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 RI-methods for HFX and K-points.
10 : !> \auhtor Augustin Bussy (01.2023)
11 : ! **************************************************************************************************
12 :
13 : MODULE hfx_ri_kp
14 : USE admm_types, ONLY: get_admm_env
15 : USE atomic_kind_types, ONLY: atomic_kind_type,&
16 : get_atomic_kind_set
17 : USE basis_set_types, ONLY: get_gto_basis_set,&
18 : gto_basis_set_p_type
19 : USE bibliography, ONLY: Bussy2024,&
20 : cite_reference
21 : USE cell_types, ONLY: cell_type,&
22 : pbc,&
23 : real_to_scaled,&
24 : scaled_to_real
25 : USE cp_array_utils, ONLY: cp_1d_logical_p_type,&
26 : cp_2d_r_p_type,&
27 : cp_3d_r_p_type
28 : USE cp_blacs_env, ONLY: cp_blacs_env_create,&
29 : cp_blacs_env_release,&
30 : cp_blacs_env_type
31 : USE cp_control_types, ONLY: dft_control_type
32 : USE cp_dbcsr_api, ONLY: &
33 : dbcsr_add, dbcsr_clear, dbcsr_copy, dbcsr_create, dbcsr_distribution_get, &
34 : dbcsr_distribution_new, dbcsr_distribution_release, dbcsr_distribution_type, dbcsr_dot, &
35 : dbcsr_filter, dbcsr_finalize, dbcsr_get_block_p, dbcsr_get_info, &
36 : dbcsr_iterator_blocks_left, dbcsr_iterator_next_block, dbcsr_iterator_start, &
37 : dbcsr_iterator_stop, dbcsr_iterator_type, dbcsr_p_type, dbcsr_put_block, dbcsr_release, &
38 : dbcsr_type, dbcsr_type_no_symmetry, dbcsr_type_symmetric
39 : USE cp_dbcsr_cholesky, ONLY: cp_dbcsr_cholesky_decompose,&
40 : cp_dbcsr_cholesky_invert
41 : USE cp_dbcsr_cp2k_link, ONLY: cp_dbcsr_alloc_block_from_nbl
42 : USE cp_dbcsr_diag, ONLY: cp_dbcsr_power
43 : USE cp_dbcsr_operations, ONLY: cp_dbcsr_dist2d_to_dist
44 : USE dbt_api, ONLY: &
45 : dbt_batched_contract_finalize, dbt_batched_contract_init, dbt_clear, dbt_contract, &
46 : dbt_copy, dbt_copy_matrix_to_tensor, dbt_copy_tensor_to_matrix, dbt_create, dbt_destroy, &
47 : dbt_distribution_destroy, dbt_distribution_new, dbt_distribution_type, dbt_filter, &
48 : dbt_finalize, dbt_get_block, dbt_get_info, dbt_get_stored_coordinates, &
49 : dbt_iterator_blocks_left, dbt_iterator_next_block, dbt_iterator_start, dbt_iterator_stop, &
50 : dbt_iterator_type, dbt_mp_environ_pgrid, dbt_pgrid_create, dbt_pgrid_destroy, &
51 : dbt_pgrid_type, dbt_put_block, dbt_scale, dbt_type
52 : USE distribution_2d_types, ONLY: distribution_2d_release,&
53 : distribution_2d_type
54 : USE hfx_ri, ONLY: get_idx_to_atom,&
55 : hfx_ri_pre_scf_calc_tensors
56 : USE hfx_types, ONLY: hfx_ri_type
57 : USE input_constants, ONLY: do_potential_short,&
58 : hfx_ri_do_2c_cholesky,&
59 : hfx_ri_do_2c_diag,&
60 : hfx_ri_do_2c_iter
61 : USE input_cp2k_hfx, ONLY: ri_pmat
62 : USE input_section_types, ONLY: section_vals_get_subs_vals,&
63 : section_vals_type,&
64 : section_vals_val_get,&
65 : section_vals_val_set
66 : USE iterate_matrix, ONLY: invert_hotelling
67 : USE kinds, ONLY: dp,&
68 : int_8
69 : USE kpoint_types, ONLY: get_kpoint_info,&
70 : kpoint_type
71 : USE libint_2c_3c, ONLY: cutoff_screen_factor
72 : USE machine, ONLY: m_flush,&
73 : m_walltime
74 : USE mathlib, ONLY: erfc_cutoff
75 : USE message_passing, ONLY: mp_cart_type,&
76 : mp_para_env_type,&
77 : mp_request_type,&
78 : mp_waitall
79 : USE particle_methods, ONLY: get_particle_set
80 : USE particle_types, ONLY: particle_type
81 : USE physcon, ONLY: angstrom
82 : USE qs_environment_types, ONLY: get_qs_env,&
83 : qs_environment_type
84 : USE qs_force_types, ONLY: qs_force_type
85 : USE qs_integral_utils, ONLY: basis_set_list_setup
86 : USE qs_interactions, ONLY: init_interaction_radii_orb_basis
87 : USE qs_kind_types, ONLY: qs_kind_type
88 : USE qs_neighbor_list_types, ONLY: get_iterator_info,&
89 : neighbor_list_iterate,&
90 : neighbor_list_iterator_create,&
91 : neighbor_list_iterator_p_type,&
92 : neighbor_list_iterator_release,&
93 : neighbor_list_set_p_type,&
94 : release_neighbor_list_sets
95 : USE qs_scf_types, ONLY: qs_scf_env_type
96 : USE qs_tensors, ONLY: &
97 : build_2c_derivatives, build_2c_neighbor_lists, build_3c_derivatives, &
98 : build_3c_neighbor_lists, get_3c_iterator_info, get_tensor_occupancy, &
99 : neighbor_list_3c_destroy, neighbor_list_3c_iterate, neighbor_list_3c_iterator_create, &
100 : neighbor_list_3c_iterator_destroy
101 : USE qs_tensors_types, ONLY: create_2c_tensor,&
102 : create_3c_tensor,&
103 : create_tensor_batches,&
104 : distribution_2d_create,&
105 : distribution_3d_create,&
106 : distribution_3d_type,&
107 : neighbor_list_3c_iterator_type,&
108 : neighbor_list_3c_type
109 : USE util, ONLY: get_limit
110 : USE virial_types, ONLY: virial_type
111 : #include "./base/base_uses.f90"
112 :
113 : !$ USE OMP_LIB, ONLY: omp_get_num_threads
114 :
115 : IMPLICIT NONE
116 : PRIVATE
117 :
118 : PUBLIC :: hfx_ri_update_ks_kp, hfx_ri_update_forces_kp
119 :
120 : CHARACTER(len=*), PARAMETER, PRIVATE :: moduleN = 'hfx_ri_kp'
121 : CONTAINS
122 :
123 : ! NOTES: for a start, we do not seek performance, but accuracy. So in this first implementation,
124 : ! we give little consideration to batching, load balance and such.
125 : ! We also put everything here, even if there is some code replication with the original RI_HFX
126 : ! We will only work in the RHO flavor
127 : ! For now, we will also always assume that there is a single para_env, and that there is no
128 : ! K-point subgroup. This might change in the future
129 :
130 : ! **************************************************************************************************
131 : !> \brief I_1nitialize the ri_data for K-point. For now, we take the normal, usual existing ri_data
132 : !> and we adapt it to our needs
133 : !> \param dbcsr_template ...
134 : !> \param ri_data ...
135 : !> \param qs_env ...
136 : ! **************************************************************************************************
137 70 : SUBROUTINE adapt_ri_data_to_kp(dbcsr_template, ri_data, qs_env)
138 : TYPE(dbcsr_type), INTENT(INOUT) :: dbcsr_template
139 : TYPE(hfx_ri_type), INTENT(INOUT) :: ri_data
140 : TYPE(qs_environment_type), POINTER :: qs_env
141 :
142 : INTEGER :: i_img, i_RI, i_spin, iatom, natom, &
143 : nblks_RI, nimg, nkind, nspins
144 70 : INTEGER, ALLOCATABLE, DIMENSION(:) :: bsizes_RI_ext, dist1, dist2, dist3
145 : TYPE(dft_control_type), POINTER :: dft_control
146 : TYPE(mp_para_env_type), POINTER :: para_env
147 :
148 70 : NULLIFY (dft_control, para_env)
149 :
150 : !The main thing that we need to do is to allocate more space for the integrals, such that there
151 : !is room for each periodic image. Note that we only go in 1D, i.e. we store (mu^0 sigma^a|P^0),
152 : !and (P^0|Q^a) => the RI basis is always in the main cell.
153 :
154 : !Get kpoint info
155 70 : CALL get_qs_env(qs_env, dft_control=dft_control, natom=natom, para_env=para_env, nkind=nkind)
156 70 : nimg = ri_data%nimg
157 :
158 : !Along the RI direction we have basis elements spread accross ncell_RI images.
159 70 : nblks_RI = SIZE(ri_data%bsizes_RI_split)
160 210 : ALLOCATE (bsizes_RI_ext(nblks_RI*ri_data%ncell_RI))
161 506 : DO i_RI = 1, ri_data%ncell_RI
162 2344 : bsizes_RI_ext((i_RI - 1)*nblks_RI + 1:i_RI*nblks_RI) = ri_data%bsizes_RI_split(:)
163 : END DO
164 :
165 4202 : ALLOCATE (ri_data%t_3c_int_ctr_1(1, nimg))
166 : CALL create_3c_tensor(ri_data%t_3c_int_ctr_1(1, 1), dist1, dist2, dist3, &
167 : ri_data%pgrid_1, ri_data%bsizes_AO_split, bsizes_RI_ext, &
168 70 : ri_data%bsizes_AO_split, [1, 2], [3], name="(AO RI | AO)")
169 :
170 1716 : DO i_img = 2, nimg
171 1716 : CALL dbt_create(ri_data%t_3c_int_ctr_1(1, 1), ri_data%t_3c_int_ctr_1(1, i_img))
172 : END DO
173 70 : DEALLOCATE (dist1, dist2, dist3)
174 :
175 770 : ALLOCATE (ri_data%t_3c_int_ctr_2(1, 1))
176 : CALL create_3c_tensor(ri_data%t_3c_int_ctr_2(1, 1), dist1, dist2, dist3, &
177 : ri_data%pgrid_1, ri_data%bsizes_AO_split, bsizes_RI_ext, &
178 70 : ri_data%bsizes_AO_split, [1], [2, 3], name="(AO RI | AO)")
179 70 : DEALLOCATE (dist1, dist2, dist3)
180 :
181 : !We use full block sizes for the 2c quantities
182 70 : DEALLOCATE (bsizes_RI_ext)
183 70 : nblks_RI = SIZE(ri_data%bsizes_RI)
184 210 : ALLOCATE (bsizes_RI_ext(nblks_RI*ri_data%ncell_RI))
185 506 : DO i_RI = 1, ri_data%ncell_RI
186 1378 : bsizes_RI_ext((i_RI - 1)*nblks_RI + 1:i_RI*nblks_RI) = ri_data%bsizes_RI(:)
187 : END DO
188 :
189 3010 : ALLOCATE (ri_data%t_2c_inv(1, natom), ri_data%t_2c_int(1, natom), ri_data%t_2c_pot(1, natom))
190 : CALL create_2c_tensor(ri_data%t_2c_inv(1, 1), dist1, dist2, ri_data%pgrid_2d, &
191 : bsizes_RI_ext, bsizes_RI_ext, &
192 70 : name="(RI | RI)")
193 70 : DEALLOCATE (dist1, dist2)
194 70 : CALL dbt_create(ri_data%t_2c_inv(1, 1), ri_data%t_2c_int(1, 1))
195 70 : CALL dbt_create(ri_data%t_2c_inv(1, 1), ri_data%t_2c_pot(1, 1))
196 140 : DO iatom = 2, natom
197 70 : CALL dbt_create(ri_data%t_2c_inv(1, 1), ri_data%t_2c_inv(1, iatom))
198 70 : CALL dbt_create(ri_data%t_2c_inv(1, 1), ri_data%t_2c_int(1, iatom))
199 140 : CALL dbt_create(ri_data%t_2c_inv(1, 1), ri_data%t_2c_pot(1, iatom))
200 : END DO
201 :
202 350 : ALLOCATE (ri_data%kp_cost(natom, natom, nimg))
203 12082 : ri_data%kp_cost = 0.0_dp
204 :
205 : !We store the density and KS matrix in tensor format
206 70 : nspins = dft_control%nspins
207 9218 : ALLOCATE (ri_data%rho_ao_t(nspins, nimg), ri_data%ks_t(nspins, nimg))
208 : CALL create_2c_tensor(ri_data%rho_ao_t(1, 1), dist1, dist2, ri_data%pgrid_2d, &
209 : ri_data%bsizes_AO_split, ri_data%bsizes_AO_split, &
210 70 : name="(AO | AO)")
211 70 : DEALLOCATE (dist1, dist2)
212 :
213 70 : CALL dbt_create(dbcsr_template, ri_data%ks_t(1, 1))
214 :
215 70 : IF (nspins == 2) THEN
216 24 : CALL dbt_create(ri_data%rho_ao_t(1, 1), ri_data%rho_ao_t(2, 1))
217 24 : CALL dbt_create(ri_data%ks_t(1, 1), ri_data%ks_t(2, 1))
218 : END IF
219 1716 : DO i_img = 2, nimg
220 3710 : DO i_spin = 1, nspins
221 1994 : CALL dbt_create(ri_data%rho_ao_t(1, 1), ri_data%rho_ao_t(i_spin, i_img))
222 3640 : CALL dbt_create(ri_data%ks_t(1, 1), ri_data%ks_t(i_spin, i_img))
223 : END DO
224 : END DO
225 :
226 210 : END SUBROUTINE adapt_ri_data_to_kp
227 :
228 : ! **************************************************************************************************
229 : !> \brief The pre-scf steps for RI-HFX k-points calculation. Namely the calculation of the integrals
230 : !> \param dbcsr_template ...
231 : !> \param ri_data ...
232 : !> \param qs_env ...
233 : ! **************************************************************************************************
234 70 : SUBROUTINE hfx_ri_pre_scf_kp(dbcsr_template, ri_data, qs_env)
235 : TYPE(dbcsr_type), INTENT(INOUT) :: dbcsr_template
236 : TYPE(hfx_ri_type), INTENT(INOUT) :: ri_data
237 : TYPE(qs_environment_type), POINTER :: qs_env
238 :
239 : CHARACTER(LEN=*), PARAMETER :: routineN = 'hfx_ri_pre_scf_kp'
240 :
241 : INTEGER :: handle, i_img, iatom, natom, nimg, nkind
242 70 : TYPE(dbcsr_type), ALLOCATABLE, DIMENSION(:) :: t_2c_op_pot, t_2c_op_RI
243 70 : TYPE(dbt_type), ALLOCATABLE, DIMENSION(:, :) :: t_3c_int
244 : TYPE(dft_control_type), POINTER :: dft_control
245 :
246 70 : NULLIFY (dft_control)
247 :
248 70 : CALL timeset(routineN, handle)
249 :
250 70 : CALL get_qs_env(qs_env, dft_control=dft_control, natom=natom, nkind=nkind)
251 :
252 70 : CALL cleanup_kp(ri_data)
253 :
254 : !We do all the checks on what we allow in this initial implementation
255 70 : IF (ri_data%flavor .NE. ri_pmat) CPABORT("K-points RI-HFX only with RHO flavor")
256 70 : IF (ri_data%same_op) ri_data%same_op = .FALSE. !force the full calculation with RI metric
257 70 : IF (ABS(ri_data%eps_pgf_orb - dft_control%qs_control%eps_pgf_orb) > 1.0E-16_dp) &
258 0 : CPABORT("RI%EPS_PGF_ORB and QS%EPS_PGF_ORB must be identical for RI-HFX k-points")
259 :
260 70 : CALL get_kp_and_ri_images(ri_data, qs_env)
261 70 : nimg = ri_data%nimg
262 :
263 : !Calculate the integrals
264 3712 : ALLOCATE (t_2c_op_pot(nimg), t_2c_op_RI(nimg))
265 4202 : ALLOCATE (t_3c_int(1, nimg))
266 70 : CALL hfx_ri_pre_scf_calc_tensors(qs_env, ri_data, t_2c_op_RI, t_2c_op_pot, t_3c_int, do_kpoints=.TRUE.)
267 :
268 : !Make sure the internals have the k-point format
269 70 : CALL adapt_ri_data_to_kp(dbcsr_template, ri_data, qs_env)
270 :
271 : !For each atom i, we calculate the inverse RI metric (P^0 | Q^0)^-1 without external bumping yet
272 : !Also store the off-diagonal integrals of the RI metric in case of forces, bumped from the left
273 210 : DO iatom = 1, natom
274 : CALL get_ext_2c_int(ri_data%t_2c_inv(1, iatom), t_2c_op_RI, iatom, iatom, 1, ri_data, qs_env, &
275 140 : do_inverse=.TRUE.)
276 : !for the forces:
277 : !off-diagonl RI metric bumped from the left
278 : CALL get_ext_2c_int(ri_data%t_2c_int(1, iatom), t_2c_op_RI, iatom, iatom, 1, ri_data, &
279 140 : qs_env, off_diagonal=.TRUE.)
280 140 : CALL apply_bump(ri_data%t_2c_int(1, iatom), iatom, ri_data, qs_env, from_left=.TRUE., from_right=.FALSE.)
281 :
282 : !RI metric with bumped off-diagonal blocks (but not inverted), depumed from left and right
283 : CALL get_ext_2c_int(ri_data%t_2c_pot(1, iatom), t_2c_op_RI, iatom, iatom, 1, ri_data, qs_env, &
284 140 : do_inverse=.TRUE., skip_inverse=.TRUE.)
285 : CALL apply_bump(ri_data%t_2c_pot(1, iatom), iatom, ri_data, qs_env, from_left=.TRUE., &
286 210 : from_right=.TRUE., debump=.TRUE.)
287 :
288 : END DO
289 :
290 1786 : DO i_img = 1, nimg
291 1786 : CALL dbcsr_release(t_2c_op_RI(i_img))
292 : END DO
293 :
294 3572 : ALLOCATE (ri_data%kp_mat_2c_pot(1, nimg))
295 1786 : DO i_img = 1, nimg
296 1716 : CALL dbcsr_create(ri_data%kp_mat_2c_pot(1, i_img), template=t_2c_op_pot(i_img))
297 1716 : CALL dbcsr_copy(ri_data%kp_mat_2c_pot(1, i_img), t_2c_op_pot(i_img))
298 1786 : CALL dbcsr_release(t_2c_op_pot(i_img))
299 : END DO
300 :
301 : !Pre-contract all 3c integrals with the bumped inverse RI metric (P^0|Q^0)^-1,
302 : !and store in ri_data%t_3c_int_ctr_1
303 70 : CALL precontract_3c_ints(t_3c_int, ri_data, qs_env)
304 :
305 : !reorder the 3c integrals such that empty images are bunched up together
306 70 : CALL reorder_3c_ints(ri_data%t_3c_int_ctr_1(1, :), ri_data)
307 :
308 70 : CALL timestop(handle)
309 :
310 1856 : END SUBROUTINE hfx_ri_pre_scf_kp
311 :
312 : ! **************************************************************************************************
313 : !> \brief clean-up the KP specific data from ri_data
314 : !> \param ri_data ...
315 : ! **************************************************************************************************
316 70 : SUBROUTINE cleanup_kp(ri_data)
317 : TYPE(hfx_ri_type), INTENT(INOUT) :: ri_data
318 :
319 : INTEGER :: i, j
320 :
321 70 : IF (ALLOCATED(ri_data%kp_cost)) DEALLOCATE (ri_data%kp_cost)
322 70 : IF (ALLOCATED(ri_data%idx_to_img)) DEALLOCATE (ri_data%idx_to_img)
323 70 : IF (ALLOCATED(ri_data%img_to_idx)) DEALLOCATE (ri_data%img_to_idx)
324 70 : IF (ALLOCATED(ri_data%present_images)) DEALLOCATE (ri_data%present_images)
325 70 : IF (ALLOCATED(ri_data%img_to_RI_cell)) DEALLOCATE (ri_data%img_to_RI_cell)
326 70 : IF (ALLOCATED(ri_data%RI_cell_to_img)) DEALLOCATE (ri_data%RI_cell_to_img)
327 :
328 70 : IF (ALLOCATED(ri_data%kp_mat_2c_pot)) THEN
329 540 : DO j = 1, SIZE(ri_data%kp_mat_2c_pot, 2)
330 1060 : DO i = 1, SIZE(ri_data%kp_mat_2c_pot, 1)
331 1040 : CALL dbcsr_release(ri_data%kp_mat_2c_pot(i, j))
332 : END DO
333 : END DO
334 20 : DEALLOCATE (ri_data%kp_mat_2c_pot)
335 : END IF
336 :
337 70 : IF (ALLOCATED(ri_data%kp_t_3c_int)) THEN
338 540 : DO i = 1, SIZE(ri_data%kp_t_3c_int)
339 540 : CALL dbt_destroy(ri_data%kp_t_3c_int(i))
340 : END DO
341 540 : DEALLOCATE (ri_data%kp_t_3c_int)
342 : END IF
343 :
344 70 : IF (ALLOCATED(ri_data%t_2c_inv)) THEN
345 160 : DO j = 1, SIZE(ri_data%t_2c_inv, 2)
346 250 : DO i = 1, SIZE(ri_data%t_2c_inv, 1)
347 180 : CALL dbt_destroy(ri_data%t_2c_inv(i, j))
348 : END DO
349 : END DO
350 160 : DEALLOCATE (ri_data%t_2c_inv)
351 : END IF
352 :
353 70 : IF (ALLOCATED(ri_data%t_2c_int)) THEN
354 160 : DO j = 1, SIZE(ri_data%t_2c_int, 2)
355 250 : DO i = 1, SIZE(ri_data%t_2c_int, 1)
356 180 : CALL dbt_destroy(ri_data%t_2c_int(i, j))
357 : END DO
358 : END DO
359 160 : DEALLOCATE (ri_data%t_2c_int)
360 : END IF
361 :
362 70 : IF (ALLOCATED(ri_data%t_2c_pot)) THEN
363 160 : DO j = 1, SIZE(ri_data%t_2c_pot, 2)
364 250 : DO i = 1, SIZE(ri_data%t_2c_pot, 1)
365 180 : CALL dbt_destroy(ri_data%t_2c_pot(i, j))
366 : END DO
367 : END DO
368 160 : DEALLOCATE (ri_data%t_2c_pot)
369 : END IF
370 :
371 70 : IF (ALLOCATED(ri_data%t_3c_int_ctr_1)) THEN
372 640 : DO j = 1, SIZE(ri_data%t_3c_int_ctr_1, 2)
373 1210 : DO i = 1, SIZE(ri_data%t_3c_int_ctr_1, 1)
374 1140 : CALL dbt_destroy(ri_data%t_3c_int_ctr_1(i, j))
375 : END DO
376 : END DO
377 640 : DEALLOCATE (ri_data%t_3c_int_ctr_1)
378 : END IF
379 :
380 70 : IF (ALLOCATED(ri_data%t_3c_int_ctr_2)) THEN
381 140 : DO j = 1, SIZE(ri_data%t_3c_int_ctr_2, 2)
382 210 : DO i = 1, SIZE(ri_data%t_3c_int_ctr_2, 1)
383 140 : CALL dbt_destroy(ri_data%t_3c_int_ctr_2(i, j))
384 : END DO
385 : END DO
386 140 : DEALLOCATE (ri_data%t_3c_int_ctr_2)
387 : END IF
388 :
389 70 : IF (ALLOCATED(ri_data%rho_ao_t)) THEN
390 640 : DO j = 1, SIZE(ri_data%rho_ao_t, 2)
391 1428 : DO i = 1, SIZE(ri_data%rho_ao_t, 1)
392 1358 : CALL dbt_destroy(ri_data%rho_ao_t(i, j))
393 : END DO
394 : END DO
395 858 : DEALLOCATE (ri_data%rho_ao_t)
396 : END IF
397 :
398 70 : IF (ALLOCATED(ri_data%ks_t)) THEN
399 640 : DO j = 1, SIZE(ri_data%ks_t, 2)
400 1428 : DO i = 1, SIZE(ri_data%ks_t, 1)
401 1358 : CALL dbt_destroy(ri_data%ks_t(i, j))
402 : END DO
403 : END DO
404 858 : DEALLOCATE (ri_data%ks_t)
405 : END IF
406 :
407 70 : END SUBROUTINE cleanup_kp
408 :
409 : ! **************************************************************************************************
410 : !> \brief Update the KS matrices for each real-space image
411 : !> \param qs_env ...
412 : !> \param ri_data ...
413 : !> \param ks_matrix ...
414 : !> \param ehfx ...
415 : !> \param rho_ao ...
416 : !> \param geometry_did_change ...
417 : !> \param nspins ...
418 : !> \param hf_fraction ...
