk_occ, nbatch_occ, ibatch - 1)
783 160 : bounds_1d = [SUM(mo_blk_size(1:occ_bo(1) - 1)) + 1, SUM(mo_blk_size(1:occ_bo(2)))]
784 :
785 32 : CALL dbt_create(mo_template(other_spin, ispin), ja_Ik)
786 : CALL dbt_contract(alpha=1.0_dp, tensor_1=ja_X(other_spin), &
787 : tensor_2=oI_Y((ispin - 1)*ndo_mo + ido_mo), &
788 : beta=0.0_dp, tensor_3=ja_Ik, contract_1=[3], &
789 : notcontract_1=[1, 2], contract_2=[2], notcontract_2=[1], &
790 32 : map_1=[1, 2], map_2=[3], bounds_3=bounds_1d)
791 :
792 : CALL calc_os_oov_contrib(parts(1), parts(2), ja_Ik, homo_evals(other_spin)%array, &
793 : lumo_evals(other_spin)%array, homo_evals(ispin)%array, &
794 32 : omega_k, c_os, nhomo(other_spin))
795 :
796 32 : CALL dbt_destroy(ja_Ik)
797 :
798 : !same-spin contribution, need to compute (ja|Ik) - (ka|Ij), all with the current spin
799 : !skip if c_ss == 0
800 64 : IF (ABS(c_ss) > EPSILON(1.0_dp)) THEN
801 :
802 : !same batching as opposite spin
803 32 : CALL dbt_create(mo_template(ispin, ispin), ja_Ik)
804 : CALL dbt_contract(alpha=1.0_dp, tensor_1=ja_X(ispin), &
805 : tensor_2=oI_Y((ispin - 1)*ndo_mo + ido_mo), &
806 : beta=0.0_dp, tensor_3=ja_Ik, contract_1=[3], &
807 : notcontract_1=[1, 2], contract_2=[2], notcontract_2=[1], &
808 32 : map_1=[1, 2], map_2=[3], bounds_3=bounds_1d)
809 :
810 96 : bounds_2d(:, 2) = bounds_1d
811 32 : bounds_2d(1, 1) = nhomo(ispin) + 1
812 32 : bounds_2d(2, 1) = nhomo(ispin) + nlumo(ispin)
813 :
814 : !the tensor difference is directly taken here
815 32 : CALL dbt_create(ja_Ik, ja_Ik_diff, map1_2d=[1], map2_2d=[2, 3])
816 32 : CALL dbt_copy(ja_Ik, ja_Ik_diff, move_data=.TRUE.)
817 :
818 : CALL dbt_contract(alpha=-1.0_dp, tensor_1=oI_Y((ispin - 1)*ndo_mo + ido_mo), &
819 : tensor_2=aj_X(ispin), beta=1.0_dp, tensor_3=ja_Ik_diff, &
820 : contract_1=[2], notcontract_1=[1], contract_2=[3], notcontract_2=[1, 2], &
821 96 : map_1=[1], map_2=[2, 3], bounds_2=[1, nhomo(ispin)], bounds_3=bounds_2d)
822 :
823 : CALL calc_ss_oov_contrib(parts(1), parts(2), ja_Ik_diff, homo_evals(ispin)%array, &
824 32 : lumo_evals(ispin)%array, omega_k, c_ss)
825 :
826 32 : CALL dbt_destroy(ja_Ik_diff)
827 32 : CALL dbt_destroy(ja_Ik)
828 : END IF !c_ss !!= 0
829 :
830 : END DO
831 :
832 64 : DO ibatch = 1, nbatch_virt
833 :
834 : !opposite-spin contribution, i.e. (a_beta j_beta| I_alpha b_alpha) and vice-versa
835 : !do the batching along b because same spin as donor MO
836 32 : virt_bo = get_limit(nblk_virt, nbatch_virt, ibatch - 1)
837 : bounds_1d = [SUM(mo_blk_size(1:nblk_occ + virt_bo(1) - 1)) + 1, &
838 256 : SUM(mo_blk_size(1:nblk_occ + virt_bo(2)))]
839 :
840 32 : CALL dbt_create(mo_template(other_spin, ispin), aj_Ib)
841 : CALL dbt_contract(alpha=1.0_dp, tensor_1=aj_X(other_spin), &
842 : tensor_2=oI_Y((ispin - 1)*ndo_mo + ido_mo), &
843 : beta=0.0_dp, tensor_3=aj_Ib, contract_1=[3], &
844 : notcontract_1=[1, 2], contract_2=[2], notcontract_2=[1], &
845 32 : map_1=[1, 2], map_2=[3], bounds_3=bounds_1d)
846 :
847 : CALL calc_os_ovv_contrib(parts(3), parts(4), aj_Ib, lumo_evals(other_spin)%array, &
848 : homo_evals(other_spin)%array, lumo_evals(ispin)%array, &
849 32 : omega_k, c_os, nhomo(other_spin), nhomo(ispin))
850 :
851 32 : CALL dbt_destroy(aj_Ib)
852 :
853 : !same-spin contribution, need to compute (aj|Ib) - (bj|Ia), all with the current spin
854 : !skip if c_ss == 0
855 64 : IF (ABS(c_ss) > EPSILON(1.0_dp)) THEN
856 :
857 : !same batching as opposite spin
858 32 : CALL dbt_create(mo_template(ispin, ispin), aj_Ib)
859 : CALL dbt_contract(alpha=1.0_dp, tensor_1=aj_X(ispin), &
860 : tensor_2=oI_Y((ispin - 1)*ndo_mo + ido_mo), &
861 : beta=0.0_dp, tensor_3=aj_Ib, contract_1=[3], &
862 : notcontract_1=[1, 2], contract_2=[2], notcontract_2=[1], &
863 32 : map_1=[1, 2], map_2=[3], bounds_3=bounds_1d)
864 :
865 32 : bounds_2d(1, 1) = 1
866 32 : bounds_2d(2, 1) = nhomo(ispin)
867 96 : bounds_2d(:, 2) = bounds_1d
868 :
869 32 : CALL dbt_create(aj_Ib, aj_Ib_diff, map1_2d=[1], map2_2d=[2, 3])
870 32 : CALL dbt_copy(aj_Ib, aj_Ib_diff, move_data=.TRUE.)
871 :
872 : CALL dbt_contract(alpha=-1.0_dp, tensor_1=oI_Y((ispin - 1)*ndo_mo + ido_mo), &
873 : tensor_2=ja_X(ispin), beta=1.0_dp, tensor_3=aj_Ib_diff, &
874 : contract_1=[2], notcontract_1=[1], contract_2=[3], &
875 : notcontract_2=[1, 2], map_1=[1], map_2=[2, 3], &
876 : bounds_2=[nhomo(ispin) + 1, nhomo(ispin) + nlumo(ispin)], &
877 96 : bounds_3=bounds_2d)
878 :
879 : CALL calc_ss_ovv_contrib(parts(3), parts(4), aj_Ib_diff, homo_evals(ispin)%array, &
880 32 : lumo_evals(ispin)%array, omega_k, c_ss)
881 :
882 32 : CALL dbt_destroy(aj_Ib_diff)
883 32 : CALL dbt_destroy(aj_Ib)
884 : END IF ! c_ss not 0
885 :
886 : END DO
887 :
888 32 : CALL para_env%sum(parts)
889 32 : s1 = parts(1); ds1 = parts(2)
890 32 : s2 = parts(3); ds2 = parts(4)
891 :
892 : !evaluate g and its derivative
893 32 : g = eps_I - omega_k + s1 + s2
894 32 : dg = -1.0_dp + ds1 + ds2
895 :
896 : !compute the diff to the new step
897 32 : diff = -g/dg
898 :
899 : !and the new omega
900 32 : omega_k = omega_k + diff
901 32 : diff = diff*evolt
902 :
903 32 : IF (output_unit > 0) THEN
904 : WRITE (UNIT=output_unit, FMT="(T21,I18,F32.6)") &
905 16 : iloop, diff
906 16 : CALL m_flush(output_unit)
907 : END IF
908 :
909 40 : IF (iloop > max_iter) THEN
910 0 : CPWARN("GW2X iteration not converged.")
911 0 : EXIT
912 : END IF
913 : END DO !while loop on eps_iter
914 :
915 : !compute the shift and update donor_state
916 8 : donor_state%gw2x_evals(ido_mo, ispin) = omega_k
917 :
918 16 : IF (output_unit > 0) THEN
919 : WRITE (UNIT=output_unit, FMT="(/T7,A,F11.6,/,T5,A,F11.6)") &
920 4 : "Final GW2X shift for this donor MO (eV):", &
921 8 : (donor_state%energy_evals(ido_mo, ispin) - omega_k)*evolt
922 : END IF
923 :
924 : END DO !ido_mo
925 :
926 12 : DEALLOCATE (mo_blk_size)
927 : END DO ! ispin
928 :
929 12 : DO ispin = 1, nspins
930 12 : CALL dbt_destroy(aj_X(ispin))
931 : END DO
932 :
933 4 : CALL timestop(handle)
934 :
935 20 : END SUBROUTINE GW2X_os_iterations
936 :
937 : ! **************************************************************************************************
938 : !> \brief Takes the 3-center integrals from the ri_ex_3c tensor and returns a full tensor. Since
939 : !> ri_ex_3c is only half filled because of symmetry, we have to add the transpose
940 : !> and scale the diagonal blocks by 0.5
941 : !> \param pq_X the full (desymmetrized) tensor containing the (pq|X) exchange integrals, in a new
942 : !> 3d distribution and optimized block sizes
943 : !> \param exat index of current excited atom
944 : !> \param xas_tdp_env ...
