Line data Source code
1 : !--------------------------------------------------------------------------------------------------!
2 : ! CP2K: A general program to perform molecular dynamics simulations !
3 : ! Copyright 2000-2026 CP2K developers group <https://cp2k.org> !
4 : ! !
5 : ! SPDX-License-Identifier: GPL-2.0-or-later !
6 : !--------------------------------------------------------------------------------------------------!
7 :
8 : ! **************************************************************************************************
9 : !> \brief All the kernel specific subroutines for XAS TDP calculations
10 : !> \author A. Bussy (03.2019)
11 : ! **************************************************************************************************
12 :
13 : MODULE xas_tdp_kernel
14 : USE basis_set_types, ONLY: gto_basis_set_p_type
15 : USE cp_dbcsr_api, ONLY: &
16 : dbcsr_add, dbcsr_complete_redistribute, dbcsr_copy, dbcsr_create, dbcsr_desymmetrize, &
17 : dbcsr_distribution_get, dbcsr_distribution_new, dbcsr_distribution_release, &
18 : dbcsr_distribution_type, dbcsr_filter, dbcsr_finalize, dbcsr_get_block_p, dbcsr_get_info, &
19 : dbcsr_get_num_blocks, dbcsr_get_stored_coordinates, dbcsr_iterator_blocks_left, &
20 : dbcsr_iterator_next_block, dbcsr_iterator_start, dbcsr_iterator_stop, dbcsr_iterator_type, &
21 : dbcsr_multiply, dbcsr_p_type, dbcsr_put_block, dbcsr_release, dbcsr_reserve_blocks, &
22 : dbcsr_set, dbcsr_transposed, dbcsr_type, dbcsr_type_no_symmetry, dbcsr_type_symmetric
23 : USE cp_dbcsr_operations, ONLY: cp_dbcsr_dist2d_to_dist,&
24 : dbcsr_deallocate_matrix_set
25 : USE dbt_api, ONLY: dbt_get_block,&
26 : dbt_iterator_blocks_left,&
27 : dbt_iterator_next_block,&
28 : dbt_iterator_start,&
29 : dbt_iterator_stop,&
30 : dbt_iterator_type,&
31 : dbt_type
32 : USE distribution_2d_types, ONLY: distribution_2d_type
33 : USE kinds, ONLY: dp
34 : USE message_passing, ONLY: mp_para_env_type
35 : USE particle_methods, ONLY: get_particle_set
36 : USE particle_types, ONLY: particle_type
37 : USE qs_environment_types, ONLY: get_qs_env,&
38 : qs_environment_type
39 : USE qs_integral_utils, ONLY: basis_set_list_setup
40 : USE qs_kind_types, ONLY: qs_kind_type
41 : USE util, ONLY: get_limit
42 : USE xas_tdp_types, ONLY: donor_state_type,&
43 : get_proc_batch_sizes,&
44 : xas_tdp_control_type,&
45 : xas_tdp_env_type
46 :
47 : !$ USE OMP_LIB, ONLY: omp_get_max_threads, omp_get_thread_num
48 : #include "./base/base_uses.f90"
49 :
50 : IMPLICIT NONE
51 : PRIVATE
52 :
53 : CHARACTER(len=*), PARAMETER, PRIVATE :: moduleN = 'xas_tdp_kernel'
54 :
55 : PUBLIC :: kernel_coulomb_xc, kernel_exchange, contract2_AO_to_doMO, &
56 : reserve_contraction_blocks, ri_all_blocks_mm
57 :
58 : CONTAINS
59 :
60 : ! **************************************************************************************************
61 : !> \brief Computes, if asked for it, the Coulomb and XC kernel matrices, in the usuall matrix format
62 : !> \param coul_ker pointer the the Coulomb kernel matrix (can be void pointer)
63 : !> \param xc_ker array of pointer to the different xc kernels (5 of them):
64 : !> 1) the restricted closed-shell singlet kernel
65 : !> 2) the restricted closed-shell triplet kernel
66 : !> 3) the spin-conserving open-shell xc kernel
67 : !> 4) the on-diagonal spin-flip open-shell xc kernel
68 : !> \param donor_state ...
69 : !> \param xas_tdp_env ...
70 : !> \param xas_tdp_control ...
71 : !> \param qs_env ...
72 : !> \note Coulomb and xc kernel are put together in the same routine because they use the same RI
73 : !> Coulomb: (aI|Jb) = (aI|P) (P|Q)^-1 (Q|Jb)
74 : !> XC : (aI|fxc|Jb) = (aI|P) (P|Q)^-1 (Q|fxc|R) (R|S)^-1 (S|Jb)
75 : !> In the above formula, a,b label the sgfs
76 : !> The routine analyses the xas_tdp_control to know which kernel must be computed and how
77 : !> (open-shell, singlet, triplet, ROKS, LSD, etc...)
78 : !> On entry, the pointers should be allocated
79 : ! **************************************************************************************************
80 156 : SUBROUTINE kernel_coulomb_xc(coul_ker, xc_ker, donor_state, xas_tdp_env, xas_tdp_control, qs_env)
81 :
82 : TYPE(dbcsr_type), INTENT(INOUT) :: coul_ker
83 : TYPE(dbcsr_p_type), DIMENSION(:), POINTER :: xc_ker
84 : TYPE(donor_state_type), POINTER :: donor_state
85 : TYPE(xas_tdp_env_type), POINTER :: xas_tdp_env
86 : TYPE(xas_tdp_control_type), POINTER :: xas_tdp_control
87 : TYPE(qs_environment_type), POINTER :: qs_env
88 :
89 : CHARACTER(len=*), PARAMETER :: routineN = 'kernel_coulomb_xc'
90 :
91 : INTEGER :: batch_size, bo(2), handle, i, ibatch, &
92 : iex, lb, natom, nbatch, ndo_mo, &
93 : ndo_so, nex_atom, nsgfp, ri_atom, &
94 : source, ub
95 78 : INTEGER, DIMENSION(:), POINTER :: blk_size
96 : LOGICAL :: do_coulomb, do_sc, do_sf, do_sg, do_tp, &
97 : do_xc, found
98 : REAL(dp), DIMENSION(:, :), POINTER :: PQ
99 : TYPE(dbcsr_distribution_type), POINTER :: dist
100 78 : TYPE(dbcsr_p_type), DIMENSION(:), POINTER :: contr1_int
101 : TYPE(mp_para_env_type), POINTER :: para_env
102 :
103 78 : NULLIFY (contr1_int, PQ, para_env, dist, blk_size)
104 :
105 : ! Initialization
106 78 : ndo_mo = donor_state%ndo_mo
107 78 : do_xc = xas_tdp_control%do_xc
108 78 : do_sg = xas_tdp_control%do_singlet
109 78 : do_tp = xas_tdp_control%do_triplet
110 78 : do_sc = xas_tdp_control%do_spin_cons
111 78 : do_sf = xas_tdp_control%do_spin_flip
112 78 : ndo_so = ndo_mo; IF (xas_tdp_control%do_uks) ndo_so = 2*ndo_mo
113 78 : ri_atom = donor_state%at_index
114 78 : CALL get_qs_env(qs_env, natom=natom, para_env=para_env)
115 78 : do_coulomb = xas_tdp_control%do_coulomb
116 78 : dist => donor_state%dbcsr_dist
117 78 : blk_size => donor_state%blk_size
118 :
119 : ! If no Coulomb nor xc, simply exit
120 78 : IF ((.NOT. do_coulomb) .AND. (.NOT. do_xc)) RETURN
121 :
122 78 : CALL timeset(routineN, handle)
123 :
124 : ! Contract the RI 3-center integrals once to get (aI|P)
125 78 : CALL contract2_AO_to_doMO(contr1_int, "COULOMB", donor_state, xas_tdp_env, xas_tdp_control, qs_env)
126 :
127 : ! Deal with the Coulomb case
128 78 : IF (do_coulomb) CALL coulomb(coul_ker, contr1_int, dist, blk_size, xas_tdp_env, &
129 78 : xas_tdp_control, qs_env)
130 :
131 : ! Deal with the XC case
132 78 : IF (do_xc) THEN
133 :
134 : ! In the end, we compute: (aI|fxc|Jb) = (aI|P) (P|Q)^-1 (Q|fxc|R) (R|S)^-1 (S|Jb)
135 : ! where fxc can take different spin contributions.
136 :
137 : ! Precompute the product (aI|P) * (P|Q)^-1 and store it in contr1_int
138 70 : PQ => xas_tdp_env%ri_inv_coul
139 70 : CALL ri_all_blocks_mm(contr1_int, PQ)
140 :
141 : ! If not already done (e.g. when multpile donor states for a given excited atom), broadcast
142 : ! the RI matrix (Q|fxc|R) on all procs
143 70 : IF (.NOT. xas_tdp_env%fxc_avail) THEN
144 : ! Find on which processor the integrals (Q|fxc|R) for this atom are stored
145 68 : nsgfp = SIZE(PQ, 1)
146 68 : CALL get_qs_env(qs_env, para_env=para_env)
147 68 : found = .FALSE.
148 68 : nex_atom = SIZE(xas_tdp_env%ex_atom_indices)
149 68 : CALL get_proc_batch_sizes(batch_size, nbatch, nex_atom, para_env%num_pe)
150 :
151 80 : DO ibatch = 0, nbatch - 1
152 :
153 80 : bo = get_limit(nex_atom, nbatch, ibatch)
154 92 : DO iex = bo(1), bo(2)
155 :
156 92 : IF (xas_tdp_env%ex_atom_indices(iex) == ri_atom) THEN
157 68 : source = ibatch*batch_size
158 : found = .TRUE. !but simply take the first
159 : EXIT
160 : END IF
161 : END DO !iex
162 0 : IF (found) EXIT
163 : END DO !ip
164 :
165 : ! Broadcast the integrals to all procs (deleted after all donor states for this atoms are treated)
166 68 : lb = 1; IF (do_sf .AND. .NOT. do_sc) lb = 4
167 68 : ub = 2; IF (do_sc) ub = 3
168 68 : IF (do_sf) ub = 4
169 212 : DO i = lb, ub
170 144 : IF (.NOT. ASSOCIATED(xas_tdp_env%ri_fxc(ri_atom, i)%array)) THEN
171 96 : ALLOCATE (xas_tdp_env%ri_fxc(ri_atom, i)%array(nsgfp, nsgfp))
172 : END IF
173 1992692 : CALL para_env%bcast(xas_tdp_env%ri_fxc(ri_atom, i)%array, source)
174 : END DO
175 :
176 68 : xas_tdp_env%fxc_avail = .TRUE.