419 : ! **************************************************************************************************
420 190 : SUBROUTINE hfx_ri_update_ks_kp(qs_env, ri_data, ks_matrix, ehfx, rho_ao, &
421 : geometry_did_change, nspins, hf_fraction)
422 :
423 : TYPE(qs_environment_type), POINTER :: qs_env
424 : TYPE(hfx_ri_type), INTENT(INOUT) :: ri_data
425 : TYPE(dbcsr_p_type), DIMENSION(:, :), POINTER :: ks_matrix
426 : REAL(KIND=dp), INTENT(OUT) :: ehfx
427 : TYPE(dbcsr_p_type), DIMENSION(:, :), POINTER :: rho_ao
428 : LOGICAL, INTENT(IN) :: geometry_did_change
429 : INTEGER, INTENT(IN) :: nspins
430 : REAL(KIND=dp), INTENT(IN) :: hf_fraction
431 :
432 : CHARACTER(LEN=*), PARAMETER :: routineN = 'hfx_ri_update_ks_kp'
433 :
434 : INTEGER :: b_img, batch_size, group_size, handle, handle2, i_batch, i_img, i_spin, iatom, &
435 : iblk, igroup, jatom, mb_img, n_batch_nze, natom, ngroups, nimg, nimg_nze
436 : INTEGER(int_8) :: nflop, nze
437 190 : INTEGER, ALLOCATABLE, DIMENSION(:) :: batch_ranges_at, batch_ranges_nze, &
438 190 : idx_to_at_AO
439 190 : INTEGER, ALLOCATABLE, DIMENSION(:, :) :: iapc_pairs
440 190 : INTEGER, ALLOCATABLE, DIMENSION(:, :, :) :: sparsity_pattern
441 : LOGICAL :: use_delta_p
442 : REAL(dp) :: etmp, fac, occ, pfac, pref, t1, t2, t3, &
443 : t4
444 : TYPE(cp_blacs_env_type), POINTER :: blacs_env_sub
445 : TYPE(dbcsr_type) :: ks_desymm, rho_desymm, tmp
446 190 : TYPE(dbcsr_type), ALLOCATABLE, DIMENSION(:) :: mat_2c_pot
447 : TYPE(dbcsr_type), POINTER :: dbcsr_template
448 190 : TYPE(dbt_type), ALLOCATABLE, DIMENSION(:) :: ks_t_split, t_2c_ao_tmp, t_2c_work, &
449 190 : t_3c_int, t_3c_work_2, t_3c_work_3
450 190 : TYPE(dbt_type), ALLOCATABLE, DIMENSION(:, :) :: ks_t, ks_t_sub, t_3c_apc, t_3c_apc_sub
451 : TYPE(mp_para_env_type), POINTER :: para_env, para_env_sub
452 : TYPE(section_vals_type), POINTER :: hfx_section
453 :
454 190 : NULLIFY (para_env, para_env_sub, blacs_env_sub, hfx_section, dbcsr_template)
455 :
456 190 : CALL cite_reference(Bussy2024)
457 :
458 190 : CALL timeset(routineN, handle)
459 :
460 190 : CALL get_qs_env(qs_env, para_env=para_env, natom=natom)
461 :
462 190 : IF (nspins == 1) THEN
463 112 : fac = 0.5_dp*hf_fraction
464 : ELSE
465 78 : fac = 1.0_dp*hf_fraction
466 : END IF
467 :
468 190 : IF (geometry_did_change) THEN
469 70 : CALL hfx_ri_pre_scf_kp(ks_matrix(1, 1)%matrix, ri_data, qs_env)
470 : END IF
471 190 : nimg = ri_data%nimg
472 190 : nimg_nze = ri_data%nimg_nze
473 :
474 : !We need to calculate the KS matrix for each periodic cell with index b: F_mu^0,nu^b
475 : !F_mu^0,nu^b = -0.5 sum_a,c P_sigma^0,lambda^c (mu^0, sigma^a| P^0) V_P^0,Q^b (Q^b| nu^b lambda^a+c)
476 : !with V_P^0,Q^b = (P^0|R^0)^-1 * (R^0|S^b) * (S^b|Q^b)^-1
477 :
478 : !We use a local RI basis set for each atom in the system, which inlcudes RI basis elements for
479 : !each neighboring atom standing within the KIND radius (decay of Gaussian with smallest exponent)
480 :
481 : !We also limit the number of periodic images we consider accorrding to the HFX potentail in the
482 : !RI basis, because if V_P^0,Q^b is zero everywhere, then image b can be ignored (RI basis less diffuse)
483 :
484 : !We manage to calculate each KS matrix doing a double loop on iamges, and a double loop on atoms
485 : !First, we pre-contract and store P_sigma^0,lambda^c (mu^0, sigma^a| P^0) (P^0|R^0)^-1 into T_mu^0,lambda^a+c,P^0
486 : !Then, we loop over b_img, iatom, jatom to get (R^0|S^b)
487 : !Finally, we do an additional loop over a+c images where we do (R^0|S^b) (S^b|Q^b)^-1 (Q^b| nu^b lambda^a+c)
488 : !and the final contraction with T_mu^0,lambda^a+c,P^0
489 :
490 : !Note that the 3-center integrals are pre-contracted with the RI metric, and that the same tensor can be used
491 : !(mu^0, sigma^a| P^0) (P^0|R^0) <===> (S^b|Q^b)^-1 (Q^b| nu^b lambda^a+c) by relabelling the images
492 :
493 190 : hfx_section => section_vals_get_subs_vals(qs_env%input, "DFT%XC%HF%RI")
494 190 : CALL section_vals_val_get(hfx_section, "KP_USE_DELTA_P", l_val=use_delta_p)
495 :
496 : !By default, build the density tensor based on the difference of this SCF P and that of the prev. SCF
497 190 : pfac = -1.0_dp
498 190 : IF (.NOT. use_delta_p) pfac = 0.0_dp
499 190 : CALL get_pmat_images(ri_data%rho_ao_t, rho_ao, pfac, ri_data, qs_env)
500 :
501 13412 : ALLOCATE (ks_t(nspins, nimg))
502 4994 : DO i_img = 1, nimg
503 11322 : DO i_spin = 1, nspins
504 11132 : CALL dbt_create(ri_data%ks_t(1, 1), ks_t(i_spin, i_img))
505 : END DO
506 : END DO
507 :
508 570 : ALLOCATE (idx_to_at_AO(SIZE(ri_data%bsizes_AO_split)))
509 190 : CALL get_idx_to_atom(idx_to_at_AO, ri_data%bsizes_AO_split, ri_data%bsizes_AO)
510 :
511 : !First we calculate and store T^1_mu^0,lambda^a+c,P = P_mu^0,lambda^c * (mu_0 sigma^a | P^0) (P^0|R^0)^-1
512 : !To avoid doing nimg**2 tiny contractions that do not scale well with a large number of CPUs,
513 : !we instead do a single loop over the a+c image index. For each a+c, we get a list of allowed
514 : !combination of a,c indices. Then we build TAS tensors P_mu^0,lambda^c with all concerned c's
515 : !and (mu^0 sigma^a | P^0)*(P^0|R^0)^-1 with all a's. Then we perform a single contraction with larger tensors,
516 : !were the sum over a,c is automatically taken care of
517 13222 : ALLOCATE (t_3c_apc(nspins, nimg))
518 4994 : DO i_img = 1, nimg
519 11322 : DO i_spin = 1, nspins
520 11132 : CALL dbt_create(ri_data%t_3c_int_ctr_2(1, 1), t_3c_apc(i_spin, i_img))
521 : END DO
522 : END DO
523 190 : CALL contract_pmat_3c(t_3c_apc, ri_data%rho_ao_t, ri_data, qs_env)
524 :
525 190 : hfx_section => section_vals_get_subs_vals(qs_env%input, "DFT%XC%HF%RI")
526 190 : CALL section_vals_val_get(hfx_section, "KP_NGROUPS", i_val=ngroups)
527 190 : CALL section_vals_val_get(hfx_section, "KP_STACK_SIZE", i_val=batch_size)
528 190 : ri_data%kp_stack_size = batch_size
529 :
530 190 : IF (MOD(para_env%num_pe, ngroups) .NE. 0) THEN
531 0 : CPWARN("KP_NGROUPS must be an integer divisor of the total number of MPI ranks. It was set to 1.")
532 0 : ngroups = 1
533 0 : CALL section_vals_val_set(hfx_section, "KP_NGROUPS", i_val=ngroups)
534 : END IF
535 190 : IF ((MOD(ngroups, natom) .NE. 0) .AND. (MOD(natom, ngroups) .NE. 0) .AND. geometry_did_change) THEN
536 0 : IF (ngroups > 1) THEN
537 0 : CPWARN("Better load balancing is reached if NGROUPS is a multiple/divisor of the number of atoms")
538 : END IF
539 : END IF
540 190 : group_size = para_env%num_pe/ngroups
541 190 : igroup = para_env%mepos/group_size
542 :
543 190 : ALLOCATE (para_env_sub)
544 190 : CALL para_env_sub%from_split(para_env, igroup)
545 190 : CALL cp_blacs_env_create(blacs_env_sub, para_env_sub)
546 :
547 : ! The sparsity pattern of each iatom, jatom pair, on each b_img, and on which subgroup
548 950 : ALLOCATE (sparsity_pattern(natom, natom, nimg))
549 190 : CALL get_sparsity_pattern(sparsity_pattern, ri_data, qs_env)
550 190 : CALL get_sub_dist(sparsity_pattern, ngroups, ri_data)
551 :
552 : !Get all the required tensors in the subgroups
553 24106 : ALLOCATE (mat_2c_pot(nimg), ks_t_sub(nspins, nimg), t_2c_ao_tmp(1), ks_t_split(2), t_2c_work(3))
554 : CALL get_subgroup_2c_tensors(mat_2c_pot, t_2c_work, t_2c_ao_tmp, ks_t_split, ks_t_sub, &
555 190 : group_size, ngroups, para_env, para_env_sub, ri_data)
556 :
557 24296 : ALLOCATE (t_3c_int(nimg), t_3c_apc_sub(nspins, nimg), t_3c_work_2(3), t_3c_work_3(3))
558 : CALL get_subgroup_3c_tensors(t_3c_int, t_3c_work_2, t_3c_work_3, t_3c_apc, t_3c_apc_sub, &
559 190 : group_size, ngroups, para_env, para_env_sub, ri_data)
560 :
561 : !We go atom by atom, therefore there is an automatic batching along that direction
562 : !Also, because we stack the 3c tensors nimg times, we naturally do some batching there too
563 570 : ALLOCATE (batch_ranges_at(natom + 1))
564 190 : batch_ranges_at(natom + 1) = SIZE(ri_data%bsizes_AO_split) + 1
565 190 : iatom = 0
566 928 : DO iblk = 1, SIZE(ri_data%bsizes_AO_split)
567 928 : IF (idx_to_at_AO(iblk) == iatom + 1) THEN
568 380 : iatom = iatom + 1
569 380 : batch_ranges_at(iatom) = iblk
570 : END IF
571 : END DO
572 :
573 190 : n_batch_nze = nimg_nze/batch_size
574 190 : IF (MODULO(nimg_nze, batch_size) .NE. 0) n_batch_nze = n_batch_nze + 1
575 570 : ALLOCATE (batch_ranges_nze(n_batch_nze + 1))
576 404 : DO i_batch = 1, n_batch_nze
577 404 : batch_ranges_nze(i_batch) = (i_batch - 1)*batch_size + 1
578 : END DO
579 190 : batch_ranges_nze(n_batch_nze + 1) = nimg_nze + 1
580 :
581 190 : CALL dbt_batched_contract_init(t_3c_work_3(1), batch_range_2=batch_ranges_at)
582 190 : CALL dbt_batched_contract_init(t_3c_work_3(2), batch_range_2=batch_ranges_at)
583 190 : CALL dbt_batched_contract_init(t_3c_work_2(1), batch_range_1=batch_ranges_at)
584 190 : CALL dbt_batched_contract_init(t_3c_work_2(2), batch_range_1=batch_ranges_at)
585 :
586 190 : t1 = m_walltime()
587 33818 : ri_data%kp_cost(:, :, :) = 0.0_dp
588 570 : ALLOCATE (iapc_pairs(nimg, 2))
589 4994 : DO b_img = 1, nimg
590 4804 : CALL dbt_batched_contract_init(ks_t_split(1))
591 4804 : CALL dbt_batched_contract_init(ks_t_split(2))
592 14412 : DO jatom = 1, natom
593 33628 : DO iatom = 1, natom
594 19216 : IF (.NOT. sparsity_pattern(iatom, jatom, b_img) == igroup) CYCLE
595 3232 : pref = 1.0_dp
596 3232 : IF (iatom == jatom .AND. b_img == 1) pref = 0.5_dp
597 :
598 : !measure the cost of the given i, j, b configuration
599 3232 : t3 = m_walltime()
600 :
601 : !Get the proper HFX potential 2c integrals (R_i^0|S_j^b)
602 3232 : CALL timeset(routineN//"_2c", handle2)
603 : CALL get_ext_2c_int(t_2c_work(1), mat_2c_pot, iatom, jatom, b_img, ri_data, qs_env, &
604 : blacs_env_ext=blacs_env_sub, para_env_ext=para_env_sub, &
605 3232 : dbcsr_template=dbcsr_template)
606 3232 : CALL dbt_copy(t_2c_work(1), t_2c_work(2), move_data=.TRUE.) !move to split blocks
607 3232 : CALL dbt_filter(t_2c_work(2), ri_data%filter_eps)
608 3232 : CALL timestop(handle2)
609 :
610 3232 : CALL dbt_batched_contract_init(t_2c_work(2))
611 3232 : CALL get_iapc_pairs(iapc_pairs, b_img, ri_data, qs_env)
612 3232 : CALL timeset(routineN//"_3c", handle2)
613 :
614 : !Stack the (S^b|Q^b)^-1 * (Q^b| nu^b lambda^a+c) integrals over a+c and multiply by (R_i^0|S_j^b)
615 7331 : DO i_batch = 1, n_batch_nze
616 : CALL fill_3c_stack(t_3c_work_3(3), t_3c_int, iapc_pairs(:, 1), 3, ri_data, &
617 : filter_at=jatom, filter_dim=2, idx_to_at=idx_to_at_AO, &
618 12297 : img_bounds=[batch_ranges_nze(i_batch), batch_ranges_nze(i_batch + 1)])
619 4099 : CALL dbt_copy(t_3c_work_3(3), t_3c_work_3(1), move_data=.TRUE.)
620 :
621 : CALL dbt_contract(1.0_dp, t_2c_work(2), t_3c_work_3(1), &
622 : 0.0_dp, t_3c_work_3(2), map_1=[1], map_2=[2, 3], &
623 : contract_1=[2], notcontract_1=[1], &
624 : contract_2=[1], notcontract_2=[2, 3], &
625 4099 : filter_eps=ri_data%filter_eps, flop=nflop)
626 4099 : ri_data%dbcsr_nflop = ri_data%dbcsr_nflop + nflop
627 4099 : CALL dbt_copy(t_3c_work_3(2), t_3c_work_2(2), order=[2, 1, 3], move_data=.TRUE.)
628 4099 : CALL dbt_copy(t_3c_work_3(3), t_3c_work_3(1))
629 :
630 : !Stack the P_sigma^a,lambda^a+c * (mu^0 sigma^a | P^0)*(P^0|R^0)^-1 integrals over a+c and contract
631 : !to get the final block of the KS matrix
632 12586 : DO i_spin = 1, nspins
633 : CALL fill_3c_stack(t_3c_work_2(3), t_3c_apc_sub(i_spin, :), iapc_pairs(:, 2), 3, &
634 : ri_data, filter_at=iatom, filter_dim=1, idx_to_at=idx_to_at_AO, &
635 15765 : img_bounds=[batch_ranges_nze(i_batch), batch_ranges_nze(i_batch + 1)])
636 5255 : CALL get_tensor_occupancy(t_3c_work_2(3), nze, occ)
637 5255 : IF (nze == 0) CYCLE
638 5235 : CALL dbt_copy(t_3c_work_2(3), t_3c_work_2(1), move_data=.TRUE.)
639 : CALL dbt_contract(-pref*fac, t_3c_work_2(1), t_3c_work_2(2), &
640 : 1.0_dp, ks_t_split(i_spin), map_1=[1], map_2=[2], &
641 : contract_1=[2, 3], notcontract_1=[1], &
642 : contract_2=[2, 3], notcontract_2=[1], &
643 : filter_eps=ri_data%filter_eps, &
644 5235 : move_data=i_spin == nspins, flop=nflop)
645 14589 : ri_data%dbcsr_nflop = ri_data%dbcsr_nflop + nflop
646 : END DO
647 : END DO !i_batch
648 3232 : CALL timestop(handle2)
649 3232 : CALL dbt_batched_contract_finalize(t_2c_work(2))
650 :
651 3232 : t4 = m_walltime()
652 35288 : ri_data%kp_cost(iatom, jatom, b_img) = t4 - t3
653 : END DO !iatom
654 : END DO !jatom
655 4804 : CALL dbt_batched_contract_finalize(ks_t_split(1))
656 4804 : CALL dbt_batched_contract_finalize(ks_t_split(2))
657 :
658 11322 : DO i_spin = 1, nspins
659 6328 : CALL dbt_copy(ks_t_split(i_spin), t_2c_ao_tmp(1), move_data=.TRUE.)
660 11132 : CALL dbt_copy(t_2c_ao_tmp(1), ks_t_sub(i_spin, b_img), summation=.TRUE.)
661 : END DO
662 : END DO !b_img
663 190 : CALL dbt_batched_contract_finalize(t_3c_work_3(1))
664 190 : CALL dbt_batched_contract_finalize(t_3c_work_3(2))
665 190 : CALL dbt_batched_contract_finalize(t_3c_work_2(1))
666 190 : CALL dbt_batched_contract_finalize(t_3c_work_2(2))
667 190 : CALL para_env%sync()
668 190 : CALL para_env%sum(ri_data%dbcsr_nflop)
669 190 : CALL para_env%sum(ri_data%kp_cost)
670 190 : t2 = m_walltime()
671 190 : ri_data%dbcsr_time = ri_data%dbcsr_time + t2 - t1
672 :
673 : !transfer KS tensor from subgroup to main group
674 190 : CALL gather_ks_matrix(ks_t, ks_t_sub, group_size, sparsity_pattern, para_env, ri_data)
675 :
676 : !Keep the 3c integrals on the subgroups to avoid communication at next SCF step
677 4994 : DO i_img = 1, nimg
678 4994 : CALL dbt_copy(t_3c_int(i_img), ri_data%kp_t_3c_int(i_img), move_data=.TRUE.)
679 : END DO
680 :
681 : !clean-up subgroup tensors
682 190 : CALL dbt_destroy(t_2c_ao_tmp(1))
683 190 : CALL dbt_destroy(ks_t_split(1))
684 190 : CALL dbt_destroy(ks_t_split(2))
685 190 : CALL dbt_destroy(t_2c_work(1))
686 190 : CALL dbt_destroy(t_2c_work(2))
687 190 : CALL dbt_destroy(t_3c_work_2(1))
688 190 : CALL dbt_destroy(t_3c_work_2(2))
689 190 : CALL dbt_destroy(t_3c_work_2(3))
690 190 : CALL dbt_destroy(t_3c_work_3(1))
691 190 : CALL dbt_destroy(t_3c_work_3(2))
692 190 : CALL dbt_destroy(t_3c_work_3(3))
693 4994 : DO i_img = 1, nimg
694 4804 : CALL dbt_destroy(t_3c_int(i_img))
695 4804 : CALL dbcsr_release(mat_2c_pot(i_img))
696 11322 : DO i_spin = 1, nspins
697 6328 : CALL dbt_destroy(t_3c_apc_sub(i_spin, i_img))
698 11132 : CALL dbt_destroy(ks_t_sub(i_spin, i_img))
699 : END DO
700 : END DO
701 190 : IF (ASSOCIATED(dbcsr_template)) THEN
702 190 : CALL dbcsr_release(dbcsr_template)
703 190 : DEALLOCATE (dbcsr_template)
704 : END IF
705 :
706 : !End of subgroup parallelization
707 190 : CALL cp_blacs_env_release(blacs_env_sub)
708 190 : CALL para_env_sub%free()
709 190 : DEALLOCATE (para_env_sub)
710 :
711 : !Currently, rho_ao_t holds the density difference (wrt to pref SCF step).
712 : !ks_t also hold that diff, while only having half the blocks => need to add to prev ks_t and symmetrize
713 : !We need the full thing for the energy, on the next SCF step
714 190 : CALL get_pmat_images(ri_data%rho_ao_t, rho_ao, 0.0_dp, ri_data, qs_env)
715 458 : DO i_spin = 1, nspins
716 6786 : DO b_img = 1, nimg
717 6328 : CALL dbt_copy(ks_t(i_spin, b_img), ri_data%ks_t(i_spin, b_img), summation=.TRUE.)
718 :
719 : !desymmetrize
720 6328 : mb_img = get_opp_index(b_img, qs_env)
721 6596 : IF (mb_img > 0 .AND. mb_img .LE. nimg) THEN
722 5708 : CALL dbt_copy(ks_t(i_spin, mb_img), ri_data%ks_t(i_spin, b_img), order=[2, 1], summation=.TRUE.)
723 : END IF
724 : END DO
725 : END DO
726 4994 : DO b_img = 1, nimg
727 11322 : DO i_spin = 1, nspins
728 11132 : CALL dbt_destroy(ks_t(i_spin, b_img))
729 : END DO
730 : END DO
731 :
732 : !calculate the energy
733 190 : CALL dbt_create(ri_data%ks_t(1, 1), t_2c_ao_tmp(1))
734 190 : CALL dbcsr_create(tmp, template=ks_matrix(1, 1)%matrix, matrix_type=dbcsr_type_symmetric)
735 190 : CALL dbcsr_create(ks_desymm, template=ks_matrix(1, 1)%matrix, matrix_type=dbcsr_type_no_symmetry)
736 190 : CALL dbcsr_create(rho_desymm, template=ks_matrix(1, 1)%matrix, matrix_type=dbcsr_type_no_symmetry)
737 190 : ehfx = 0.0_dp
738 4994 : DO i_img = 1, nimg
739 11322 : DO i_spin = 1, nspins
740 6328 : CALL dbt_filter(ri_data%ks_t(i_spin, i_img), ri_data%filter_eps)
741 6328 : CALL dbt_copy(ri_data%ks_t(i_spin, i_img), t_2c_ao_tmp(1))
742 6328 : CALL dbt_copy_tensor_to_matrix(t_2c_ao_tmp(1), ks_desymm)
743 6328 : CALL dbt_copy_tensor_to_matrix(t_2c_ao_tmp(1), tmp)
744 6328 : CALL dbcsr_add(ks_matrix(i_spin, i_img)%matrix, tmp, 1.0_dp, 1.0_dp)
745 :
746 6328 : CALL dbt_copy(ri_data%rho_ao_t(i_spin, i_img), t_2c_ao_tmp(1))
747 6328 : CALL dbt_copy_tensor_to_matrix(t_2c_ao_tmp(1), rho_desymm)
748 :
749 6328 : CALL dbcsr_dot(ks_desymm, rho_desymm, etmp)
750 6328 : ehfx = ehfx + 0.5_dp*etmp
751 :
752 11132 : IF (.NOT. use_delta_p) CALL dbt_clear(ri_data%ks_t(i_spin, i_img))
753 : END DO
754 : END DO
755 190 : CALL dbcsr_release(rho_desymm)
756 190 : CALL dbcsr_release(ks_desymm)
757 190 : CALL dbcsr_release(tmp)
758 190 : CALL dbt_destroy(t_2c_ao_tmp(1))
759 :
760 190 : CALL timestop(handle)
761 :
762 33346 : END SUBROUTINE hfx_ri_update_ks_kp
763 :
764 : ! **************************************************************************************************
765 : !> \brief Update the K-points RI-HFX forces
766 : !> \param qs_env ...
767 : !> \param ri_data ...
768 : !> \param nspins ...
769 : !> \param hf_fraction ...
770 : !> \param rho_ao ...
771 : !> \param use_virial ...
772 : !> \note Because this routine uses stored quantities calculated in the energy calculation, they should
773 : !> always be called by pairs, and with the same input densities
774 : ! **************************************************************************************************
775 42 : SUBROUTINE hfx_ri_update_forces_kp(qs_env, ri_data, nspins, hf_fraction, rho_ao, use_virial)
776 :
777 : TYPE(qs_environment_type), POINTER :: qs_env
778 : TYPE(hfx_ri_type), INTENT(INOUT) :: ri_data
779 : INTEGER, INTENT(IN) :: nspins
780 : REAL(KIND=dp), INTENT(IN) :: hf_fraction
781 : TYPE(dbcsr_p_type), DIMENSION(:, :), POINTER :: rho_ao
782 : LOGICAL, INTENT(IN), OPTIONAL :: use_virial
783 :
784 : CHARACTER(LEN=*), PARAMETER :: routineN = 'hfx_ri_update_forces_kp'
785 :
786 : INTEGER :: b_img, batch_size, group_size, handle, handle2, i_batch, i_img, i_loop, i_spin, &
787 : i_xyz, iatom, iblk, igroup, j_xyz, jatom, k_xyz, n_batch, natom, ngroups, nimg, nimg_nze
788 : INTEGER(int_8) :: nflop, nze
789 42 : INTEGER, ALLOCATABLE, DIMENSION(:) :: atom_of_kind, batch_ranges_at, &
790 42 : batch_ranges_nze, dist1, dist2, &
791 42 : i_images, idx_to_at_AO, idx_to_at_RI, &
792 42 : kind_of
793 42 : INTEGER, ALLOCATABLE, DIMENSION(:, :) :: iapc_pairs
794 42 : INTEGER, ALLOCATABLE, DIMENSION(:, :, :) :: force_pattern, sparsity_pattern
795 : INTEGER, DIMENSION(2, 1) :: bounds_iat, bounds_jat
796 : LOGICAL :: use_virial_prv
797 : REAL(dp) :: fac, occ, pref, t1, t2
798 : REAL(dp), DIMENSION(3, 3) :: work_virial
799 42 : TYPE(atomic_kind_type), DIMENSION(:), POINTER :: atomic_kind_set
800 : TYPE(cell_type), POINTER :: cell
801 : TYPE(cp_blacs_env_type), POINTER :: blacs_env_sub
802 42 : TYPE(dbcsr_type), ALLOCATABLE, DIMENSION(:) :: mat_2c_pot
803 42 : TYPE(dbcsr_type), ALLOCATABLE, DIMENSION(:, :) :: mat_der_pot, mat_der_pot_sub
804 : TYPE(dbcsr_type), POINTER :: dbcsr_template
805 714 : TYPE(dbt_type) :: t_2c_R, t_2c_R_split
806 42 : TYPE(dbt_type), ALLOCATABLE, DIMENSION(:) :: t_2c_bint, t_2c_binv, t_2c_der_pot, &
807 84 : t_2c_inv, t_2c_metric, t_2c_work, &
808 42 : t_3c_der_stack, t_3c_work_2, &
809 42 : t_3c_work_3
810 42 : TYPE(dbt_type), ALLOCATABLE, DIMENSION(:, :) :: rho_ao_t, rho_ao_t_sub, t_2c_der_metric, &
811 84 : t_2c_der_metric_sub, t_3c_apc, t_3c_apc_sub, t_3c_der_AO, t_3c_der_AO_sub, t_3c_der_RI, &
812 42 : t_3c_der_RI_sub
813 : TYPE(mp_para_env_type), POINTER :: para_env, para_env_sub
814 42 : TYPE(particle_type), DIMENSION(:), POINTER :: particle_set
815 42 : TYPE(qs_force_type), DIMENSION(:), POINTER :: force
816 : TYPE(section_vals_type), POINTER :: hfx_section
817 : TYPE(virial_type), POINTER :: virial
818 :
819 42 : NULLIFY (para_env, para_env_sub, hfx_section, blacs_env_sub, dbcsr_template, force, atomic_kind_set, &
820 42 : virial, particle_set, cell)
821 :
822 42 : CALL timeset(routineN, handle)
823 :
824 42 : use_virial_prv = .FALSE.