945 : !> \param qs_env ...
946 : ! **************************************************************************************************
947 34 : SUBROUTINE get_full_pqX_from_3c_ex(pq_X, exat, xas_tdp_env, qs_env)
948 :
949 : TYPE(dbt_type), INTENT(INOUT) :: pq_X
950 : INTEGER, INTENT(IN) :: exat
951 : TYPE(xas_tdp_env_type), POINTER :: xas_tdp_env
952 : TYPE(qs_environment_type), POINTER :: qs_env
953 :
954 : INTEGER :: i, ind(3), natom, nblk_ri, nsgf_x
955 34 : INTEGER, ALLOCATABLE, DIMENSION(:) :: orb_blk_size, proc_dist_1, proc_dist_2, &
956 34 : proc_dist_3
957 : INTEGER, DIMENSION(3) :: pdims
958 : LOGICAL :: found
959 34 : REAL(dp), ALLOCATABLE, DIMENSION(:, :, :) :: pblock
960 306 : TYPE(dbt_distribution_type) :: t_dist
961 : TYPE(dbt_iterator_type) :: iter
962 102 : TYPE(dbt_pgrid_type) :: t_pgrid
963 612 : TYPE(dbt_type) :: pq_X_tmp, work
964 : TYPE(mp_para_env_type), POINTER :: para_env
965 :
966 34 : NULLIFY (para_env)
967 :
968 : !create work tensor with same 2D dist as pq_X, but only keep excited atom along RI direction
969 34 : CALL get_qs_env(qs_env, para_env=para_env, natom=natom)
970 34 : CALL dbt_get_info(xas_tdp_env%ri_3c_ex, pdims=pdims)
971 34 : nsgf_x = SIZE(xas_tdp_env%ri_inv_ex, 1)
972 34 : nblk_ri = 1
973 :
974 34 : CALL dbt_pgrid_create(para_env, pdims, t_pgrid)
975 170 : ALLOCATE (proc_dist_1(natom), proc_dist_2(natom), orb_blk_size(natom))
976 : CALL dbt_get_info(xas_tdp_env%ri_3c_ex, proc_dist_1=proc_dist_1, proc_dist_2=proc_dist_2, &
977 34 : blk_size_1=orb_blk_size)
978 : CALL dbt_distribution_new(t_dist, t_pgrid, nd_dist_1=proc_dist_1, nd_dist_2=proc_dist_2, &
979 136 : nd_dist_3=[(0, i=1, nblk_ri)])
980 :
981 : CALL dbt_create(work, name="(pq|X)", dist=t_dist, map1_2d=[1], map2_2d=[2, 3], &
982 68 : blk_size_1=orb_blk_size, blk_size_2=orb_blk_size, blk_size_3=[nsgf_x])
983 34 : CALL dbt_distribution_destroy(t_dist)
984 :
985 : !dist of 3c_ex and work match, can simply copy blocks over. Diagonal with factor 0.5
986 :
987 : !$OMP PARALLEL DEFAULT(NONE) SHARED(xas_tdp_env,exat,work,orb_blk_size,nsgf_x) &
988 34 : !$OMP PRIVATE(iter,ind,pblock,found)
989 : CALL dbt_iterator_start(iter, xas_tdp_env%ri_3c_ex)
990 : DO WHILE (dbt_iterator_blocks_left(iter))
991 : CALL dbt_iterator_next_block(iter, ind)
992 : CALL dbt_get_block(xas_tdp_env%ri_3c_ex, ind, pblock, found)
993 :
994 : IF (ind(1) == ind(2)) pblock = 0.5_dp*pblock
995 : IF (ind(3) /= exat) CYCLE
996 :
997 : CALL dbt_put_block(work, [ind(1), ind(2), 1], &
998 : [orb_blk_size(ind(1)), orb_blk_size(ind(2)), nsgf_x], pblock)
999 :
1000 : DEALLOCATE (pblock)
1001 : END DO
1002 : CALL dbt_iterator_stop(iter)
1003 : !$OMP END PARALLEL
1004 34 : CALL dbt_finalize(work)
1005 :
1006 : !create (pq|X) based on work and copy over
1007 34 : CALL dbt_create(work, pq_X_tmp)
1008 34 : CALL dbt_copy(work, pq_X_tmp)
1009 34 : CALL dbt_copy(work, pq_X_tmp, order=[2, 1, 3], summation=.TRUE., move_data=.TRUE.)
1010 :
1011 34 : CALL dbt_destroy(work)
1012 :
1013 : !create the pgrid, based on the 2D dbcsr grid
1014 34 : CALL dbt_pgrid_destroy(t_pgrid)
1015 34 : pdims = 0
1016 136 : CALL dbt_pgrid_create(para_env, pdims, t_pgrid, tensor_dims=[natom, natom, 1])
1017 :
1018 : !cyclic distribution accross all directions.
1019 34 : ALLOCATE (proc_dist_3(nblk_ri))
1020 34 : CALL dbt_default_distvec(natom, pdims(1), orb_blk_size, proc_dist_1)
1021 34 : CALL dbt_default_distvec(natom, pdims(2), orb_blk_size, proc_dist_2)
1022 68 : CALL dbt_default_distvec(nblk_ri, pdims(3), [nsgf_x], proc_dist_3)
1023 : CALL dbt_distribution_new(t_dist, t_pgrid, nd_dist_1=proc_dist_1, nd_dist_2=proc_dist_2, &
1024 34 : nd_dist_3=proc_dist_3)
1025 :
1026 : CALL dbt_create(pq_X, name="(pq|X)", dist=t_dist, map1_2d=[2, 3], map2_2d=[1], &
1027 68 : blk_size_1=orb_blk_size, blk_size_2=orb_blk_size, blk_size_3=[nsgf_x])
1028 34 : CALL dbt_copy(pq_X_tmp, pq_X, move_data=.TRUE.)
1029 :
1030 34 : CALL dbt_distribution_destroy(t_dist)
1031 34 : CALL dbt_pgrid_destroy(t_pgrid)
1032 34 : CALL dbt_destroy(pq_X_tmp)
1033 :
1034 68 : END SUBROUTINE get_full_pqX_from_3c_ex
1035 :
1036 : ! **************************************************************************************************
1037 : !> \brief Contracts (pq|X) and (rI|Y) from AOs to MOs to (ja|X) and (oI|Y) respectively, where
1038 : !> j is a occupied MO, a is a virtual MO and o is a general MO
1039 : !> \param ja_X partial contraction over occupied MOs j, virtual MOs a: (ja|X), for both spins (alpha-alpha or beta-beta)
1040 : !> \param oI_Y partial contraction over all MOs o and donor MOs I (can be more than 1 if 2p or open-shell)
1041 : !> \param ja_Io_template template to be able to build tensors after calling this routine, for each spin combination
1042 : !> \param mo_coeffs ...
1043 : !> \param nocc ...
1044 : !> \param nvirt ...
1045 : !> \param donor_state ...
1046 : !> \param xas_tdp_env ...
1047 : !> \param xas_tdp_control ...
1048 : !> \param qs_env ...