177 : END IF
178 :
179 : ! Case study on the calculation type
180 70 : IF (do_sg .OR. do_tp) THEN
181 : CALL rcs_xc(xc_ker(1)%matrix, xc_ker(2)%matrix, contr1_int, dist, blk_size, &
182 62 : donor_state, xas_tdp_env, xas_tdp_control, qs_env)
183 : END IF
184 :
185 70 : IF (do_sc) THEN
186 : CALL sc_os_xc(xc_ker(3)%matrix, contr1_int, dist, blk_size, donor_state, &
187 8 : xas_tdp_env, xas_tdp_control, qs_env)
188 : END IF
189 :
190 70 : IF (do_sf) THEN
191 : CALL ondiag_sf_os_xc(xc_ker(4)%matrix, contr1_int, dist, blk_size, donor_state, &
192 0 : xas_tdp_env, xas_tdp_control, qs_env)
193 : END IF
194 :
195 : END IF ! do_xc
196 :
197 : ! Clean-up
198 78 : CALL dbcsr_deallocate_matrix_set(contr1_int)
199 :
200 78 : CALL timestop(handle)
201 :
202 78 : END SUBROUTINE kernel_coulomb_xc
203 :
204 : ! **************************************************************************************************
205 : !> \brief Create the matrix containing the Coulomb kernel, which is:
206 : !> (aI_sigma|J_tau b) ~= (aI_sigma|P) * (P|Q) * (Q|J_tau b)
207 : !> \param coul_ker the Coulomb kernel
208 : !> \param contr1_int the once contracted RI integrals (aI|P)
209 : !> \param dist the inherited dbcsr ditribution
210 : !> \param blk_size the inherited block sizes
211 : !> \param xas_tdp_env ...
212 : !> \param xas_tdp_control ...
213 : !> \param qs_env ...
214 : ! **************************************************************************************************
215 78 : SUBROUTINE coulomb(coul_ker, contr1_int, dist, blk_size, xas_tdp_env, xas_tdp_control, qs_env)
216 :
217 : TYPE(dbcsr_type), INTENT(INOUT) :: coul_ker
218 : TYPE(dbcsr_p_type), DIMENSION(:), POINTER :: contr1_int
219 : TYPE(dbcsr_distribution_type), POINTER :: dist
220 : INTEGER, DIMENSION(:), POINTER :: blk_size
221 : TYPE(xas_tdp_env_type), POINTER :: xas_tdp_env
222 : TYPE(xas_tdp_control_type), POINTER :: xas_tdp_control
223 : TYPE(qs_environment_type), POINTER :: qs_env
224 :
225 : LOGICAL :: quadrants(3)
226 : REAL(dp), DIMENSION(:, :), POINTER :: PQ
227 78 : TYPE(dbcsr_p_type), DIMENSION(:), POINTER :: lhs_int, rhs_int
228 : TYPE(dbcsr_type) :: work_mat
229 :
230 78 : NULLIFY (PQ, rhs_int, lhs_int)
231 :
232 : ! Get the inver RI coulomb
233 78 : PQ => xas_tdp_env%ri_inv_coul
234 :
235 : ! Create a normal type work matrix
236 : CALL dbcsr_create(work_mat, name="WORK", matrix_type=dbcsr_type_no_symmetry, dist=dist, &
237 78 : row_blk_size=blk_size, col_blk_size=blk_size)
238 :
239 : ! Compute the product (aI|P) * (P|Q)^-1 * (Q|Jb) = (aI|Jb)
240 78 : rhs_int => contr1_int ! the incoming contr1_int is not modified
241 336 : ALLOCATE (lhs_int(SIZE(contr1_int)))
242 78 : CALL copy_ri_contr_int(lhs_int, rhs_int) ! RHS containts (Q|JB)^T
243 78 : CALL ri_all_blocks_mm(lhs_int, PQ) ! LHS contatins (aI|P)*(P|Q)^-1
244 :
245 : !In the special case of ROKS, same MOs for each spin => put same (aI|Jb) product on the
246 : !alpha-alpha, alpha-beta and beta-beta quadrants of the kernel matrix.
247 78 : IF (xas_tdp_control%do_roks) THEN
248 2 : quadrants = [.TRUE., .TRUE., .TRUE.]
249 : ELSE
250 76 : quadrants = [.TRUE., .FALSE., .FALSE.]
251 : END IF
252 : CALL ri_int_product(work_mat, lhs_int, rhs_int, quadrants, qs_env, &
253 78 : eps_filter=xas_tdp_control%eps_filter)
254 78 : CALL dbcsr_finalize(work_mat)
255 :
256 : !Create the symmetric kernel matrix and redistribute work_mat into it
257 : CALL dbcsr_create(coul_ker, name="COULOMB KERNEL", matrix_type=dbcsr_type_symmetric, dist=dist, &
258 78 : row_blk_size=blk_size, col_blk_size=blk_size)
259 78 : CALL dbcsr_complete_redistribute(work_mat, coul_ker)
260 :
261 : !clean-up
262 78 : CALL dbcsr_release(work_mat)
263 78 : CALL dbcsr_deallocate_matrix_set(lhs_int)
264 :
265 78 : END SUBROUTINE coulomb
266 :
267 : ! **************************************************************************************************
268 : !> \brief Create the matrix containing the XC kenrel in the spin-conserving open-shell case:
269 : !> (aI_sigma|fxc|J_tau b) ~= (aI_sigma|P) (P|Q)^-1 (Q|fxc|R) (R|S)^-1 (S|J_tau b)
270 : !> \param xc_ker the kernel matrix
271 : !> \param contr1_int_PQ the once contracted RI integrals, with inverse coulomb: (aI_sigma|P) (P|Q)^-1
272 : !> \param dist inherited dbcsr dist
273 : !> \param blk_size inherited block sizes
274 : !> \param donor_state ...
275 : !> \param xas_tdp_env ...
276 : !> \param xas_tdp_control ...
277 : !> \param qs_env ...
278 : !> note Prior to calling this function, the (Q|fxc|R) integral must be brodcasted to all procs
279 : ! **************************************************************************************************
280 8 : SUBROUTINE sc_os_xc(xc_ker, contr1_int_PQ, dist, blk_size, donor_state, xas_tdp_env, &
281 : xas_tdp_control, qs_env)
282 :
283 : TYPE(dbcsr_type), INTENT(INOUT) :: xc_ker
284 : TYPE(dbcsr_p_type), DIMENSION(:), POINTER :: contr1_int_PQ
285 : TYPE(dbcsr_distribution_type), POINTER :: dist
286 : INTEGER, DIMENSION(:), POINTER :: blk_size
287 : TYPE(donor_state_type), POINTER :: donor_state
288 : TYPE(xas_tdp_env_type), POINTER :: xas_tdp_env
289 : TYPE(xas_tdp_control_type), POINTER :: xas_tdp_control
290 : TYPE(qs_environment_type), POINTER :: qs_env
291 :
292 : INTEGER :: ndo_mo, ri_atom
293 : LOGICAL :: quadrants(3)
294 8 : TYPE(dbcsr_p_type), DIMENSION(:), POINTER :: lhs_int, rhs_int
295 : TYPE(dbcsr_type) :: work_mat
296 :
297 8 : NULLIFY (lhs_int, rhs_int)
298 :
299 : ! Initialization
300 8 : ndo_mo = donor_state%ndo_mo
301 8 : ri_atom = donor_state%at_index
302 : !normal type work matrix such that distribution of all spin quadrants match
303 : CALL dbcsr_create(work_mat, name="WORK", matrix_type=dbcsr_type_no_symmetry, dist=dist, &
304 8 : row_blk_size=blk_size, col_blk_size=blk_size)
305 :
306 8 : rhs_int => contr1_int_PQ ! contains [ (aI|P)*(P|Q)^-1 ]^T
307 38 : ALLOCATE (lhs_int(SIZE(contr1_int_PQ))) ! will contain (aI|P)*(P|Q)^-1 * (Q|fxc|R)
308 :
309 : ! Case study: UKS or ROKS ?
310 8 : IF (xas_tdp_control%do_uks) THEN
311 :
312 : ! In the case of UKS, donor MOs might be different for different spins. Moreover, the
313 : ! fxc itself might change since fxc = fxc_sigma,tau
314 : ! => Carfully treat each spin-quadrant separately
315 :
316 : ! alpha-alpha spin quadrant (upper-lefet)
317 6 : quadrants = [.TRUE., .FALSE., .FALSE.]
318 :
319 : ! Copy the alpha part into lhs_int, multiply by the alpha-alpha (Q|fxc|R) and then
320 : ! by the alpha part of rhs_int
321 6 : CALL copy_ri_contr_int(lhs_int(1:ndo_mo), rhs_int(1:ndo_mo))
322 6 : CALL ri_all_blocks_mm(lhs_int(1:ndo_mo), xas_tdp_env%ri_fxc(ri_atom, 1)%array)
323 : CALL ri_int_product(work_mat, lhs_int(1:ndo_mo), rhs_int(1:ndo_mo), quadrants, qs_env, &
324 6 : eps_filter=xas_tdp_control%eps_filter)
325 :
326 : ! alpha-beta spin quadrant (upper-right)
327 6 : quadrants = [.FALSE., .TRUE., .FALSE.]
328 :
329 : !Copy the alpha part into LHS, multiply by the alpha-beta kernel and the beta part of RHS
330 6 : CALL copy_ri_contr_int(lhs_int(1:ndo_mo), rhs_int(1:ndo_mo))
331 6 : CALL ri_all_blocks_mm(lhs_int(1:ndo_mo), xas_tdp_env%ri_fxc(ri_atom, 2)%array)
332 : CALL ri_int_product(work_mat, lhs_int(1:ndo_mo), rhs_int(ndo_mo + 1:2*ndo_mo), &
333 6 : quadrants, qs_env, eps_filter=xas_tdp_control%eps_filter)
334 :
335 : ! beta-beta spin quadrant (lower-right)
336 6 : quadrants = [.FALSE., .FALSE., .TRUE.]
337 :
338 : !Copy the beta part into LHS, multiply by the beta-beta kernel and the beta part of RHS
339 6 : CALL copy_ri_contr_int(lhs_int(ndo_mo + 1:2*ndo_mo), rhs_int(ndo_mo + 1:2*ndo_mo))
340 6 : CALL ri_all_blocks_mm(lhs_int(ndo_mo + 1:2*ndo_mo), xas_tdp_env%ri_fxc(ri_atom, 3)%array)
341 : CALL ri_int_product(work_mat, lhs_int(ndo_mo + 1:2*ndo_mo), rhs_int(ndo_mo + 1:2*ndo_mo), &
342 6 : quadrants, qs_env, eps_filter=xas_tdp_control%eps_filter)
343 :
344 2 : ELSE IF (xas_tdp_control%do_roks) THEN
345 :
346 : ! In the case of ROKS, fxc = fxc_sigma,tau is different for each spin quadrant, but the
347 : ! donor MOs remain the same
348 :
349 : ! alpha-alpha kernel in the upper left quadrant
350 2 : quadrants = [.TRUE., .FALSE., .FALSE.]