825 42 : IF (PRESENT(use_virial)) use_virial_prv = use_virial
826 :
827 42 : IF (nspins == 1) THEN
828 28 : fac = 0.5_dp*hf_fraction
829 : ELSE
830 14 : fac = 1.0_dp*hf_fraction
831 : END IF
832 :
833 : CALL get_qs_env(qs_env, natom=natom, para_env=para_env, force=force, cell=cell, virial=virial, &
834 42 : atomic_kind_set=atomic_kind_set, particle_set=particle_set)
835 42 : CALL get_atomic_kind_set(atomic_kind_set, kind_of=kind_of, atom_of_kind=atom_of_kind)
836 :
837 126 : ALLOCATE (idx_to_at_AO(SIZE(ri_data%bsizes_AO_split)))
838 42 : CALL get_idx_to_atom(idx_to_at_AO, ri_data%bsizes_AO_split, ri_data%bsizes_AO)
839 :
840 126 : ALLOCATE (idx_to_at_RI(SIZE(ri_data%bsizes_RI_split)))
841 42 : CALL get_idx_to_atom(idx_to_at_RI, ri_data%bsizes_RI_split, ri_data%bsizes_RI)
842 :
843 42 : nimg = ri_data%nimg
844 11190 : ALLOCATE (t_3c_der_RI(nimg, 3), t_3c_der_AO(nimg, 3), mat_der_pot(nimg, 3), t_2c_der_metric(natom, 3))
845 :
846 : !We assume that the integrals are available from the SCF
847 : !pre-calculate the derivs. 3c tensors as (P^0| sigma^a mu^0), with t_3c_der_AO holding deriv wrt mu^0
848 42 : CALL precalc_derivatives(t_3c_der_RI, t_3c_der_AO, mat_der_pot, t_2c_der_metric, ri_data, qs_env)
849 :
850 : !Calculate the density matrix at each image
851 2690 : ALLOCATE (rho_ao_t(nspins, nimg))
852 : CALL create_2c_tensor(rho_ao_t(1, 1), dist1, dist2, ri_data%pgrid_2d, &
853 : ri_data%bsizes_AO_split, ri_data%bsizes_AO_split, &
854 42 : name="(AO | AO)")
855 42 : DEALLOCATE (dist1, dist2)
856 42 : IF (nspins == 2) CALL dbt_create(rho_ao_t(1, 1), rho_ao_t(2, 1))
857 1010 : DO i_img = 2, nimg
858 2130 : DO i_spin = 1, nspins
859 2088 : CALL dbt_create(rho_ao_t(1, 1), rho_ao_t(i_spin, i_img))
860 : END DO
861 : END DO
862 42 : CALL get_pmat_images(rho_ao_t, rho_ao, 0.0_dp, ri_data, qs_env)
863 :
864 : !Contract integrals with the density matrix
865 2690 : ALLOCATE (t_3c_apc(nspins, nimg))
866 1052 : DO i_img = 1, nimg
867 2228 : DO i_spin = 1, nspins
868 2186 : CALL dbt_create(ri_data%t_3c_int_ctr_2(1, 1), t_3c_apc(i_spin, i_img))
869 : END DO
870 : END DO
871 42 : CALL contract_pmat_3c(t_3c_apc, rho_ao_t, ri_data, qs_env)
872 :
873 : !Setup the subgroups
874 42 : hfx_section => section_vals_get_subs_vals(qs_env%input, "DFT%XC%HF%RI")
875 42 : CALL section_vals_val_get(hfx_section, "KP_NGROUPS", i_val=ngroups)
876 42 : group_size = para_env%num_pe/ngroups
877 42 : igroup = para_env%mepos/group_size
878 :
879 42 : ALLOCATE (para_env_sub)
880 42 : CALL para_env_sub%from_split(para_env, igroup)
881 42 : CALL cp_blacs_env_create(blacs_env_sub, para_env_sub)
882 :
883 : !Get the ususal sparsity pattern
884 210 : ALLOCATE (sparsity_pattern(natom, natom, nimg))
885 42 : CALL get_sparsity_pattern(sparsity_pattern, ri_data, qs_env)
886 42 : CALL get_sub_dist(sparsity_pattern, ngroups, ri_data)
887 :
888 : !Get the 2-center quantities in the subgroups (note: main group derivs are deleted wihtin)
889 0 : ALLOCATE (t_2c_inv(natom), mat_2c_pot(nimg), rho_ao_t_sub(nspins, nimg), t_2c_work(5), &
890 0 : t_2c_der_metric_sub(natom, 3), mat_der_pot_sub(nimg, 3), t_2c_bint(natom), &
891 10300 : t_2c_metric(natom), t_2c_binv(natom))
892 : CALL get_subgroup_2c_derivs(t_2c_inv, t_2c_bint, t_2c_metric, mat_2c_pot, t_2c_work, rho_ao_t, &
893 : rho_ao_t_sub, t_2c_der_metric, t_2c_der_metric_sub, mat_der_pot, &
894 42 : mat_der_pot_sub, group_size, ngroups, para_env, para_env_sub, ri_data)
895 42 : CALL dbt_create(t_2c_work(1), t_2c_R) !nRI x nRI
896 42 : CALL dbt_create(t_2c_work(5), t_2c_R_split) !nRI x nRI with split blocks
897 :
898 504 : ALLOCATE (t_2c_der_pot(3))
899 168 : DO i_xyz = 1, 3
900 168 : CALL dbt_create(t_2c_R, t_2c_der_pot(i_xyz))
901 : END DO
902 :
903 : !Get the 3-center quantities in the subgroups. The integrals and t_3c_apc already there
904 0 : ALLOCATE (t_3c_work_2(3), t_3c_work_3(4), t_3c_der_stack(6), t_3c_der_AO_sub(nimg, 3), &
905 11312 : t_3c_der_RI_sub(nimg, 3), t_3c_apc_sub(nspins, nimg))
906 : CALL get_subgroup_3c_derivs(t_3c_work_2, t_3c_work_3, t_3c_der_AO, t_3c_der_AO_sub, &
907 : t_3c_der_RI, t_3c_der_RI_sub, t_3c_apc, t_3c_apc_sub, t_3c_der_stack, &
908 42 : group_size, ngroups, para_env, para_env_sub, ri_data)
909 :
910 : !Set up batched contraction (go atom by atom)
911 126 : ALLOCATE (batch_ranges_at(natom + 1))
912 42 : batch_ranges_at(natom + 1) = SIZE(ri_data%bsizes_AO_split) + 1
913 42 : iatom = 0
914 212 : DO iblk = 1, SIZE(ri_data%bsizes_AO_split)
915 212 : IF (idx_to_at_AO(iblk) == iatom + 1) THEN
916 84 : iatom = iatom + 1
917 84 : batch_ranges_at(iatom) = iblk
918 : END IF
919 : END DO
920 :
921 42 : CALL dbt_batched_contract_init(t_3c_work_3(1), batch_range_2=batch_ranges_at)
922 42 : CALL dbt_batched_contract_init(t_3c_work_3(2), batch_range_2=batch_ranges_at)
923 42 : CALL dbt_batched_contract_init(t_3c_work_3(3), batch_range_2=batch_ranges_at)
924 42 : CALL dbt_batched_contract_init(t_3c_work_2(1), batch_range_1=batch_ranges_at)
925 42 : CALL dbt_batched_contract_init(t_3c_work_2(2), batch_range_1=batch_ranges_at)
926 :
927 : !Preparing for the stacking of 3c tensors
928 42 : nimg_nze = ri_data%nimg_nze
929 42 : batch_size = ri_data%kp_stack_size
930 42 : n_batch = nimg_nze/batch_size
931 42 : IF (MODULO(nimg_nze, batch_size) .NE. 0) n_batch = n_batch + 1
932 126 : ALLOCATE (batch_ranges_nze(n_batch + 1))
933 94 : DO i_batch = 1, n_batch
934 94 : batch_ranges_nze(i_batch) = (i_batch - 1)*batch_size + 1
935 : END DO
936 42 : batch_ranges_nze(n_batch + 1) = nimg_nze + 1
937 :
938 : !Applying the external bump to ((P|Q)_D + B*(P|Q)_OD*B)^-1 from left and right
939 : !And keep the bump on LHS only version as well, with B*M^-1 = (M^-1*B)^T
940 126 : DO iatom = 1, natom
941 84 : CALL dbt_create(t_2c_inv(iatom), t_2c_binv(iatom))
942 84 : CALL dbt_copy(t_2c_inv(iatom), t_2c_binv(iatom))
943 84 : CALL apply_bump(t_2c_binv(iatom), iatom, ri_data, qs_env, from_left=.TRUE., from_right=.FALSE.)
944 126 : CALL apply_bump(t_2c_inv(iatom), iatom, ri_data, qs_env, from_left=.TRUE., from_right=.TRUE.)
945 : END DO
946 :
947 42 : t1 = m_walltime()
948 42 : work_virial = 0.0_dp
949 210 : ALLOCATE (iapc_pairs(nimg, 2), i_images(nimg))
950 210 : ALLOCATE (force_pattern(natom, natom, nimg))
951 7112 : force_pattern(:, :, :) = -1
952 : !We proceed with 2 loops: one over the sparsity pattern from the SCF, one over the rest
953 : !We use the SCF cost model for the first loop, while we calculate the cost of the upcoming loop
954 126 : DO i_loop = 1, 2
955 2104 : DO b_img = 1, nimg
956 6144 : DO jatom = 1, natom
957 14140 : DO iatom = 1, natom
958 :
959 8080 : pref = -0.5_dp*fac
960 8080 : IF (i_loop == 1 .AND. (.NOT. sparsity_pattern(iatom, jatom, b_img) == igroup)) CYCLE
961 4619 : IF (i_loop == 2 .AND. (.NOT. force_pattern(iatom, jatom, b_img) == igroup)) CYCLE
962 :
963 : !Get the proper HFX potential 2c integrals (R_i^0|S_j^b), times (S_j^b|Q_j^b)^-1
964 1098 : CALL timeset(routineN//"_2c_1", handle2)
965 : CALL get_ext_2c_int(t_2c_work(1), mat_2c_pot, iatom, jatom, b_img, ri_data, qs_env, &
966 : blacs_env_ext=blacs_env_sub, para_env_ext=para_env_sub, &
967 1098 : dbcsr_template=dbcsr_template)
968 : CALL dbt_contract(1.0_dp, t_2c_work(1), t_2c_inv(jatom), &
969 : 0.0_dp, t_2c_work(2), map_1=[1], map_2=[2], &
970 : contract_1=[2], notcontract_1=[1], &
971 : contract_2=[1], notcontract_2=[2], &
972 1098 : filter_eps=ri_data%filter_eps, flop=nflop)
973 1098 : CALL dbt_copy(t_2c_work(2), t_2c_work(5), move_data=.TRUE.) !move to split blocks
974 1098 : CALL dbt_filter(t_2c_work(5), ri_data%filter_eps)
975 1098 : CALL timestop(handle2)
976 :
977 1098 : CALL timeset(routineN//"_3c", handle2)
978 5488 : bounds_iat(:, 1) = [SUM(ri_data%bsizes_AO(1:iatom - 1)) + 1, SUM(ri_data%bsizes_AO(1:iatom))]
979 5452 : bounds_jat(:, 1) = [SUM(ri_data%bsizes_AO(1:jatom - 1)) + 1, SUM(ri_data%bsizes_AO(1:jatom))]
980 1098 : CALL dbt_clear(t_2c_R_split)
981 :
982 2435 : DO i_spin = 1, nspins
983 2435 : CALL dbt_batched_contract_init(rho_ao_t_sub(i_spin, b_img))
984 : END DO
985 :
986 1098 : CALL get_iapc_pairs(iapc_pairs, b_img, ri_data, qs_env, i_images) !i = a+c-b
987 2667 : DO i_batch = 1, n_batch
988 :
989 : !Stack the 3c derivatives to take the trace later on
990 6276 : DO i_xyz = 1, 3
991 4707 : CALL dbt_clear(t_3c_der_stack(i_xyz))
992 : CALL fill_3c_stack(t_3c_der_stack(i_xyz), t_3c_der_RI_sub(:, i_xyz), &
993 : iapc_pairs(:, 1), 3, ri_data, filter_at=jatom, &
994 : filter_dim=2, idx_to_at=idx_to_at_AO, &
995 14121 : img_bounds=[batch_ranges_nze(i_batch), batch_ranges_nze(i_batch + 1)])
996 :
997 4707 : CALL dbt_clear(t_3c_der_stack(3 + i_xyz))
998 : CALL fill_3c_stack(t_3c_der_stack(3 + i_xyz), t_3c_der_AO_sub(:, i_xyz), &
999 : iapc_pairs(:, 1), 3, ri_data, filter_at=jatom, &
1000 : filter_dim=2, idx_to_at=idx_to_at_AO, &
1001 15690 : img_bounds=[batch_ranges_nze(i_batch), batch_ranges_nze(i_batch + 1)])
1002 : END DO
1003 :
1004 4475 : DO i_spin = 1, nspins
1005 : !stack the t_3c_apc tensors
1006 1808 : CALL dbt_clear(t_3c_work_2(3))
1007 : CALL fill_3c_stack(t_3c_work_2(3), t_3c_apc_sub(i_spin, :), iapc_pairs(:, 2), 3, &
1008 : ri_data, filter_at=iatom, filter_dim=1, idx_to_at=idx_to_at_AO, &
1009 5424 : img_bounds=[batch_ranges_nze(i_batch), batch_ranges_nze(i_batch + 1)])
1010 1808 : CALL get_tensor_occupancy(t_3c_work_2(3), nze, occ)
1011 1808 : IF (nze == 0) CYCLE
1012 1808 : CALL dbt_copy(t_3c_work_2(3), t_3c_work_2(1), move_data=.TRUE.)
1013 :
1014 : !Contract with the second density matrix: P_mu^0,nu^b * t_3c_apc,
1015 : !where t_3c_apc = P_sigma^a,lambda^a+c (mu^0 P^0 sigma^a) *(P^0|R^0)^-1 (stacked along a+c)
1016 : CALL dbt_contract(1.0_dp, rho_ao_t_sub(i_spin, b_img), t_3c_work_2(1), &
1017 : 0.0_dp, t_3c_work_2(2), map_1=[1], map_2=[2, 3], &
1018 : contract_1=[1], notcontract_1=[2], &
1019 : contract_2=[1], notcontract_2=[2, 3], &
1020 : bounds_1=bounds_iat, bounds_2=bounds_jat, &
1021 1808 : filter_eps=ri_data%filter_eps, flop=nflop)
1022 1808 : ri_data%dbcsr_nflop = ri_data%dbcsr_nflop + nflop
1023 :
1024 1808 : CALL get_tensor_occupancy(t_3c_work_2(2), nze, occ)
1025 1808 : IF (nze == 0) CYCLE
1026 :
1027 : !Contract with V_PQ so that we can take the trace with (Q^b|nu^b lmabda^a+c)^(x)
1028 1666 : CALL dbt_copy(t_3c_work_2(2), t_3c_work_3(1), order=[2, 1, 3], move_data=.TRUE.)
1029 1666 : CALL dbt_batched_contract_init(t_2c_work(5))
1030 : CALL dbt_contract(1.0_dp, t_2c_work(5), t_3c_work_3(1), &
1031 : 0.0_dp, t_3c_work_3(2), map_1=[1], map_2=[2, 3], &
1032 : contract_1=[1], notcontract_1=[2], &
1033 : contract_2=[1], notcontract_2=[2, 3], &
1034 1666 : filter_eps=ri_data%filter_eps, flop=nflop)
1035 1666 : ri_data%dbcsr_nflop = ri_data%dbcsr_nflop + nflop
1036 1666 : CALL dbt_batched_contract_finalize(t_2c_work(5))
1037 :
1038 : !Contract with the 3c derivatives to get the force/virial
1039 1666 : CALL dbt_copy(t_3c_work_3(2), t_3c_work_3(4), move_data=.TRUE.)
1040 1666 : IF (use_virial_prv) THEN
1041 : CALL get_force_from_3c_trace(force, t_3c_work_3(4), t_3c_der_stack(1:3), &
1042 : t_3c_der_stack(4:6), atom_of_kind, kind_of, &
1043 : idx_to_at_RI, idx_to_at_AO, i_images, &
1044 : batch_ranges_nze(i_batch), 2.0_dp*pref, &
1045 257 : ri_data, qs_env, work_virial, cell, particle_set)
1046 : ELSE
1047 : CALL get_force_from_3c_trace(force, t_3c_work_3(4), t_3c_der_stack(1:3), &
1048 : t_3c_der_stack(4:6), atom_of_kind, kind_of, &
1049 : idx_to_at_RI, idx_to_at_AO, i_images, &
1050 : batch_ranges_nze(i_batch), 2.0_dp*pref, &
1051 1409 : ri_data, qs_env)
1052 : END IF
1053 1666 : CALL dbt_clear(t_3c_work_3(4))
1054 :
1055 : !Contract with the 3-center integrals in order to have a matrix R_PQ such that
1056 : !we can take the trace sum_PQ R_PQ (P^0|Q^b)^(x)
1057 1666 : IF (i_loop == 2) CYCLE
1058 :
1059 : !Stack the 3c integrals
1060 : CALL fill_3c_stack(t_3c_work_3(4), ri_data%kp_t_3c_int, iapc_pairs(:, 1), 3, ri_data, &
1061 : filter_at=jatom, filter_dim=2, idx_to_at=idx_to_at_AO, &
1062 2634 : img_bounds=[batch_ranges_nze(i_batch), batch_ranges_nze(i_batch + 1)])
1063 878 : CALL dbt_copy(t_3c_work_3(4), t_3c_work_3(3), move_data=.TRUE.)
1064 :
1065 878 : CALL dbt_batched_contract_init(t_2c_R_split)
1066 : CALL dbt_contract(1.0_dp, t_3c_work_3(1), t_3c_work_3(3), &
1067 : 1.0_dp, t_2c_R_split, map_1=[1], map_2=[2], &
1068 : contract_1=[2, 3], notcontract_1=[1], &
1069 : contract_2=[2, 3], notcontract_2=[1], &
1070 878 : filter_eps=ri_data%filter_eps, flop=nflop)
1071 878 : ri_data%dbcsr_nflop = ri_data%dbcsr_nflop + nflop
1072 878 : CALL dbt_batched_contract_finalize(t_2c_R_split)
1073 6063 : CALL dbt_copy(t_3c_work_3(4), t_3c_work_3(1))
1074 : END DO
1075 : END DO
1076 2435 : DO i_spin = 1, nspins
1077 2435 : CALL dbt_batched_contract_finalize(rho_ao_t_sub(i_spin, b_img))
1078 : END DO
1079 1098 : CALL timestop(handle2)
1080 :
1081 1098 : IF (i_loop == 2) CYCLE
1082 579 : pref = 2.0_dp*pref
1083 579 : IF (iatom == jatom .AND. b_img == 1) pref = 0.5_dp*pref
1084 :
1085 579 : CALL timeset(routineN//"_2c_2", handle2)
1086 : !Note that the derivatives are in atomic block format (not split)
1087 579 : CALL dbt_copy(t_2c_R_split, t_2c_R, move_data=.TRUE.)
1088 :
1089 : CALL get_ext_2c_int(t_2c_work(1), mat_2c_pot, iatom, jatom, b_img, ri_data, qs_env, &
1090 : blacs_env_ext=blacs_env_sub, para_env_ext=para_env_sub, &
1091 579 : dbcsr_template=dbcsr_template)
1092 :
1093 : !We have to calculate: S^-1(iat) * R_PQ * S^-1(jat) to trace with HFX pot der
1094 : ! + R_PQ * S^-1(jat) * pot^T to trace with S^(x) (iat)
1095 : ! + pot^T * S^-1(iat) *R_PQ to trace with S^(x) (jat)
1096 :
1097 : !Because 3c tensors are all precontracted with the inverse RI metric,
1098 : !t_2c_R is currently implicitely multiplied by S^-1(iat) from the left
1099 : !and S^-1(jat) from the right, directly in the proper format for the trace
1100 : !with the HFX potential derivative
1101 :
1102 : !Trace with HFX pot deriv, that we need to build first
1103 2316 : DO i_xyz = 1, 3
1104 : CALL get_ext_2c_int(t_2c_der_pot(i_xyz), mat_der_pot_sub(:, i_xyz), iatom, jatom, &
1105 : b_img, ri_data, qs_env, blacs_env_ext=blacs_env_sub, &
1106 2316 : para_env_ext=para_env_sub, dbcsr_template=dbcsr_template)
1107 : END DO
1108 :
1109 579 : IF (use_virial_prv) THEN
1110 : CALL get_2c_der_force(force, t_2c_R, t_2c_der_pot, atom_of_kind, kind_of, &
1111 113 : b_img, pref, ri_data, qs_env, work_virial, cell, particle_set)
1112 : ELSE
1113 : CALL get_2c_der_force(force, t_2c_R, t_2c_der_pot, atom_of_kind, kind_of, &
1114 466 : b_img, pref, ri_data, qs_env)
1115 : END IF
1116 :
1117 2316 : DO i_xyz = 1, 3
1118 2316 : CALL dbt_clear(t_2c_der_pot(i_xyz))
1119 : END DO
1120 :
1121 : !R_PQ * S^-1(jat) * pot^T (=A)
1122 : CALL dbt_contract(1.0_dp, t_2c_metric(iatom), t_2c_R, & !get rid of implicit S^-1(iat)
1123 : 0.0_dp, t_2c_work(2), map_1=[1], map_2=[2], &
1124 : contract_1=[2], notcontract_1=[1], &
1125 : contract_2=[1], notcontract_2=[2], &
1126 579 : filter_eps=ri_data%filter_eps, flop=nflop)
1127 579 : ri_data%dbcsr_nflop = ri_data%dbcsr_nflop + nflop
1128 : CALL dbt_contract(1.0_dp, t_2c_work(2), t_2c_work(1), &
1129 : 0.0_dp, t_2c_work(3), map_1=[1], map_2=[2], &
1130 : contract_1=[2], notcontract_1=[1], &
1131 : contract_2=[2], notcontract_2=[1], &
1132 579 : filter_eps=ri_data%filter_eps, flop=nflop)
1133 579 : ri_data%dbcsr_nflop = ri_data%dbcsr_nflop + nflop
1134 :
1135 : !With the RI bump function, things get more complex. M = (S|P)_D + B*(S|P)_OD*B
1136 : !Calculate M^-1*B*A + A*B*M^-1 to contract with B^x. A is in t_2c_work(3)
1137 : CALL dbt_contract(1.0_dp, t_2c_work(3), t_2c_binv(iatom), &
1138 : 0.0_dp, t_2c_work(2), map_1=[1], map_2=[2], &
1139 : contract_1=[2], notcontract_1=[1], &
1140 : contract_2=[1], notcontract_2=[2], &
1141 579 : filter_eps=ri_data%filter_eps, flop=nflop)
1142 579 : ri_data%dbcsr_nflop = ri_data%dbcsr_nflop + nflop
1143 :
1144 : CALL dbt_contract(1.0_dp, t_2c_binv(iatom), t_2c_work(3), & !use transpose of B*M^-1 = M^-1*B
1145 : 0.0_dp, t_2c_work(4), map_1=[1], map_2=[2], &
1146 : contract_1=[1], notcontract_1=[2], &
1147 : contract_2=[1], notcontract_2=[2], &
1148 579 : filter_eps=ri_data%filter_eps, flop=nflop)
1149 579 : ri_data%dbcsr_nflop = ri_data%dbcsr_nflop + nflop
1150 :
1151 579 : CALL dbt_copy(t_2c_work(2), t_2c_work(4), summation=.TRUE.)
1152 : CALL get_2c_bump_forces(force, t_2c_work(4), iatom, atom_of_kind, kind_of, pref, &
1153 579 : ri_data, qs_env, work_virial)
1154 :
1155 : !Calculate -M^-1*B*A*B*M^-1 to contracte with diagonal RI metric deriv. t_2c_work(2) holds A*B*M^-1
1156 : CALL dbt_contract(1.0_dp, t_2c_binv(iatom), t_2c_work(2), &
1157 : 0.0_dp, t_2c_work(4), map_1=[1], map_2=[2], &
1158 : contract_1=[1], notcontract_1=[2], &
1159 : contract_2=[1], notcontract_2=[2], &
1160 579 : filter_eps=ri_data%filter_eps, flop=nflop)
1161 579 : ri_data%dbcsr_nflop = ri_data%dbcsr_nflop + nflop
1162 :
1163 579 : IF (use_virial_prv) THEN
1164 : CALL get_2c_der_force(force, t_2c_work(4), t_2c_der_metric_sub(iatom, :), atom_of_kind, &
1165 : kind_of, 1, -pref, ri_data, qs_env, work_virial, cell, particle_set, &
1166 113 : diag=.TRUE., offdiag=.FALSE.)