1049 : !> \note the multiplication by (X|Y)^-1 is included in the final (oI|Y) tensor. Only integrals with the
1050 : !> same spin on one center are non-zero, i.e. (oI|Y) is non zero only if both o and Y have the same spin
1051 : ! **************************************************************************************************
1052 30 : SUBROUTINE contract_AOs_to_MOs(ja_X, oI_Y, ja_Io_template, mo_coeffs, nocc, nvirt, &
1053 : donor_state, xas_tdp_env, xas_tdp_control, qs_env)
1054 :
1055 : TYPE(dbt_type), ALLOCATABLE, DIMENSION(:), &
1056 : INTENT(INOUT) :: ja_X, oI_Y
1057 : TYPE(dbt_type), ALLOCATABLE, DIMENSION(:, :), &
1058 : INTENT(INOUT) :: ja_Io_template
1059 : TYPE(cp_fm_type), DIMENSION(:), INTENT(INOUT) :: mo_coeffs
1060 : INTEGER, INTENT(IN) :: nocc(2), nvirt(2)
1061 : TYPE(donor_state_type), POINTER :: donor_state
1062 : TYPE(xas_tdp_env_type), POINTER :: xas_tdp_env
1063 : TYPE(xas_tdp_control_type), POINTER :: xas_tdp_control
1064 : TYPE(qs_environment_type), POINTER :: qs_env
1065 :
1066 : CHARACTER(len=*), PARAMETER :: routineN = 'contract_AOs_to_MOs'
1067 :
1068 : INTEGER :: bo(2), handle, i, ispin, jspin, &
1069 : nblk_aos, nblk_mos(2), nblk_occ(2), &
1070 : nblk_pqX(3), nblk_ri, nblk_virt(2), &
1071 : nspins
1072 : INTEGER, DIMENSION(3) :: pdims
1073 30 : INTEGER, DIMENSION(:), POINTER :: ao_blk_size, ao_col_dist, ao_row_dist, &
1074 60 : mo_dist_3, ri_blk_size, ri_dist_3
1075 30 : INTEGER, DIMENSION(:, :), POINTER :: mat_pgrid
1076 30 : TYPE(cp_1d_i_p_type), ALLOCATABLE, DIMENSION(:) :: mo_blk_size, mo_col_dist, mo_row_dist
1077 : TYPE(dbcsr_distribution_type) :: mat_dist
1078 : TYPE(dbcsr_distribution_type), POINTER :: std_mat_dist
1079 : TYPE(dbcsr_type) :: dbcsr_mo_coeffs
1080 210 : TYPE(dbt_distribution_type) :: t_dist
1081 90 : TYPE(dbt_pgrid_type) :: t_pgrid
1082 630 : TYPE(dbt_type) :: jq_X, pq_X, t_mo_coeffs
1083 : TYPE(mp_para_env_type), POINTER :: para_env
1084 :
1085 30 : NULLIFY (ao_blk_size, ao_col_dist, ao_row_dist, mo_dist_3, ri_blk_size, ri_dist_3, mat_pgrid, &
1086 30 : para_env, std_mat_dist)
1087 :
1088 30 : CALL timeset(routineN, handle)
1089 :
1090 30 : nspins = 1; IF (xas_tdp_control%do_uks .OR. xas_tdp_control%do_roks) nspins = 2
1091 :
1092 : !There are 2 contractions to do for the first tensor: (pq|X) --> (jq|X) --> (ja|X)
1093 : !Because memory is the main concern, we move_data everytime at the cost of extra copies
1094 :
1095 : !Some quantities need to be stored for both spins, because they are later combined
1096 30 : CALL get_qs_env(qs_env, para_env=para_env)
1097 222 : ALLOCATE (mo_blk_size(nspins), mo_row_dist(nspins), mo_col_dist(nspins))
1098 274 : ALLOCATE (ja_X(nspins))
1099 312 : ALLOCATE (oI_Y(nspins*donor_state%ndo_mo))
1100 :
1101 64 : DO ispin = 1, nspins
1102 :
1103 : !First, we need a fully populated pq_X (spin-independent)
1104 34 : CALL get_full_pqX_from_3c_ex(pq_X, donor_state%at_index, xas_tdp_env, qs_env)
1105 :
1106 : !Create the tensor pgrid. AOs and RI independent from spin
1107 34 : IF (ispin == 1) THEN
1108 30 : CALL dbt_get_info(pq_X, pdims=pdims, nblks_total=nblk_pqX)
1109 30 : CALL dbt_pgrid_create(para_env, pdims, t_pgrid)
1110 30 : nblk_aos = nblk_pqX(1)
1111 30 : nblk_ri = nblk_pqX(3)
1112 : END IF
1113 :
1114 : !Define MO block sizes, at worst, take one block per proc
1115 34 : nblk_occ(ispin) = MAX(pdims(1), nocc(ispin)/16)
1116 34 : nblk_virt(ispin) = MAX(pdims(2), nvirt(ispin)/16)
1117 34 : nblk_mos(ispin) = nblk_occ(ispin) + nblk_virt(ispin)
1118 102 : ALLOCATE (mo_blk_size(ispin)%array(nblk_mos(ispin)))
1119 102 : DO i = 1, nblk_occ(ispin)
1120 68 : bo = get_limit(nocc(ispin), nblk_occ(ispin), i - 1)
1121 102 : mo_blk_size(ispin)%array(i) = bo(2) - bo(1) + 1
1122 : END DO
1123 96 : DO i = 1, nblk_virt(ispin)
1124 62 : bo = get_limit(nvirt(ispin), nblk_virt(ispin), i - 1)
1125 96 : mo_blk_size(ispin)%array(nblk_occ(ispin) + i) = bo(2) - bo(1) + 1
1126 : END DO
1127 :
1128 : !Convert the fm mo_coeffs into a dbcsr matrix and then a tensor
1129 34 : CALL get_qs_env(qs_env, dbcsr_dist=std_mat_dist)
1130 34 : CALL dbcsr_distribution_get(std_mat_dist, pgrid=mat_pgrid)
1131 170 : ALLOCATE (ao_blk_size(nblk_aos), ri_blk_size(nblk_ri))
1132 34 : CALL dbt_get_info(pq_X, blk_size_1=ao_blk_size, blk_size_3=ri_blk_size)
1133 :
1134 : !we opt for a cyclic dist for the MOs (since they should be rather dense anyways)
1135 102 : ALLOCATE (ao_row_dist(nblk_aos), mo_col_dist(ispin)%array(nblk_mos(ispin)))
1136 34 : CALL dbt_default_distvec(nblk_aos, SIZE(mat_pgrid, 1), ao_blk_size, ao_row_dist)
1137 : CALL dbt_default_distvec(nblk_mos(ispin), SIZE(mat_pgrid, 2), mo_blk_size(ispin)%array, &
1138 34 : mo_col_dist(ispin)%array)
1139 : CALL dbcsr_distribution_new(mat_dist, group=para_env%get_handle(), pgrid=mat_pgrid, &
1140 34 : row_dist=ao_row_dist, col_dist=mo_col_dist(ispin)%array)
1141 :
1142 : CALL dbcsr_create(dbcsr_mo_coeffs, name="MO coeffs", matrix_type="N", dist=mat_dist, &
1143 34 : row_blk_size=ao_blk_size, col_blk_size=mo_blk_size(ispin)%array)
1144 34 : CALL copy_fm_to_dbcsr(mo_coeffs(ispin), dbcsr_mo_coeffs)
1145 :
1146 34 : CALL dbt_create(dbcsr_mo_coeffs, t_mo_coeffs)
1147 34 : CALL dbt_copy_matrix_to_tensor(dbcsr_mo_coeffs, t_mo_coeffs)
1148 :
1149 : !prepare the (jq|X) tensor for the first contraction (over occupied MOs)
1150 136 : ALLOCATE (mo_row_dist(ispin)%array(nblk_mos(ispin)), ao_col_dist(nblk_aos), ri_dist_3(nblk_ri))
1151 34 : CALL dbt_default_distvec(nblk_mos(ispin), pdims(1), mo_blk_size(ispin)%array, mo_row_dist(ispin)%array)
1152 34 : CALL dbt_default_distvec(nblk_aos, pdims(2), ao_blk_size, ao_col_dist)
1153 34 : CALL dbt_default_distvec(nblk_ri, pdims(3), ri_blk_size, ri_dist_3)
1154 : CALL dbt_distribution_new(t_dist, t_pgrid, nd_dist_1=mo_row_dist(ispin)%array, &
1155 34 : nd_dist_2=ao_col_dist, nd_dist_3=ri_dist_3)
1156 :
1157 : CALL dbt_create(jq_X, name="(jq|X)", dist=t_dist, map1_2d=[1, 3], map2_2d=[2], &
1158 34 : blk_size_1=mo_blk_size(ispin)%array, blk_size_2=ao_blk_size, blk_size_3=ri_blk_size)
1159 34 : CALL dbt_distribution_destroy(t_dist)
1160 :
1161 : !contract (pq|X) into (jq|X)
1162 : CALL dbt_contract(alpha=1.0_dp, tensor_1=pq_X, tensor_2=t_mo_coeffs, &
1163 : beta=0.0_dp, tensor_3=jq_X, contract_1=[1], &
1164 : notcontract_1=[2, 3], contract_2=[1], notcontract_2=[2], &
1165 : map_1=[2, 3], map_2=[1], bounds_3=[1, nocc(ispin)], &!only want occupied MOs for j
1166 102 : move_data=.TRUE.)
1167 :
1168 34 : CALL dbt_destroy(pq_X)
1169 34 : CALL dbt_copy_matrix_to_tensor(dbcsr_mo_coeffs, t_mo_coeffs)
1170 :
1171 : !prepare (ja|X) tensor for the second contraction (over virtual MOs)
1172 : !only virtual-occupied bit of the first 2 indices is occupied + it should be dense
1173 : !take blk dist such that blocks are evenly distributed
1174 : CALL dbt_default_distvec(nblk_occ(ispin), pdims(1), mo_blk_size(ispin)%array(1:nblk_occ(ispin)), &
1175 34 : mo_row_dist(ispin)%array(1:nblk_occ(ispin)))
1176 : CALL dbt_default_distvec(nblk_virt(ispin), pdims(1), &
1177 : mo_blk_size(ispin)%array(nblk_occ(ispin) + 1:nblk_mos(ispin)), &
1178 34 : mo_row_dist(ispin)%array(nblk_occ(ispin) + 1:nblk_mos(ispin)))
1179 : CALL dbt_default_distvec(nblk_occ(ispin), pdims(2), mo_blk_size(ispin)%array(1:nblk_occ(ispin)), &
1180 34 : mo_col_dist(ispin)%array(1:nblk_occ(ispin)))
1181 : CALL dbt_default_distvec(nblk_virt(ispin), pdims(2), &
1182 : mo_blk_size(ispin)%array(nblk_occ(ispin) + 1:nblk_mos(ispin)), &
1183 34 : mo_col_dist(ispin)%array(nblk_occ(ispin) + 1:nblk_mos(ispin)))
1184 : CALL dbt_distribution_new(t_dist, t_pgrid, nd_dist_1=mo_row_dist(ispin)%array, &
1185 34 : nd_dist_2=mo_col_dist(ispin)%array, nd_dist_3=ri_dist_3)
1186 :
1187 : CALL dbt_create(ja_X(ispin), name="(ja|X)", dist=t_dist, map1_2d=[1, 2], map2_2d=[3], &
1188 : blk_size_1=mo_blk_size(ispin)%array, blk_size_2=mo_blk_size(ispin)%array, &
1189 34 : blk_size_3=ri_blk_size)
1190 34 : CALL dbt_distribution_destroy(t_dist)
1191 :
1192 : !contract (jq|X) into (ja|X)
1193 : CALL dbt_contract(alpha=1.0_dp, tensor_1=jq_X, tensor_2=t_mo_coeffs, &
1194 : beta=0.0_dp, tensor_3=ja_X(ispin), contract_1=[2], &
1195 : notcontract_1=[1, 3], contract_2=[1], notcontract_2=[2], &
1196 : map_1=[1, 3], map_2=[2], move_data=.TRUE., &
1197 102 : bounds_3=[nocc(ispin) + 1, nocc(ispin) + nvirt(ispin)])
1198 :
1199 34 : CALL dbt_destroy(jq_X)
1200 34 : CALL dbt_copy_matrix_to_tensor(dbcsr_mo_coeffs, t_mo_coeffs)
1201 :
1202 : !Finally, get the oI_Y tensors
1203 : CALL get_oIY_tensors(oI_Y, ispin, ao_blk_size, mo_blk_size(ispin)%array, ri_blk_size, &
1204 34 : t_mo_coeffs, donor_state, xas_tdp_env, xas_tdp_control, qs_env)
1205 :
1206 : !intermediate clen-up
1207 34 : CALL dbt_destroy(t_mo_coeffs)
1208 34 : CALL dbcsr_distribution_release(mat_dist)
1209 34 : CALL dbcsr_release(dbcsr_mo_coeffs)
1210 64 : DEALLOCATE (ao_col_dist, ri_dist_3, ri_blk_size, ao_blk_size, ao_row_dist)
1211 :
1212 : END DO !ispin
1213 :
1214 : !create a empty tensor template for the fully contracted (ja|Io) MO integrals, for all spin
1215 : !configureations: alpha-alpha|alpha-alpha, alpha-alpha|beta-beta, etc.