351 :
352 : !Copy the LHS and multiply by alpha-alpha kernel
353 2 : CALL copy_ri_contr_int(lhs_int, rhs_int)
354 2 : CALL ri_all_blocks_mm(lhs_int, xas_tdp_env%ri_fxc(ri_atom, 1)%array)
355 : CALL ri_int_product(work_mat, lhs_int, rhs_int, quadrants, qs_env, &
356 2 : eps_filter=xas_tdp_control%eps_filter)
357 :
358 : ! alpha-beta kernel in the upper-right quadrant
359 2 : quadrants = [.FALSE., .TRUE., .FALSE.]
360 :
361 : !Copy LHS and multiply by the alpha-beta kernel
362 2 : CALL copy_ri_contr_int(lhs_int, rhs_int)
363 2 : CALL ri_all_blocks_mm(lhs_int, xas_tdp_env%ri_fxc(ri_atom, 2)%array)
364 : CALL ri_int_product(work_mat, lhs_int, rhs_int, quadrants, qs_env, &
365 2 : eps_filter=xas_tdp_control%eps_filter)
366 :
367 : ! beta-beta kernel in the lower-right quadrant
368 2 : quadrants = [.FALSE., .FALSE., .TRUE.]
369 :
370 : !Copy the LHS and multiply by the beta-beta kernel
371 2 : CALL copy_ri_contr_int(lhs_int, rhs_int)
372 2 : CALL ri_all_blocks_mm(lhs_int, xas_tdp_env%ri_fxc(ri_atom, 3)%array)
373 : CALL ri_int_product(work_mat, lhs_int, rhs_int, quadrants, qs_env, &
374 2 : eps_filter=xas_tdp_control%eps_filter)
375 :
376 : END IF
377 8 : CALL dbcsr_finalize(work_mat)
378 :
379 : ! Create a symmetric kernel matrix and redistribute the normal work matrix into it
380 : CALL dbcsr_create(xc_ker, name="SC OS XC KERNEL", matrix_type=dbcsr_type_symmetric, dist=dist, &
381 8 : row_blk_size=blk_size, col_blk_size=blk_size)
382 8 : CALL dbcsr_complete_redistribute(work_mat, xc_ker)
383 :
384 : !clean-up
385 8 : CALL dbcsr_deallocate_matrix_set(lhs_int)
386 8 : CALL dbcsr_release(work_mat)
387 :
388 8 : END SUBROUTINE sc_os_xc
389 :
390 : ! **************************************************************************************************
391 : !> \brief Create the matrix containing the on-diagonal spin-flip XC kernel (open-shell), which is:
392 : !> (a I_sigma|fxc|J_tau b) * delta_sigma,tau, fxc = 1/(rhoa-rhob) * (dE/drhoa - dE/drhob)
393 : !> with RI: (a I_sigma|fxc|J_tau b) ~= (aI_sigma|P) (P|Q)^-1 (Q|fxc|R) (R|S)^-1 (S|J_tau b)
394 : !> \param xc_ker the kernel matrix
395 : !> \param contr1_int_PQ the once contracted RI integrals with coulomb product: (aI_sigma|P) (P|Q)^-1
396 : !> \param dist inherited dbcsr dist
397 : !> \param blk_size inherited block sizes
398 : !> \param donor_state ...
399 : !> \param xas_tdp_env ...
400 : !> \param xas_tdp_control ...
401 : !> \param qs_env ...
402 : !> \note It must be later on multiplied by the spin-swapped Q projector
403 : !> Prior to calling this function, the (Q|fxc|R) integral must be brodcasted to all procs
404 : ! **************************************************************************************************
405 0 : SUBROUTINE ondiag_sf_os_xc(xc_ker, contr1_int_PQ, dist, blk_size, donor_state, xas_tdp_env, &
406 : xas_tdp_control, qs_env)
407 :
408 : TYPE(dbcsr_type), INTENT(INOUT) :: xc_ker
409 : TYPE(dbcsr_p_type), DIMENSION(:), POINTER :: contr1_int_PQ
410 : TYPE(dbcsr_distribution_type), POINTER :: dist
411 : INTEGER, DIMENSION(:), POINTER :: blk_size
412 : TYPE(donor_state_type), POINTER :: donor_state
413 : TYPE(xas_tdp_env_type), POINTER :: xas_tdp_env
414 : TYPE(xas_tdp_control_type), POINTER :: xas_tdp_control
415 : TYPE(qs_environment_type), POINTER :: qs_env
416 :
417 : INTEGER :: ndo_mo, ri_atom
418 : LOGICAL :: quadrants(3)
419 0 : TYPE(dbcsr_p_type), DIMENSION(:), POINTER :: lhs_int, rhs_int
420 : TYPE(dbcsr_type) :: work_mat
421 :
422 0 : NULLIFY (lhs_int, rhs_int)
423 :
424 : ! Initialization
425 0 : ndo_mo = donor_state%ndo_mo
426 0 : ri_atom = donor_state%at_index
427 : !normal type work matrix such that distribution of all spin quadrants match
428 : CALL dbcsr_create(work_mat, name="WORK", matrix_type=dbcsr_type_no_symmetry, dist=dist, &
429 0 : row_blk_size=blk_size, col_blk_size=blk_size)
430 :
431 : !Create a lhs_int, in which the whole (aI_sigma|P) (P|Q)^-1 (Q|fxc|R) will be put
432 : !because in spin-flip fxc is spin-independent, can take the product once and for all
433 0 : rhs_int => contr1_int_PQ
434 0 : ALLOCATE (lhs_int(SIZE(contr1_int_PQ)))
435 0 : CALL copy_ri_contr_int(lhs_int, rhs_int)
436 0 : CALL ri_all_blocks_mm(lhs_int, xas_tdp_env%ri_fxc(ri_atom, 4)%array)
437 :
438 : ! Case study: UKS or ROKS ?
439 0 : IF (xas_tdp_control%do_uks) THEN
440 :
441 : ! In the case of UKS, donor MOs might be different for different spins
442 : ! => Carfully treat each spin-quadrant separately
443 : ! NO alpha-beta because of the delta_sigma,tau
444 :
445 : ! alpha-alpha spin quadrant (upper-lefet)
446 0 : quadrants = [.TRUE., .FALSE., .FALSE.]
447 : CALL ri_int_product(work_mat, lhs_int(1:ndo_mo), rhs_int(1:ndo_mo), quadrants, qs_env, &
448 0 : eps_filter=xas_tdp_control%eps_filter)
449 :
450 : ! beta-beta spin quadrant (lower-right)
451 0 : quadrants = [.FALSE., .FALSE., .TRUE.]
452 : CALL ri_int_product(work_mat, lhs_int(ndo_mo + 1:2*ndo_mo), rhs_int(ndo_mo + 1:2*ndo_mo), &
453 0 : quadrants, qs_env, eps_filter=xas_tdp_control%eps_filter)
454 :
455 0 : ELSE IF (xas_tdp_control%do_roks) THEN
456 :
457 : ! In the case of ROKS, same donor MOs for both spins => can do it all at once
458 : ! But NOT the alpha-beta quadrant because of delta_sigma,tau
459 :
460 0 : quadrants = [.TRUE., .FALSE., .TRUE.]
461 : CALL ri_int_product(work_mat, lhs_int, rhs_int, quadrants, qs_env, &
462 0 : eps_filter=xas_tdp_control%eps_filter)
463 :
464 : END IF
465 0 : CALL dbcsr_finalize(work_mat)
466 :
467 : ! Create a symmetric kernel matrix and redistribute the normal work matrix into it
468 : CALL dbcsr_create(xc_ker, name="ON-DIAG SF OS XC KERNEL", matrix_type=dbcsr_type_symmetric, &
469 0 : dist=dist, row_blk_size=blk_size, col_blk_size=blk_size)
470 0 : CALL dbcsr_complete_redistribute(work_mat, xc_ker)
471 :
472 : !clean-up
473 0 : CALL dbcsr_deallocate_matrix_set(lhs_int)
474 0 : CALL dbcsr_release(work_mat)
475 :
476 0 : END SUBROUTINE ondiag_sf_os_xc
477 :
478 : ! **************************************************************************************************
479 : !> \brief Create the matrix containing the XC kernel in the restricted closed-shell case, for
480 : !> singlets: (aI|fxc|Jb) = (aI|P) (P|Q)^-1 (Q|fxc_alp,alp+fxc_alp,bet|R) (R|S)^-1 (S|Jb)
481 : !> triplets: (aI|fxc|Jb) = (aI|P) (P|Q)^-1 (Q|fxc_alp,alp-fxc_alp,bet|R) (R|S)^-1 (S|Jb)
482 : !> \param sg_xc_ker the singlet kernel matrix
483 : !> \param tp_xc_ker the triplet kernel matrix
484 : !> \param contr1_int_PQ the once contracted RI integrals including inverse 2-center Coulomb prodcut:
485 : !> (aI|P)*(P|Q)^-1
486 : !> \param dist inherited dbcsr dist
487 : !> \param blk_size inherited block sizes
488 : !> \param donor_state ...
489 : !> \param xas_tdp_env ...
490 : !> \param xas_tdp_control ...
491 : !> \param qs_env ...
492 : ! **************************************************************************************************
493 62 : SUBROUTINE rcs_xc(sg_xc_ker, tp_xc_ker, contr1_int_PQ, dist, blk_size, donor_state, &
494 : xas_tdp_env, xas_tdp_control, qs_env)
495 :
496 : TYPE(dbcsr_type), INTENT(INOUT) :: sg_xc_ker, tp_xc_ker
497 : TYPE(dbcsr_p_type), DIMENSION(:), POINTER :: contr1_int_PQ
498 : TYPE(dbcsr_distribution_type), POINTER :: dist
499 : INTEGER, DIMENSION(:), POINTER :: blk_size
500 : TYPE(donor_state_type), POINTER :: donor_state
501 : TYPE(xas_tdp_env_type), POINTER :: xas_tdp_env
502 : TYPE(xas_tdp_control_type), POINTER :: xas_tdp_control
503 : TYPE(qs_environment_type), POINTER :: qs_env
504 :
505 : INTEGER :: nsgfp, ri_atom
506 : LOGICAL :: quadrants(3)
507 : REAL(dp), ALLOCATABLE, DIMENSION(:, :) :: fxc
508 62 : TYPE(dbcsr_p_type), DIMENSION(:), POINTER :: lhs_int, rhs_int
509 : TYPE(dbcsr_type) :: work_mat
510 :
511 62 : NULLIFY (lhs_int, rhs_int)
512 :
513 : ! Initialization
514 62 : ri_atom = donor_state%at_index
515 62 : nsgfp = SIZE(xas_tdp_env%ri_fxc(ri_atom, 1)%array, 1)
516 62 : rhs_int => contr1_int_PQ ! RHS contains [ (aI|P)*(P|Q)^-1 ]^T
517 248 : ALLOCATE (lhs_int(SIZE(contr1_int_PQ))) ! LHS will contatin (aI|P)*(P|Q)^-1 * (Q|fxc|R)
518 :
519 : ! Work structures
520 248 : ALLOCATE (fxc(nsgfp, nsgfp))
521 : CALL dbcsr_create(work_mat, name="WORK", matrix_type=dbcsr_type_no_symmetry, dist=dist, &
522 62 : row_blk_size=blk_size, col_blk_size=blk_size)
523 :
524 : ! Case study: singlet and/or triplet ?