1167 : ELSE
1168 : CALL get_2c_der_force(force, t_2c_work(4), t_2c_der_metric_sub(iatom, :), atom_of_kind, &
1169 466 : kind_of, 1, -pref, ri_data, qs_env, diag=.TRUE., offdiag=.FALSE.)
1170 : END IF
1171 :
1172 : !Calculate -B*M^-1*B*A*B*M^-1*B to contract with off-diagonal RI metric derivs
1173 579 : CALL dbt_copy(t_2c_work(4), t_2c_work(2))
1174 579 : CALL apply_bump(t_2c_work(2), iatom, ri_data, qs_env, from_left=.TRUE., from_right=.TRUE.)
1175 :
1176 579 : IF (use_virial_prv) THEN
1177 : CALL get_2c_der_force(force, t_2c_work(2), t_2c_der_metric_sub(iatom, :), atom_of_kind, &
1178 : kind_of, 1, -pref, ri_data, qs_env, work_virial, cell, particle_set, &
1179 113 : diag=.FALSE., offdiag=.TRUE.)
1180 : ELSE
1181 : CALL get_2c_der_force(force, t_2c_work(2), t_2c_der_metric_sub(iatom, :), atom_of_kind, &
1182 466 : kind_of, 1, -pref, ri_data, qs_env, diag=.FALSE., offdiag=.TRUE.)
1183 : END IF
1184 :
1185 : !Calculate -O*B*M^-1*B*A*B*M^-1 - M^-1*B*A*B*M^-1*B*O, where O is off-diagonal integrals
1186 : !t_2c_work(4) holds M^-1*B*A*B*M^-1, and exploit transpose of B*O (stored in t_2c_bint)
1187 : CALL dbt_contract(1.0_dp, t_2c_work(4), t_2c_bint(iatom), &
1188 : 0.0_dp, t_2c_work(2), map_1=[1], map_2=[2], &
1189 : contract_1=[2], notcontract_1=[1], &
1190 : contract_2=[1], notcontract_2=[2], &
1191 579 : filter_eps=ri_data%filter_eps, flop=nflop)
1192 579 : ri_data%dbcsr_nflop = ri_data%dbcsr_nflop + nflop
1193 :
1194 : CALL dbt_contract(1.0_dp, t_2c_bint(iatom), t_2c_work(4), &
1195 : 1.0_dp, t_2c_work(2), map_1=[1], map_2=[2], &
1196 : contract_1=[1], notcontract_1=[2], &
1197 : contract_2=[1], notcontract_2=[2], &
1198 579 : filter_eps=ri_data%filter_eps, flop=nflop)
1199 579 : ri_data%dbcsr_nflop = ri_data%dbcsr_nflop + nflop
1200 :
1201 : CALL get_2c_bump_forces(force, t_2c_work(2), iatom, atom_of_kind, kind_of, -pref, &
1202 579 : ri_data, qs_env, work_virial)
1203 :
1204 : ! pot^T * S^-1(iat) * R_PQ (=A)
1205 : CALL dbt_contract(1.0_dp, t_2c_work(1), t_2c_R, &
1206 : 0.0_dp, t_2c_work(2), map_1=[1], map_2=[2], &
1207 : contract_1=[1], notcontract_1=[2], &
1208 : contract_2=[1], notcontract_2=[2], &
1209 579 : filter_eps=ri_data%filter_eps, flop=nflop)
1210 579 : ri_data%dbcsr_nflop = ri_data%dbcsr_nflop + nflop
1211 :
1212 : CALL dbt_contract(1.0_dp, t_2c_work(2), t_2c_metric(jatom), & !get rid of implicit S^-1(jat)
1213 : 0.0_dp, t_2c_work(3), map_1=[1], map_2=[2], &
1214 : contract_1=[2], notcontract_1=[1], &
1215 : contract_2=[1], notcontract_2=[2], &
1216 579 : filter_eps=ri_data%filter_eps, flop=nflop)
1217 579 : ri_data%dbcsr_nflop = ri_data%dbcsr_nflop + nflop
1218 :
1219 : !Do the same shenanigans with the S^(x) (jatom)
1220 : !Calculate M^-1*B*A + A*B*M^-1 to contract with B^x. A is in t_2c_work(3)
1221 : CALL dbt_contract(1.0_dp, t_2c_work(3), t_2c_binv(jatom), &
1222 : 0.0_dp, t_2c_work(2), map_1=[1], map_2=[2], &
1223 : contract_1=[2], notcontract_1=[1], &
1224 : contract_2=[1], notcontract_2=[2], &
1225 579 : filter_eps=ri_data%filter_eps, flop=nflop)
1226 579 : ri_data%dbcsr_nflop = ri_data%dbcsr_nflop + nflop
1227 :
1228 : CALL dbt_contract(1.0_dp, t_2c_binv(jatom), t_2c_work(3), & !use transpose of B*M^-1 = M^-1*B
1229 : 0.0_dp, t_2c_work(4), map_1=[1], map_2=[2], &
1230 : contract_1=[1], notcontract_1=[2], &
1231 : contract_2=[1], notcontract_2=[2], &
1232 579 : filter_eps=ri_data%filter_eps, flop=nflop)
1233 579 : ri_data%dbcsr_nflop = ri_data%dbcsr_nflop + nflop
1234 :
1235 579 : CALL dbt_copy(t_2c_work(2), t_2c_work(4), summation=.TRUE.)
1236 : CALL get_2c_bump_forces(force, t_2c_work(4), jatom, atom_of_kind, kind_of, pref, &
1237 579 : ri_data, qs_env, work_virial)
1238 :
1239 : !Calculate -M^-1*B*A*B*M^-1 to contracte with diagonal RI metric deriv. t_2c_work(2) holds A*B*M^-1
1240 : CALL dbt_contract(1.0_dp, t_2c_binv(jatom), t_2c_work(2), &
1241 : 0.0_dp, t_2c_work(4), map_1=[1], map_2=[2], &
1242 : contract_1=[1], notcontract_1=[2], &
1243 : contract_2=[1], notcontract_2=[2], &
1244 579 : filter_eps=ri_data%filter_eps, flop=nflop)
1245 579 : ri_data%dbcsr_nflop = ri_data%dbcsr_nflop + nflop
1246 :
1247 579 : IF (use_virial_prv) THEN
1248 : CALL get_2c_der_force(force, t_2c_work(4), t_2c_der_metric_sub(jatom, :), atom_of_kind, &
1249 : kind_of, 1, -pref, ri_data, qs_env, work_virial, cell, particle_set, &
1250 113 : diag=.TRUE., offdiag=.FALSE.)
1251 : ELSE
1252 : CALL get_2c_der_force(force, t_2c_work(4), t_2c_der_metric_sub(jatom, :), atom_of_kind, &
1253 466 : kind_of, 1, -pref, ri_data, qs_env, diag=.TRUE., offdiag=.FALSE.)
1254 : END IF
1255 :
1256 : !Calculate -B*M^-1*B*A*B*M^-1*B to contract with off-diagonal RI metric derivs
1257 579 : CALL dbt_copy(t_2c_work(4), t_2c_work(2))
1258 579 : CALL apply_bump(t_2c_work(2), jatom, ri_data, qs_env, from_left=.TRUE., from_right=.TRUE.)
1259 :
1260 579 : IF (use_virial_prv) THEN
1261 : CALL get_2c_der_force(force, t_2c_work(2), t_2c_der_metric_sub(jatom, :), atom_of_kind, &
1262 : kind_of, 1, -pref, ri_data, qs_env, work_virial, cell, particle_set, &
1263 113 : diag=.FALSE., offdiag=.TRUE.)
1264 : ELSE
1265 : CALL get_2c_der_force(force, t_2c_work(2), t_2c_der_metric_sub(jatom, :), atom_of_kind, &
1266 466 : kind_of, 1, -pref, ri_data, qs_env, diag=.FALSE., offdiag=.TRUE.)
1267 : END IF
1268 :
1269 : !Calculate -O*B*M^-1*B*A*B*M^-1 - M^-1*B*A*B*M^-1*B*O, where O is off-diagonal integrals
1270 : !t_2c_work(4) holds M^-1*B*A*B*M^-1, and exploit transpose of B*O (stored in t_2c_bint)
1271 : CALL dbt_contract(1.0_dp, t_2c_work(4), t_2c_bint(jatom), &
1272 : 0.0_dp, t_2c_work(2), map_1=[1], map_2=[2], &
1273 : contract_1=[2], notcontract_1=[1], &
1274 : contract_2=[1], notcontract_2=[2], &
1275 579 : filter_eps=ri_data%filter_eps, flop=nflop)
1276 579 : ri_data%dbcsr_nflop = ri_data%dbcsr_nflop + nflop
1277 :
1278 : CALL dbt_contract(1.0_dp, t_2c_bint(jatom), t_2c_work(4), &
1279 : 1.0_dp, t_2c_work(2), map_1=[1], map_2=[2], &
1280 : contract_1=[1], notcontract_1=[2], &
1281 : contract_2=[1], notcontract_2=[2], &
1282 579 : filter_eps=ri_data%filter_eps, flop=nflop)
1283 579 : ri_data%dbcsr_nflop = ri_data%dbcsr_nflop + nflop
1284 :
1285 : CALL get_2c_bump_forces(force, t_2c_work(2), jatom, atom_of_kind, kind_of, -pref, &
1286 579 : ri_data, qs_env, work_virial)
1287 :
1288 14376 : CALL timestop(handle2)
1289 : END DO !iatom
1290 : END DO !jatom
1291 : END DO !b_img
1292 :
1293 126 : IF (i_loop == 1) THEN
1294 42 : CALL update_pattern_to_forces(force_pattern, sparsity_pattern, ngroups, ri_data, qs_env)
1295 : END IF
1296 : END DO !i_loop
1297 :
1298 42 : CALL dbt_batched_contract_finalize(t_3c_work_3(1))
1299 42 : CALL dbt_batched_contract_finalize(t_3c_work_3(2))
1300 42 : CALL dbt_batched_contract_finalize(t_3c_work_3(3))
1301 42 : CALL dbt_batched_contract_finalize(t_3c_work_2(1))
1302 42 : CALL dbt_batched_contract_finalize(t_3c_work_2(2))
1303 :
1304 42 : IF (use_virial_prv) THEN
1305 32 : DO k_xyz = 1, 3
1306 104 : DO j_xyz = 1, 3
1307 312 : DO i_xyz = 1, 3
1308 : virial%pv_fock_4c(i_xyz, j_xyz) = virial%pv_fock_4c(i_xyz, j_xyz) &
1309 288 : + work_virial(i_xyz, k_xyz)*cell%hmat(j_xyz, k_xyz)
1310 : END DO
1311 : END DO
1312 : END DO
1313 : END IF
1314 :
1315 : !End of subgroup parallelization
1316 42 : CALL cp_blacs_env_release(blacs_env_sub)
1317 42 : CALL para_env_sub%free()
1318 42 : DEALLOCATE (para_env_sub)
1319 :
1320 42 : CALL para_env%sync()
1321 42 : t2 = m_walltime()
1322 42 : ri_data%dbcsr_time = ri_data%dbcsr_time + t2 - t1
1323 :
1324 : !clean-up
1325 42 : IF (ASSOCIATED(dbcsr_template)) THEN
1326 42 : CALL dbcsr_release(dbcsr_template)
1327 42 : DEALLOCATE (dbcsr_template)
1328 : END IF
1329 42 : CALL dbt_destroy(t_2c_R)
1330 42 : CALL dbt_destroy(t_2c_R_split)
1331 42 : CALL dbt_destroy(t_2c_work(1))
1332 42 : CALL dbt_destroy(t_2c_work(2))
1333 42 : CALL dbt_destroy(t_2c_work(3))
1334 42 : CALL dbt_destroy(t_2c_work(4))
1335 42 : CALL dbt_destroy(t_2c_work(5))
1336 42 : CALL dbt_destroy(t_3c_work_2(1))
1337 42 : CALL dbt_destroy(t_3c_work_2(2))
1338 42 : CALL dbt_destroy(t_3c_work_2(3))
1339 42 : CALL dbt_destroy(t_3c_work_3(1))
1340 42 : CALL dbt_destroy(t_3c_work_3(2))
1341 42 : CALL dbt_destroy(t_3c_work_3(3))
1342 42 : CALL dbt_destroy(t_3c_work_3(4))
1343 42 : CALL dbt_destroy(t_3c_der_stack(1))
1344 42 : CALL dbt_destroy(t_3c_der_stack(2))
1345 42 : CALL dbt_destroy(t_3c_der_stack(3))
1346 42 : CALL dbt_destroy(t_3c_der_stack(4))
1347 42 : CALL dbt_destroy(t_3c_der_stack(5))
1348 42 : CALL dbt_destroy(t_3c_der_stack(6))
1349 168 : DO i_xyz = 1, 3
1350 168 : CALL dbt_destroy(t_2c_der_pot(i_xyz))
1351 : END DO
1352 126 : DO iatom = 1, natom
1353 84 : CALL dbt_destroy(t_2c_inv(iatom))
1354 84 : CALL dbt_destroy(t_2c_binv(iatom))
1355 84 : CALL dbt_destroy(t_2c_bint(iatom))
1356 84 : CALL dbt_destroy(t_2c_metric(iatom))
1357 378 : DO i_xyz = 1, 3
1358 336 : CALL dbt_destroy(t_2c_der_metric_sub(iatom, i_xyz))
1359 : END DO
1360 : END DO
1361 1052 : DO i_img = 1, nimg
1362 1010 : CALL dbcsr_release(mat_2c_pot(i_img))
1363 2228 : DO i_spin = 1, nspins
1364 1176 : CALL dbt_destroy(rho_ao_t_sub(i_spin, i_img))
1365 2186 : CALL dbt_destroy(t_3c_apc_sub(i_spin, i_img))
1366 : END DO
1367 : END DO
1368 168 : DO i_xyz = 1, 3
1369 3198 : DO i_img = 1, nimg
1370 3030 : CALL dbt_destroy(t_3c_der_RI_sub(i_img, i_xyz))
1371 3030 : CALL dbt_destroy(t_3c_der_AO_sub(i_img, i_xyz))
1372 3156 : CALL dbcsr_release(mat_der_pot_sub(i_img, i_xyz))
1373 : END DO
1374 : END DO
1375 :
1376 42 : CALL timestop(handle)
1377 :
1378 18756 : END SUBROUTINE hfx_ri_update_forces_kp
1379 :
1380 : ! **************************************************************************************************
1381 : !> \brief A routine the applies the RI bump matrix from the left and/or the right, given an input
1382 : !> matrix and the central RI atom. We assume atomic block sizes
1383 : !> \param t_2c_inout ...
1384 : !> \param atom_i ...
1385 : !> \param ri_data ...
1386 : !> \param qs_env ...
1387 : !> \param from_left ...
1388 : !> \param from_right ...
1389 : !> \param debump ...
1390 : ! **************************************************************************************************
1391 1746 : SUBROUTINE apply_bump(t_2c_inout, atom_i, ri_data, qs_env, from_left, from_right, debump)
1392 : TYPE(dbt_type), INTENT(INOUT) :: t_2c_inout
1393 : INTEGER, INTENT(IN) :: atom_i
1394 : TYPE(hfx_ri_type), INTENT(INOUT) :: ri_data
1395 : TYPE(qs_environment_type), POINTER :: qs_env
1396 : LOGICAL, INTENT(IN), OPTIONAL :: from_left, from_right, debump
1397 :
1398 : INTEGER :: i_img, i_RI, iatom, ind(2), j_img, j_RI, &
1399 : jatom, natom, nblks(2), nimg, nkind
1400 1746 : INTEGER, DIMENSION(:, :), POINTER :: index_to_cell
1401 1746 : INTEGER, DIMENSION(:, :, :), POINTER :: cell_to_index
1402 : LOGICAL :: found, my_debump, my_left, my_right
1403 : REAL(dp) :: bval, r0, r1, ri(3), rj(3), rref(3), &
1404 : scoord(3)
1405 1746 : REAL(dp), ALLOCATABLE, DIMENSION(:, :) :: blk
1406 : TYPE(cell_type), POINTER :: cell
1407 : TYPE(dbt_iterator_type) :: iter
1408 : TYPE(kpoint_type), POINTER :: kpoints
1409 1746 : TYPE(particle_type), DIMENSION(:), POINTER :: particle_set
1410 1746 : TYPE(qs_kind_type), DIMENSION(:), POINTER :: qs_kind_set
1411 :
1412 1746 : NULLIFY (qs_kind_set, particle_set, kpoints, index_to_cell, cell_to_index, cell)
1413 :
1414 : CALL get_qs_env(qs_env, natom=natom, nkind=nkind, qs_kind_set=qs_kind_set, cell=cell, &
1415 1746 : kpoints=kpoints, particle_set=particle_set)
1416 1746 : CALL get_kpoint_info(kpoints, cell_to_index=cell_to_index, index_to_cell=index_to_cell)
1417 :
1418 1746 : my_debump = .FALSE.
1419 1746 : IF (PRESENT(debump)) my_debump = debump
1420 :
1421 1746 : my_left = .FALSE.
1422 1746 : IF (PRESENT(from_left)) my_left = from_left
1423 :
1424 1746 : my_right = .FALSE.
1425 1746 : IF (PRESENT(from_right)) my_right = from_right
1426 1746 : CPASSERT(my_left .OR. my_right)
1427 :
1428 1746 : CALL dbt_get_info(t_2c_inout, nblks_total=nblks)
1429 1746 : CPASSERT(nblks(1) == ri_data%ncell_RI*natom)
1430 1746 : CPASSERT(nblks(2) == ri_data%ncell_RI*natom)
1431 :
1432 1746 : nimg = ri_data%nimg
1433 :
1434 : !Loop over the RI cells and atoms, and apply bump accordingly
1435 1746 : r1 = ri_data%kp_RI_range
1436 1746 : r0 = ri_data%kp_bump_rad
1437 1746 : rref = pbc(particle_set(atom_i)%r, cell)
1438 :
1439 : !$OMP PARALLEL DEFAULT(NONE) SHARED(t_2c_inout,natom,ri_data,cell,particle_set,index_to_cell,my_left, &
1440 : !$OMP my_right,r0,r1,rref,my_debump) &
1441 1746 : !$OMP PRIVATE(iter,ind,blk,found,i_RI,i_img,iatom,j_RI,j_img,jatom,scoord,ri,rj,bval)
1442 : CALL dbt_iterator_start(iter, t_2c_inout)
1443 : DO WHILE (dbt_iterator_blocks_left(iter))
1444 : CALL dbt_iterator_next_block(iter, ind)
1445 : CALL dbt_get_block(t_2c_inout, ind, blk, found)
1446 : IF (.NOT. found) CYCLE
1447 :
1448 : i_RI = (ind(1) - 1)/natom + 1
1449 : i_img = ri_data%RI_cell_to_img(i_RI)
1450 : iatom = ind(1) - (i_RI - 1)*natom
1451 :
1452 : CALL real_to_scaled(scoord, pbc(particle_set(iatom)%r, cell), cell)
1453 : CALL scaled_to_real(ri, scoord(:) + index_to_cell(:, i_img), cell)
1454 :
1455 : j_RI = (ind(2) - 1)/natom + 1
1456 : j_img = ri_data%RI_cell_to_img(j_RI)
1457 : jatom = ind(2) - (j_RI - 1)*natom
1458 :
1459 : CALL real_to_scaled(scoord, pbc(particle_set(jatom)%r, cell), cell)
1460 : CALL scaled_to_real(rj, scoord(:) + index_to_cell(:, j_img), cell)
1461 :
1462 : IF (.NOT. my_debump) THEN
1463 : IF (my_left) blk(:, :) = blk(:, :)*bump(NORM2(ri - rref), r0, r1)
1464 : IF (my_right) blk(:, :) = blk(:, :)*bump(NORM2(rj - rref), r0, r1)
1465 : ELSE
1466 : !Note: by construction, the bump function is never quite zero, as its range is the same
1467 : ! as that of the extended RI basis (but we are safe)
1468 : bval = bump(NORM2(ri - rref), r0, r1)
1469 : IF (my_left .AND. bval > EPSILON(1.0_dp)) blk(:, :) = blk(:, :)/bval
1470 : bval = bump(NORM2(rj - rref), r0, r1)
1471 : IF (my_right .AND. bval > EPSILON(1.0_dp)) blk(:, :) = blk(:, :)/bval
1472 : END IF
1473 :
1474 : CALL dbt_put_block(t_2c_inout, ind, SHAPE(blk), blk)
1475 :
1476 : DEALLOCATE (blk)
1477 : END DO
1478 : CALL dbt_iterator_stop(iter)
1479 : !$OMP END PARALLEL
1480 1746 : CALL dbt_filter(t_2c_inout, ri_data%filter_eps)
1481 :
1482 3492 : END SUBROUTINE apply_bump
1483 :
1484 : ! **************************************************************************************************
1485 : !> \brief A routine that calculates the forces due to the derivative of the bump function
1486 : !> \param force ...
1487 : !> \param t_2c_in ...
1488 : !> \param atom_i ...
1489 : !> \param atom_of_kind ...
1490 : !> \param kind_of ...
1491 : !> \param pref ...
1492 : !> \param ri_data ...
1493 : !> \param qs_env ...
1494 : !> \param work_virial ...
1495 : ! **************************************************************************************************
1496 2316 : SUBROUTINE get_2c_bump_forces(force, t_2c_in, atom_i, atom_of_kind, kind_of, pref, ri_data, &
1497 : qs_env, work_virial)
1498 : TYPE(qs_force_type), DIMENSION(:), POINTER :: force
1499 : TYPE(dbt_type), INTENT(INOUT) :: t_2c_in
1500 : INTEGER, INTENT(IN) :: atom_i
1501 : INTEGER, DIMENSION(:), INTENT(IN) :: atom_of_kind, kind_of
1502 : REAL(dp), INTENT(IN) :: pref
1503 : TYPE(hfx_ri_type), INTENT(INOUT) :: ri_data
1504 : TYPE(qs_environment_type), POINTER :: qs_env
1505 : REAL(dp), DIMENSION(3, 3), INTENT(INOUT) :: work_virial
1506 :
1507 : INTEGER :: i, i_img, i_RI, i_xyz, iat_of_kind, iatom, ikind, ind(2), j_img, j_RI, j_xyz, &
1508 : jat_of_kind, jatom, jkind, natom, nblks(2), nimg, nkind
1509 2316 : INTEGER, DIMENSION(:, :), POINTER :: index_to_cell
1510 2316 : INTEGER, DIMENSION(:, :, :), POINTER :: cell_to_index
1511 : LOGICAL :: found
1512 : REAL(dp) :: new_force, r0, r1, ri(3), rj(3), &
1513 : rref(3), scoord(3), x
1514 2316 : REAL(dp), ALLOCATABLE, DIMENSION(:, :) :: blk
1515 : TYPE(cell_type), POINTER :: cell
1516 : TYPE(dbt_iterator_type) :: iter
1517 : TYPE(kpoint_type), POINTER :: kpoints
1518 2316 : TYPE(particle_type), DIMENSION(:), POINTER :: particle_set
1519 2316 : TYPE(qs_kind_type), DIMENSION(:), POINTER :: qs_kind_set
1520 :
1521 2316 : NULLIFY (qs_kind_set, particle_set, kpoints, index_to_cell, cell_to_index, cell)
1522 :
1523 : CALL get_qs_env(qs_env, natom=natom, nkind=nkind, qs_kind_set=qs_kind_set, cell=cell, &
1524 2316 : kpoints=kpoints, particle_set=particle_set)
1525 2316 : CALL get_kpoint_info(kpoints, cell_to_index=cell_to_index, index_to_cell=index_to_cell)
1526 :
1527 2316 : CALL dbt_get_info(t_2c_in, nblks_total=nblks)
1528 2316 : CPASSERT(nblks(1) == ri_data%ncell_RI*natom)
1529 2316 : CPASSERT(nblks(2) == ri_data%ncell_RI*natom)
1530 :
1531 2316 : nimg = ri_data%nimg
1532 :
1533 : !Loop over the RI cells and atoms, and apply bump accordingly
1534 2316 : r1 = ri_data%kp_RI_range
1535 2316 : r0 = ri_data%kp_bump_rad
1536 2316 : rref = pbc(particle_set(atom_i)%r, cell)
1537 :
1538 2316 : iat_of_kind = atom_of_kind(atom_i)
1539 2316 : ikind = kind_of(atom_i)
1540 :
1541 : !$OMP PARALLEL DEFAULT(NONE) SHARED(t_2c_in,natom,ri_data,cell,particle_set,index_to_cell,pref, &
1542 : !$OMP force,r0,r1,rref,atom_of_kind,kind_of,iat_of_kind,ikind,work_virial) &
1543 : !$OMP PRIVATE(iter,ind,blk,found,i_RI,i_img,iatom,j_RI,j_img,jatom,scoord,ri,rj,jkind,jat_of_kind, &
1544 2316 : !$OMP new_force,i_xyz,i,x,j_xyz)
1545 : CALL dbt_iterator_start(iter, t_2c_in)
1546 : DO WHILE (dbt_iterator_blocks_left(iter))
1547 : CALL dbt_iterator_next_block(iter, ind)
1548 : IF (ind(1) .NE. ind(2)) CYCLE !bump matrix is diagonal
1549 :
1550 : CALL dbt_get_block(t_2c_in, ind, blk, found)
1551 : IF (.NOT. found) CYCLE
1552 :
1553 : !bump is a function of x = SQRT((R - Rref)^2). We refer to R as jatom, and Rref as atom_i
1554 : j_RI = (ind(2) - 1)/natom + 1
1555 : j_img = ri_data%RI_cell_to_img(j_RI)
1556 : jatom = ind(2) - (j_RI - 1)*natom
1557 : jat_of_kind = atom_of_kind(jatom)
1558 : jkind = kind_of(jatom)
1559 :
1560 : CALL real_to_scaled(scoord, pbc(particle_set(jatom)%r, cell), cell)
1561 : CALL scaled_to_real(rj, scoord(:) + index_to_cell(:, j_img), cell)
1562 : x = NORM2(rj - rref)
1563 : IF (x < r0 .OR. x > r1) CYCLE
1564 :
1565 : new_force = 0.0_dp
1566 : DO i = 1, SIZE(blk, 1)
1567 : new_force = new_force + blk(i, i)
1568 : END DO
1569 : new_force = pref*new_force*dbump(x, r0, r1)
1570 :
1571 : !x = SQRT((R - Rref)^2), so we multiply by dx/dR and dx/dRref
1572 : DO i_xyz = 1, 3
1573 : !Force acting on second atom
1574 : !$OMP ATOMIC
1575 : force(jkind)%fock_4c(i_xyz, jat_of_kind) = force(jkind)%fock_4c(i_xyz, jat_of_kind) + &
1576 : new_force*(rj(i_xyz) - rref(i_xyz))/x
1577 :
1578 : !virial acting on second atom
1579 : CALL real_to_scaled(scoord, rj, cell)
1580 : DO j_xyz = 1, 3
1581 : !$OMP ATOMIC
1582 : work_virial(i_xyz, j_xyz) = work_virial(i_xyz, j_xyz) &
1583 : + new_force*scoord(j_xyz)*(rj(i_xyz) - rref(i_xyz))/x
1584 : END DO
1585 :
1586 : !Force acting on reference atom, defining the RI basis
1587 : !$OMP ATOMIC
1588 : force(ikind)%fock_4c(i_xyz, iat_of_kind) = force(ikind)%fock_4c(i_xyz, iat_of_kind) - &
1589 : new_force*(rj(i_xyz) - rref(i_xyz))/x
1590 :
1591 : !virial of ref atom
1592 : CALL real_to_scaled(scoord, rref, cell)
1593 : DO j_xyz = 1, 3
1594 : !$OMP ATOMIC
1595 : work_virial(i_xyz, j_xyz) = work_virial(i_xyz, j_xyz) &
1596 : - new_force*scoord(j_xyz)*(rj(i_xyz) - rref(i_xyz))/x
1597 : END DO
1598 : END DO !i_xyz
1599 :
1600 : DEALLOCATE (blk)
1601 : END DO
1602 : CALL dbt_iterator_stop(iter)
1603 : !$OMP END PARALLEL
1604 :
1605 4632 : END SUBROUTINE get_2c_bump_forces
1606 :
1607 : ! **************************************************************************************************
1608 : !> \brief The bumb function as defined by Juerg
1609 : !> \param x ...