1216 376 : ALLOCATE (ja_Io_template(nspins, nspins))
1217 64 : DO ispin = 1, nspins
1218 106 : DO jspin = 1, nspins
1219 126 : ALLOCATE (mo_dist_3(nblk_mos(jspin)))
1220 : CALL dbt_default_distvec(nblk_occ(jspin), pdims(3), mo_blk_size(jspin)%array(1:nblk_occ(jspin)), &
1221 42 : mo_dist_3(1:nblk_occ(jspin)))
1222 : CALL dbt_default_distvec(nblk_virt(jspin), pdims(3), &
1223 : mo_blk_size(jspin)%array(nblk_occ(jspin) + 1:nblk_mos(jspin)), &
1224 42 : mo_dist_3(nblk_occ(jspin) + 1:nblk_mos(jspin)))
1225 : CALL dbt_distribution_new(t_dist, t_pgrid, nd_dist_1=mo_row_dist(ispin)%array, &
1226 42 : nd_dist_2=mo_col_dist(ispin)%array, nd_dist_3=mo_dist_3)
1227 :
1228 : CALL dbt_create(ja_Io_template(ispin, jspin), name="(ja|Io)", dist=t_dist, map1_2d=[1, 2], &
1229 : map2_2d=[3], blk_size_1=mo_blk_size(ispin)%array, &
1230 42 : blk_size_2=mo_blk_size(ispin)%array, blk_size_3=mo_blk_size(jspin)%array)
1231 42 : CALL dbt_distribution_destroy(t_dist)
1232 76 : DEALLOCATE (mo_dist_3)
1233 : END DO
1234 : END DO
1235 :
1236 : !clean-up
1237 30 : CALL dbt_pgrid_destroy(t_pgrid)
1238 64 : DO ispin = 1, nspins
1239 34 : DEALLOCATE (mo_blk_size(ispin)%array)
1240 34 : DEALLOCATE (mo_col_dist(ispin)%array)
1241 64 : DEALLOCATE (mo_row_dist(ispin)%array)
1242 : END DO
1243 :
1244 30 : CALL timestop(handle)
1245 :
1246 150 : END SUBROUTINE contract_AOs_to_MOs
1247 :
1248 : ! **************************************************************************************************
1249 : !> \brief Contracts the (oI|Y) tensors, for each donor MO
1250 : !> \param oI_Y the contracted tensr. It is assumed to be allocated outside of this routine
1251 : !> \param ispin ...
1252 : !> \param ao_blk_size ...
1253 : !> \param mo_blk_size ...
1254 : !> \param ri_blk_size ...
1255 : !> \param t_mo_coeffs ...
1256 : !> \param donor_state ...
1257 : !> \param xas_tdp_env ...
1258 : !> \param xas_tdp_control ...
1259 : !> \param qs_env ...
1260 : ! **************************************************************************************************
1261 34 : SUBROUTINE get_oIY_tensors(oI_Y, ispin, ao_blk_size, mo_blk_size, ri_blk_size, t_mo_coeffs, &
1262 : donor_state, xas_tdp_env, xas_tdp_control, qs_env)
1263 :
1264 : TYPE(dbt_type), ALLOCATABLE, DIMENSION(:), &
1265 : INTENT(INOUT) :: oI_Y
1266 : INTEGER, INTENT(IN) :: ispin
1267 : INTEGER, DIMENSION(:), POINTER :: ao_blk_size, mo_blk_size, ri_blk_size
1268 : TYPE(dbt_type), INTENT(inout) :: t_mo_coeffs
1269 : TYPE(donor_state_type), POINTER :: donor_state
1270 : TYPE(xas_tdp_env_type), POINTER :: xas_tdp_env
1271 : TYPE(xas_tdp_control_type), POINTER :: xas_tdp_control
1272 : TYPE(qs_environment_type), POINTER :: qs_env
1273 :
1274 : CHARACTER(len=*), PARAMETER :: routineN = 'get_oIY_tensors'
1275 :
1276 : INTEGER :: bo(2), handle, i, ido_mo, ind(2), natom, &
1277 : nblk_aos, nblk_mos, nblk_ri, ndo_mo, &
1278 : pdims_2d(2), proc_id
1279 34 : INTEGER, DIMENSION(:), POINTER :: ao_row_dist, mo_row_dist, ri_col_dist
1280 34 : INTEGER, DIMENSION(:, :), POINTER :: mat_pgrid
1281 : LOGICAL :: found
1282 34 : REAL(dp), ALLOCATABLE, DIMENSION(:, :) :: pblock
1283 : TYPE(dbcsr_distribution_type), POINTER :: std_mat_dist
1284 34 : TYPE(dbcsr_p_type), DIMENSION(:), POINTER :: pI_Y
1285 306 : TYPE(dbt_distribution_type) :: t_dist
1286 : TYPE(dbt_iterator_type) :: iter
1287 102 : TYPE(dbt_pgrid_type) :: t_pgrid
1288 578 : TYPE(dbt_type) :: t_pI_Y, t_work
1289 : TYPE(mp_para_env_type), POINTER :: para_env
1290 :
1291 34 : CALL timeset(routineN, handle)
1292 :
1293 34 : CALL get_qs_env(qs_env, natom=natom, para_env=para_env, dbcsr_dist=std_mat_dist)
1294 34 : ndo_mo = donor_state%ndo_mo
1295 34 : nblk_aos = SIZE(ao_blk_size)
1296 34 : nblk_mos = SIZE(mo_blk_size)
1297 34 : nblk_ri = SIZE(ri_blk_size)
1298 :
1299 : !We first contract (pq|X) over q into I using kernel routines (goes over all MOs and spins)
1300 34 : CALL contract2_AO_to_doMO(pI_Y, "EXCHANGE", donor_state, xas_tdp_env, xas_tdp_control, qs_env)
1301 :
1302 : !multiply by (X|Y)^-1
1303 34 : CALL ri_all_blocks_mm(pI_Y, xas_tdp_env%ri_inv_ex)
1304 :
1305 : !get standaed 2d matrix proc grid
1306 34 : CALL dbcsr_distribution_get(std_mat_dist, pgrid=mat_pgrid)
1307 :
1308 : !Loop over donor MOs of this spin
1309 76 : DO ido_mo = (ispin - 1)*ndo_mo + 1, ispin*ndo_mo
1310 :
1311 : !cast the matrix into a tensor
1312 42 : CALL dbt_create(pI_Y(ido_mo)%matrix, t_work)
1313 42 : CALL dbt_copy_matrix_to_tensor(pI_Y(ido_mo)%matrix, t_work)
1314 :
1315 : !find col proc_id of the only populated column of t_work
1316 126 : ALLOCATE (ri_col_dist(natom))
1317 42 : CALL dbt_get_info(t_work, proc_dist_2=ri_col_dist)
1318 42 : proc_id = ri_col_dist(donor_state%at_index)
1319 42 : DEALLOCATE (ri_col_dist)
1320 :
1321 : !preapre (oI_Y) tensor and (pI|Y) tensor in proper dist and blk sizes
1322 42 : pdims_2d(1) = SIZE(mat_pgrid, 1); pdims_2d(2) = SIZE(mat_pgrid, 2)
1323 42 : CALL dbt_pgrid_create(para_env, pdims_2d, t_pgrid)
1324 :
1325 294 : ALLOCATE (ri_col_dist(nblk_ri), ao_row_dist(nblk_aos), mo_row_dist(nblk_mos))
1326 42 : CALL dbt_get_info(t_work, proc_dist_1=ao_row_dist)
1327 84 : ri_col_dist = proc_id
1328 :
1329 42 : CALL dbt_distribution_new(t_dist, t_pgrid, nd_dist_1=ao_row_dist, nd_dist_2=ri_col_dist)
1330 : CALL dbt_create(t_pI_Y, name="(pI|Y)", dist=t_dist, map1_2d=[1], map2_2d=[2], &
1331 42 : blk_size_1=ao_blk_size, blk_size_2=ri_blk_size)
1332 42 : CALL dbt_distribution_destroy(t_dist)
1333 :
1334 : !copy block by block, dist match
1335 :
1336 : !$OMP PARALLEL DEFAULT(NONE) SHARED(t_work,t_pI_Y,nblk_ri,ri_blk_size,ao_blk_size) &
1337 42 : !$OMP PRIVATE(iter,ind,pblock,found,bo)
1338 : CALL dbt_iterator_start(iter, t_work)
1339 : DO WHILE (dbt_iterator_blocks_left(iter))
1340 : CALL dbt_iterator_next_block(iter, ind)
1341 : CALL dbt_get_block(t_work, ind, pblock, found)
1342 :
1343 : DO i = 1, nblk_ri
1344 : bo(1) = SUM(ri_blk_size(1:i - 1)) + 1
1345 : bo(2) = bo(1) + ri_blk_size(i) - 1
1346 : CALL dbt_put_block(t_pI_Y, [ind(1), i], [ao_blk_size(ind(1)), ri_blk_size(i)], &
1347 : pblock(:, bo(1):bo(2)))
1348 : END DO
1349 :
1350 : DEALLOCATE (pblock)
1351 : END DO
1352 : CALL dbt_iterator_stop(iter)
1353 : !$OMP END PARALLEL
1354 42 : CALL dbt_finalize(t_pI_Y)
1355 :
1356 : !get optimal pgrid for (oI|Y)
1357 42 : CALL dbt_pgrid_destroy(t_pgrid)
1358 42 : pdims_2d = 0
1359 126 : CALL dbt_pgrid_create(para_env, pdims_2d, t_pgrid, tensor_dims=[nblk_mos, nblk_ri])
1360 :
1361 42 : CALL dbt_default_distvec(nblk_aos, pdims_2d(1), ao_blk_size, ao_row_dist)
1362 42 : CALL dbt_default_distvec(nblk_mos, pdims_2d(1), mo_blk_size, mo_row_dist)
1363 42 : CALL dbt_default_distvec(nblk_ri, pdims_2d(2), ri_blk_size, ri_col_dist)
1364 :
1365 : !transfer pI_Y to the correct pgrid
1366 42 : CALL dbt_destroy(t_work)
1367 42 : CALL dbt_distribution_new(t_dist, t_pgrid, nd_dist_1=ao_row_dist, nd_dist_2=ri_col_dist)
1368 : CALL dbt_create(t_work, name="t_pI_Y", dist=t_dist, map1_2d=[1], map2_2d=[2], &
1369 42 : blk_size_1=ao_blk_size, blk_size_2=ri_blk_size)
1370 42 : CALL dbt_copy(t_pI_Y, t_work, move_data=.TRUE.)