525 62 : IF (xas_tdp_control%do_singlet) THEN
526 :
527 : ! Take the sum of fxc for alpha-alpha and alpha-beta
528 62 : CALL dcopy(nsgfp*nsgfp, xas_tdp_env%ri_fxc(ri_atom, 1)%array, 1, fxc, 1)
529 62 : CALL daxpy(nsgfp*nsgfp, 1.0_dp, xas_tdp_env%ri_fxc(ri_atom, 2)%array, 1, fxc, 1)
530 :
531 : ! Copy the fresh lhs_int = (aI|P) (P|Q)^-1 and multiply by (Q|fxc|R)
532 62 : CALL copy_ri_contr_int(lhs_int, rhs_int)
533 62 : CALL ri_all_blocks_mm(lhs_int, fxc)
534 :
535 : ! Compute the final LHS RHS product => spin-restricted, only upper-left quadrant
536 62 : quadrants = [.TRUE., .FALSE., .FALSE.]
537 : CALL ri_int_product(work_mat, lhs_int, rhs_int, quadrants, qs_env, &
538 62 : eps_filter=xas_tdp_control%eps_filter)
539 62 : CALL dbcsr_finalize(work_mat)
540 :
541 : !Create the symmetric kernel matrix and redistribute work_mat into it
542 : CALL dbcsr_create(sg_xc_ker, name="XC SINGLET KERNEL", matrix_type=dbcsr_type_symmetric, &
543 62 : dist=dist, row_blk_size=blk_size, col_blk_size=blk_size)
544 62 : CALL dbcsr_complete_redistribute(work_mat, sg_xc_ker)
545 :
546 : END IF
547 :
548 62 : IF (xas_tdp_control%do_triplet) THEN
549 :
550 : ! Take the difference of fxc for alpha-alpha and alpha-beta
551 0 : CALL dcopy(nsgfp*nsgfp, xas_tdp_env%ri_fxc(ri_atom, 1)%array, 1, fxc, 1)
552 0 : CALL daxpy(nsgfp*nsgfp, -1.0_dp, xas_tdp_env%ri_fxc(ri_atom, 2)%array, 1, fxc, 1)
553 :
554 : ! Copy the fresh lhs_int = (aI|P) (P|Q)^-1 and multiply by (Q|fxc|R)
555 0 : CALL copy_ri_contr_int(lhs_int, rhs_int)
556 0 : CALL ri_all_blocks_mm(lhs_int, fxc)
557 :
558 : ! Compute the final LHS RHS product => spin-restricted, only upper-left quadrant
559 0 : quadrants = [.TRUE., .FALSE., .FALSE.]
560 : CALL ri_int_product(work_mat, lhs_int, rhs_int, quadrants, qs_env, &
561 0 : eps_filter=xas_tdp_control%eps_filter)
562 0 : CALL dbcsr_finalize(work_mat)
563 :
564 : !Create the symmetric kernel matrix and redistribute work_mat into it
565 : CALL dbcsr_create(tp_xc_ker, name="XC TRIPLET KERNEL", matrix_type=dbcsr_type_symmetric, &
566 0 : dist=dist, row_blk_size=blk_size, col_blk_size=blk_size)
567 0 : CALL dbcsr_complete_redistribute(work_mat, tp_xc_ker)
568 :
569 : END IF
570 :
571 : ! clean-up
572 62 : CALL dbcsr_deallocate_matrix_set(lhs_int)
573 62 : CALL dbcsr_release(work_mat)
574 62 : DEALLOCATE (fxc)
575 :
576 62 : END SUBROUTINE rcs_xc
577 :
578 : ! **************************************************************************************************
579 : !> \brief Computes the exact exchange kernel matrix using RI. Returns an array of 2 matrices,
580 : !> which are:
581 : !> 1) the on-diagonal kernel: (ab|I_sigma J_tau) * delta_sigma,tau
582 : !> 2) the off-diagonal spin-conserving kernel: (aJ_sigma|I_tau b) * delta_sigma,tau
583 : !> An internal analysis determines which of the above are computed (can range from 0 to 2),
584 : !> \param ex_ker ...
585 : !> \param donor_state ...
586 : !> \param xas_tdp_env ...
587 : !> \param xas_tdp_control ...
588 : !> \param qs_env ...
589 : !> \note In the case of spin-conserving excitation, the kernel must later be multiplied by the
590 : !> usual Q projector. In the case of spin-flip, one needs to project the excitations coming
591 : !> from alpha donor MOs on the unoccupied beta MOs. This is done by multiplying by a Q
592 : !> projector where the alpha-alpha and beta-beta quadrants are swapped
593 : !> The ex_ker array should be allocated on entry (not the internals)
594 : ! **************************************************************************************************
595 132 : SUBROUTINE kernel_exchange(ex_ker, donor_state, xas_tdp_env, xas_tdp_control, qs_env)
596 :
597 : TYPE(dbcsr_p_type), DIMENSION(:), POINTER :: ex_ker
598 : TYPE(donor_state_type), POINTER :: donor_state
599 : TYPE(xas_tdp_env_type), POINTER :: xas_tdp_env
600 : TYPE(xas_tdp_control_type), POINTER :: xas_tdp_control
601 : TYPE(qs_environment_type), POINTER :: qs_env
602 :
603 : CHARACTER(len=*), PARAMETER :: routineN = 'kernel_exchange'
604 :
605 : INTEGER :: handle
606 78 : INTEGER, DIMENSION(:), POINTER :: blk_size
607 : LOGICAL :: do_off_sc
608 : TYPE(dbcsr_distribution_type), POINTER :: dist
609 78 : TYPE(dbcsr_p_type), DIMENSION(:), POINTER :: contr1_int
610 :
611 78 : NULLIFY (contr1_int, dist, blk_size)
612 :
613 : !Don't do anything if no hfx
614 24 : IF (.NOT. xas_tdp_control%do_hfx) RETURN
615 :
616 54 : CALL timeset(routineN, handle)
617 :
618 54 : dist => donor_state%dbcsr_dist
619 54 : blk_size => donor_state%blk_size
620 :
621 : !compute the off-diag spin-conserving only if not TDA and anything that is spin-conserving
622 : do_off_sc = (.NOT. xas_tdp_control%tamm_dancoff) .AND. &
623 54 : (xas_tdp_control%do_spin_cons .OR. xas_tdp_control%do_singlet .OR. xas_tdp_control%do_triplet)
624 :
625 : ! Need the once contracted integrals (aI|P)
626 54 : CALL contract2_AO_to_doMO(contr1_int, "EXCHANGE", donor_state, xas_tdp_env, xas_tdp_control, qs_env)
627 :
628 : ! The on-diagonal exchange : (ab|P) * (P|Q)^-1 * (Q|I_sigma J_tau) * delta_sigma,tau
629 : CALL ondiag_ex(ex_ker(1)%matrix, contr1_int, dist, blk_size, donor_state, xas_tdp_env, &
630 54 : xas_tdp_control, qs_env)
631 :
632 : ! The off-diag spin-conserving case: (aJ_sigma|P) * (P|Q)^-1 * (Q|I_tau b) * delta_sigma,tau
633 54 : IF (do_off_sc) THEN
634 : CALL offdiag_ex_sc(ex_ker(2)%matrix, contr1_int, dist, blk_size, donor_state, &
635 6 : xas_tdp_env, xas_tdp_control, qs_env)
636 : END IF
637 :
638 : !clean-up
639 54 : CALL dbcsr_deallocate_matrix_set(contr1_int)
640 :
641 54 : CALL timestop(handle)
642 :
643 78 : END SUBROUTINE kernel_exchange
644 :
645 : ! **************************************************************************************************
646 : !> \brief Create the matrix containing the on-diagonal exact exchange kernel, which is:
647 : !> (ab|I_sigma J_tau) * delta_sigma,tau, where a,b are AOs, I_sigma and J_tau are the donor
648 : !> spin-orbitals. A RI is done: (ab|I_sigma J_tau) = (ab|P) * (P|Q)^-1 * (Q|I_sigma J_tau)
649 : !> \param ondiag_ex_ker the on-diagonal exchange kernel in dbcsr format
650 : !> \param contr1_int the already once-contracted RI 3-center integrals (aI_sigma|P)
651 : !> where each matrix of the array contains the contraction for the donor spin-orbital I_sigma
652 : !> \param dist the inherited dbcsr distribution
653 : !> \param blk_size the inherited dbcsr block sizes
654 : !> \param donor_state ...
655 : !> \param xas_tdp_env ...
656 : !> \param xas_tdp_control ...
657 : !> \param qs_env ...
658 : !> \note In the presence of a RI metric, we have instead M^-1 * (P|Q) * M^-1
659 : ! **************************************************************************************************
660 54 : SUBROUTINE ondiag_ex(ondiag_ex_ker, contr1_int, dist, blk_size, donor_state, xas_tdp_env, &
661 : xas_tdp_control, qs_env)
662 :
663 : TYPE(dbcsr_type), INTENT(INOUT) :: ondiag_ex_ker
664 : TYPE(dbcsr_p_type), DIMENSION(:), POINTER :: contr1_int
665 : TYPE(dbcsr_distribution_type), POINTER :: dist
666 : INTEGER, DIMENSION(:), POINTER :: blk_size
667 : TYPE(donor_state_type), POINTER :: donor_state
668 : TYPE(xas_tdp_env_type), POINTER :: xas_tdp_env
669 : TYPE(xas_tdp_control_type), POINTER :: xas_tdp_control
670 : TYPE(qs_environment_type), POINTER :: qs_env
671 :
672 : INTEGER :: group, iblk, iso, jblk, jso, nblk, &
673 : ndo_mo, ndo_so, nsgfa, nsgfp, ri_atom, &
674 : source
675 54 : INTEGER, DIMENSION(:), POINTER :: col_dist, col_dist_work, row_dist, &
676 54 : row_dist_work
677 54 : INTEGER, DIMENSION(:, :), POINTER :: pgrid
678 : LOGICAL :: do_roks, do_uks, found
679 54 : REAL(dp), ALLOCATABLE, DIMENSION(:, :) :: coeffs, ri_coeffs
680 54 : REAL(dp), DIMENSION(:, :), POINTER :: aIQ, pblock, PQ
681 : TYPE(dbcsr_distribution_type) :: opt_dbcsr_dist, work_dbcsr_dist
682 : TYPE(dbcsr_iterator_type) :: iter
683 54 : TYPE(dbcsr_p_type), DIMENSION(:), POINTER :: matrix_s
684 : TYPE(dbcsr_type) :: abIJ, mats_desymm, work_mat
685 : TYPE(mp_para_env_type), POINTER :: para_env
686 :
687 54 : NULLIFY (para_env, matrix_s, pblock, aIQ, row_dist, col_dist, row_dist_work, col_dist_work, pgrid)
688 :
689 : ! We want to compute (ab|I_sigma J_tau) = (ab|P) * (P|Q)^-1 * (Q|I_sigma J_tau)
690 : ! Already have (cJ_tau|P) stored in contr1_int. Need to further contract the
691 : ! AOs with the coeff of the I_alpha spin-orbital.