1610 : !> \param r0 ...
1611 : !> \param r1 ...
1612 : !> \return ...
1613 : ! **************************************************************************************************
1614 25181 : FUNCTION bump(x, r0, r1) RESULT(b)
1615 : REAL(dp), INTENT(IN) :: x, r0, r1
1616 : REAL(dp) :: b
1617 :
1618 : REAL(dp) :: r
1619 :
1620 : !Head-Gordon
1621 : !b = 1.0_dp/(1.0_dp+EXP((r1-r0)/(r1-x)-(r1-r0)/(x-r0)))
1622 : !Juerg
1623 25181 : r = (x - r0)/(r1 - r0)
1624 25181 : b = -6.0_dp*r**5 + 15.0_dp*r**4 - 10.0_dp*r**3 + 1.0_dp
1625 25181 : IF (x .GE. r1) b = 0.0_dp
1626 25181 : IF (x .LE. r0) b = 1.0_dp
1627 :
1628 25181 : END FUNCTION bump
1629 :
1630 : ! **************************************************************************************************
1631 : !> \brief The derivative of the bump function
1632 : !> \param x ...
1633 : !> \param r0 ...
1634 : !> \param r1 ...
1635 : !> \return ...
1636 : ! **************************************************************************************************
1637 525 : FUNCTION dbump(x, r0, r1) RESULT(b)
1638 : REAL(dp), INTENT(IN) :: x, r0, r1
1639 : REAL(dp) :: b
1640 :
1641 : REAL(dp) :: r
1642 :
1643 525 : r = (x - r0)/(r1 - r0)
1644 525 : b = (-30.0_dp*r**4 + 60.0_dp*r**3 - 30.0_dp*r**2)/(r1 - r0)
1645 525 : IF (x .GE. r1) b = 0.0_dp
1646 525 : IF (x .LE. r0) b = 0.0_dp
1647 :
1648 525 : END FUNCTION dbump
1649 :
1650 : ! **************************************************************************************************
1651 : !> \brief return the cell index a+c corresponding to given cell index i and b, with i = a+c-b
1652 : !> \param i_index ...
1653 : !> \param b_index ...
1654 : !> \param qs_env ...
1655 : !> \return ...
1656 : ! **************************************************************************************************
1657 151439 : FUNCTION get_apc_index_from_ib(i_index, b_index, qs_env) RESULT(apc_index)
1658 : INTEGER, INTENT(IN) :: i_index, b_index
1659 : TYPE(qs_environment_type), POINTER :: qs_env
1660 : INTEGER :: apc_index
1661 :
1662 : INTEGER, DIMENSION(3) :: cell_apc
1663 151439 : INTEGER, DIMENSION(:, :), POINTER :: index_to_cell
1664 151439 : INTEGER, DIMENSION(:, :, :), POINTER :: cell_to_index
1665 : TYPE(kpoint_type), POINTER :: kpoints
1666 :
1667 151439 : CALL get_qs_env(qs_env, kpoints=kpoints)
1668 151439 : CALL get_kpoint_info(kpoints, cell_to_index=cell_to_index, index_to_cell=index_to_cell)
1669 :
1670 : !i = a+c-b => a+c = i+b
1671 605756 : cell_apc(:) = index_to_cell(:, i_index) + index_to_cell(:, b_index)
1672 :
1673 1037869 : IF (ANY([cell_apc(1), cell_apc(2), cell_apc(3)] < LBOUND(cell_to_index)) .OR. &
1674 : ANY([cell_apc(1), cell_apc(2), cell_apc(3)] > UBOUND(cell_to_index))) THEN
1675 :
1676 : apc_index = 0
1677 : ELSE
1678 132550 : apc_index = cell_to_index(cell_apc(1), cell_apc(2), cell_apc(3))
1679 : END IF
1680 :
1681 151439 : END FUNCTION get_apc_index_from_ib
1682 :
1683 : ! **************************************************************************************************
1684 : !> \brief return the cell index i corresponding to the summ of cell_a and cell_c
1685 : !> \param a_index ...
1686 : !> \param c_index ...
1687 : !> \param qs_env ...
1688 : !> \return ...
1689 : ! **************************************************************************************************
1690 0 : FUNCTION get_apc_index(a_index, c_index, qs_env) RESULT(i_index)
1691 : INTEGER, INTENT(IN) :: a_index, c_index
1692 : TYPE(qs_environment_type), POINTER :: qs_env
1693 : INTEGER :: i_index
1694 :
1695 : INTEGER, DIMENSION(3) :: cell_i
1696 0 : INTEGER, DIMENSION(:, :), POINTER :: index_to_cell
1697 0 : INTEGER, DIMENSION(:, :, :), POINTER :: cell_to_index
1698 : TYPE(kpoint_type), POINTER :: kpoints
1699 :
1700 0 : CALL get_qs_env(qs_env, kpoints=kpoints)
1701 0 : CALL get_kpoint_info(kpoints, cell_to_index=cell_to_index, index_to_cell=index_to_cell)
1702 :
1703 0 : cell_i(:) = index_to_cell(:, a_index) + index_to_cell(:, c_index)
1704 :
1705 0 : IF (ANY([cell_i(1), cell_i(2), cell_i(3)] < LBOUND(cell_to_index)) .OR. &
1706 : ANY([cell_i(1), cell_i(2), cell_i(3)] > UBOUND(cell_to_index))) THEN
1707 :
1708 : i_index = 0
1709 : ELSE
1710 0 : i_index = cell_to_index(cell_i(1), cell_i(2), cell_i(3))
1711 : END IF
1712 :
1713 0 : END FUNCTION get_apc_index
1714 :
1715 : ! **************************************************************************************************
1716 : !> \brief return the cell index i corresponding to the summ of cell_a + cell_c - cell_b
1717 : !> \param apc_index ...
1718 : !> \param b_index ...
1719 : !> \param qs_env ...
1720 : !> \return ...
1721 : ! **************************************************************************************************
1722 510764 : FUNCTION get_i_index(apc_index, b_index, qs_env) RESULT(i_index)
1723 : INTEGER, INTENT(IN) :: apc_index, b_index
1724 : TYPE(qs_environment_type), POINTER :: qs_env
1725 : INTEGER :: i_index
1726 :
1727 : INTEGER, DIMENSION(3) :: cell_i
1728 510764 : INTEGER, DIMENSION(:, :), POINTER :: index_to_cell
1729 510764 : INTEGER, DIMENSION(:, :, :), POINTER :: cell_to_index
1730 : TYPE(kpoint_type), POINTER :: kpoints
1731 :
1732 510764 : CALL get_qs_env(qs_env, kpoints=kpoints)
1733 510764 : CALL get_kpoint_info(kpoints, cell_to_index=cell_to_index, index_to_cell=index_to_cell)
1734 :
1735 2043056 : cell_i(:) = index_to_cell(:, apc_index) - index_to_cell(:, b_index)
1736 :
1737 3491940 : IF (ANY([cell_i(1), cell_i(2), cell_i(3)] < LBOUND(cell_to_index)) .OR. &
1738 : ANY([cell_i(1), cell_i(2), cell_i(3)] > UBOUND(cell_to_index))) THEN
1739 :
1740 : i_index = 0
1741 : ELSE
1742 438348 : i_index = cell_to_index(cell_i(1), cell_i(2), cell_i(3))
1743 : END IF
1744 :
1745 510764 : END FUNCTION get_i_index
1746 :
1747 : ! **************************************************************************************************
1748 : !> \brief A routine that returns all allowed a,c pairs such that a+c images corresponds to the value
1749 : !> of the apc_index input. Takes into account that image a corresponds to 3c integrals, which
1750 : !> are ordered in their own way
1751 : !> \param ac_pairs ...
1752 : !> \param apc_index ...
1753 : !> \param ri_data ...
1754 : !> \param qs_env ...
1755 : ! **************************************************************************************************
1756 15228 : SUBROUTINE get_ac_pairs(ac_pairs, apc_index, ri_data, qs_env)
1757 : INTEGER, DIMENSION(:, :), INTENT(INOUT) :: ac_pairs
1758 : INTEGER, INTENT(IN) :: apc_index
1759 : TYPE(hfx_ri_type), INTENT(INOUT) :: ri_data
1760 : TYPE(qs_environment_type), POINTER :: qs_env
1761 :
1762 : INTEGER :: a_index, actual_img, c_index, nimg
1763 :
1764 15228 : nimg = SIZE(ac_pairs, 1)
1765 :
1766 1067212 : ac_pairs(:, :) = 0
1767 : !$OMP PARALLEL DO DEFAULT(NONE) SHARED(ac_pairs,nimg,ri_data,qs_env,apc_index) &
1768 15228 : !$OMP PRIVATE(a_index,actual_img,c_index)
1769 : DO a_index = 1, nimg
1770 : actual_img = ri_data%idx_to_img(a_index)
1771 : !c = a+c - a
1772 : c_index = get_i_index(apc_index, actual_img, qs_env)
1773 : ac_pairs(a_index, 1) = a_index
1774 : ac_pairs(a_index, 2) = c_index
1775 : END DO
1776 : !$OMP END PARALLEL DO
1777 :
1778 15228 : END SUBROUTINE get_ac_pairs
1779 :
1780 : ! **************************************************************************************************
1781 : !> \brief A routine that returns all allowed i,a+c pairs such that, for the given value of b, we have
1782 : !> i = a+c-b. Takes into account that image i corrsponds to the 3c ints, which are ordered in
1783 : !> their own way
1784 : !> \param iapc_pairs ...
1785 : !> \param b_index ...
1786 : !> \param ri_data ...
1787 : !> \param qs_env ...
1788 : !> \param actual_i_img ...
1789 : ! **************************************************************************************************
1790 4330 : SUBROUTINE get_iapc_pairs(iapc_pairs, b_index, ri_data, qs_env, actual_i_img)
1791 : INTEGER, DIMENSION(:, :), INTENT(INOUT) :: iapc_pairs
1792 : INTEGER, INTENT(IN) :: b_index
1793 : TYPE(hfx_ri_type), INTENT(INOUT) :: ri_data
1794 : TYPE(qs_environment_type), POINTER :: qs_env
1795 : INTEGER, DIMENSION(:), INTENT(INOUT), OPTIONAL :: actual_i_img
1796 :
1797 : INTEGER :: actual_img, apc_index, i_index, nimg
1798 :
1799 4330 : nimg = SIZE(iapc_pairs, 1)
1800 43396 : IF (PRESENT(actual_i_img)) actual_i_img(:) = 0
1801 :
1802 315868 : iapc_pairs(:, :) = 0
1803 : !$OMP PARALLEL DO DEFAULT(NONE) SHARED(iapc_pairs,nimg,ri_data,qs_env,b_index,actual_i_img) &
1804 4330 : !$OMP PRIVATE(i_index,actual_img,apc_index)
1805 : DO i_index = 1, nimg
1806 : actual_img = ri_data%idx_to_img(i_index)
1807 : apc_index = get_apc_index_from_ib(actual_img, b_index, qs_env)
1808 : IF (apc_index == 0) CYCLE
1809 : iapc_pairs(i_index, 1) = i_index
1810 : iapc_pairs(i_index, 2) = apc_index
1811 : IF (PRESENT(actual_i_img)) actual_i_img(i_index) = actual_img
1812 : END DO
1813 :
1814 4330 : END SUBROUTINE get_iapc_pairs
1815 :
1816 : ! **************************************************************************************************
1817 : !> \brief A function that, given a cell index a, returun the index corresponding to -a, and zero if
1818 : !> if out of bounds
1819 : !> \param a_index ...
1820 : !> \param qs_env ...
1821 : !> \return ...
1822 : ! **************************************************************************************************
1823 61137 : FUNCTION get_opp_index(a_index, qs_env) RESULT(opp_index)
1824 : INTEGER, INTENT(IN) :: a_index
1825 : TYPE(qs_environment_type), POINTER :: qs_env
1826 : INTEGER :: opp_index
1827 :
1828 : INTEGER, DIMENSION(3) :: opp_cell
1829 61137 : INTEGER, DIMENSION(:, :), POINTER :: index_to_cell
1830 61137 : INTEGER, DIMENSION(:, :, :), POINTER :: cell_to_index
1831 : TYPE(kpoint_type), POINTER :: kpoints
1832 :
1833 61137 : NULLIFY (kpoints, cell_to_index, index_to_cell)
1834 :
1835 61137 : CALL get_qs_env(qs_env, kpoints=kpoints)
1836 61137 : CALL get_kpoint_info(kpoints, cell_to_index=cell_to_index, index_to_cell=index_to_cell)
1837 :
1838 244548 : opp_cell(:) = -index_to_cell(:, a_index)
1839 :
1840 427959 : IF (ANY([opp_cell(1), opp_cell(2), opp_cell(3)] < LBOUND(cell_to_index)) .OR. &
1841 : ANY([opp_cell(1), opp_cell(2), opp_cell(3)] > UBOUND(cell_to_index))) THEN
1842 :
1843 : opp_index = 0
1844 : ELSE
1845 61137 : opp_index = cell_to_index(opp_cell(1), opp_cell(2), opp_cell(3))
1846 : END IF
1847 :
1848 61137 : END FUNCTION get_opp_index
1849 :
1850 : ! **************************************************************************************************
1851 : !> \brief A routine that returns the actual non-symemtric density matrix for each image, by Fourier
1852 : !> transforming the kpoint density matrix
1853 : !> \param rho_ao_t ...
1854 : !> \param rho_ao ...
1855 : !> \param scale_prev_p ...
1856 : !> \param ri_data ...
1857 : !> \param qs_env ...
1858 : ! **************************************************************************************************
1859 422 : SUBROUTINE get_pmat_images(rho_ao_t, rho_ao, scale_prev_p, ri_data, qs_env)
1860 : TYPE(dbt_type), DIMENSION(:, :), INTENT(INOUT) :: rho_ao_t
1861 : TYPE(dbcsr_p_type), DIMENSION(:, :), INTENT(INOUT) :: rho_ao
1862 : REAL(dp), INTENT(IN) :: scale_prev_p
1863 : TYPE(hfx_ri_type), INTENT(INOUT) :: ri_data
1864 : TYPE(qs_environment_type), POINTER :: qs_env
1865 :
1866 : INTEGER :: cell_j(3), i_img, i_spin, iatom, icol, &
1867 : irow, j_img, jatom, mi_img, mj_img, &
1868 : nimg, nspins
1869 422 : INTEGER, DIMENSION(:, :, :), POINTER :: cell_to_index
1870 : LOGICAL :: found
1871 : REAL(dp) :: fac
1872 422 : REAL(dp), DIMENSION(:, :), POINTER :: pblock, pblock_desymm
1873 422 : TYPE(dbcsr_p_type), DIMENSION(:, :), POINTER :: matrix_ks, rho_desymm
1874 3798 : TYPE(dbt_type) :: tmp
1875 : TYPE(dft_control_type), POINTER :: dft_control
1876 : TYPE(kpoint_type), POINTER :: kpoints
1877 : TYPE(neighbor_list_iterator_p_type), &
1878 422 : DIMENSION(:), POINTER :: nl_iterator
1879 : TYPE(neighbor_list_set_p_type), DIMENSION(:), &
1880 422 : POINTER :: sab_nl, sab_nl_nosym
1881 : TYPE(qs_scf_env_type), POINTER :: scf_env
1882 :
1883 422 : NULLIFY (rho_desymm, kpoints, sab_nl_nosym, scf_env, matrix_ks, dft_control, &
1884 422 : sab_nl, nl_iterator, cell_to_index, pblock, pblock_desymm)
1885 :
1886 422 : CALL get_qs_env(qs_env, kpoints=kpoints, scf_env=scf_env, matrix_ks_kp=matrix_ks, dft_control=dft_control)
1887 422 : CALL get_kpoint_info(kpoints, sab_nl_nosym=sab_nl_nosym, cell_to_index=cell_to_index, sab_nl=sab_nl)
1888 :
1889 422 : IF (dft_control%do_admm) THEN
1890 204 : CALL get_admm_env(qs_env%admm_env, matrix_ks_aux_fit_kp=matrix_ks)
1891 : END IF
1892 :
1893 422 : nspins = SIZE(matrix_ks, 1)
1894 422 : nimg = ri_data%nimg
1895 :
1896 26138 : ALLOCATE (rho_desymm(nspins, nimg))
1897 11040 : DO i_img = 1, nimg
1898 24872 : DO i_spin = 1, nspins
1899 13832 : ALLOCATE (rho_desymm(i_spin, i_img)%matrix)
1900 : CALL dbcsr_create(rho_desymm(i_spin, i_img)%matrix, template=matrix_ks(i_spin, i_img)%matrix, &
1901 13832 : matrix_type=dbcsr_type_no_symmetry)
1902 24450 : CALL cp_dbcsr_alloc_block_from_nbl(rho_desymm(i_spin, i_img)%matrix, sab_nl_nosym)
1903 : END DO
1904 : END DO
1905 422 : CALL dbt_create(rho_desymm(1, 1)%matrix, tmp)
1906 :
1907 : !We transfor the symmtric typed (but not actually symmetric: P_ab^i = P_ba^-i) real-spaced density
1908 : !matrix into proper non-symemtric ones (using the same nl for consistency)
1909 422 : CALL neighbor_list_iterator_create(nl_iterator, sab_nl)
1910 18275 : DO WHILE (neighbor_list_iterate(nl_iterator) == 0)
1911 17853 : CALL get_iterator_info(nl_iterator, iatom=iatom, jatom=jatom, cell=cell_j)
1912 17853 : j_img = cell_to_index(cell_j(1), cell_j(2), cell_j(3))
1913 17853 : IF (j_img > nimg .OR. j_img < 1) CYCLE
1914 :
1915 12737 : fac = 1.0_dp
1916 12737 : IF (iatom == jatom) fac = 0.5_dp
1917 12737 : mj_img = get_opp_index(j_img, qs_env)
1918 : !if no opposite image, then no sum of P^j + P^-j => need full diag
1919 12737 : IF (mj_img == 0) fac = 1.0_dp
1920 :
1921 12737 : irow = iatom
1922 12737 : icol = jatom
1923 12737 : IF (iatom > jatom) THEN
1924 : !because symmetric nl. Value for atom pair i,j is actually stored in j,i if i > j
1925 3997 : irow = jatom
1926 3997 : icol = iatom
1927 : END IF
1928 :
1929 30064 : DO i_spin = 1, nspins
1930 16905 : CALL dbcsr_get_block_p(rho_ao(i_spin, j_img)%matrix, irow, icol, pblock, found)
1931 16905 : IF (.NOT. found) CYCLE
1932 :
1933 : !distribution of symm and non-symm matrix match in that way
1934 16905 : CALL dbcsr_get_block_p(rho_desymm(i_spin, j_img)%matrix, iatom, jatom, pblock_desymm, found)
1935 16905 : IF (.NOT. found) CYCLE
1936 :
1937 68568 : IF (iatom > jatom) THEN
1938 656678 : pblock_desymm(:, :) = fac*TRANSPOSE(pblock(:, :))
1939 : ELSE
1940 1568460 : pblock_desymm(:, :) = fac*pblock(:, :)
1941 : END IF
1942 : END DO
1943 : END DO
1944 422 : CALL neighbor_list_iterator_release(nl_iterator)
1945 :
1946 11040 : DO i_img = 1, nimg
1947 24872 : DO i_spin = 1, nspins
1948 13832 : CALL dbt_scale(rho_ao_t(i_spin, i_img), scale_prev_p)
1949 :
1950 13832 : CALL dbt_copy_matrix_to_tensor(rho_desymm(i_spin, i_img)%matrix, tmp)
1951 13832 : CALL dbt_copy(tmp, rho_ao_t(i_spin, i_img), summation=.TRUE., move_data=.TRUE.)
1952 :
1953 : !symmetrize by addin transpose of opp img
1954 13832 : mi_img = get_opp_index(i_img, qs_env)
1955 13832 : IF (mi_img > 0 .AND. mi_img .LE. nimg) THEN
1956 12464 : CALL dbt_copy_matrix_to_tensor(rho_desymm(i_spin, mi_img)%matrix, tmp)
1957 12464 : CALL dbt_copy(tmp, rho_ao_t(i_spin, i_img), order=[2, 1], summation=.TRUE., move_data=.TRUE.)
1958 : END IF
1959 24450 : CALL dbt_filter(rho_ao_t(i_spin, i_img), ri_data%filter_eps)
1960 : END DO
1961 : END DO
1962 :
1963 11040 : DO i_img = 1, nimg
1964 24872 : DO i_spin = 1, nspins
1965 13832 : CALL dbcsr_release(rho_desymm(i_spin, i_img)%matrix)
1966 24450 : DEALLOCATE (rho_desymm(i_spin, i_img)%matrix)
1967 : END DO
1968 : END DO
1969 :
1970 422 : CALL dbt_destroy(tmp)
1971 422 : DEALLOCATE (rho_desymm)
1972 :
1973 844 : END SUBROUTINE get_pmat_images
1974 :
1975 : ! **************************************************************************************************
1976 : !> \brief A routine that, given a cell index b and atom indices ij, returns a 2c tensor with the HFX
1977 : !> potential (P_i^0|Q_j^b), within the extended RI basis
1978 : !> \param t_2c_pot ...
1979 : !> \param mat_orig ...
1980 : !> \param atom_i ...
1981 : !> \param atom_j ...
1982 : !> \param img_b ...
1983 : !> \param ri_data ...
1984 : !> \param qs_env ...
1985 : !> \param do_inverse ...
1986 : !> \param para_env_ext ...
1987 : !> \param blacs_env_ext ...
1988 : !> \param dbcsr_template ...
1989 : !> \param off_diagonal ...
1990 : !> \param skip_inverse ...