1371 42 : CALL dbt_distribution_destroy(t_dist)
1372 :
1373 : !create (oI|Y)
1374 42 : CALL dbt_distribution_new(t_dist, t_pgrid, nd_dist_1=mo_row_dist, nd_dist_2=ri_col_dist)
1375 : CALL dbt_create(oI_Y(ido_mo), name="(oI|Y)", dist=t_dist, map1_2d=[1], map2_2d=[2], &
1376 42 : blk_size_1=mo_blk_size, blk_size_2=ri_blk_size)
1377 42 : CALL dbt_distribution_destroy(t_dist)
1378 :
1379 : !contract (pI|Y) into (oI|Y)
1380 : CALL dbt_contract(alpha=1.0_dp, tensor_1=t_work, tensor_2=t_mo_coeffs, &
1381 : beta=0.0_dp, tensor_3=oI_Y(ido_mo), contract_1=[1], &
1382 : notcontract_1=[2], contract_2=[1], notcontract_2=[2], &
1383 42 : map_1=[2], map_2=[1]) !no bound, all MOs needed
1384 :
1385 : !intermediate clean-up
1386 42 : CALL dbt_destroy(t_work)
1387 42 : CALL dbt_destroy(t_pI_Y)
1388 42 : CALL dbt_pgrid_destroy(t_pgrid)
1389 76 : DEALLOCATE (ri_col_dist, ao_row_dist, mo_row_dist)
1390 :
1391 : END DO !ido_mo
1392 :
1393 : !final clean-up
1394 34 : CALL dbcsr_deallocate_matrix_set(pI_Y)
1395 :
1396 34 : CALL timestop(handle)
1397 :
1398 68 : END SUBROUTINE get_oIY_tensors
1399 :
1400 : ! **************************************************************************************************
1401 : !> \brief Computes the same spin, occupied-occupied-virtual MO contribution to the electron propagator:
1402 : !> 0.5 * sum_ajk |<Ia||jk>|^2/(omega + eps_a - epsj_j - eps_k) and its 1st derivative wrt omega:
1403 : !> -0.5 * sum_ajk |<Ia||jk>|^2/(omega + eps_a - epsj_j - eps_k)**2
1404 : !> \param contrib ...
1405 : !> \param dev the first derivative
1406 : !> \param ja_Ik_diff ... contains the (ja|Ik) - (ka|Ij) tensor
1407 : !> \param occ_evals ...
1408 : !> \param virt_evals ...
1409 : !> \param omega ...
1410 : !> \param c_ss ...
1411 : !> \note since the is same-spin, there is only one possibility for occ_evals and virt_evals
1412 : ! **************************************************************************************************
1413 174 : SUBROUTINE calc_ss_oov_contrib(contrib, dev, ja_Ik_diff, occ_evals, virt_evals, omega, c_ss)
1414 :
1415 : REAL(dp), INTENT(inout) :: contrib, dev
1416 : TYPE(dbt_type), INTENT(inout) :: ja_Ik_diff
1417 : REAL(dp), DIMENSION(:), INTENT(IN) :: occ_evals, virt_evals
1418 : REAL(dp), INTENT(in) :: omega, c_ss
1419 :
1420 : CHARACTER(len=*), PARAMETER :: routineN = 'calc_ss_oov_contrib'
1421 :
1422 : INTEGER :: a, boff(3), bsize(3), handle, idx1, &
1423 : idx2, idx3, ind(3), j, k, nocc
1424 : LOGICAL :: found
1425 : REAL(dp) :: denom, tmp
1426 174 : REAL(dp), ALLOCATABLE, DIMENSION(:, :, :) :: tensor_blk
1427 : TYPE(dbt_iterator_type) :: iter
1428 :
1429 174 : CALL timeset(routineN, handle)
1430 :
1431 : !<Ia||jk> = <Ia|jk> - <Ia|kj> = (Ij|ak) - (Ik|aj) = (ka|Ij) - (ja|Ik)
1432 : !Note: the same spin contribution only involve spib-orbitals that are all of the same spin
1433 :
1434 174 : nocc = SIZE(occ_evals, 1)
1435 :
1436 : !Iterate over the tensors and sum. Both tensors have same dist
1437 :
1438 : !$OMP PARALLEL DEFAULT(NONE) REDUCTION(+:contrib,dev) &
1439 : !$OMP SHARED(ja_Ik_diff,occ_evals,virt_evals,omega,c_ss,nocc) &
1440 174 : !$OMP PRIVATE(iter,ind,boff,bsize,tensor_blk,found,idx1,idx2,idx3,j,A,k,denom,tmp)
1441 : CALL dbt_iterator_start(iter, ja_Ik_diff)
1442 : DO WHILE (dbt_iterator_blocks_left(iter))
1443 : CALL dbt_iterator_next_block(iter, ind, blk_offset=boff, blk_size=bsize)
1444 : CALL dbt_get_block(ja_Ik_diff, ind, tensor_blk, found)
1445 :
1446 : IF (found) THEN
1447 :
1448 : DO idx3 = 1, bsize(3)
1449 : DO idx2 = 1, bsize(2)
1450 : DO idx1 = 1, bsize(1)
1451 :
1452 : !get proper MO indices
1453 : j = boff(1) + idx1 - 1
1454 : a = boff(2) + idx2 - 1 - nocc
1455 : k = boff(3) + idx3 - 1
1456 :
1457 : !the denominator
1458 : denom = omega + virt_evals(a) - occ_evals(j) - occ_evals(k)
1459 :
1460 : !the same spin contribution
1461 : tmp = c_ss*tensor_blk(idx1, idx2, idx3)**2
1462 :
1463 : contrib = contrib + 0.5_dp*tmp/denom
1464 : dev = dev - 0.5_dp*tmp/denom**2
1465 :
1466 : END DO
1467 : END DO
1468 : END DO
1469 : END IF
1470 : DEALLOCATE (tensor_blk)
1471 : END DO
1472 : CALL dbt_iterator_stop(iter)
1473 : !$OMP END PARALLEL
1474 :
1475 174 : CALL timestop(handle)
1476 :
1477 348 : END SUBROUTINE calc_ss_oov_contrib
1478 :
1479 : ! **************************************************************************************************
1480 : !> \brief Computes the opposite spin, occupied-occupied-virtual MO contribution to the electron propagator:
1481 : !> 0.5 * sum_ajk |<Ia||jk>|^2/(omega + eps_a - epsj_j - eps_k) and its 1st derivative wrt omega:
1482 : !> -0.5 * sum_ajk |<Ia||jk>|^2/(omega + eps_a - epsj_j - eps_k)**2
1483 : !> \param contrib ...
1484 : !> \param dev the first derivative
1485 : !> \param ja_Ik ...
1486 : !> \param j_evals ocucpied evals for j MO
1487 : !> \param a_evals virtual evals for a MO
1488 : !> \param k_evals ocucpied evals for k MO
1489 : !> \param omega ...
1490 : !> \param c_os ...