692 :
693 : ! Initialization
694 54 : ndo_mo = donor_state%ndo_mo
695 54 : ri_atom = donor_state%at_index
696 54 : do_roks = xas_tdp_control%do_roks
697 54 : do_uks = xas_tdp_control%do_uks
698 6 : ndo_so = ndo_mo; IF (do_uks) ndo_so = 2*ndo_mo !if not UKS, same donor MOs for both spins
699 54 : PQ => xas_tdp_env%ri_inv_ex
700 :
701 54 : CALL get_qs_env(qs_env, para_env=para_env, matrix_s=matrix_s, natom=nblk)
702 54 : nsgfp = SIZE(PQ, 1)
703 54 : nsgfa = SIZE(donor_state%contract_coeffs, 1)
704 324 : ALLOCATE (coeffs(nsgfp, ndo_so), ri_coeffs(nsgfp, ndo_so))
705 :
706 : ! a and b need to overlap for non-zero (ab|IJ) => same block structure as overlap S
707 : ! need compatible distribution_2d with 3c tensor + normal type
708 54 : CALL cp_dbcsr_dist2d_to_dist(xas_tdp_env%opt_dist2d_ex, opt_dbcsr_dist)
709 :
710 54 : CALL dbcsr_desymmetrize(matrix_s(1)%matrix, mats_desymm)
711 :
712 54 : CALL dbcsr_create(abIJ, template=mats_desymm, name="(ab|IJ)", dist=opt_dbcsr_dist)
713 54 : CALL dbcsr_complete_redistribute(mats_desymm, abIJ)
714 :
715 54 : CALL dbcsr_release(mats_desymm)
716 :
717 : ! Create a work distribution based on opt_dbcsr_dist, but for full size matrices
718 : CALL dbcsr_distribution_get(opt_dbcsr_dist, row_dist=row_dist, col_dist=col_dist, group=group, &
719 54 : pgrid=pgrid)
720 :
721 162 : ALLOCATE (row_dist_work(ndo_so*nblk))
722 108 : ALLOCATE (col_dist_work(ndo_so*nblk))
723 126 : DO iso = 1, ndo_so
724 632 : row_dist_work((iso - 1)*nblk + 1:iso*nblk) = row_dist(:)
725 686 : col_dist_work((iso - 1)*nblk + 1:iso*nblk) = col_dist(:)
726 : END DO
727 :
728 : CALL dbcsr_distribution_new(work_dbcsr_dist, group=group, pgrid=pgrid, row_dist=row_dist_work, &
729 54 : col_dist=col_dist_work)
730 :
731 : CALL dbcsr_create(work_mat, name="WORK", matrix_type=dbcsr_type_no_symmetry, dist=work_dbcsr_dist, &
732 54 : row_blk_size=blk_size, col_blk_size=blk_size)
733 :
734 : ! Loop over donor spin-orbitals. End matrix is symmetric => span only upper half
735 126 : DO iso = 1, ndo_so
736 :
737 : ! take the (aI|Q) block in contr1_int that has a centered on the excited atom
738 72 : CALL dbcsr_get_stored_coordinates(contr1_int(iso)%matrix, ri_atom, ri_atom, source)
739 72 : IF (para_env%mepos == source) THEN
740 36 : CALL dbcsr_get_block_p(contr1_int(iso)%matrix, ri_atom, ri_atom, aIQ, found)
741 : ELSE
742 144 : ALLOCATE (aIQ(nsgfa, nsgfp))
743 : END IF
744 188164 : CALL para_env%bcast(aIQ, source)
745 :
746 : ! get the contraction (Q|IJ) by taking (Q|Ia)*contract_coeffs and put it in coeffs
747 : CALL dgemm('T', 'N', nsgfp, ndo_so, nsgfa, 1.0_dp, aIQ, nsgfa, donor_state%contract_coeffs, &
748 72 : nsgfa, 0.0_dp, coeffs, nsgfp)
749 :
750 : ! take (P|Q)^-1 * (Q|IJ) and put that in ri_coeffs
751 : CALL dgemm('N', 'N', nsgfp, ndo_so, nsgfp, 1.0_dp, PQ, nsgfp, coeffs, nsgfp, 0.0_dp, &
752 72 : ri_coeffs, nsgfp)
753 :
754 72 : IF (.NOT. para_env%mepos == source) DEALLOCATE (aIQ)
755 :
756 310 : DO jso = iso, ndo_so
757 :
758 : ! There is no alpha-beta exchange. In case of UKS, iso,jso span all spin-orbitals
759 : ! => CYCLE if iso and jso are indexing MOs with different spin (and we have UKS)
760 112 : IF (do_uks .AND. (iso <= ndo_mo .AND. jso > ndo_mo)) CYCLE
761 :
762 : ! compute (ab|IJ) = sum_P (ab|P) * (P|Q)^-1 * (Q|IJ)
763 90 : CALL dbcsr_set(abIJ, 0.0_dp)
764 90 : CALL contract3_RI_to_doMOs(xas_tdp_env%ri_3c_ex, ri_coeffs(:, jso), abIJ, ri_atom)
765 :
766 : ! Loop over (ab|IJ) and copy into work. OK because dist are made to match
767 90 : CALL dbcsr_iterator_start(iter, abIJ)
768 685 : DO WHILE (dbcsr_iterator_blocks_left(iter))
769 :
770 595 : CALL dbcsr_iterator_next_block(iter, row=iblk, column=jblk)
771 595 : IF (iso == jso .AND. jblk < iblk) CYCLE
772 :
773 363 : CALL dbcsr_get_block_p(abIJ, iblk, jblk, pblock, found)
774 :
775 453 : IF (found) THEN
776 363 : CALL dbcsr_put_block(work_mat, (iso - 1)*nblk + iblk, (jso - 1)*nblk + jblk, pblock)
777 :
778 : !In case of ROKS, we have (ab|IJ) for alpha-alpha spin, but it is the same for
779 : !beta-beta => replicate the blocks (alpha-beta is zero)
780 363 : IF (do_roks) THEN
781 : !the beta-beta block
782 : CALL dbcsr_put_block(work_mat, (ndo_so + iso - 1)*nblk + iblk, &
783 0 : (ndo_so + jso - 1)*nblk + jblk, pblock)
784 : END IF
785 : END IF
786 :
787 : END DO !iterator
788 274 : CALL dbcsr_iterator_stop(iter)
789 :
790 : END DO !jso
791 : END DO !iso
792 :
793 54 : CALL dbcsr_finalize(work_mat)
794 : CALL dbcsr_create(ondiag_ex_ker, name="ONDIAG EX KERNEL", matrix_type=dbcsr_type_symmetric, &
795 54 : dist=dist, row_blk_size=blk_size, col_blk_size=blk_size)
796 54 : CALL dbcsr_complete_redistribute(work_mat, ondiag_ex_ker)
797 :
798 : !Clean-up
799 54 : CALL dbcsr_release(work_mat)
800 54 : CALL dbcsr_release(abIJ)
801 54 : CALL dbcsr_distribution_release(opt_dbcsr_dist)
802 54 : CALL dbcsr_distribution_release(work_dbcsr_dist)
803 54 : DEALLOCATE (col_dist_work, row_dist_work)
804 :
805 270 : END SUBROUTINE ondiag_ex
806 :
807 : ! **************************************************************************************************
808 : !> \brief Create the matrix containing the off-diagonal exact exchange kernel in the spin-conserving
809 : !> case (which also includes excitations from the closed=shell ref state ) This matrix reads:
810 : !> (aJ_sigma|I_tau b) * delta_sigma,tau , where a, b are AOs and J_sigma, I_tau are the donor
811 : !> spin-orbital. A RI is done: (aJ_sigma|I_tau b) = (aJ_sigma|P) * (P|Q)^-1 * (Q|I_tau b)
812 : !> \param offdiag_ex_ker the off-diagonal, spin-conserving exchange kernel in dbcsr format
813 : !> \param contr1_int the once-contracted RI integrals: (aJ_sigma|P)
814 : !> \param dist the inherited dbcsr ditribution
815 : !> \param blk_size the inherited block sizes
816 : !> \param donor_state ...
817 : !> \param xas_tdp_env ...
818 : !> \param xas_tdp_control ...
819 : !> \param qs_env ...
820 : ! **************************************************************************************************
821 6 : SUBROUTINE offdiag_ex_sc(offdiag_ex_ker, contr1_int, dist, blk_size, donor_state, xas_tdp_env, &
822 : xas_tdp_control, qs_env)
823 :
824 : TYPE(dbcsr_type), INTENT(INOUT) :: offdiag_ex_ker
825 : TYPE(dbcsr_p_type), DIMENSION(:), POINTER :: contr1_int
826 : TYPE(dbcsr_distribution_type), POINTER :: dist
827 : INTEGER, DIMENSION(:), POINTER :: blk_size
828 : TYPE(donor_state_type), POINTER :: donor_state
829 : TYPE(xas_tdp_env_type), POINTER :: xas_tdp_env
830 : TYPE(xas_tdp_control_type), POINTER :: xas_tdp_control
831 : TYPE(qs_environment_type), POINTER :: qs_env
832 :
833 : INTEGER :: ndo_mo
834 : LOGICAL :: do_roks, do_uks, quadrants(3)
835 : REAL(dp), DIMENSION(:, :), POINTER :: PQ
836 6 : TYPE(dbcsr_p_type), DIMENSION(:), POINTER :: lhs_int, rhs_int
837 : TYPE(dbcsr_type) :: work_mat
838 :
839 6 : NULLIFY (PQ, lhs_int, rhs_int)
840 :
841 : !Initialization
842 6 : ndo_mo = donor_state%ndo_mo
843 6 : do_roks = xas_tdp_control%do_roks
844 6 : do_uks = xas_tdp_control%do_uks
845 6 : PQ => xas_tdp_env%ri_inv_ex
846 :
847 6 : rhs_int => contr1_int
848 24 : ALLOCATE (lhs_int(SIZE(contr1_int)))
849 6 : CALL copy_ri_contr_int(lhs_int, rhs_int)
850 6 : CALL ri_all_blocks_mm(lhs_int, PQ)
851 :
852 : !Given the lhs_int and rhs_int, all we need to do is multiply elements from the former by
853 : !the transpose of the later, and put the result in the correct spin quadrants
854 :
855 : !Create a normal type work matrix
856 : CALL dbcsr_create(work_mat, name="WORK", matrix_type=dbcsr_type_no_symmetry, dist=dist, &
857 6 : row_blk_size=blk_size, col_blk_size=blk_size)
858 :
859 : !Case study on closed-shell, ROKS or UKS
860 6 : IF (do_roks) THEN
861 : !In ROKS, the donor MOs for each spin are the same => copy the product in both the
862 : !alpha-alpha and the beta-beta quadrants
863 0 : quadrants = [.TRUE., .FALSE., .TRUE.]