1991 : ! **************************************************************************************************
1992 7318 : SUBROUTINE get_ext_2c_int(t_2c_pot, mat_orig, atom_i, atom_j, img_b, ri_data, qs_env, do_inverse, &
1993 : para_env_ext, blacs_env_ext, dbcsr_template, off_diagonal, skip_inverse)
1994 : TYPE(dbt_type), INTENT(INOUT) :: t_2c_pot
1995 : TYPE(dbcsr_type), DIMENSION(:), INTENT(INOUT) :: mat_orig
1996 : INTEGER, INTENT(IN) :: atom_i, atom_j, img_b
1997 : TYPE(hfx_ri_type), INTENT(INOUT) :: ri_data
1998 : TYPE(qs_environment_type), POINTER :: qs_env
1999 : LOGICAL, INTENT(IN), OPTIONAL :: do_inverse
2000 : TYPE(mp_para_env_type), OPTIONAL, POINTER :: para_env_ext
2001 : TYPE(cp_blacs_env_type), OPTIONAL, POINTER :: blacs_env_ext
2002 : TYPE(dbcsr_type), OPTIONAL, POINTER :: dbcsr_template
2003 : LOGICAL, INTENT(IN), OPTIONAL :: off_diagonal, skip_inverse
2004 :
2005 : CHARACTER(LEN=*), PARAMETER :: routineN = 'get_ext_2c_int'
2006 :
2007 : INTEGER :: blk, group, handle, handle2, i_img, i_RI, iatom, iblk, ikind, img_tot, j_img, &
2008 : j_RI, jatom, jblk, jkind, n_dependent, natom, nblks_RI, nimg, nkind
2009 7318 : INTEGER, ALLOCATABLE, DIMENSION(:) :: dist1, dist2
2010 7318 : INTEGER, ALLOCATABLE, DIMENSION(:, :) :: present_atoms_i, present_atoms_j
2011 : INTEGER, DIMENSION(3) :: cell_b, cell_i, cell_j, cell_tot
2012 7318 : INTEGER, DIMENSION(:), POINTER :: col_dist, col_dist_ext, ri_blk_size_ext, &
2013 7318 : row_dist, row_dist_ext
2014 7318 : INTEGER, DIMENSION(:, :), POINTER :: index_to_cell, pgrid
2015 7318 : INTEGER, DIMENSION(:, :, :), POINTER :: cell_to_index
2016 : LOGICAL :: do_inverse_prv, found, my_offd, &
2017 : skip_inverse_prv, use_template
2018 : REAL(dp) :: bfac, dij, r0, r1, threshold
2019 : REAL(dp), DIMENSION(3) :: ri, rij, rj, rref, scoord
2020 7318 : REAL(dp), DIMENSION(:, :), POINTER :: pblock
2021 : TYPE(cell_type), POINTER :: cell
2022 : TYPE(cp_blacs_env_type), POINTER :: blacs_env
2023 : TYPE(dbcsr_distribution_type) :: dbcsr_dist, dbcsr_dist_ext
2024 : TYPE(dbcsr_iterator_type) :: dbcsr_iter
2025 : TYPE(dbcsr_type) :: work, work_tight, work_tight_inv
2026 51226 : TYPE(dbt_type) :: t_2c_tmp
2027 : TYPE(distribution_2d_type), POINTER :: dist_2d
2028 : TYPE(gto_basis_set_p_type), ALLOCATABLE, &
2029 7318 : DIMENSION(:), TARGET :: basis_set_RI
2030 : TYPE(kpoint_type), POINTER :: kpoints
2031 : TYPE(mp_para_env_type), POINTER :: para_env
2032 : TYPE(neighbor_list_iterator_p_type), &
2033 7318 : DIMENSION(:), POINTER :: nl_iterator
2034 : TYPE(neighbor_list_set_p_type), DIMENSION(:), &
2035 7318 : POINTER :: nl_2c
2036 7318 : TYPE(particle_type), DIMENSION(:), POINTER :: particle_set
2037 7318 : TYPE(qs_kind_type), DIMENSION(:), POINTER :: qs_kind_set
2038 :
2039 7318 : NULLIFY (qs_kind_set, nl_2c, nl_iterator, cell, kpoints, cell_to_index, index_to_cell, dist_2d, &
2040 7318 : para_env, pblock, blacs_env, particle_set, col_dist, row_dist, pgrid, &
2041 7318 : col_dist_ext, row_dist_ext)
2042 :
2043 7318 : CALL timeset(routineN, handle)
2044 :
2045 : !Idea: run over the neighbor list once for i and once for j, and record in which cell the MIC
2046 : ! atoms are. Then loop over the atoms and only take the pairs the we need
2047 :
2048 : CALL get_qs_env(qs_env, natom=natom, nkind=nkind, qs_kind_set=qs_kind_set, cell=cell, &
2049 7318 : kpoints=kpoints, para_env=para_env, blacs_env=blacs_env, particle_set=particle_set)
2050 7318 : CALL get_kpoint_info(kpoints, cell_to_index=cell_to_index, index_to_cell=index_to_cell)
2051 :
2052 7318 : do_inverse_prv = .FALSE.
2053 7318 : IF (PRESENT(do_inverse)) do_inverse_prv = do_inverse
2054 280 : IF (do_inverse_prv) THEN
2055 280 : CPASSERT(atom_i == atom_j)
2056 : END IF
2057 :
2058 7318 : skip_inverse_prv = .FALSE.
2059 7318 : IF (PRESENT(skip_inverse)) skip_inverse_prv = skip_inverse
2060 :
2061 7318 : my_offd = .FALSE.
2062 7318 : IF (PRESENT(off_diagonal)) my_offd = off_diagonal
2063 :
2064 7318 : IF (PRESENT(para_env_ext)) para_env => para_env_ext
2065 7318 : IF (PRESENT(blacs_env_ext)) blacs_env => blacs_env_ext
2066 :
2067 7318 : nimg = SIZE(mat_orig)
2068 :
2069 7318 : CALL timeset(routineN//"_nl_iter", handle2)
2070 :
2071 : !create our own dist_2d in the subgroup
2072 29272 : ALLOCATE (dist1(natom), dist2(natom))
2073 21954 : DO iatom = 1, natom
2074 14636 : dist1(iatom) = MOD(iatom, blacs_env%num_pe(1))
2075 21954 : dist2(iatom) = MOD(iatom, blacs_env%num_pe(2))
2076 : END DO
2077 7318 : CALL distribution_2d_create(dist_2d, dist1, dist2, nkind, particle_set, blacs_env_ext=blacs_env)
2078 :
2079 33055 : ALLOCATE (basis_set_RI(nkind))
2080 7318 : CALL basis_set_list_setup(basis_set_RI, ri_data%ri_basis_type, qs_kind_set)
2081 :
2082 : CALL build_2c_neighbor_lists(nl_2c, basis_set_RI, basis_set_RI, ri_data%ri_metric, &
2083 7318 : "HFX_2c_nl_RI", qs_env, sym_ij=.FALSE., dist_2d=dist_2d)
2084 :
2085 43908 : ALLOCATE (present_atoms_i(natom, nimg), present_atoms_j(natom, nimg))
2086 746851 : present_atoms_i = 0
2087 746851 : present_atoms_j = 0
2088 :
2089 7318 : CALL neighbor_list_iterator_create(nl_iterator, nl_2c)
2090 291602 : DO WHILE (neighbor_list_iterate(nl_iterator) == 0)
2091 : CALL get_iterator_info(nl_iterator, iatom=iatom, jatom=jatom, r=rij, cell=cell_j, &
2092 284284 : ikind=ikind, jkind=jkind)
2093 :
2094 1137136 : dij = NORM2(rij)
2095 :
2096 284284 : j_img = cell_to_index(cell_j(1), cell_j(2), cell_j(3))
2097 284284 : IF (j_img > nimg .OR. j_img < 1) CYCLE
2098 :
2099 282070 : IF (iatom == atom_i .AND. dij .LE. ri_data%kp_RI_range) present_atoms_i(jatom, j_img) = 1
2100 289388 : IF (iatom == atom_j .AND. dij .LE. ri_data%kp_RI_range) present_atoms_j(jatom, j_img) = 1
2101 : END DO
2102 7318 : CALL neighbor_list_iterator_release(nl_iterator)
2103 7318 : CALL release_neighbor_list_sets(nl_2c)
2104 7318 : CALL distribution_2d_release(dist_2d)
2105 7318 : CALL timestop(handle2)
2106 :
2107 7318 : CALL para_env%sum(present_atoms_i)
2108 7318 : CALL para_env%sum(present_atoms_j)
2109 :
2110 : !Need to build a work matrix with matching distribution to mat_orig
2111 : !If template is provided, use it. If not, we create it.
2112 7318 : use_template = .FALSE.
2113 7318 : IF (PRESENT(dbcsr_template)) THEN
2114 6646 : IF (ASSOCIATED(dbcsr_template)) use_template = .TRUE.
2115 : END IF
2116 :
2117 : IF (use_template) THEN
2118 6414 : CALL dbcsr_create(work, template=dbcsr_template)
2119 : ELSE
2120 904 : CALL dbcsr_get_info(mat_orig(1), distribution=dbcsr_dist)
2121 904 : CALL dbcsr_distribution_get(dbcsr_dist, row_dist=row_dist, col_dist=col_dist, group=group, pgrid=pgrid)
2122 3616 : ALLOCATE (row_dist_ext(ri_data%ncell_RI*natom), col_dist_ext(ri_data%ncell_RI*natom))
2123 1808 : ALLOCATE (ri_blk_size_ext(ri_data%ncell_RI*natom))
2124 6428 : DO i_RI = 1, ri_data%ncell_RI
2125 27620 : row_dist_ext((i_RI - 1)*natom + 1:i_RI*natom) = row_dist(:)
2126 27620 : col_dist_ext((i_RI - 1)*natom + 1:i_RI*natom) = col_dist(:)
2127 17476 : RI_blk_size_ext((i_RI - 1)*natom + 1:i_RI*natom) = ri_data%bsizes_RI(:)
2128 : END DO
2129 :
2130 : CALL dbcsr_distribution_new(dbcsr_dist_ext, group=group, pgrid=pgrid, &
2131 904 : row_dist=row_dist_ext, col_dist=col_dist_ext)
2132 : CALL dbcsr_create(work, dist=dbcsr_dist_ext, name="RI_ext", matrix_type=dbcsr_type_no_symmetry, &
2133 904 : row_blk_size=RI_blk_size_ext, col_blk_size=RI_blk_size_ext)
2134 904 : CALL dbcsr_distribution_release(dbcsr_dist_ext)
2135 904 : DEALLOCATE (col_dist_ext, row_dist_ext, RI_blk_size_ext)
2136 :
2137 2712 : IF (PRESENT(dbcsr_template)) THEN
2138 232 : ALLOCATE (dbcsr_template)
2139 232 : CALL dbcsr_create(dbcsr_template, template=work)
2140 : END IF
2141 : END IF !use_template
2142 :
2143 29272 : cell_b(:) = index_to_cell(:, img_b)
2144 253829 : DO i_img = 1, nimg
2145 246511 : i_RI = ri_data%img_to_RI_cell(i_img)
2146 246511 : IF (i_RI == 0) CYCLE
2147 195364 : cell_i(:) = index_to_cell(:, i_img)
2148 2062970 : DO j_img = 1, nimg
2149 2006811 : j_RI = ri_data%img_to_RI_cell(j_img)
2150 2006811 : IF (j_RI == 0) CYCLE
2151 1693732 : cell_j(:) = index_to_cell(:, j_img)
2152 1693732 : cell_tot = cell_j - cell_i + cell_b
2153 :
2154 2924091 : IF (ANY([cell_tot(1), cell_tot(2), cell_tot(3)] < LBOUND(cell_to_index)) .OR. &
2155 : ANY([cell_tot(1), cell_tot(2), cell_tot(3)] > UBOUND(cell_to_index))) CYCLE
2156 387929 : img_tot = cell_to_index(cell_tot(1), cell_tot(2), cell_tot(3))
2157 387929 : IF (img_tot > nimg .OR. img_tot < 1) CYCLE
2158 :
2159 263923 : CALL dbcsr_iterator_start(dbcsr_iter, mat_orig(img_tot))
2160 754095 : DO WHILE (dbcsr_iterator_blocks_left(dbcsr_iter))
2161 490172 : CALL dbcsr_iterator_next_block(dbcsr_iter, row=iatom, column=jatom, blk=blk)
2162 490172 : IF (present_atoms_i(iatom, i_img) == 0) CYCLE
2163 184998 : IF (present_atoms_j(jatom, j_img) == 0) CYCLE
2164 80543 : IF (my_offd .AND. (i_RI - 1)*natom + iatom == (j_RI - 1)*natom + jatom) CYCLE
2165 :
2166 80254 : CALL dbcsr_get_block_p(mat_orig(img_tot), iatom, jatom, pblock, found)
2167 80254 : IF (.NOT. found) CYCLE
2168 :
2169 754095 : CALL dbcsr_put_block(work, (i_RI - 1)*natom + iatom, (j_RI - 1)*natom + jatom, pblock)
2170 :
2171 : END DO
2172 2481741 : CALL dbcsr_iterator_stop(dbcsr_iter)
2173 :
2174 : END DO !j_img
2175 : END DO !i_img
2176 7318 : CALL dbcsr_finalize(work)
2177 :
2178 7318 : IF (do_inverse_prv) THEN
2179 :
2180 280 : r1 = ri_data%kp_RI_range
2181 280 : r0 = ri_data%kp_bump_rad
2182 :
2183 : !Because there are a lot of empty rows/cols in work, we need to get rid of them for inversion
2184 20872 : nblks_RI = SUM(present_atoms_i)
2185 1400 : ALLOCATE (col_dist_ext(nblks_RI), row_dist_ext(nblks_RI), RI_blk_size_ext(nblks_RI))
2186 280 : iblk = 0
2187 7144 : DO i_img = 1, nimg
2188 6864 : i_RI = ri_data%img_to_RI_cell(i_img)
2189 6864 : IF (i_RI == 0) CYCLE
2190 5512 : DO iatom = 1, natom
2191 3488 : IF (present_atoms_i(iatom, i_img) == 0) CYCLE
2192 1156 : iblk = iblk + 1
2193 1156 : col_dist_ext(iblk) = col_dist(iatom)
2194 1156 : row_dist_ext(iblk) = row_dist(iatom)
2195 10352 : RI_blk_size_ext(iblk) = ri_data%bsizes_RI(iatom)
2196 : END DO
2197 : END DO
2198 :
2199 : CALL dbcsr_distribution_new(dbcsr_dist_ext, group=group, pgrid=pgrid, &
2200 280 : row_dist=row_dist_ext, col_dist=col_dist_ext)
2201 : CALL dbcsr_create(work_tight, dist=dbcsr_dist_ext, name="RI_ext", matrix_type=dbcsr_type_no_symmetry, &
2202 280 : row_blk_size=RI_blk_size_ext, col_blk_size=RI_blk_size_ext)
2203 : CALL dbcsr_create(work_tight_inv, dist=dbcsr_dist_ext, name="RI_ext", matrix_type=dbcsr_type_no_symmetry, &
2204 280 : row_blk_size=RI_blk_size_ext, col_blk_size=RI_blk_size_ext)
2205 280 : CALL dbcsr_distribution_release(dbcsr_dist_ext)
2206 280 : DEALLOCATE (col_dist_ext, row_dist_ext, RI_blk_size_ext)
2207 :
2208 : !We apply a bump function to the RI metric inverse for smooth RI basis extension:
2209 : ! S^-1 = B * ((P|Q)_D + B*(P|Q)_OD*B)^-1 * B, with D block-diagonal blocks and OD off-diagonal
2210 280 : rref = pbc(particle_set(atom_i)%r, cell)
2211 :
2212 280 : iblk = 0
2213 7144 : DO i_img = 1, nimg
2214 6864 : i_RI = ri_data%img_to_RI_cell(i_img)
2215 6864 : IF (i_RI == 0) CYCLE
2216 5512 : DO iatom = 1, natom
2217 3488 : IF (present_atoms_i(iatom, i_img) == 0) CYCLE
2218 1156 : iblk = iblk + 1
2219 :
2220 1156 : CALL real_to_scaled(scoord, pbc(particle_set(iatom)%r, cell), cell)
2221 4624 : CALL scaled_to_real(ri, scoord(:) + index_to_cell(:, i_img), cell)
2222 :
2223 1156 : jblk = 0
2224 42764 : DO j_img = 1, nimg
2225 34744 : j_RI = ri_data%img_to_RI_cell(j_img)
2226 34744 : IF (j_RI == 0) CYCLE
2227 29504 : DO jatom = 1, natom
2228 17344 : IF (present_atoms_j(jatom, j_img) == 0) CYCLE
2229 5580 : jblk = jblk + 1
2230 :
2231 5580 : CALL real_to_scaled(scoord, pbc(particle_set(jatom)%r, cell), cell)
2232 22320 : CALL scaled_to_real(rj, scoord(:) + index_to_cell(:, j_img), cell)
2233 :
2234 5580 : CALL dbcsr_get_block_p(work, (i_RI - 1)*natom + iatom, (j_RI - 1)*natom + jatom, pblock, found)
2235 5580 : IF (.NOT. found) CYCLE
2236 :
2237 2508 : bfac = 1.0_dp
2238 14088 : IF (iblk .NE. jblk) bfac = bump(NORM2(ri - rref), r0, r1)*bump(NORM2(rj - rref), r0, r1)
2239 5075024 : CALL dbcsr_put_block(work_tight, iblk, jblk, bfac*pblock(:, :))
2240 : END DO
2241 : END DO
2242 : END DO
2243 : END DO
2244 280 : CALL dbcsr_finalize(work_tight)
2245 280 : CALL dbcsr_clear(work)
2246 :
2247 280 : IF (.NOT. skip_inverse_prv) THEN
2248 140 : SELECT CASE (ri_data%t2c_method)
2249 : CASE (hfx_ri_do_2c_iter)
2250 0 : threshold = MAX(ri_data%filter_eps, 1.0e-12_dp)
2251 0 : CALL invert_hotelling(work_tight_inv, work_tight, threshold=threshold, silent=.FALSE.)
2252 : CASE (hfx_ri_do_2c_cholesky)
2253 140 : CALL dbcsr_copy(work_tight_inv, work_tight)
2254 140 : CALL cp_dbcsr_cholesky_decompose(work_tight_inv, para_env=para_env, blacs_env=blacs_env)
2255 : CALL cp_dbcsr_cholesky_invert(work_tight_inv, para_env=para_env, blacs_env=blacs_env, &
2256 140 : upper_to_full=.TRUE.)
2257 : CASE (hfx_ri_do_2c_diag)
2258 0 : CALL dbcsr_copy(work_tight_inv, work_tight)
2259 : CALL cp_dbcsr_power(work_tight_inv, -1.0_dp, ri_data%eps_eigval, n_dependent, &
2260 140 : para_env, blacs_env, verbose=ri_data%unit_nr_dbcsr > 0)
2261 : END SELECT
2262 : ELSE
2263 140 : CALL dbcsr_copy(work_tight_inv, work_tight)
2264 : END IF
2265 :
2266 : !move back data to standard extended RI pattern
2267 : !Note: we apply the external bump to ((P|Q)_D + B*(P|Q)_OD*B)^-1 later, because this matrix
2268 : ! is required for forces
2269 280 : iblk = 0
2270 7144 : DO i_img = 1, nimg
2271 6864 : i_RI = ri_data%img_to_RI_cell(i_img)
2272 6864 : IF (i_RI == 0) CYCLE
2273 5512 : DO iatom = 1, natom
2274 3488 : IF (present_atoms_i(iatom, i_img) == 0) CYCLE
2275 1156 : iblk = iblk + 1
2276 :
2277 1156 : jblk = 0
2278 42764 : DO j_img = 1, nimg
2279 34744 : j_RI = ri_data%img_to_RI_cell(j_img)
2280 34744 : IF (j_RI == 0) CYCLE
2281 29504 : DO jatom = 1, natom
2282 17344 : IF (present_atoms_j(jatom, j_img) == 0) CYCLE
2283 5580 : jblk = jblk + 1
2284 :
2285 5580 : CALL dbcsr_get_block_p(work_tight_inv, iblk, jblk, pblock, found)
2286 5580 : IF (.NOT. found) CYCLE
2287 :
2288 54737 : CALL dbcsr_put_block(work, (i_RI - 1)*natom + iatom, (j_RI - 1)*natom + jatom, pblock)
2289 : END DO
2290 : END DO
2291 : END DO
2292 : END DO
2293 280 : CALL dbcsr_finalize(work)
2294 :
2295 280 : CALL dbcsr_release(work_tight)
2296 560 : CALL dbcsr_release(work_tight_inv)
2297 : END IF
2298 :
2299 7318 : CALL dbt_create(work, t_2c_tmp)
2300 7318 : CALL dbt_copy_matrix_to_tensor(work, t_2c_tmp)
2301 7318 : CALL dbt_copy(t_2c_tmp, t_2c_pot, move_data=.TRUE.)
2302 7318 : CALL dbt_filter(t_2c_pot, ri_data%filter_eps)
2303 :
2304 7318 : CALL dbt_destroy(t_2c_tmp)
2305 7318 : CALL dbcsr_release(work)
2306 :
2307 7318 : CALL timestop(handle)
2308 :
2309 29272 : END SUBROUTINE get_ext_2c_int
2310 :
2311 : ! **************************************************************************************************
2312 : !> \brief Pre-contract the density matrices with the 3-center integrals:
2313 : !> P_sigma^a,lambda^a+c (mu^0 sigma^a| P^0)
2314 : !> \param t_3c_apc ...
2315 : !> \param rho_ao_t ...
2316 : !> \param ri_data ...
2317 : !> \param qs_env ...
2318 : ! **************************************************************************************************
2319 232 : SUBROUTINE contract_pmat_3c(t_3c_apc, rho_ao_t, ri_data, qs_env)
2320 : TYPE(dbt_type), DIMENSION(:, :), INTENT(INOUT) :: t_3c_apc, rho_ao_t
2321 : TYPE(hfx_ri_type), INTENT(INOUT) :: ri_data
2322 : TYPE(qs_environment_type), POINTER :: qs_env
2323 :
2324 : CHARACTER(len=*), PARAMETER :: routineN = 'contract_pmat_3c'
2325 :
2326 : INTEGER :: apc_img, batch_size, handle, i_batch, &
2327 : i_img, i_spin, j_batch, n_batch_img, &
2328 : n_batch_nze, nimg, nimg_nze, nspins
2329 : INTEGER(int_8) :: nflop, nze
2330 232 : INTEGER, ALLOCATABLE, DIMENSION(:) :: batch_ranges_img, batch_ranges_nze, &
2331 232 : int_indices
2332 232 : INTEGER, ALLOCATABLE, DIMENSION(:, :) :: ac_pairs
2333 : REAL(dp) :: occ, t1, t2
2334 2088 : TYPE(dbt_type) :: t_3c_tmp
2335 232 : TYPE(dbt_type), ALLOCATABLE, DIMENSION(:) :: ints_stack, res_stack, rho_stack
2336 : TYPE(dft_control_type), POINTER :: dft_control
2337 :
2338 232 : CALL timeset(routineN, handle)
2339 :
2340 232 : CALL get_qs_env(qs_env, dft_control=dft_control)
2341 :
2342 232 : nimg = ri_data%nimg
2343 232 : nimg_nze = ri_data%nimg_nze
2344 232 : nspins = dft_control%nspins
2345 :
2346 232 : CALL dbt_create(t_3c_apc(1, 1), t_3c_tmp)
2347 :
2348 232 : batch_size = nimg/ri_data%n_mem
2349 :
2350 : !batching over all images
2351 232 : n_batch_img = nimg/batch_size
2352 232 : IF (MODULO(nimg, batch_size) .NE. 0) n_batch_img = n_batch_img + 1
2353 696 : ALLOCATE (batch_ranges_img(n_batch_img + 1))
2354 810 : DO i_batch = 1, n_batch_img
2355 810 : batch_ranges_img(i_batch) = (i_batch - 1)*batch_size + 1
2356 : END DO
2357 232 : batch_ranges_img(n_batch_img + 1) = nimg + 1
2358 :
2359 : !batching over images with non-zero 3c integrals
2360 232 : n_batch_nze = nimg_nze/batch_size
2361 232 : IF (MODULO(nimg_nze, batch_size) .NE. 0) n_batch_nze = n_batch_nze + 1
2362 696 : ALLOCATE (batch_ranges_nze(n_batch_nze + 1))
2363 752 : DO i_batch = 1, n_batch_nze
2364 752 : batch_ranges_nze(i_batch) = (i_batch - 1)*batch_size + 1
2365 : END DO
2366 232 : batch_ranges_nze(n_batch_nze + 1) = nimg_nze + 1
2367 :
2368 : !Create the stack tensors in the approriate distribution
2369 7192 : ALLOCATE (rho_stack(2), ints_stack(2), res_stack(2))
2370 : CALL get_stack_tensors(res_stack, rho_stack, ints_stack, rho_ao_t(1, 1), &
2371 232 : ri_data%t_3c_int_ctr_1(1, 1), batch_size, ri_data, qs_env)
2372 :
2373 1160 : ALLOCATE (ac_pairs(nimg, 2), int_indices(nimg_nze))
2374 4554 : DO i_img = 1, nimg_nze
2375 4554 : int_indices(i_img) = i_img
2376 : END DO
2377 :
2378 232 : t1 = m_walltime()
2379 752 : DO j_batch = 1, n_batch_nze
2380 : !First batch is over the integrals. They are always in the same order, consistent with get_ac_pairs
2381 : CALL fill_3c_stack(ints_stack(1), ri_data%t_3c_int_ctr_1(1, :), int_indices, 3, ri_data, &
2382 1560 : img_bounds=[batch_ranges_nze(j_batch), batch_ranges_nze(j_batch + 1)])
2383 520 : CALL dbt_copy(ints_stack(1), ints_stack(2), move_data=.TRUE.)
2384 :
2385 1492 : DO i_spin = 1, nspins
2386 3152 : DO i_batch = 1, n_batch_img
2387 : !Second batch is over the P matrix. Here we fill the stacked rho tensors col by col
2388 17120 : DO apc_img = batch_ranges_img(i_batch), batch_ranges_img(i_batch + 1) - 1
2389 15228 : CALL get_ac_pairs(ac_pairs, apc_img, ri_data, qs_env)
2390 : CALL fill_2c_stack(rho_stack(1), rho_ao_t(i_spin, :), ac_pairs(:, 2), 1, ri_data, &
2391 : img_bounds=[batch_ranges_nze(j_batch), batch_ranges_nze(j_batch + 1)], &
2392 47576 : shift=apc_img - batch_ranges_img(i_batch) + 1)
2393 :
2394 : END DO !apc_img
2395 1892 : CALL get_tensor_occupancy(rho_stack(1), nze, occ)
2396 1892 : IF (nze == 0) CYCLE
2397 1650 : CALL dbt_copy(rho_stack(1), rho_stack(2), move_data=.TRUE.)