1491 : !> \param a_offset the number of occupied MOs for the same spin as a MOs
1492 : !> \note since this is opposite-spin, evals might be different for different spins
1493 : ! **************************************************************************************************
1494 174 : SUBROUTINE calc_os_oov_contrib(contrib, dev, ja_Ik, j_evals, a_evals, k_evals, omega, c_os, a_offset)
1495 :
1496 : REAL(dp), INTENT(inout) :: contrib, dev
1497 : TYPE(dbt_type), INTENT(inout) :: ja_Ik
1498 : REAL(dp), DIMENSION(:), INTENT(IN) :: j_evals, a_evals, k_evals
1499 : REAL(dp), INTENT(in) :: omega, c_os
1500 : INTEGER, INTENT(IN) :: a_offset
1501 :
1502 : CHARACTER(len=*), PARAMETER :: routineN = 'calc_os_oov_contrib'
1503 :
1504 : INTEGER :: a, boff(3), bsize(3), handle, idx1, &
1505 : idx2, idx3, ind(3), j, k
1506 : LOGICAL :: found
1507 : REAL(dp) :: denom, tmp
1508 174 : REAL(dp), ALLOCATABLE, DIMENSION(:, :, :) :: ja_Ik_blk
1509 : TYPE(dbt_iterator_type) :: iter
1510 :
1511 174 : CALL timeset(routineN, handle)
1512 :
1513 : !<Ia||jk> = <Ia|jk> - <Ia|kj> = (Ij|ak) - (Ik|aj) = (ka|Ij) - (ja|Ik)
1514 : !Note: the opposite spin contribution comes in 2 parts, once (ka|Ij) and once (ja|Ik) only,
1515 : ! where both spin-orbitals on one center have the same spin, but it is the opposite of
1516 : ! the spin on the other center. Because it is eventually summed, can consider only one
1517 : ! of the 2 terms, but with a factor 2
1518 :
1519 : !Iterate over the tensor and sum
1520 :
1521 : !$OMP PARALLEL DEFAULT(NONE) REDUCTION(+:contrib,dev) &
1522 : !$OMP SHARED(ja_Ik,j_evals,a_evals,k_evals,omega,c_os,a_offset) &
1523 174 : !$OMP PRIVATE(iter,ind,boff,bsize,ja_Ik_blk,found,idx1,idx2,idx3,j,A,k,denom,tmp)
1524 : CALL dbt_iterator_start(iter, ja_Ik)
1525 : DO WHILE (dbt_iterator_blocks_left(iter))
1526 : CALL dbt_iterator_next_block(iter, ind, blk_offset=boff, blk_size=bsize)
1527 : CALL dbt_get_block(ja_Ik, ind, ja_Ik_blk, found)
1528 :
1529 : IF (found) THEN
1530 :
1531 : DO idx3 = 1, bsize(3)
1532 : DO idx2 = 1, bsize(2)
1533 : DO idx1 = 1, bsize(1)
1534 :
1535 : !get proper MO indices
1536 : j = boff(1) + idx1 - 1
1537 : a = boff(2) + idx2 - 1 - a_offset
1538 : k = boff(3) + idx3 - 1
1539 :
1540 : !the denominator
1541 : denom = omega + a_evals(a) - j_evals(j) - k_evals(k)
1542 :
1543 : !the opposite spin contribution
1544 : tmp = c_os*ja_Ik_blk(idx1, idx2, idx3)**2
1545 :
1546 : !take factor 2 into acocunt (2 x 0.5 = 1)
1547 : contrib = contrib + tmp/denom
1548 : dev = dev - tmp/denom**2
1549 :
1550 : END DO
1551 : END DO
1552 : END DO
1553 : END IF
1554 : DEALLOCATE (ja_Ik_blk)
1555 : END DO
1556 : CALL dbt_iterator_stop(iter)
1557 : !$OMP END PARALLEL
1558 :
1559 174 : CALL timestop(handle)
1560 :
1561 348 : END SUBROUTINE calc_os_oov_contrib
1562 :
1563 : ! **************************************************************************************************
1564 : !> \brief Computes the same-spin occupied-virtual-virtual MO contribution to the electron propagator:
1565 : !> 0.5 * sum_abj |<Ij||ab>|^2/(omega + eps_j - eps_a - eps_b) as well as its first derivative:
1566 : !> -0.5 * sum_abj |<Ij||ab>|^2/(omega + eps_j - eps_a - eps_b)**2
1567 : !> \param contrib ...
1568 : !> \param dev the first derivative
1569 : !> \param aj_Ib_diff contatins the (aj|Ib) - (bj|Ia) tensor
1570 : !> \param occ_evals ...
1571 : !> \param virt_evals ...
1572 : !> \param omega ...
1573 : !> \param c_ss ...
1574 : !> \note since the is same-spin, there is only one possibility for occ_evals and virt_evals
1575 : ! **************************************************************************************************
1576 166 : SUBROUTINE calc_ss_ovv_contrib(contrib, dev, aj_Ib_diff, occ_evals, virt_evals, omega, c_ss)
1577 :
1578 : REAL(dp), INTENT(inout) :: contrib, dev
1579 : TYPE(dbt_type), INTENT(inout) :: aj_Ib_diff
1580 : REAL(dp), DIMENSION(:), INTENT(IN) :: occ_evals, virt_evals
1581 : REAL(dp), INTENT(in) :: omega, c_ss
1582 :
1583 : CHARACTER(len=*), PARAMETER :: routineN = 'calc_ss_ovv_contrib'
1584 :
1585 : INTEGER :: a, b, boff(3), bsize(3), handle, idx1, &
1586 : idx2, idx3, ind(3), j, nocc
1587 : LOGICAL :: found
1588 : REAL(dp) :: denom, tmp
1589 166 : REAL(dp), ALLOCATABLE, DIMENSION(:, :, :) :: tensor_blk
1590 : TYPE(dbt_iterator_type) :: iter
1591 :
1592 166 : CALL timeset(routineN, handle)
1593 :
1594 : !<Ij||ab> = <Ij|ab> - <Ij|ba> = (Ia|jb) - (Ib|ja) = (jb|Ia) - (ja|Ib)
1595 : !Notes: only non-zero contribution if all MOs have the same spin
1596 :
1597 166 : nocc = SIZE(occ_evals, 1)
1598 :
1599 : !tensors have matching distributions, can do that safely
1600 :
1601 : !$OMP PARALLEL DEFAULT(NONE) REDUCTION(+:contrib,dev) &
1602 : !$OMP SHARED(aj_Ib_diff,occ_evals,virt_evals,omega,c_ss,nocc) &
1603 166 : !$OMP PRIVATE(iter,ind,boff,bsize,tensor_blk,found,idx1,idx2,idx3,j,A,b,denom,tmp)
1604 : CALL dbt_iterator_start(iter, aj_Ib_diff)
1605 : DO WHILE (dbt_iterator_blocks_left(iter))
1606 : CALL dbt_iterator_next_block(iter, ind, blk_offset=boff, blk_size=bsize)
1607 : CALL dbt_get_block(aj_Ib_diff, ind, tensor_blk, found)
1608 :
1609 : IF (found) THEN
1610 :
1611 : DO idx3 = 1, bsize(3)
1612 : DO idx2 = 1, bsize(2)
1613 : DO idx1 = 1, bsize(1)
1614 :
1615 : !get proper MO indices
1616 : a = boff(1) + idx1 - 1 - nocc
1617 : j = boff(2) + idx2 - 1
1618 : b = boff(3) + idx3 - 1 - nocc
1619 :
1620 : !the common denominator
1621 : denom = omega + occ_evals(j) - virt_evals(a) - virt_evals(b)
1622 :
1623 : !the same spin contribution
1624 : tmp = c_ss*tensor_blk(idx1, idx2, idx3)**2
1625 :
1626 : contrib = contrib + 0.5_dp*tmp/denom
1627 : dev = dev - 0.5_dp*tmp/denom**2
1628 :
1629 : END DO
1630 : END DO
1631 : END DO
1632 : END IF
1633 : DEALLOCATE (tensor_blk)
1634 : END DO
1635 : CALL dbt_iterator_stop(iter)
1636 : !$OMP END PARALLEL
1637 :
1638 166 : CALL timestop(handle)
1639 :
1640 332 : END SUBROUTINE calc_ss_ovv_contrib
1641 :
1642 : ! **************************************************************************************************
1643 : !> \brief Computes the opposite-spin occupied-virtual-virtual MO contribution to the electron propagator:
1644 : !> 0.5 * sum_abj |<Ij||ab>|^2/(omega + eps_j - eps_a - eps_b) as well as its first derivative:
1645 : !> -0.5 * sum_abj |<Ij||ab>|^2/(omega + eps_j - eps_a - eps_b)**2
1646 : !> \param contrib ...
1647 : !> \param dev the first derivative
1648 : !> \param aj_Ib ...
1649 : !> \param a_evals virtual evals for a MO
1650 : !> \param j_evals occupied evals for j MO
1651 : !> \param b_evals virtual evals for b MO
1652 : !> \param omega ...
1653 : !> \param c_os ...