864 : CALL ri_int_product(work_mat, lhs_int, rhs_int, quadrants, qs_env, &
865 0 : eps_filter=xas_tdp_control%eps_filter, mo_transpose=.TRUE.)
866 :
867 6 : ELSE IF (do_uks) THEN
868 : !In UKS, the donor MOs are possibly different for each spin => start with the
869 : !alpha-alpha product and the perform the beta-beta product separately
870 0 : quadrants = [.TRUE., .FALSE., .FALSE.]
871 : CALL ri_int_product(work_mat, lhs_int(1:ndo_mo), rhs_int(1:ndo_mo), quadrants, &
872 0 : qs_env, eps_filter=xas_tdp_control%eps_filter, mo_transpose=.TRUE.)
873 :
874 0 : quadrants = [.FALSE., .FALSE., .TRUE.]
875 : CALL ri_int_product(work_mat, lhs_int(ndo_mo + 1:2*ndo_mo), rhs_int(ndo_mo + 1:2*ndo_mo), &
876 0 : quadrants, qs_env, eps_filter=xas_tdp_control%eps_filter, mo_transpose=.TRUE.)
877 : ELSE
878 : !In the restricted closed-shell case, only have one spin and a single qudarant
879 6 : quadrants = [.TRUE., .FALSE., .FALSE.]
880 : CALL ri_int_product(work_mat, lhs_int, rhs_int, quadrants, qs_env, &
881 6 : eps_filter=xas_tdp_control%eps_filter, mo_transpose=.TRUE.)
882 : END IF
883 6 : CALL dbcsr_finalize(work_mat)
884 :
885 : !Create the symmetric kernel matrix and redistribute work_mat into it
886 : CALL dbcsr_create(offdiag_ex_ker, name="OFFDIAG EX KERNEL", matrix_type=dbcsr_type_symmetric, &
887 6 : dist=dist, row_blk_size=blk_size, col_blk_size=blk_size)
888 6 : CALL dbcsr_complete_redistribute(work_mat, offdiag_ex_ker)
889 :
890 : !clean-up
891 6 : CALL dbcsr_release(work_mat)
892 6 : CALL dbcsr_deallocate_matrix_set(lhs_int)
893 :
894 6 : END SUBROUTINE offdiag_ex_sc
895 :
896 : ! **************************************************************************************************
897 : !> \brief Reserves the blocks in of a dbcsr matrix as needed for RI 3-center contraction (aI|P)
898 : !> \param matrices the matrices for which blocks are reserved
899 : !> \param ri_atom the index of the atom on which RI is done (= all coeffs of I are there, and P too)
900 : !> \param qs_env ...
901 : !> \note the end product are normal type matrices that are possibly slightly spraser as matrix_s
902 : ! **************************************************************************************************
903 166 : SUBROUTINE reserve_contraction_blocks(matrices, ri_atom, qs_env)
904 :
905 : TYPE(dbcsr_p_type), DIMENSION(:), POINTER :: matrices
906 : INTEGER, INTENT(IN) :: ri_atom
907 : TYPE(qs_environment_type), POINTER :: qs_env
908 :
909 : INTEGER :: i, iblk, jblk, max_nblks, nblks
910 166 : INTEGER, ALLOCATABLE, DIMENSION(:) :: reserve_cols, reserve_rows
911 : TYPE(dbcsr_distribution_type) :: dist
912 : TYPE(dbcsr_iterator_type) :: iter
913 166 : TYPE(dbcsr_p_type), DIMENSION(:), POINTER :: matrix_s
914 : TYPE(dbcsr_type) :: template, work
915 :
916 166 : NULLIFY (matrix_s)
917 :
918 : ! Initialization
919 166 : CALL get_qs_env(qs_env, matrix_s=matrix_s)
920 166 : CALL dbcsr_get_info(matrices(1)%matrix, distribution=dist)
921 :
922 : ! Need to redistribute matrix_s in the distribution of matrices
923 166 : CALL dbcsr_create(work, template=matrix_s(1)%matrix, dist=dist)
924 166 : CALL dbcsr_complete_redistribute(matrix_s(1)%matrix, work)
925 :
926 : ! Need to desymmetrize matrix as as a template
927 166 : CALL dbcsr_desymmetrize(work, template)
928 :
929 : ! Allocate space for block indicies to reserve.
930 166 : max_nblks = dbcsr_get_num_blocks(template)
931 648 : ALLOCATE (reserve_rows(max_nblks), reserve_cols(max_nblks))
932 :
933 : ! Loop over matrix_s as need a,b to overlap
934 166 : nblks = 0
935 166 : CALL dbcsr_iterator_start(iter, template)
936 19522 : DO WHILE (dbcsr_iterator_blocks_left(iter))
937 19356 : CALL dbcsr_iterator_next_block(iter, row=iblk, column=jblk)
938 : !only have a,b pair if one of them is the ri_atom
939 19522 : IF (iblk == ri_atom .OR. jblk == ri_atom) THEN
940 875 : nblks = nblks + 1
941 875 : reserve_rows(nblks) = iblk
942 875 : reserve_cols(nblks) = jblk
943 : END IF
944 : END DO
945 166 : CALL dbcsr_iterator_stop(iter)
946 :
947 498 : DO i = 1, SIZE(matrices)
948 498 : CALL dbcsr_reserve_blocks(matrices(i)%matrix, rows=reserve_rows(1:nblks), cols=reserve_cols(1:nblks))
949 : END DO
950 :
951 : ! Clean-up
952 166 : CALL dbcsr_release(template)
953 166 : CALL dbcsr_release(work)
954 :
955 498 : END SUBROUTINE reserve_contraction_blocks
956 :
957 : ! **************************************************************************************************
958 : !> \brief Contract the ri 3-center integrals stored in a tensor with repect to the donor MOs coeffs,
959 : !> for a given excited atom k => (aI|k) = sum_b c_Ib (ab|k)
960 : !> \param contr_int the contracted integrals as array of dbcsr matrices
961 : !> \param op_type for which operator type we contract (COULOMB or EXCHANGE)
962 : !> \param donor_state ...
963 : !> \param xas_tdp_env ...
964 : !> \param xas_tdp_control ...
965 : !> \param qs_env ...
966 : !> \note In the output matrices, (aI_b|k) is stored at block a,b where I_b is the partial
967 : !> contraction that only includes coeffs from atom b. Note that the contracted matrix is
968 : !> not symmetric. To get the fully contracted matrix over b, one need to add the block
969 : !> columns of (aI_b|k) (get an array of size nao*nsgfp). This step is unnessary in our case
970 : !> because we assume locality of donor state, and only one column od (aI_b|k) is pouplated
971 : ! **************************************************************************************************
972 166 : SUBROUTINE contract2_AO_to_doMO(contr_int, op_type, donor_state, xas_tdp_env, xas_tdp_control, qs_env)
973 :
974 : TYPE(dbcsr_p_type), DIMENSION(:), POINTER :: contr_int
975 : CHARACTER(len=*), INTENT(IN) :: op_type
976 : TYPE(donor_state_type), POINTER :: donor_state
977 : TYPE(xas_tdp_env_type), POINTER :: xas_tdp_env
978 : TYPE(xas_tdp_control_type), POINTER :: xas_tdp_control
979 : TYPE(qs_environment_type), POINTER :: qs_env
980 :
981 : CHARACTER(len=*), PARAMETER :: routineN = 'contract2_AO_to_doMO'
982 :
983 : INTEGER :: handle, i, imo, ispin, katom, kkind, &
984 : natom, ndo_mo, ndo_so, nkind, nspins
985 166 : INTEGER, DIMENSION(:), POINTER :: ri_blk_size, std_blk_size
986 : LOGICAL :: do_uks
987 166 : REAL(dp), DIMENSION(:, :), POINTER :: coeffs
988 : TYPE(dbcsr_distribution_type) :: opt_dbcsr_dist
989 166 : TYPE(dbcsr_p_type), DIMENSION(:), POINTER :: matrices, matrix_s
990 : TYPE(dbcsr_type), POINTER :: aI_P, P_Ib, work
991 : TYPE(dbt_type), POINTER :: pq_X
992 : TYPE(distribution_2d_type), POINTER :: opt_dist2d
993 166 : TYPE(gto_basis_set_p_type), DIMENSION(:), POINTER :: ri_basis
994 : TYPE(mp_para_env_type), POINTER :: para_env
995 166 : TYPE(particle_type), DIMENSION(:), POINTER :: particle_set
996 166 : TYPE(qs_kind_type), DIMENSION(:), POINTER :: qs_kind_set
997 :
998 166 : NULLIFY (matrix_s, std_blk_size, ri_blk_size, qs_kind_set, ri_basis, pq_X)
999 166 : NULLIFY (aI_P, P_Ib, work, matrices, coeffs, opt_dist2d, particle_set)
1000 :
1001 166 : CALL timeset(routineN, handle)
1002 :
1003 : ! Initialization
1004 166 : CALL get_qs_env(qs_env, natom=natom, matrix_s=matrix_s, qs_kind_set=qs_kind_set, para_env=para_env)
1005 166 : ndo_mo = donor_state%ndo_mo
1006 166 : kkind = donor_state%kind_index
1007 166 : katom = donor_state%at_index
1008 : !by default contract for Coulomb
1009 166 : pq_X => xas_tdp_env%ri_3c_coul
1010 166 : opt_dist2d => xas_tdp_env%opt_dist2d_coul
1011 166 : IF (op_type == "EXCHANGE") THEN
1012 88 : CPASSERT(ASSOCIATED(xas_tdp_env%ri_3c_ex))
1013 88 : pq_X => xas_tdp_env%ri_3c_ex
1014 88 : opt_dist2d => xas_tdp_env%opt_dist2d_ex
1015 : END IF
1016 166 : do_uks = xas_tdp_control%do_uks
1017 166 : nspins = 1; IF (do_uks) nspins = 2
1018 166 : ndo_so = nspins*ndo_mo
1019 :
1020 : ! contracted integrals block sizes
1021 166 : CALL dbcsr_get_info(matrix_s(1)%matrix, col_blk_size=std_blk_size)
1022 : ! getting the block dimensions for the RI basis
1023 166 : CALL get_qs_env(qs_env, particle_set=particle_set, nkind=nkind)
1024 1110 : ALLOCATE (ri_basis(nkind), ri_blk_size(natom))
1025 166 : CALL basis_set_list_setup(ri_basis, "RI_XAS", qs_kind_set)
1026 166 : CALL get_particle_set(particle_set, qs_kind_set, nsgf=ri_blk_size, basis=ri_basis)