2398 :
2399 : !The actual contraction
2400 1650 : CALL dbt_batched_contract_init(rho_stack(2))
2401 : CALL dbt_contract(1.0_dp, ints_stack(2), rho_stack(2), &
2402 : 0.0_dp, res_stack(2), map_1=[1, 2], map_2=[3], &
2403 : contract_1=[3], notcontract_1=[1, 2], &
2404 : contract_2=[1], notcontract_2=[2], &
2405 1650 : filter_eps=ri_data%filter_eps, flop=nflop)
2406 1650 : ri_data%dbcsr_nflop = ri_data%dbcsr_nflop + nflop
2407 1650 : CALL dbt_batched_contract_finalize(rho_stack(2))
2408 1650 : CALL dbt_copy(res_stack(2), res_stack(1), move_data=.TRUE.)
2409 :
2410 18680 : DO apc_img = batch_ranges_img(i_batch), batch_ranges_img(i_batch + 1) - 1
2411 : !Destack the resulting tensor and put it in t_3c_apc with correct apc_img
2412 14398 : CALL unstack_t_3c_apc(t_3c_tmp, res_stack(1), apc_img - batch_ranges_img(i_batch) + 1)
2413 16290 : CALL dbt_copy(t_3c_tmp, t_3c_apc(i_spin, apc_img), summation=.TRUE., move_data=.TRUE.)
2414 : END DO
2415 :
2416 : END DO !i_batch
2417 : END DO !i_spin
2418 : END DO !j_batch
2419 232 : DEALLOCATE (batch_ranges_img)
2420 232 : DEALLOCATE (batch_ranges_nze)
2421 232 : t2 = m_walltime()
2422 232 : ri_data%dbcsr_time = ri_data%dbcsr_time + t2 - t1
2423 :
2424 232 : CALL dbt_destroy(rho_stack(1))
2425 232 : CALL dbt_destroy(rho_stack(2))
2426 232 : CALL dbt_destroy(ints_stack(1))
2427 232 : CALL dbt_destroy(ints_stack(2))
2428 232 : CALL dbt_destroy(res_stack(1))
2429 232 : CALL dbt_destroy(res_stack(2))
2430 232 : CALL dbt_destroy(t_3c_tmp)
2431 :
2432 232 : CALL timestop(handle)
2433 :
2434 2320 : END SUBROUTINE contract_pmat_3c
2435 :
2436 : ! **************************************************************************************************
2437 : !> \brief Pre-contract 3-center integrals with the bumped invrse RI metric, for each atom
2438 : !> \param t_3c_int ...
2439 : !> \param ri_data ...
2440 : !> \param qs_env ...
2441 : ! **************************************************************************************************
2442 70 : SUBROUTINE precontract_3c_ints(t_3c_int, ri_data, qs_env)
2443 : TYPE(dbt_type), DIMENSION(:, :), INTENT(INOUT) :: t_3c_int
2444 : TYPE(hfx_ri_type), INTENT(INOUT) :: ri_data
2445 : TYPE(qs_environment_type), POINTER :: qs_env
2446 :
2447 : CHARACTER(len=*), PARAMETER :: routineN = 'precontract_3c_ints'
2448 :
2449 : INTEGER :: batch_size, handle, i_batch, i_img, &
2450 : i_RI, iatom, is, n_batch, natom, &
2451 : nblks, nblks_3c(3), nimg
2452 : INTEGER(int_8) :: nflop
2453 70 : INTEGER, ALLOCATABLE, DIMENSION(:) :: batch_ranges, bsizes_RI_ext, bsizes_RI_ext_split, &
2454 70 : bsizes_stack, dist1, dist2, dist3, dist_stack3, idx_to_at_AO, int_indices
2455 630 : TYPE(dbt_distribution_type) :: t_dist
2456 14630 : TYPE(dbt_type) :: t_2c_RI_tmp(2), t_3c_tmp(3)
2457 :
2458 70 : CALL timeset(routineN, handle)
2459 :
2460 70 : CALL get_qs_env(qs_env, natom=natom)
2461 :
2462 70 : nimg = ri_data%nimg
2463 210 : ALLOCATE (int_indices(nimg))
2464 1786 : DO i_img = 1, nimg
2465 1786 : int_indices(i_img) = i_img
2466 : END DO
2467 :
2468 210 : ALLOCATE (idx_to_at_AO(SIZE(ri_data%bsizes_AO_split)))
2469 70 : CALL get_idx_to_atom(idx_to_at_AO, ri_data%bsizes_AO_split, ri_data%bsizes_AO)
2470 :
2471 70 : nblks = SIZE(ri_data%bsizes_RI_split)
2472 210 : ALLOCATE (bsizes_RI_ext(ri_data%ncell_RI*natom))
2473 210 : ALLOCATE (bsizes_RI_ext_split(ri_data%ncell_RI*nblks))
2474 506 : DO i_RI = 1, ri_data%ncell_RI
2475 1308 : bsizes_RI_ext((i_RI - 1)*natom + 1:i_RI*natom) = ri_data%bsizes_RI(:)
2476 2344 : bsizes_RI_ext_split((i_RI - 1)*nblks + 1:i_RI*nblks) = ri_data%bsizes_RI_split(:)
2477 : END DO
2478 : CALL create_2c_tensor(t_2c_RI_tmp(1), dist1, dist2, ri_data%pgrid_2d, &
2479 : bsizes_RI_ext, bsizes_RI_ext, &
2480 : name="(RI | RI)")
2481 70 : DEALLOCATE (dist1, dist2)
2482 : CALL create_2c_tensor(t_2c_RI_tmp(2), dist1, dist2, ri_data%pgrid_2d, &
2483 : bsizes_RI_ext_split, bsizes_RI_ext_split, &
2484 : name="(RI | RI)")
2485 70 : DEALLOCATE (dist1, dist2)
2486 :
2487 : !For more efficiency, we stack multiple images of the 3-center integrals into a single tensor
2488 70 : batch_size = nimg/ri_data%n_mem
2489 70 : n_batch = nimg/batch_size
2490 70 : IF (MODULO(nimg, batch_size) .NE. 0) n_batch = n_batch + 1
2491 210 : ALLOCATE (batch_ranges(n_batch + 1))
2492 246 : DO i_batch = 1, n_batch
2493 246 : batch_ranges(i_batch) = (i_batch - 1)*batch_size + 1
2494 : END DO
2495 70 : batch_ranges(n_batch + 1) = nimg + 1
2496 :
2497 70 : nblks = SIZE(ri_data%bsizes_AO_split)
2498 210 : ALLOCATE (bsizes_stack(batch_size*nblks))
2499 910 : DO is = 1, batch_size
2500 3502 : bsizes_stack((is - 1)*nblks + 1:is*nblks) = ri_data%bsizes_AO_split(:)
2501 : END DO
2502 :
2503 70 : CALL dbt_get_info(t_3c_int(1, 1), nblks_total=nblks_3c)
2504 630 : ALLOCATE (dist1(nblks_3c(1)), dist2(nblks_3c(2)), dist3(nblks_3c(3)), dist_stack3(batch_size*nblks_3c(3)))
2505 70 : CALL dbt_get_info(t_3c_int(1, 1), proc_dist_1=dist1, proc_dist_2=dist2, proc_dist_3=dist3)
2506 910 : DO is = 1, batch_size
2507 3502 : dist_stack3((is - 1)*nblks_3c(3) + 1:is*nblks_3c(3)) = dist3(:)
2508 : END DO
2509 :
2510 70 : CALL dbt_distribution_new(t_dist, ri_data%pgrid, dist1, dist2, dist_stack3)
2511 : CALL dbt_create(t_3c_tmp(1), "ints_stack", t_dist, [1], [2, 3], bsizes_RI_ext_split, &
2512 70 : ri_data%bsizes_AO_split, bsizes_stack)
2513 70 : CALL dbt_distribution_destroy(t_dist)
2514 70 : DEALLOCATE (dist1, dist2, dist3, dist_stack3)
2515 :
2516 70 : CALL dbt_create(t_3c_tmp(1), t_3c_tmp(2))
2517 70 : CALL dbt_create(t_3c_int(1, 1), t_3c_tmp(3))
2518 :
2519 210 : DO iatom = 1, natom
2520 140 : CALL dbt_copy(ri_data%t_2c_inv(1, iatom), t_2c_RI_tmp(1))
2521 140 : CALL apply_bump(t_2c_RI_tmp(1), iatom, ri_data, qs_env, from_left=.TRUE., from_right=.TRUE.)
2522 140 : CALL dbt_copy(t_2c_RI_tmp(1), t_2c_RI_tmp(2), move_data=.TRUE.)
2523 :
2524 140 : CALL dbt_batched_contract_init(t_2c_RI_tmp(2))
2525 492 : DO i_batch = 1, n_batch
2526 :
2527 : CALL fill_3c_stack(t_3c_tmp(1), t_3c_int(1, :), int_indices, 3, ri_data, &
2528 : img_bounds=[batch_ranges(i_batch), batch_ranges(i_batch + 1)], &
2529 1056 : filter_at=iatom, filter_dim=2, idx_to_at=idx_to_at_AO)
2530 :
2531 : CALL dbt_contract(1.0_dp, t_2c_RI_tmp(2), t_3c_tmp(1), &
2532 : 0.0_dp, t_3c_tmp(2), map_1=[1], map_2=[2, 3], &
2533 : contract_1=[2], notcontract_1=[1], &
2534 : contract_2=[1], notcontract_2=[2, 3], &
2535 352 : filter_eps=ri_data%filter_eps, flop=nflop)
2536 352 : ri_data%dbcsr_nflop = ri_data%dbcsr_nflop + nflop
2537 :
2538 3784 : DO i_img = batch_ranges(i_batch), batch_ranges(i_batch + 1) - 1
2539 3432 : CALL unstack_t_3c_apc(t_3c_tmp(3), t_3c_tmp(2), i_img - batch_ranges(i_batch) + 1)
2540 : CALL dbt_copy(t_3c_tmp(3), ri_data%t_3c_int_ctr_1(1, i_img), summation=.TRUE., &
2541 3784 : order=[2, 1, 3], move_data=.TRUE.)
2542 : END DO
2543 492 : CALL dbt_clear(t_3c_tmp(1))
2544 : END DO
2545 210 : CALL dbt_batched_contract_finalize(t_2c_RI_tmp(2))
2546 :
2547 : END DO
2548 70 : CALL dbt_destroy(t_2c_RI_tmp(1))
2549 70 : CALL dbt_destroy(t_2c_RI_tmp(2))
2550 70 : CALL dbt_destroy(t_3c_tmp(1))
2551 70 : CALL dbt_destroy(t_3c_tmp(2))
2552 70 : CALL dbt_destroy(t_3c_tmp(3))
2553 :
2554 1786 : DO i_img = 1, nimg
2555 1786 : CALL dbt_destroy(t_3c_int(1, i_img))
2556 : END DO
2557 :
2558 70 : CALL timestop(handle)
2559 :
2560 350 : END SUBROUTINE precontract_3c_ints
2561 :
2562 : ! **************************************************************************************************
2563 : !> \brief Copy the data of a 2D tensor living in the main MPI group to a sub-group, given the proc
2564 : !> mapping from one to the other (e.g. for a proc idx in the subgroup, we get the idx in the main)
2565 : !> \param t2c_sub ...
2566 : !> \param t2c_main ...
2567 : !> \param group_size ...
2568 : !> \param ngroups ...
2569 : !> \param para_env ...
2570 : ! **************************************************************************************************
2571 8000 : SUBROUTINE copy_2c_to_subgroup(t2c_sub, t2c_main, group_size, ngroups, para_env)
2572 : TYPE(dbt_type), INTENT(INOUT) :: t2c_sub, t2c_main
2573 : INTEGER, INTENT(IN) :: group_size, ngroups
2574 : TYPE(mp_para_env_type), POINTER :: para_env
2575 :
2576 : INTEGER :: batch_size, i, i_batch, i_msg, iblk, &
2577 : igroup, iproc, ir, is, jblk, n_batch, &
2578 : nocc, tag
2579 8000 : INTEGER, ALLOCATABLE, DIMENSION(:) :: bsizes1, bsizes2
2580 8000 : INTEGER, ALLOCATABLE, DIMENSION(:, :) :: block_dest, block_source
2581 8000 : INTEGER, ALLOCATABLE, DIMENSION(:, :, :) :: current_dest
2582 : INTEGER, DIMENSION(2) :: ind, nblks
2583 : LOGICAL :: found
2584 8000 : REAL(dp), ALLOCATABLE, DIMENSION(:, :) :: blk
2585 8000 : TYPE(cp_2d_r_p_type), ALLOCATABLE, DIMENSION(:) :: recv_buff, send_buff
2586 : TYPE(dbt_iterator_type) :: iter
2587 8000 : TYPE(mp_request_type), ALLOCATABLE, DIMENSION(:) :: recv_req, send_req
2588 :
2589 : !Stategy: we loop over the main tensor, and send all the data. Then we loop over the sub tensor
2590 : ! and receive it. We do all of it with async MPI communication. The sub tensor needs
2591 : ! to have blocks pre-reserved though
2592 :
2593 8000 : CALL dbt_get_info(t2c_main, nblks_total=nblks)
2594 :
2595 : !Loop over the main tensor, count how many blocks are there, which ones, and on which proc
2596 32000 : ALLOCATE (block_source(nblks(1), nblks(2)))
2597 166468 : block_source = -1
2598 8000 : nocc = 0
2599 8000 : !$OMP PARALLEL DEFAULT(NONE) SHARED(t2c_main,para_env,nocc,block_source) PRIVATE(iter,ind,blk,found)
2600 : CALL dbt_iterator_start(iter, t2c_main)
2601 : DO WHILE (dbt_iterator_blocks_left(iter))
2602 : CALL dbt_iterator_next_block(iter, ind)
2603 : CALL dbt_get_block(t2c_main, ind, blk, found)
2604 : IF (.NOT. found) CYCLE
2605 :
2606 : block_source(ind(1), ind(2)) = para_env%mepos
2607 : !$OMP ATOMIC
2608 : nocc = nocc + 1
2609 : DEALLOCATE (blk)
2610 : END DO
2611 : CALL dbt_iterator_stop(iter)
2612 : !$OMP END PARALLEL
2613 :
2614 8000 : CALL para_env%sum(nocc)
2615 8000 : CALL para_env%sum(block_source)
2616 166468 : block_source = block_source + para_env%num_pe - 1
2617 8000 : IF (nocc == 0) RETURN
2618 :
2619 : !Loop over the sub tensor, get the block destination
2620 7820 : igroup = para_env%mepos/group_size
2621 23460 : ALLOCATE (block_dest(nblks(1), nblks(2)))
2622 165208 : block_dest = -1
2623 29840 : DO jblk = 1, nblks(2)
2624 165208 : DO iblk = 1, nblks(1)
2625 135368 : IF (block_source(iblk, jblk) == -1) CYCLE
2626 :
2627 98868 : CALL dbt_get_stored_coordinates(t2c_sub, [iblk, jblk], iproc)
2628 157388 : block_dest(iblk, jblk) = igroup*group_size + iproc !mapping of iproc in subgroup to main group idx
2629 : END DO
2630 : END DO
2631 :
2632 39100 : ALLOCATE (bsizes1(nblks(1)), bsizes2(nblks(2)))
2633 7820 : CALL dbt_get_info(t2c_main, blk_size_1=bsizes1, blk_size_2=bsizes2)
2634 :
2635 39100 : ALLOCATE (current_dest(nblks(1), nblks(2), 0:ngroups - 1))
2636 23460 : DO igroup = 0, ngroups - 1
2637 : !for a given subgroup, need to make the destination available to everyone in the main group
2638 330416 : current_dest(:, :, igroup) = block_dest(:, :)
2639 23460 : CALL para_env%bcast(current_dest(:, :, igroup), source=igroup*group_size) !bcast from first proc in sub-group
2640 : END DO
2641 :
2642 : !We go by batches, which cannot be larger than the maximum MPI tag value
2643 7820 : batch_size = MIN(para_env%get_tag_ub(), 128000, nocc*ngroups)
2644 7820 : n_batch = (nocc*ngroups)/batch_size
2645 7820 : IF (MODULO(nocc*ngroups, batch_size) .NE. 0) n_batch = n_batch + 1
2646 :
2647 15640 : DO i_batch = 1, n_batch
2648 : !Loop over groups, blocks and send/receive
2649 163104 : ALLOCATE (send_buff(batch_size), recv_buff(batch_size))
2650 163104 : ALLOCATE (send_req(batch_size), recv_req(batch_size))
2651 : ir = 0
2652 : is = 0
2653 : i_msg = 0
2654 29840 : DO jblk = 1, nblks(2)
2655 165208 : DO iblk = 1, nblks(1)
2656 428124 : DO igroup = 0, ngroups - 1
2657 270736 : IF (block_source(iblk, jblk) == -1) CYCLE
2658 :
2659 65912 : i_msg = i_msg + 1
2660 65912 : IF (i_msg < (i_batch - 1)*batch_size + 1 .OR. i_msg > i_batch*batch_size) CYCLE
2661 :
2662 : !a unique tag per block, within this batch
2663 65912 : tag = i_msg - (i_batch - 1)*batch_size
2664 :
2665 65912 : found = .FALSE.
2666 65912 : IF (para_env%mepos == block_source(iblk, jblk)) THEN
2667 98868 : CALL dbt_get_block(t2c_main, [iblk, jblk], blk, found)
2668 : END IF
2669 :
2670 : !If blocks live on same proc, simply copy. Else MPI send/recv
2671 65912 : IF (block_source(iblk, jblk) == current_dest(iblk, jblk, igroup)) THEN
2672 98868 : IF (found) CALL dbt_put_block(t2c_sub, [iblk, jblk], SHAPE(blk), blk)
2673 : ELSE
2674 32956 : IF (para_env%mepos == block_source(iblk, jblk) .AND. found) THEN
2675 65912 : ALLOCATE (send_buff(tag)%array(bsizes1(iblk), bsizes2(jblk)))
2676 21062150 : send_buff(tag)%array(:, :) = blk(:, :)
2677 16478 : is = is + 1
2678 : CALL para_env%isend(msgin=send_buff(tag)%array, dest=current_dest(iblk, jblk, igroup), &
2679 16478 : request=send_req(is), tag=tag)
2680 : END IF
2681 :
2682 32956 : IF (para_env%mepos == current_dest(iblk, jblk, igroup)) THEN
2683 65912 : ALLOCATE (recv_buff(tag)%array(bsizes1(iblk), bsizes2(jblk)))
2684 16478 : ir = ir + 1
2685 : CALL para_env%irecv(msgout=recv_buff(tag)%array, source=block_source(iblk, jblk), &
2686 16478 : request=recv_req(ir), tag=tag)
2687 : END IF
2688 : END IF
2689 :
2690 201280 : IF (found) DEALLOCATE (blk)
2691 : END DO
2692 : END DO
2693 : END DO
2694 :
2695 7820 : CALL mp_waitall(send_req(1:is))
2696 7820 : CALL mp_waitall(recv_req(1:ir))
2697 : !clean-up
2698 73732 : DO i = 1, batch_size
2699 73732 : IF (ASSOCIATED(send_buff(i)%array)) DEALLOCATE (send_buff(i)%array)
2700 : END DO
2701 :
2702 : !Finally copy the data from the buffer to the sub-tensor
2703 : i_msg = 0
2704 29840 : DO jblk = 1, nblks(2)
2705 165208 : DO iblk = 1, nblks(1)
2706 428124 : DO igroup = 0, ngroups - 1
2707 270736 : IF (block_source(iblk, jblk) == -1) CYCLE
2708 :
2709 65912 : i_msg = i_msg + 1
2710 65912 : IF (i_msg < (i_batch - 1)*batch_size + 1 .OR. i_msg > i_batch*batch_size) CYCLE
2711 :
2712 : !a unique tag per block, within this batch
2713 65912 : tag = i_msg - (i_batch - 1)*batch_size
2714 :
2715 65912 : IF (para_env%mepos == current_dest(iblk, jblk, igroup) .AND. &
2716 135368 : block_source(iblk, jblk) .NE. current_dest(iblk, jblk, igroup)) THEN
2717 :
2718 65912 : ALLOCATE (blk(bsizes1(iblk), bsizes2(jblk)))
2719 21062150 : blk(:, :) = recv_buff(tag)%array(:, :)
2720 82390 : CALL dbt_put_block(t2c_sub, [iblk, jblk], SHAPE(blk), blk)
2721 16478 : DEALLOCATE (blk)
2722 : END IF
2723 : END DO
2724 : END DO
2725 : END DO
2726 :
2727 : !clean-up
2728 73732 : DO i = 1, batch_size
2729 73732 : IF (ASSOCIATED(recv_buff(i)%array)) DEALLOCATE (recv_buff(i)%array)
2730 : END DO
2731 15640 : DEALLOCATE (send_buff, recv_buff, send_req, recv_req)
2732 : END DO !i_batch
2733 7820 : CALL dbt_finalize(t2c_sub)
2734 :
2735 16000 : END SUBROUTINE copy_2c_to_subgroup
2736 :
2737 : ! **************************************************************************************************
2738 : !> \brief Copy the data of a 3D tensor living in the main MPI group to a sub-group, given the proc
2739 : !> mapping from one to the other (e.g. for a proc idx in the subgroup, we get the idx in the main)
2740 : !> \param t3c_sub ...
2741 : !> \param t3c_main ...
2742 : !> \param group_size ...
2743 : !> \param ngroups ...
2744 : !> \param para_env ...
2745 : !> \param iatom_to_subgroup ...
2746 : !> \param dim_at ...
2747 : !> \param idx_to_at ...
2748 : ! **************************************************************************************************
2749 11690 : SUBROUTINE copy_3c_to_subgroup(t3c_sub, t3c_main, group_size, ngroups, para_env, iatom_to_subgroup, &
2750 11690 : dim_at, idx_to_at)
2751 : TYPE(dbt_type), INTENT(INOUT) :: t3c_sub, t3c_main
2752 : INTEGER, INTENT(IN) :: group_size, ngroups
2753 : TYPE(mp_para_env_type), POINTER :: para_env
2754 : TYPE(cp_1d_logical_p_type), DIMENSION(:), &
2755 : INTENT(INOUT), OPTIONAL :: iatom_to_subgroup
2756 : INTEGER, INTENT(IN), OPTIONAL :: dim_at
2757 : INTEGER, DIMENSION(:), OPTIONAL :: idx_to_at
2758 :
2759 : INTEGER :: batch_size, i, i_batch, i_msg, iatom, &
2760 : iblk, igroup, iproc, ir, is, jblk, &
2761 : kblk, n_batch, nocc, tag
2762 11690 : INTEGER, ALLOCATABLE, DIMENSION(:) :: bsizes1, bsizes2, bsizes3
2763 11690 : INTEGER, ALLOCATABLE, DIMENSION(:, :, :) :: block_dest, block_source
2764 11690 : INTEGER, ALLOCATABLE, DIMENSION(:, :, :, :) :: current_dest
2765 : INTEGER, DIMENSION(3) :: ind, nblks
2766 : LOGICAL :: filter_at, found
2767 11690 : REAL(dp), ALLOCATABLE, DIMENSION(:, :, :) :: blk
2768 11690 : TYPE(cp_3d_r_p_type), ALLOCATABLE, DIMENSION(:) :: recv_buff, send_buff
2769 : TYPE(dbt_iterator_type) :: iter
2770 11690 : TYPE(mp_request_type), ALLOCATABLE, DIMENSION(:) :: recv_req, send_req
2771 :
2772 : !Stategy: we loop over the main tensor, and send all the data. Then we loop over the sub tensor
2773 : ! and receive it. We do all of it with async MPI communication. The sub tensor needs
2774 : ! to have blocks pre-reserved though
2775 :
2776 11690 : CALL dbt_get_info(t3c_main, nblks_total=nblks)
2777 :
2778 : !in some cases, only copy a fraction of the 3c tensor to a given subgroup (corresponding to some atoms)
2779 11690 : filter_at = .FALSE.
2780 11690 : IF (PRESENT(iatom_to_subgroup) .AND. PRESENT(dim_at) .AND. PRESENT(idx_to_at)) THEN
2781 6614 : filter_at = .TRUE.