1654 : !> \param a_offset number of occupied MOs for the same spin as a MO
1655 : !> \param b_offset number of occupied MOs for the same spin as b MO
1656 : !> \note since this is opposite-spin, evals might be different for different spins
1657 : ! **************************************************************************************************
1658 166 : SUBROUTINE calc_os_ovv_contrib(contrib, dev, aj_Ib, a_evals, j_evals, b_evals, omega, c_os, &
1659 : a_offset, b_offset)
1660 :
1661 : REAL(dp), INTENT(inout) :: contrib, dev
1662 : TYPE(dbt_type), INTENT(inout) :: aj_Ib
1663 : REAL(dp), DIMENSION(:), INTENT(IN) :: a_evals, j_evals, b_evals
1664 : REAL(dp), INTENT(in) :: omega, c_os
1665 : INTEGER, INTENT(IN) :: a_offset, b_offset
1666 :
1667 : CHARACTER(len=*), PARAMETER :: routineN = 'calc_os_ovv_contrib'
1668 :
1669 : INTEGER :: a, b, boff(3), bsize(3), handle, idx1, &
1670 : idx2, idx3, ind(3), j
1671 : LOGICAL :: found
1672 : REAL(dp) :: denom, tmp
1673 166 : REAL(dp), ALLOCATABLE, DIMENSION(:, :, :) :: aj_Ib_blk
1674 : TYPE(dbt_iterator_type) :: iter
1675 :
1676 166 : CALL timeset(routineN, handle)
1677 :
1678 : !<Ij||ab> = <Ij|ab> - <Ij|ba> = (Ia|jb) - (Ib|ja) = (jb|Ia) - (ja|Ib)
1679 : !Notes: only 2 distinct contributions, once from (jb|Ia) and once form (ja|Ib) only, when the 2
1680 : ! MOs on one center have one spin and the 2 MOs on the other center have another spin
1681 : ! In the end, the sum is such that can take one of those with a factor 2
1682 :
1683 : !$OMP PARALLEL DEFAULT(NONE) REDUCTION(+:contrib,dev) &
1684 : !$OMP SHARED(aj_Ib,a_evals,j_evals,b_evals,omega,c_os,a_offset,b_offset) &
1685 166 : !$OMP PRIVATE(iter,ind,boff,bsize,aj_Ib_blk,found,idx1,idx2,idx3,j,A,b,denom,tmp)
1686 : CALL dbt_iterator_start(iter, aj_Ib)
1687 : DO WHILE (dbt_iterator_blocks_left(iter))
1688 : CALL dbt_iterator_next_block(iter, ind, blk_offset=boff, blk_size=bsize)
1689 : CALL dbt_get_block(aj_Ib, ind, aj_Ib_blk, found)
1690 :
1691 : IF (found) THEN
1692 :
1693 : DO idx3 = 1, bsize(3)
1694 : DO idx2 = 1, bsize(2)
1695 : DO idx1 = 1, bsize(1)
1696 :
1697 : !get proper MO indices
1698 : a = boff(1) + idx1 - 1 - a_offset
1699 : j = boff(2) + idx2 - 1
1700 : b = boff(3) + idx3 - 1 - b_offset
1701 :
1702 : !the denominator
1703 : denom = omega + j_evals(j) - a_evals(a) - b_evals(b)
1704 :
1705 : !the opposite-spin contribution. Factor 2 taken into account (2 x 0.5 = 1)
1706 : tmp = c_os*(aj_Ib_blk(idx1, idx2, idx3))**2
1707 :
1708 : contrib = contrib + tmp/denom
1709 : dev = dev - tmp/denom**2
1710 :
1711 : END DO
1712 : END DO
1713 : END DO
1714 : END IF
1715 : DEALLOCATE (aj_Ib_blk)
1716 : END DO
1717 : CALL dbt_iterator_stop(iter)
1718 : !$OMP END PARALLEL
1719 :
1720 166 : CALL timestop(handle)
1721 :
1722 332 : END SUBROUTINE calc_os_ovv_contrib
1723 :
1724 : ! **************************************************************************************************
1725 : !> \brief We try to compute the spin-orbit splitting via perturbation theory. We keep it
1726 : !>\ cheap by only inculding the degenerate states (2p, 3d, 3p, etc.).
1727 : !> \param soc_shifts the SOC corrected orbital shifts to apply to original energies, for both spins
1728 : !> \param donor_state ...
1729 : !> \param xas_tdp_env ...
1730 : !> \param xas_tdp_control ...
1731 : !> \param qs_env ...
1732 : ! **************************************************************************************************
1733 4 : SUBROUTINE get_soc_splitting(soc_shifts, donor_state, xas_tdp_env, xas_tdp_control, qs_env)
1734 :
1735 : REAL(dp), ALLOCATABLE, DIMENSION(:, :), &
1736 : INTENT(out) :: soc_shifts
1737 : TYPE(donor_state_type), POINTER :: donor_state
1738 : TYPE(xas_tdp_env_type), POINTER :: xas_tdp_env
1739 : TYPE(xas_tdp_control_type), POINTER :: xas_tdp_control
1740 : TYPE(qs_environment_type), POINTER :: qs_env
1741 :
1742 : CHARACTER(len=*), PARAMETER :: routineN = 'get_soc_splitting'
1743 :
1744 4 : COMPLEX(dp), ALLOCATABLE, DIMENSION(:, :) :: evecs, hami
1745 : INTEGER :: beta_spin, handle, ialpha, ibeta, &
1746 : ido_mo, ispin, nao, ndo_mo, ndo_so, &
1747 : nspins
1748 : REAL(dp) :: alpha_tot_contrib, beta_tot_contrib
1749 4 : REAL(dp), ALLOCATABLE, DIMENSION(:) :: evals
1750 4 : REAL(dp), ALLOCATABLE, DIMENSION(:, :) :: tmp_shifts
1751 : TYPE(cp_blacs_env_type), POINTER :: blacs_env
1752 : TYPE(cp_cfm_type) :: hami_cfm
1753 : TYPE(cp_fm_struct_type), POINTER :: ao_domo_struct, domo_domo_struct, &
1754 : doso_doso_struct
1755 : TYPE(cp_fm_type) :: alpha_gs_coeffs, ao_domo_work, &
1756 : beta_gs_coeffs, domo_domo_work, &
1757 : img_fm, real_fm
1758 4 : TYPE(dbcsr_p_type), DIMENSION(:), POINTER :: matrix_ks
1759 : TYPE(dbcsr_type), POINTER :: orb_soc_x, orb_soc_y, orb_soc_z
1760 : TYPE(mp_para_env_type), POINTER :: para_env
1761 :
1762 4 : NULLIFY (matrix_ks, para_env, blacs_env, ao_domo_struct, domo_domo_struct, &
1763 4 : doso_doso_struct, orb_soc_x, orb_soc_y, orb_soc_z)
1764 :
1765 4 : CALL timeset(routineN, handle)
1766 :
1767 : ! Idea: we compute the SOC matrix in the space of the degenerate spin-orbitals, add it to
1768 : ! the KS matrix in the same basis, diagonalize the whole thing and get the corrected energies
1769 : ! for SOC
1770 :
1771 4 : CALL get_qs_env(qs_env, matrix_ks=matrix_ks, para_env=para_env, blacs_env=blacs_env)
1772 :
1773 4 : orb_soc_x => xas_tdp_env%orb_soc(1)%matrix
1774 4 : orb_soc_y => xas_tdp_env%orb_soc(2)%matrix
1775 4 : orb_soc_z => xas_tdp_env%orb_soc(3)%matrix
1776 :
1777 : ! Whether it is open-shell or not, we have 2*ndo_mo spin-orbitals
1778 4 : nspins = 2
1779 4 : ndo_mo = donor_state%ndo_mo
1780 4 : ndo_so = nspins*ndo_mo
1781 4 : CALL dbcsr_get_info(matrix_ks(1)%matrix, nfullrows_total=nao)
1782 :
1783 : ! Build the fm infrastructure
1784 : CALL cp_fm_struct_create(ao_domo_struct, context=blacs_env, para_env=para_env, &
1785 4 : nrow_global=nao, ncol_global=ndo_mo)
1786 : CALL cp_fm_struct_create(domo_domo_struct, context=blacs_env, para_env=para_env, &
1787 4 : nrow_global=ndo_mo, ncol_global=ndo_mo)
1788 : CALL cp_fm_struct_create(doso_doso_struct, context=blacs_env, para_env=para_env, &
1789 4 : nrow_global=ndo_so, ncol_global=ndo_so)
1790 :
1791 4 : CALL cp_fm_create(alpha_gs_coeffs, ao_domo_struct)
1792 4 : CALL cp_fm_create(beta_gs_coeffs, ao_domo_struct)
1793 4 : CALL cp_fm_create(ao_domo_work, ao_domo_struct)
1794 4 : CALL cp_fm_create(domo_domo_work, domo_domo_struct)
1795 4 : CALL cp_fm_create(real_fm, doso_doso_struct)
1796 4 : CALL cp_fm_create(img_fm, doso_doso_struct)
1797 :
1798 : ! Put the gs_coeffs in the correct format.