1027 :
1028 : ! Create work matrices. Everything that goes into a 3c routine must be compatible with the optimal dist_2d
1029 166 : CALL cp_dbcsr_dist2d_to_dist(opt_dist2d, opt_dbcsr_dist)
1030 :
1031 498 : ALLOCATE (aI_P, P_Ib, work, matrices(2))
1032 : CALL dbcsr_create(aI_P, dist=opt_dbcsr_dist, matrix_type=dbcsr_type_no_symmetry, name="(aI|P)", &
1033 166 : row_blk_size=std_blk_size, col_blk_size=ri_blk_size)
1034 :
1035 : CALL dbcsr_create(P_Ib, dist=opt_dbcsr_dist, matrix_type=dbcsr_type_no_symmetry, name="(P|Ib)", &
1036 166 : row_blk_size=ri_blk_size, col_blk_size=std_blk_size)
1037 :
1038 : !reserve the blocks (needed for 3c contraction routines)
1039 166 : matrices(1)%matrix => aI_P; matrices(2)%matrix => P_Ib
1040 166 : CALL reserve_contraction_blocks(matrices, katom, qs_env)
1041 166 : DEALLOCATE (matrices)
1042 :
1043 : ! Create the contracted integral matrices
1044 722 : ALLOCATE (contr_int(ndo_so))
1045 390 : DO i = 1, ndo_so
1046 224 : ALLOCATE (contr_int(i)%matrix)
1047 : CALL dbcsr_create(matrix=contr_int(i)%matrix, template=matrix_s(1)%matrix, &
1048 : matrix_type=dbcsr_type_no_symmetry, row_blk_size=std_blk_size, &
1049 390 : col_blk_size=ri_blk_size)
1050 : END DO
1051 :
1052 : ! Only take the coeffs for atom on which MOs I,J are localized
1053 166 : coeffs => donor_state%contract_coeffs
1054 :
1055 358 : DO ispin = 1, nspins
1056 :
1057 : ! Loop over the donor MOs and contract
1058 582 : DO imo = 1, ndo_mo
1059 :
1060 : ! do the contraction
1061 224 : CALL dbcsr_set(aI_P, 0.0_dp); CALL dbcsr_set(P_Ib, 0.0_dp)
1062 224 : CALL contract2_AO_to_doMO_low(pq_X, coeffs(:, (ispin - 1)*ndo_mo + imo), aI_P, P_Ib, katom)
1063 :
1064 : ! Get the full (aI|P) contracted integrals
1065 224 : CALL dbcsr_transposed(work, P_Ib)
1066 224 : CALL dbcsr_add(work, aI_P, 1.0_dp, 1.0_dp)
1067 224 : CALL dbcsr_complete_redistribute(work, contr_int((ispin - 1)*ndo_mo + imo)%matrix)
1068 224 : CALL dbcsr_filter(contr_int((ispin - 1)*ndo_mo + imo)%matrix, 1.0E-16_dp)
1069 :
1070 416 : CALL dbcsr_release(work)
1071 : END DO !imo
1072 : END DO !ispin
1073 :
1074 : ! Clean-up
1075 166 : CALL dbcsr_release(aI_P)
1076 166 : CALL dbcsr_release(P_Ib)
1077 166 : CALL dbcsr_distribution_release(opt_dbcsr_dist)
1078 166 : DEALLOCATE (ri_blk_size, aI_P, P_Ib, work, ri_basis)
1079 :
1080 166 : CALL timestop(handle)
1081 :
1082 332 : END SUBROUTINE contract2_AO_to_doMO
1083 :
1084 : ! **************************************************************************************************
1085 : !> \brief Contraction of the 3-center integrals (ab|Q) over the RI basis elements Q to get donor MOS
1086 : !> => (ab|IJ) = sum_X (ab|Q) coeffs_Q
1087 : !> \param ab_Q the tensor holding the integrals
1088 : !> \param vec the contraction coefficients
1089 : !> \param mat_abIJ the matrix holding the (ab|IJ) integrals (blocks must be reserved)
1090 : !> \param atom_k the atom for which we contract, i.e. we only take RI basis Q centered on atom_k
1091 : !> \note By construction, distribution of tensor and matrix match, also for OMP threads
1092 : ! **************************************************************************************************
1093 90 : SUBROUTINE contract3_RI_to_doMOs(ab_Q, vec, mat_abIJ, atom_k)
1094 :
1095 : TYPE(dbt_type) :: ab_Q
1096 : REAL(dp), DIMENSION(:), INTENT(IN) :: vec
1097 : TYPE(dbcsr_type) :: mat_abIJ
1098 : INTEGER, INTENT(IN) :: atom_k
1099 :
1100 : CHARACTER(len=*), PARAMETER :: routineN = 'contract3_RI_to_doMOs'
1101 :
1102 : INTEGER :: handle, i, iatom, ind(3), j, jatom, katom
1103 : LOGICAL :: found, t_found
1104 : REAL(dp) :: prefac
1105 90 : REAL(dp), ALLOCATABLE, DIMENSION(:, :, :) :: iabc
1106 90 : REAL(KIND=dp), DIMENSION(:, :), POINTER :: pblock
1107 : TYPE(dbcsr_type) :: work
1108 : TYPE(dbt_iterator_type) :: iter
1109 :
1110 90 : NULLIFY (pblock)
1111 :
1112 90 : CALL timeset(routineN, handle)
1113 :
1114 : !$OMP PARALLEL DEFAULT(NONE) &
1115 : !$OMP SHARED(ab_Q,vec,mat_abIJ,atom_k) &
1116 90 : !$OMP PRIVATE(iter,ind,iatom,jatom,katom,prefac,iabc,t_found,found,pblock,i,j)
1117 : CALL dbt_iterator_start(iter, ab_Q)
1118 : DO WHILE (dbt_iterator_blocks_left(iter))
1119 : CALL dbt_iterator_next_block(iter, ind)
1120 :
1121 : iatom = ind(1)
1122 : jatom = ind(2)
1123 : katom = ind(3)
1124 :
1125 : IF (.NOT. atom_k == katom) CYCLE
1126 :
1127 : prefac = 1.0_dp
1128 : IF (iatom == jatom) prefac = 0.5_dp
1129 :
1130 : CALL dbt_get_block(ab_Q, ind, iabc, t_found)
1131 :
1132 : CALL dbcsr_get_block_p(mat_abIJ, iatom, jatom, pblock, found)
1133 : IF ((.NOT. found) .OR. (.NOT. t_found)) CYCLE
1134 :
1135 : DO i = 1, SIZE(pblock, 1)
1136 : DO j = 1, SIZE(pblock, 2)
1137 : !$OMP ATOMIC
1138 : pblock(i, j) = pblock(i, j) + prefac*DOT_PRODUCT(vec(:), iabc(i, j, :))
1139 : END DO
1140 : END DO
1141 :
1142 : DEALLOCATE (iabc)
1143 : END DO !iter
1144 : CALL dbt_iterator_stop(iter)
1145 : !$OMP END PARALLEL
1146 :
1147 : !matrix only half filled => need to add its transpose
1148 90 : CALL dbcsr_create(work, template=mat_abIJ)
1149 90 : CALL dbcsr_transposed(work, mat_abIJ)
1150 90 : CALL dbcsr_add(mat_abIJ, work, 1.0_dp, 1.0_dp)
1151 90 : CALL dbcsr_release(work)
1152 :
1153 90 : CALL timestop(handle)
1154 :
1155 180 : END SUBROUTINE contract3_RI_to_doMOs
1156 :
1157 : ! **************************************************************************************************
1158 : !> \brief Contraction of the 3-center integrals over index 1 and 2, for a given atom_k. The results
1159 : !> are stored in two matrices, such that (a,b are block indices):
1160 : !> mat_aIb(ab) = mat_aIb(ab) + sum j_b (i_aj_b|k)*v(j_b) and
1161 : !> mat_bIa(ba) = mat_bIa(ba) + sum i_a (i_aj_b|k)*v(i_a)
1162 : !> The block size of the columns of mat_aIb and the rows of mat_bIa are the size of k (RI)
1163 : !> \param ab_Q the tensor containing the 3-center integrals
1164 : !> \param vec the contraction coefficients
1165 : !> \param mat_aIb normal type dbcsr matrix
1166 : !> \param mat_bIa normal type dbcsr matrix
1167 : !> \param atom_k the atom for which we contract
1168 : !> It is assumed that the contraction coefficients for MO I are all on atom_k
1169 : !> We do the classic thing when we fill half the matrix and add its transposed to get the full
1170 : !> one, but here, the matrix is not symmetric, hence we explicitely have 2 input matrices
1171 : !> The distribution of the integrals and the normal dbcsr matrix are compatible out of the box
1172 : ! **************************************************************************************************
1173 224 : SUBROUTINE contract2_AO_to_doMO_low(ab_Q, vec, mat_aIb, mat_bIa, atom_k)
1174 :
1175 : TYPE(dbt_type) :: ab_Q
1176 : REAL(dp), DIMENSION(:), INTENT(IN) :: vec
1177 : TYPE(dbcsr_type), INTENT(INOUT) :: mat_aIb, mat_bIa
1178 : INTEGER, INTENT(IN) :: atom_k
1179 :
1180 : CHARACTER(LEN=*), PARAMETER :: routineN = 'contract2_AO_to_doMO_low'
1181 :
1182 : INTEGER :: handle, i, iatom, ind(3), j, jatom, &
1183 : katom, s1, s2
1184 224 : INTEGER, DIMENSION(:), POINTER :: atom_blk_size
1185 : LOGICAL :: found, t_found
1186 224 : REAL(dp), ALLOCATABLE, DIMENSION(:, :, :) :: iabc
1187 224 : REAL(dp), DIMENSION(:, :), POINTER :: pblock
1188 : TYPE(dbt_iterator_type) :: iter
1189 :
1190 224 : NULLIFY (atom_blk_size, pblock)
1191 :
1192 224 : CALL timeset(routineN, handle)
1193 :
1194 224 : CALL dbcsr_get_info(mat_aIb, row_blk_size=atom_blk_size)
1195 :
1196 : !$OMP PARALLEL DEFAULT(NONE) &
1197 : !$OMP SHARED(ab_Q,vec,mat_aIb,mat_bIa,atom_k,atom_blk_size) &
1198 224 : !$OMP PRIVATE(iter,ind,iatom,jatom,katom,iabc,t_found,found,s1,s2,j,i,pblock)
1199 : CALL dbt_iterator_start(iter, ab_Q)
1200 : DO WHILE (dbt_iterator_blocks_left(iter))
1201 : CALL dbt_iterator_next_block(iter, ind)
1202 :
1203 : iatom = ind(1)
1204 : jatom = ind(2)
1205 : katom = ind(3)
1206 :
1207 : IF (atom_k /= katom) CYCLE
1208 :
1209 : CALL dbt_get_block(ab_Q, ind, iabc, t_found)
1210 : IF (.NOT. t_found) CYCLE
1211 :
1212 : ! Deal with mat_aIb
1213 : IF (jatom == atom_k) THEN
1214 : s1 = atom_blk_size(iatom)
1215 : s2 = SIZE(iabc, 3)
1216 :
1217 : CALL dbcsr_get_block_p(matrix=mat_aIb, row=iatom, col=jatom, BLOCK=pblock, found=found)
1218 :
1219 : IF (found) THEN
1220 : DO i = 1, s1
1221 : DO j = 1, s2
1222 : !$OMP ATOMIC
1223 : pblock(i, j) = pblock(i, j) + DOT_PRODUCT(vec, iabc(i, :, j))
1224 : END DO
1225 : END DO
1226 : END IF
1227 : END IF ! jatom == atom_k
1228 :
1229 : ! Deal with mat_bIa, keep block diagonal empty
1230 : IF (iatom == jatom) CYCLE
1231 : IF (iatom == atom_k) THEN
1232 : s1 = SIZE(iabc, 3)
1233 : s2 = atom_blk_size(jatom)
1234 :
1235 : CALL dbcsr_get_block_p(matrix=mat_bIa, row=iatom, col=jatom, BLOCK=pblock, found=found)
1236 :
1237 : IF (found) THEN
1238 : DO i = 1, s1
1239 : DO j = 1, s2
1240 : !$OMP ATOMIC
1241 : pblock(i, j) = pblock(i, j) + DOT_PRODUCT(vec, iabc(:, j, i))
1242 : END DO
1243 : END DO
1244 : END IF
1245 : END IF !iatom== atom_k
1246 :
1247 : DEALLOCATE (iabc)
1248 : END DO !iter
1249 : CALL dbt_iterator_stop(iter)
1250 : !$OMP END PARALLEL
1251 :
1252 224 : CALL timestop(handle)
1253 :
1254 448 : END SUBROUTINE contract2_AO_to_doMO_low
1255 :
1256 : ! **************************************************************************************************
1257 : !> \brief Multiply all the blocks of a contracted RI integral (aI|P) by a matrix of type (P|...|Q)
1258 : !> \param contr_int the integral array
1259 : !> \param PQ the smaller matrix to multiply all blocks
1260 : !> \note It is assumed that all non-zero blocks have the same number of columns. Can pass partial
1261 : !> arrays, e.g. contr_int(1:3)
1262 : ! **************************************************************************************************
1263 274 : SUBROUTINE ri_all_blocks_mm(contr_int, PQ)
1264 :
1265 : TYPE(dbcsr_p_type), DIMENSION(:) :: contr_int
1266 : REAL(dp), DIMENSION(:, :), INTENT(IN) :: PQ
1267 :
1268 : INTEGER :: iblk, imo, jblk, ndo_mo, s1, s2
1269 : LOGICAL :: found
1270 274 : REAL(dp), ALLOCATABLE, DIMENSION(:, :) :: work
1271 274 : REAL(dp), DIMENSION(:, :), POINTER :: pblock
1272 : TYPE(dbcsr_iterator_type) :: iter
1273 :
1274 274 : NULLIFY (pblock)
1275 :
1276 274 : ndo_mo = SIZE(contr_int)
1277 :
1278 594 : DO imo = 1, ndo_mo
1279 320 : CALL dbcsr_iterator_start(iter, contr_int(imo)%matrix)
1280 1368 : DO WHILE (dbcsr_iterator_blocks_left(iter))
1281 :
1282 1048 : CALL dbcsr_iterator_next_block(iter, row=iblk, column=jblk)
1283 1048 : CALL dbcsr_get_block_p(contr_int(imo)%matrix, iblk, jblk, pblock, found)
1284 :
1285 1368 : IF (found) THEN
1286 1048 : s1 = SIZE(pblock, 1)
1287 1048 : s2 = SIZE(pblock, 2)
1288 4192 : ALLOCATE (work(s1, s2))
1289 1048 : CALL dgemm('N', 'N', s1, s2, s2, 1.0_dp, pblock, s1, PQ, s2, 0.0_dp, work, s1)
1290 1048 : CALL dcopy(s1*s2, work, 1, pblock, 1)
1291 1048 : DEALLOCATE (work)
1292 : END IF
1293 :
1294 : END DO ! dbcsr iterator
1295 914 : CALL dbcsr_iterator_stop(iter)
1296 : END DO !imo
1297 :
1298 548 : END SUBROUTINE ri_all_blocks_mm
1299 :
1300 : ! **************************************************************************************************
1301 : !> \brief Copies an (partial) array of contracted RI integrals into anoter one
1302 : !> \param new_int where the copy is stored
1303 : !> \param ref_int what is copied
1304 : !> \note Allocate the matrices of new_int if not done already
1305 : ! **************************************************************************************************
1306 170 : SUBROUTINE copy_ri_contr_int(new_int, ref_int)
1307 :
1308 : TYPE(dbcsr_p_type), DIMENSION(:), INTENT(INOUT) :: new_int
1309 : TYPE(dbcsr_p_type), DIMENSION(:), INTENT(IN) :: ref_int
1310 :
1311 : INTEGER :: iso, ndo_so
1312 :
1313 170 : CPASSERT(SIZE(new_int) == SIZE(ref_int))
1314 170 : ndo_so = SIZE(ref_int)
1315 :
1316 364 : DO iso = 1, ndo_so
1317 194 : IF (.NOT. ASSOCIATED(new_int(iso)%matrix)) ALLOCATE (new_int(iso)%matrix)
1318 364 : CALL dbcsr_copy(new_int(iso)%matrix, ref_int(iso)%matrix)
1319 : END DO
1320 :
1321 170 : END SUBROUTINE copy_ri_contr_int
1322 :
1323 : ! **************************************************************************************************
1324 : !> \brief Takes the product of contracted integrals and put them in a kernel matrix
1325 : !> \param kernel the matrix where the products are stored
1326 : !> \param lhs_int the left-hand side contracted integrals
1327 : !> \param rhs_int the right-hand side contracted integrals
1328 : !> \param quadrants on which quadrant(s) on the kernel matrix the product is stored
1329 : !> \param qs_env ...
1330 : !> \param eps_filter filter for dbcsr matrix multiplication
1331 : !> \param mo_transpose whether the MO blocks should be transpose, i.e. (aI|Jb) => (aJ|Ib)
1332 : !> \note It is assumed that the kerenl matrix is NOT symmetric
1333 : !> There are three quadrants, corresponding to 1: the upper-left (diagonal), 2: the
1334 : !> upper-right (off-diagonal) and 3: the lower-right (diagonal).
1335 : !> Need to finalize the kernel matrix after calling this routine (possibly multiple times)
1336 : ! **************************************************************************************************
1337 170 : SUBROUTINE ri_int_product(kernel, lhs_int, rhs_int, quadrants, qs_env, eps_filter, mo_transpose)
1338 :
1339 : TYPE(dbcsr_type), INTENT(INOUT) :: kernel
1340 : TYPE(dbcsr_p_type), DIMENSION(:), INTENT(IN) :: lhs_int, rhs_int
1341 : LOGICAL, DIMENSION(3), INTENT(IN) :: quadrants
1342 : TYPE(qs_environment_type), POINTER :: qs_env
1343 : REAL(dp), INTENT(IN), OPTIONAL :: eps_filter
1344 : LOGICAL, INTENT(IN), OPTIONAL :: mo_transpose
1345 :
1346 : INTEGER :: i, iblk, iso, j, jblk, jso, nblk, ndo_so
1347 : LOGICAL :: found, my_mt
1348 170 : REAL(dp), DIMENSION(:, :), POINTER :: pblock
1349 : TYPE(dbcsr_iterator_type) :: iter
1350 170 : TYPE(dbcsr_p_type), DIMENSION(:), POINTER :: matrix_s
1351 : TYPE(dbcsr_type) :: prod
1352 :
1353 170 : NULLIFY (matrix_s, pblock)
1354 :
1355 : ! Initialization
1356 0 : CPASSERT(SIZE(lhs_int) == SIZE(rhs_int))
1357 194 : CPASSERT(ANY(quadrants))
1358 170 : ndo_so = SIZE(lhs_int)
1359 170 : CALL get_qs_env(qs_env, matrix_s=matrix_s, natom=nblk)
1360 170 : CALL dbcsr_create(prod, template=matrix_s(1)%matrix, matrix_type=dbcsr_type_no_symmetry)
1361 170 : my_mt = .FALSE.
1362 170 : IF (PRESENT(mo_transpose)) my_mt = mo_transpose
1363 :
1364 : ! The kernel matrix is symmetric (even if normal type) => only fill upper half on diagonal
1365 : ! quadrants, but the whole thing on upper-right quadrant
1366 364 : DO iso = 1, ndo_so
1367 650 : DO jso = 1, ndo_so
1368 :
1369 : ! If on-diagonal quadrants only, can skip jso < iso
1370 286 : IF (.NOT. quadrants(2) .AND. jso < iso) CYCLE
1371 :
1372 240 : i = iso; j = jso
1373 240 : IF (my_mt) THEN
1374 6 : i = jso; j = iso
1375 : END IF
1376 :
1377 : ! Take the product lhs*rhs^T
1378 : CALL dbcsr_multiply('N', 'T', 1.0_dp, lhs_int(i)%matrix, rhs_int(j)%matrix, &
1379 240 : 0.0_dp, prod, filter_eps=eps_filter)
1380 :
1381 : ! Loop over blocks of prod and fill kernel matrix => ok cuz same (but replicated) dist
1382 240 : CALL dbcsr_iterator_start(iter, prod)
1383 8201 : DO WHILE (dbcsr_iterator_blocks_left(iter))
1384 :
1385 7961 : CALL dbcsr_iterator_next_block(iter, row=iblk, column=jblk)
1386 7961 : IF ((iso == jso .AND. jblk < iblk) .AND. .NOT. quadrants(2)) CYCLE
1387 :
1388 4227 : CALL dbcsr_get_block_p(prod, iblk, jblk, pblock, found)
1389 :
1390 4467 : IF (found) THEN
1391 :
1392 : ! Case study on quadrant
1393 : !upper-left
1394 4227 : IF (quadrants(1)) THEN
1395 4123 : CALL dbcsr_put_block(kernel, (iso - 1)*nblk + iblk, (jso - 1)*nblk + jblk, pblock)
1396 : END IF
1397 :
1398 : !upper-right
1399 4227 : IF (quadrants(2)) THEN
1400 80 : CALL dbcsr_put_block(kernel, (iso - 1)*nblk + iblk, (ndo_so + jso - 1)*nblk + jblk, pblock)
1401 : END IF
1402 :
1403 : !lower-right
1404 4227 : IF (quadrants(3)) THEN
1405 56 : CALL dbcsr_put_block(kernel, (ndo_so + iso - 1)*nblk + iblk, (ndo_so + jso - 1)*nblk + jblk, pblock)
1406 : END IF
1407 :
1408 : END IF
1409 :
1410 : END DO ! dbcsr iterator
1411 720 : CALL dbcsr_iterator_stop(iter)
1412 :
1413 : END DO !jso
1414 : END DO !iso
1415 :
1416 : ! Clean-up
1417 170 : CALL dbcsr_release(prod)
1418 :
1419 170 : END SUBROUTINE ri_int_product
1420 :
1421 : END MODULE xas_tdp_kernel
|