1799 4 : IF (xas_tdp_control%do_uks) THEN
1800 :
1801 : CALL cp_fm_to_fm_submat(msource=donor_state%gs_coeffs, mtarget=alpha_gs_coeffs, nrow=nao, &
1802 0 : ncol=ndo_mo, s_firstrow=1, s_firstcol=1, t_firstrow=1, t_firstcol=1)
1803 : CALL cp_fm_to_fm_submat(msource=donor_state%gs_coeffs, mtarget=beta_gs_coeffs, nrow=nao, &
1804 0 : ncol=ndo_mo, s_firstrow=1, s_firstcol=ndo_mo + 1, t_firstrow=1, t_firstcol=1)
1805 :
1806 : ELSE
1807 :
1808 4 : CALL cp_fm_to_fm(donor_state%gs_coeffs, alpha_gs_coeffs)
1809 4 : CALL cp_fm_to_fm(donor_state%gs_coeffs, beta_gs_coeffs)
1810 : END IF
1811 :
1812 : ! Compute the KS matrix in this basis, add it to the real part of the final matrix
1813 : !alpha-alpha block in upper left quadrant
1814 4 : CALL cp_dbcsr_sm_fm_multiply(matrix_ks(1)%matrix, alpha_gs_coeffs, ao_domo_work, ncol=ndo_mo)
1815 : CALL parallel_gemm('T', 'N', ndo_mo, ndo_mo, nao, 1.0_dp, alpha_gs_coeffs, ao_domo_work, 0.0_dp, &
1816 4 : domo_domo_work)
1817 : CALL cp_fm_to_fm_submat(msource=domo_domo_work, mtarget=real_fm, nrow=ndo_mo, ncol=ndo_mo, &
1818 4 : s_firstrow=1, s_firstcol=1, t_firstrow=1, t_firstcol=1)
1819 :
1820 : !beta-beta block in lower right quadrant
1821 4 : beta_spin = 1; IF (xas_tdp_control%do_uks .OR. xas_tdp_control%do_roks) beta_spin = 2
1822 4 : CALL cp_dbcsr_sm_fm_multiply(matrix_ks(beta_spin)%matrix, beta_gs_coeffs, ao_domo_work, ncol=ndo_mo)
1823 : CALL parallel_gemm('T', 'N', ndo_mo, ndo_mo, nao, 1.0_dp, beta_gs_coeffs, ao_domo_work, 0.0_dp, &
1824 4 : domo_domo_work)
1825 : CALL cp_fm_to_fm_submat(msource=domo_domo_work, mtarget=real_fm, nrow=ndo_mo, ncol=ndo_mo, &
1826 4 : s_firstrow=1, s_firstcol=1, t_firstrow=ndo_mo + 1, t_firstcol=ndo_mo + 1)
1827 :
1828 : ! Compute the SOC matrix elements and add them to the real or imaginary part of the matrix
1829 : ! alpha-alpha block, only Hz not zero, purely imaginary, addition
1830 4 : CALL cp_dbcsr_sm_fm_multiply(orb_soc_z, alpha_gs_coeffs, ao_domo_work, ncol=ndo_mo)
1831 : CALL parallel_gemm('T', 'N', ndo_mo, ndo_mo, nao, 1.0_dp, alpha_gs_coeffs, ao_domo_work, 0.0_dp, &
1832 4 : domo_domo_work)
1833 : CALL cp_fm_to_fm_submat(msource=domo_domo_work, mtarget=img_fm, nrow=ndo_mo, ncol=ndo_mo, &
1834 4 : s_firstrow=1, s_firstcol=1, t_firstrow=1, t_firstcol=1)
1835 :
1836 : ! beta-beta block, only Hz not zero, purely imaginary, substraciton
1837 4 : CALL cp_dbcsr_sm_fm_multiply(orb_soc_z, beta_gs_coeffs, ao_domo_work, ncol=ndo_mo)
1838 : CALL parallel_gemm('T', 'N', ndo_mo, ndo_mo, nao, -1.0_dp, beta_gs_coeffs, ao_domo_work, 0.0_dp, &
1839 4 : domo_domo_work)
1840 : CALL cp_fm_to_fm_submat(msource=domo_domo_work, mtarget=img_fm, nrow=ndo_mo, ncol=ndo_mo, &
1841 4 : s_firstrow=1, s_firstcol=1, t_firstrow=ndo_mo + 1, t_firstcol=ndo_mo + 1)
1842 :
1843 : ! alpha-beta block, two non-zero terms in Hx and Hy
1844 : ! Hx term, purely imaginary, addition
1845 4 : CALL cp_dbcsr_sm_fm_multiply(orb_soc_x, beta_gs_coeffs, ao_domo_work, ncol=ndo_mo)
1846 : CALL parallel_gemm('T', 'N', ndo_mo, ndo_mo, nao, 1.0_dp, alpha_gs_coeffs, ao_domo_work, 0.0_dp, &
1847 4 : domo_domo_work)
1848 : CALL cp_fm_to_fm_submat(msource=domo_domo_work, mtarget=img_fm, nrow=ndo_mo, ncol=ndo_mo, &
1849 4 : s_firstrow=1, s_firstcol=1, t_firstrow=1, t_firstcol=ndo_mo + 1)
1850 : ! Hy term, purely real, addition
1851 4 : CALL cp_dbcsr_sm_fm_multiply(orb_soc_y, beta_gs_coeffs, ao_domo_work, ncol=ndo_mo)
1852 : CALL parallel_gemm('T', 'N', ndo_mo, ndo_mo, nao, 1.0_dp, alpha_gs_coeffs, ao_domo_work, 0.0_dp, &
1853 4 : domo_domo_work)
1854 : CALL cp_fm_to_fm_submat(msource=domo_domo_work, mtarget=real_fm, nrow=ndo_mo, ncol=ndo_mo, &
1855 4 : s_firstrow=1, s_firstcol=1, t_firstrow=1, t_firstcol=ndo_mo + 1)
1856 :
1857 : ! beta-alpha block, two non-zero terms in Hx and Hy
1858 : ! Hx term, purely imaginary, addition
1859 4 : CALL cp_dbcsr_sm_fm_multiply(orb_soc_x, alpha_gs_coeffs, ao_domo_work, ncol=ndo_mo)
1860 : CALL parallel_gemm('T', 'N', ndo_mo, ndo_mo, nao, 1.0_dp, beta_gs_coeffs, ao_domo_work, 0.0_dp, &
1861 4 : domo_domo_work)
1862 : CALL cp_fm_to_fm_submat(msource=domo_domo_work, mtarget=img_fm, nrow=ndo_mo, ncol=ndo_mo, &
1863 4 : s_firstrow=1, s_firstcol=1, t_firstrow=ndo_mo + 1, t_firstcol=1)
1864 : ! Hy term, purely real, substraction
1865 4 : CALL cp_dbcsr_sm_fm_multiply(orb_soc_y, alpha_gs_coeffs, ao_domo_work, ncol=ndo_mo)
1866 : CALL parallel_gemm('T', 'N', ndo_mo, ndo_mo, nao, -1.0_dp, beta_gs_coeffs, ao_domo_work, 0.0_dp, &
1867 4 : domo_domo_work)
1868 : CALL cp_fm_to_fm_submat(msource=domo_domo_work, mtarget=real_fm, nrow=ndo_mo, ncol=ndo_mo, &
1869 4 : s_firstrow=1, s_firstcol=1, t_firstrow=ndo_mo + 1, t_firstcol=1)
1870 :
1871 : ! Cast everything in complex fm format
1872 4 : CALL cp_cfm_create(hami_cfm, doso_doso_struct)
1873 4 : CALL cp_fm_to_cfm(real_fm, img_fm, hami_cfm)
1874 :
1875 : ! And diagonalize. Since tiny matrix (6x6), diagonalize locally
1876 32 : ALLOCATE (evals(ndo_so), evecs(ndo_so, ndo_so), hami(ndo_so, ndo_so))
1877 4 : CALL cp_cfm_get_submatrix(hami_cfm, hami)
1878 4 : CALL complex_diag(hami, evecs, evals)
1879 :
1880 : !The SOC corrected KS eigenvalues
1881 12 : ALLOCATE (tmp_shifts(ndo_mo, 2))
1882 :
1883 4 : ialpha = 1; ibeta = 1;
1884 28 : DO ido_mo = 1, ndo_so
1885 : !need to find out whether the eigenvalue corresponds to an alpha or beta spin-orbtial
1886 96 : alpha_tot_contrib = REAL(DOT_PRODUCT(evecs(1:ndo_mo, ido_mo), evecs(1:ndo_mo, ido_mo)))
1887 96 : beta_tot_contrib = REAL(DOT_PRODUCT(evecs(ndo_mo + 1:ndo_so, ido_mo), evecs(ndo_mo + 1:ndo_so, ido_mo)))
1888 :
1889 28 : IF (alpha_tot_contrib > beta_tot_contrib) THEN
1890 12 : tmp_shifts(ialpha, 1) = evals(ido_mo)
1891 12 : ialpha = ialpha + 1
1892 : ELSE
1893 12 : tmp_shifts(ibeta, 2) = evals(ido_mo)
1894 12 : ibeta = ibeta + 1
1895 : END IF
1896 : END DO
1897 :
1898 : !compute shift from KS evals
1899 16 : ALLOCATE (soc_shifts(ndo_mo, SIZE(donor_state%energy_evals, 2)))
1900 8 : DO ispin = 1, SIZE(donor_state%energy_evals, 2)
1901 20 : soc_shifts(:, ispin) = tmp_shifts(:, ispin) - donor_state%energy_evals(:, ispin)
1902 : END DO
1903 :
1904 : ! clean-up
1905 4 : CALL cp_fm_release(alpha_gs_coeffs)
1906 4 : CALL cp_fm_release(beta_gs_coeffs)
1907 4 : CALL cp_fm_release(ao_domo_work)
1908 4 : CALL cp_fm_release(domo_domo_work)
1909 4 : CALL cp_fm_release(real_fm)
1910 4 : CALL cp_fm_release(img_fm)
1911 :
1912 4 : CALL cp_cfm_release(hami_cfm)
1913 :
1914 4 : CALL cp_fm_struct_release(ao_domo_struct)
1915 4 : CALL cp_fm_struct_release(domo_domo_struct)
1916 4 : CALL cp_fm_struct_release(doso_doso_struct)
1917 :
1918 4 : CALL timestop(handle)
1919 :
1920 20 : END SUBROUTINE get_soc_splitting
1921 :
1922 : END MODULE xas_tdp_correction
|