2782 6614 : CPASSERT(nblks(dim_at) == SIZE(idx_to_at))
2783 : END IF
2784 :
2785 : !Loop over the main tensor, count how many blocks are there, which ones, and on which proc
2786 58450 : ALLOCATE (block_source(nblks(1), nblks(2), nblks(3)))
2787 640562 : block_source = -1
2788 11690 : nocc = 0
2789 11690 : !$OMP PARALLEL DEFAULT(NONE) SHARED(t3c_main,para_env,nocc,block_source) PRIVATE(iter,ind,blk,found)
2790 : CALL dbt_iterator_start(iter, t3c_main)
2791 : DO WHILE (dbt_iterator_blocks_left(iter))
2792 : CALL dbt_iterator_next_block(iter, ind)
2793 : CALL dbt_get_block(t3c_main, ind, blk, found)
2794 : IF (.NOT. found) CYCLE
2795 :
2796 : block_source(ind(1), ind(2), ind(3)) = para_env%mepos
2797 : !$OMP ATOMIC
2798 : nocc = nocc + 1
2799 : DEALLOCATE (blk)
2800 : END DO
2801 : CALL dbt_iterator_stop(iter)
2802 : !$OMP END PARALLEL
2803 :
2804 11690 : CALL para_env%sum(nocc)
2805 11690 : CALL para_env%sum(block_source)
2806 640562 : block_source = block_source + para_env%num_pe - 1
2807 11690 : IF (nocc == 0) RETURN
2808 :
2809 : !Loop over the sub tensor, get the block destination
2810 11690 : igroup = para_env%mepos/group_size
2811 46760 : ALLOCATE (block_dest(nblks(1), nblks(2), nblks(3)))
2812 640562 : block_dest = -1
2813 35070 : DO kblk = 1, nblks(3)
2814 163314 : DO jblk = 1, nblks(2)
2815 628872 : DO iblk = 1, nblks(1)
2816 477248 : IF (block_source(iblk, jblk, kblk) == -1) CYCLE
2817 :
2818 491240 : CALL dbt_get_stored_coordinates(t3c_sub, [iblk, jblk, kblk], iproc)
2819 605492 : block_dest(iblk, jblk, kblk) = igroup*group_size + iproc !mapping of iproc in subgroup to main group idx
2820 : END DO
2821 : END DO
2822 : END DO
2823 :
2824 81830 : ALLOCATE (bsizes1(nblks(1)), bsizes2(nblks(2)), bsizes3(nblks(3)))
2825 11690 : CALL dbt_get_info(t3c_main, blk_size_1=bsizes1, blk_size_2=bsizes2, blk_size_3=bsizes3)
2826 :
2827 70140 : ALLOCATE (current_dest(nblks(1), nblks(2), nblks(3), 0:ngroups - 1))
2828 35070 : DO igroup = 0, ngroups - 1
2829 : !for a given subgroup, need to make the destination available to everyone in the main group
2830 1281124 : current_dest(:, :, :, igroup) = block_dest(:, :, :)
2831 35070 : CALL para_env%bcast(current_dest(:, :, :, igroup), source=igroup*group_size) !bcast from first proc in subgroup
2832 : END DO
2833 :
2834 : !We go by batches, which cannot be larger than the maximum MPI tag value
2835 11690 : batch_size = MIN(para_env%get_tag_ub(), 128000, nocc*ngroups)
2836 11690 : n_batch = (nocc*ngroups)/batch_size
2837 11690 : IF (MODULO(nocc*ngroups, batch_size) .NE. 0) n_batch = n_batch + 1
2838 :
2839 23380 : DO i_batch = 1, n_batch
2840 : !Loop over groups, blocks and send/receive
2841 538000 : ALLOCATE (send_buff(batch_size), recv_buff(batch_size))
2842 538000 : ALLOCATE (send_req(batch_size), recv_req(batch_size))
2843 : ir = 0
2844 : is = 0
2845 : i_msg = 0
2846 35070 : DO kblk = 1, nblks(3)
2847 163314 : DO jblk = 1, nblks(2)
2848 628872 : DO iblk = 1, nblks(1)
2849 1559988 : DO igroup = 0, ngroups - 1
2850 954496 : IF (block_source(iblk, jblk, kblk) == -1) CYCLE
2851 :
2852 245620 : i_msg = i_msg + 1
2853 245620 : IF (i_msg < (i_batch - 1)*batch_size + 1 .OR. i_msg > i_batch*batch_size) CYCLE
2854 :
2855 : !a unique tag per block, within this batch
2856 245620 : tag = i_msg - (i_batch - 1)*batch_size
2857 :
2858 245620 : IF (filter_at) THEN
2859 693136 : ind(:) = [iblk, jblk, kblk]
2860 173284 : iatom = idx_to_at(ind(dim_at))
2861 173284 : IF (.NOT. iatom_to_subgroup(iatom)%array(igroup + 1)) CYCLE
2862 : END IF
2863 :
2864 158978 : found = .FALSE.
2865 158978 : IF (para_env%mepos == block_source(iblk, jblk, kblk)) THEN
2866 317956 : CALL dbt_get_block(t3c_main, [iblk, jblk, kblk], blk, found)
2867 : END IF
2868 :
2869 : !If blocks live on same proc, simply copy. Else MPI send/recv
2870 158978 : IF (block_source(iblk, jblk, kblk) == current_dest(iblk, jblk, kblk, igroup)) THEN
2871 329744 : IF (found) CALL dbt_put_block(t3c_sub, [iblk, jblk, kblk], SHAPE(blk), blk)
2872 : ELSE
2873 76542 : IF (para_env%mepos == block_source(iblk, jblk, kblk) .AND. found) THEN
2874 191355 : ALLOCATE (send_buff(tag)%array(bsizes1(iblk), bsizes2(jblk), bsizes3(kblk)))
2875 187303822 : send_buff(tag)%array(:, :, :) = blk(:, :, :)
2876 38271 : is = is + 1
2877 : CALL para_env%isend(msgin=send_buff(tag)%array, &
2878 : dest=current_dest(iblk, jblk, kblk, igroup), &
2879 38271 : request=send_req(is), tag=tag)
2880 : END IF
2881 :
2882 76542 : IF (para_env%mepos == current_dest(iblk, jblk, kblk, igroup)) THEN
2883 191355 : ALLOCATE (recv_buff(tag)%array(bsizes1(iblk), bsizes2(jblk), bsizes3(kblk)))
2884 38271 : ir = ir + 1
2885 : CALL para_env%irecv(msgout=recv_buff(tag)%array, source=block_source(iblk, jblk, kblk), &
2886 38271 : request=recv_req(ir), tag=tag)
2887 : END IF
2888 : END IF
2889 :
2890 636226 : IF (found) DEALLOCATE (blk)
2891 : END DO
2892 : END DO
2893 : END DO
2894 : END DO
2895 :
2896 11690 : CALL mp_waitall(send_req(1:is))
2897 11690 : CALL mp_waitall(recv_req(1:ir))
2898 : !clean-up
2899 257310 : DO i = 1, batch_size
2900 257310 : IF (ASSOCIATED(send_buff(i)%array)) DEALLOCATE (send_buff(i)%array)
2901 : END DO
2902 :
2903 : !Finally copy the data from the buffer to the sub-tensor
2904 : i_msg = 0
2905 35070 : DO kblk = 1, nblks(3)
2906 163314 : DO jblk = 1, nblks(2)
2907 628872 : DO iblk = 1, nblks(1)
2908 1559988 : DO igroup = 0, ngroups - 1
2909 954496 : IF (block_source(iblk, jblk, kblk) == -1) CYCLE
2910 :
2911 245620 : i_msg = i_msg + 1
2912 245620 : IF (i_msg < (i_batch - 1)*batch_size + 1 .OR. i_msg > i_batch*batch_size) CYCLE
2913 :
2914 : !a unique tag per block, within this batch
2915 245620 : tag = i_msg - (i_batch - 1)*batch_size
2916 :
2917 245620 : IF (filter_at) THEN
2918 693136 : ind(:) = [iblk, jblk, kblk]
2919 173284 : iatom = idx_to_at(ind(dim_at))
2920 173284 : IF (.NOT. iatom_to_subgroup(iatom)%array(igroup + 1)) CYCLE
2921 : END IF
2922 :
2923 158978 : IF (para_env%mepos == current_dest(iblk, jblk, kblk, igroup) .AND. &
2924 477248 : block_source(iblk, jblk, kblk) .NE. current_dest(iblk, jblk, kblk, igroup)) THEN
2925 :
2926 191355 : ALLOCATE (blk(bsizes1(iblk), bsizes2(jblk), bsizes3(kblk)))
2927 187303822 : blk(:, :, :) = recv_buff(tag)%array(:, :, :)
2928 267897 : CALL dbt_put_block(t3c_sub, [iblk, jblk, kblk], SHAPE(blk), blk)
2929 38271 : DEALLOCATE (blk)
2930 : END IF
2931 : END DO
2932 : END DO
2933 : END DO
2934 : END DO
2935 :
2936 : !clean-up
2937 257310 : DO i = 1, batch_size
2938 257310 : IF (ASSOCIATED(recv_buff(i)%array)) DEALLOCATE (recv_buff(i)%array)
2939 : END DO
2940 23380 : DEALLOCATE (send_buff, recv_buff, send_req, recv_req)
2941 : END DO !i_batch
2942 11690 : CALL dbt_finalize(t3c_sub)
2943 :
2944 23380 : END SUBROUTINE copy_3c_to_subgroup
2945 :
2946 : ! **************************************************************************************************
2947 : !> \brief A routine that gather the pieces of the KS matrix accross the subgroup and puts it in the
2948 : !> main group. Each b_img, iatom, jatom tuple is one a single CPU
2949 : !> \param ks_t ...
2950 : !> \param ks_t_sub ...
2951 : !> \param group_size ...
2952 : !> \param sparsity_pattern ...
2953 : !> \param para_env ...
2954 : !> \param ri_data ...
2955 : ! **************************************************************************************************
2956 190 : SUBROUTINE gather_ks_matrix(ks_t, ks_t_sub, group_size, sparsity_pattern, para_env, ri_data)
2957 : TYPE(dbt_type), DIMENSION(:, :), INTENT(INOUT) :: ks_t, ks_t_sub
2958 : INTEGER, INTENT(IN) :: group_size
2959 : INTEGER, DIMENSION(:, :, :), INTENT(IN) :: sparsity_pattern
2960 : TYPE(mp_para_env_type), POINTER :: para_env
2961 : TYPE(hfx_ri_type), INTENT(INOUT) :: ri_data
2962 :
2963 : CHARACTER(len=*), PARAMETER :: routineN = 'gather_ks_matrix'
2964 :
2965 : INTEGER :: b_img, dest, handle, i, i_spin, iatom, &
2966 : igroup, ir, is, jatom, n_mess, natom, &
2967 : nimg, nspins, source, tag
2968 : LOGICAL :: found
2969 190 : REAL(dp), ALLOCATABLE, DIMENSION(:, :) :: blk
2970 190 : TYPE(cp_2d_r_p_type), ALLOCATABLE, DIMENSION(:) :: recv_buff, send_buff
2971 190 : TYPE(mp_request_type), ALLOCATABLE, DIMENSION(:) :: recv_req, send_req
2972 :
2973 190 : CALL timeset(routineN, handle)
2974 :
2975 190 : nimg = SIZE(sparsity_pattern, 3)
2976 190 : natom = SIZE(sparsity_pattern, 2)
2977 190 : nspins = SIZE(ks_t, 1)
2978 :
2979 4994 : DO b_img = 1, nimg
2980 : n_mess = 0
2981 11132 : DO i_spin = 1, nspins
2982 23788 : DO jatom = 1, natom
2983 44296 : DO iatom = 1, natom
2984 37968 : IF (sparsity_pattern(iatom, jatom, b_img) > -1) n_mess = n_mess + 1
2985 : END DO
2986 : END DO
2987 : END DO
2988 :
2989 37136 : ALLOCATE (send_buff(n_mess), recv_buff(n_mess))
2990 41940 : ALLOCATE (send_req(n_mess), recv_req(n_mess))
2991 4804 : ir = 0
2992 4804 : is = 0
2993 4804 : n_mess = 0
2994 4804 : tag = 0
2995 :
2996 11132 : DO i_spin = 1, nspins
2997 23788 : DO jatom = 1, natom
2998 44296 : DO iatom = 1, natom
2999 25312 : IF (sparsity_pattern(iatom, jatom, b_img) < 0) CYCLE
3000 8776 : n_mess = n_mess + 1
3001 8776 : tag = tag + 1
3002 :
3003 : !sending the message
3004 26328 : CALL dbt_get_stored_coordinates(ks_t(i_spin, b_img), [iatom, jatom], dest)
3005 26328 : CALL dbt_get_stored_coordinates(ks_t_sub(i_spin, b_img), [iatom, jatom], source) !source within sub
3006 8776 : igroup = sparsity_pattern(iatom, jatom, b_img)
3007 8776 : source = source + igroup*group_size
3008 8776 : IF (para_env%mepos == source) THEN
3009 13164 : CALL dbt_get_block(ks_t_sub(i_spin, b_img), [iatom, jatom], blk, found)
3010 4388 : IF (source == dest) THEN
3011 3547 : IF (found) CALL dbt_put_block(ks_t(i_spin, b_img), [iatom, jatom], SHAPE(blk), blk)
3012 : ELSE
3013 13764 : ALLOCATE (send_buff(n_mess)%array(ri_data%bsizes_AO(iatom), ri_data%bsizes_AO(jatom)))
3014 282773 : send_buff(n_mess)%array(:, :) = 0.0_dp
3015 3441 : IF (found) THEN
3016 194070 : send_buff(n_mess)%array(:, :) = blk(:, :)
3017 : END IF
3018 3441 : is = is + 1
3019 : CALL para_env%isend(msgin=send_buff(n_mess)%array, dest=dest, &
3020 3441 : request=send_req(is), tag=tag)
3021 : END IF
3022 4388 : DEALLOCATE (blk)
3023 : END IF
3024 :
3025 : !receiving the message
3026 21432 : IF (para_env%mepos == dest .AND. source .NE. dest) THEN
3027 13764 : ALLOCATE (recv_buff(n_mess)%array(ri_data%bsizes_AO(iatom), ri_data%bsizes_AO(jatom)))
3028 3441 : ir = ir + 1
3029 : CALL para_env%irecv(msgout=recv_buff(n_mess)%array, source=source, &
3030 3441 : request=recv_req(ir), tag=tag)
3031 : END IF
3032 : END DO !iatom
3033 : END DO !jatom
3034 : END DO !ispin
3035 :
3036 4804 : CALL mp_waitall(send_req(1:is))
3037 4804 : CALL mp_waitall(recv_req(1:ir))
3038 :
3039 : !Copy the messages received into the KS matrix
3040 4804 : n_mess = 0
3041 11132 : DO i_spin = 1, nspins
3042 23788 : DO jatom = 1, natom
3043 44296 : DO iatom = 1, natom
3044 25312 : IF (sparsity_pattern(iatom, jatom, b_img) < 0) CYCLE
3045 8776 : n_mess = n_mess + 1
3046 :
3047 26328 : CALL dbt_get_stored_coordinates(ks_t(i_spin, b_img), [iatom, jatom], dest)
3048 21432 : IF (para_env%mepos == dest) THEN
3049 4388 : IF (.NOT. ASSOCIATED(recv_buff(n_mess)%array)) CYCLE
3050 13764 : ALLOCATE (blk(ri_data%bsizes_AO(iatom), ri_data%bsizes_AO(jatom)))
3051 282773 : blk(:, :) = recv_buff(n_mess)%array(:, :)
3052 17205 : CALL dbt_put_block(ks_t(i_spin, b_img), [iatom, jatom], SHAPE(blk), blk)
3053 3441 : DEALLOCATE (blk)
3054 : END IF
3055 : END DO
3056 : END DO
3057 : END DO
3058 :
3059 : !clean-up
3060 13580 : DO i = 1, n_mess
3061 8776 : IF (ASSOCIATED(send_buff(i)%array)) DEALLOCATE (send_buff(i)%array)
3062 13580 : IF (ASSOCIATED(recv_buff(i)%array)) DEALLOCATE (recv_buff(i)%array)
3063 : END DO
3064 4994 : DEALLOCATE (send_buff, recv_buff, send_req, recv_req)
3065 : END DO !b_img
3066 :
3067 190 : CALL timestop(handle)
3068 :
3069 190 : END SUBROUTINE gather_ks_matrix
3070 :
3071 : ! **************************************************************************************************
3072 : !> \brief copy all required 2c tensors from the main MPI group to the subgroups
3073 : !> \param mat_2c_pot ...
3074 : !> \param t_2c_work ...
3075 : !> \param t_2c_ao_tmp ...
3076 : !> \param ks_t_split ...
3077 : !> \param ks_t_sub ...
3078 : !> \param group_size ...
3079 : !> \param ngroups ...
3080 : !> \param para_env ...
3081 : !> \param para_env_sub ...
3082 : !> \param ri_data ...
3083 : ! **************************************************************************************************
3084 190 : SUBROUTINE get_subgroup_2c_tensors(mat_2c_pot, t_2c_work, t_2c_ao_tmp, ks_t_split, ks_t_sub, &
3085 : group_size, ngroups, para_env, para_env_sub, ri_data)
3086 : TYPE(dbcsr_type), DIMENSION(:), INTENT(INOUT) :: mat_2c_pot
3087 : TYPE(dbt_type), DIMENSION(:), INTENT(INOUT) :: t_2c_work, t_2c_ao_tmp, ks_t_split
3088 : TYPE(dbt_type), DIMENSION(:, :), INTENT(INOUT) :: ks_t_sub
3089 : INTEGER, INTENT(IN) :: group_size, ngroups
3090 : TYPE(mp_para_env_type), POINTER :: para_env, para_env_sub
3091 : TYPE(hfx_ri_type), INTENT(INOUT) :: ri_data
3092 :
3093 : CHARACTER(len=*), PARAMETER :: routineN = 'get_subgroup_2c_tensors'
3094 :
3095 : INTEGER :: handle, i, i_img, i_RI, i_spin, iproc, &
3096 : j, natom, nblks, nimg, nspins
3097 : INTEGER(int_8) :: nze
3098 : INTEGER, ALLOCATABLE, DIMENSION(:) :: bsizes_RI_ext, bsizes_RI_ext_split, &
3099 190 : dist1, dist2
3100 : INTEGER, DIMENSION(2) :: pdims_2d
3101 380 : INTEGER, DIMENSION(:), POINTER :: col_dist, RI_blk_size, row_dist
3102 190 : INTEGER, DIMENSION(:, :), POINTER :: dbcsr_pgrid
3103 : REAL(dp) :: occ
3104 : TYPE(dbcsr_distribution_type) :: dbcsr_dist_sub
3105 570 : TYPE(dbt_pgrid_type) :: pgrid_2d
3106 2470 : TYPE(dbt_type) :: work, work_sub
3107 :
3108 190 : CALL timeset(routineN, handle)
3109 :
3110 : !Create the 2d pgrid
3111 190 : pdims_2d = 0
3112 190 : CALL dbt_pgrid_create(para_env_sub, pdims_2d, pgrid_2d)
3113 :
3114 190 : natom = SIZE(ri_data%bsizes_RI)
3115 190 : nblks = SIZE(ri_data%bsizes_RI_split)
3116 570 : ALLOCATE (bsizes_RI_ext(ri_data%ncell_RI*natom))
3117 570 : ALLOCATE (bsizes_RI_ext_split(ri_data%ncell_RI*nblks))
3118 1334 : DO i_RI = 1, ri_data%ncell_RI
3119 3432 : bsizes_RI_ext((i_RI - 1)*natom + 1:i_RI*natom) = ri_data%bsizes_RI(:)
3120 6064 : bsizes_RI_ext_split((i_RI - 1)*nblks + 1:i_RI*nblks) = ri_data%bsizes_RI_split(:)
3121 : END DO
3122 :
3123 : !nRI x nRI 2c tensors
3124 : CALL create_2c_tensor(t_2c_work(1), dist1, dist2, pgrid_2d, &
3125 : bsizes_RI_ext, bsizes_RI_ext, &
3126 : name="(RI | RI)")
3127 190 : DEALLOCATE (dist1, dist2)
3128 :
3129 : CALL create_2c_tensor(t_2c_work(2), dist1, dist2, pgrid_2d, &
3130 : bsizes_RI_ext_split, bsizes_RI_ext_split, &
3131 190 : name="(RI | RI)")
3132 190 : DEALLOCATE (dist1, dist2)
3133 :
3134 : !the AO based tensors
3135 : CALL create_2c_tensor(ks_t_split(1), dist1, dist2, pgrid_2d, &
3136 : ri_data%bsizes_AO_split, ri_data%bsizes_AO_split, &
3137 : name="(AO | AO)")
3138 190 : DEALLOCATE (dist1, dist2)
3139 190 : CALL dbt_create(ks_t_split(1), ks_t_split(2))
3140 :
3141 : CALL create_2c_tensor(t_2c_ao_tmp(1), dist1, dist2, pgrid_2d, &
3142 : ri_data%bsizes_AO, ri_data%bsizes_AO, &
3143 : name="(AO | AO)")
3144 190 : DEALLOCATE (dist1, dist2)
3145 :
3146 190 : nspins = SIZE(ks_t_sub, 1)
3147 190 : nimg = SIZE(ks_t_sub, 2)
3148 4994 : DO i_img = 1, nimg
3149 11322 : DO i_spin = 1, nspins
3150 11132 : CALL dbt_create(t_2c_ao_tmp(1), ks_t_sub(i_spin, i_img))
3151 : END DO
3152 : END DO
3153 :
3154 : !Finally the HFX potential matrices
3155 : !For now, we do a convoluted things where we go to tensors first, then back to matrices.
3156 : CALL create_2c_tensor(work_sub, dist1, dist2, pgrid_2d, &
3157 : ri_data%bsizes_RI, ri_data%bsizes_RI, &
3158 : name="(RI | RI)")
3159 190 : CALL dbt_create(ri_data%kp_mat_2c_pot(1, 1), work)
3160 :
3161 760 : ALLOCATE (dbcsr_pgrid(0:pdims_2d(1) - 1, 0:pdims_2d(2) - 1))
3162 190 : iproc = 0
3163 380 : DO i = 0, pdims_2d(1) - 1
3164 570 : DO j = 0, pdims_2d(2) - 1
3165 190 : dbcsr_pgrid(i, j) = iproc
3166 380 : iproc = iproc + 1
3167 : END DO
3168 : END DO
3169 :
3170 : !We need to have the same exact 2d block dist as the tensors
3171 760 : ALLOCATE (col_dist(natom), row_dist(natom))
3172 570 : row_dist(:) = dist1(:)
3173 570 : col_dist(:) = dist2(:)
3174 :
3175 380 : ALLOCATE (RI_blk_size(natom))
3176 570 : RI_blk_size(:) = ri_data%bsizes_RI(:)
3177 :
3178 : CALL dbcsr_distribution_new(dbcsr_dist_sub, group=para_env_sub%get_handle(), pgrid=dbcsr_pgrid, &
3179 190 : row_dist=row_dist, col_dist=col_dist)
3180 : CALL dbcsr_create(mat_2c_pot(1), dist=dbcsr_dist_sub, name="sub", matrix_type=dbcsr_type_no_symmetry, &
3181 190 : row_blk_size=RI_blk_size, col_blk_size=RI_blk_size)
3182 :
3183 4994 : DO i_img = 1, nimg
3184 4804 : IF (i_img > 1) CALL dbcsr_create(mat_2c_pot(i_img), template=mat_2c_pot(1))
3185 4804 : CALL dbt_copy_matrix_to_tensor(ri_data%kp_mat_2c_pot(1, i_img), work)
3186 4804 : CALL get_tensor_occupancy(work, nze, occ)
3187 4804 : IF (nze == 0) CYCLE
3188 :
3189 3708 : CALL copy_2c_to_subgroup(work_sub, work, group_size, ngroups, para_env)
3190 3708 : CALL dbt_copy_tensor_to_matrix(work_sub, mat_2c_pot(i_img))
3191 3708 : CALL dbcsr_filter(mat_2c_pot(i_img), ri_data%filter_eps)
3192 8702 : CALL dbt_clear(work_sub)
3193 : END DO
3194 :
3195 190 : CALL dbt_destroy(work)
3196 190 : CALL dbt_destroy(work_sub)
3197 190 : CALL dbt_pgrid_destroy(pgrid_2d)
3198 190 : CALL dbcsr_distribution_release(dbcsr_dist_sub)
3199 190 : DEALLOCATE (col_dist, row_dist, RI_blk_size, dbcsr_pgrid)
3200 190 : CALL timestop(handle)
3201 :
3202 1710 : END SUBROUTINE get_subgroup_2c_tensors
3203 :
3204 : ! **************************************************************************************************
3205 : !> \brief copy all required 3c tensors from the main MPI group to the subgroups
3206 : !> \param t_3c_int ...
3207 : !> \param t_3c_work_2 ...
3208 : !> \param t_3c_work_3 ...
3209 : !> \param t_3c_apc ...
3210 : !> \param t_3c_apc_sub ...
3211 : !> \param group_size ...
3212 : !> \param ngroups ...
3213 : !> \param para_env ...
3214 : !> \param para_env_sub ...
3215 : !> \param ri_data ...
3216 : ! **************************************************************************************************
3217 190 : SUBROUTINE get_subgroup_3c_tensors(t_3c_int, t_3c_work_2, t_3c_work_3, t_3c_apc, t_3c_apc_sub, &
3218 : group_size, ngroups, para_env, para_env_sub, ri_data)
3219 : TYPE(dbt_type), DIMENSION(:), INTENT(INOUT) :: t_3c_int, t_3c_work_2, t_3c_work_3
3220 : TYPE(dbt_type), DIMENSION(:, :), INTENT(INOUT) :: t_3c_apc, t_3c_apc_sub
3221 : INTEGER, INTENT(IN) :: group_size, ngroups
3222 : TYPE(mp_para_env_type), POINTER :: para_env, para_env_sub
3223 : TYPE(hfx_ri_type), INTENT(INOUT) :: ri_data
3224 :
3225 : CHARACTER(len=*), PARAMETER :: routineN = 'get_subgroup_3c_tensors'
3226 :
3227 : INTEGER :: batch_size, bfac, bo(2), handle, &
3228 : handle2, i_blk, i_img, i_RI, i_spin, &
3229 : ib, natom, nblks_AO, nblks_RI, nimg, &
3230 : nspins
3231 : INTEGER(int_8) :: nze
3232 190 : INTEGER, ALLOCATABLE, DIMENSION(:) :: bsizes_RI_ext, bsizes_RI_ext_split, &
3233 190 : bsizes_stack, bsizes_tmp, dist1, &
3234 190 : dist2, dist3, dist_stack, idx_to_at
3235 : INTEGER, DIMENSION(3) :: pdims
3236 : REAL(dp) :: occ
3237 1710 : TYPE(dbt_distribution_type) :: t_dist
3238 570 : TYPE(dbt_pgrid_type) :: pgrid
3239 4750 : TYPE(dbt_type) :: tmp, work_atom_block, work_atom_block_sub
3240 :
3241 190 : CALL timeset(routineN, handle)
3242 :
3243 190 : nblks_RI = SIZE(ri_data%bsizes_RI_split)
3244 570 : ALLOCATE (bsizes_RI_ext_split(ri_data%ncell_RI*nblks_RI))
3245 1334 : DO i_RI = 1, ri_data%ncell_RI
3246 6064 : bsizes_RI_ext_split((i_RI - 1)*nblks_RI + 1:i_RI*nblks_RI) = ri_data%bsizes_RI_split(:)
3247 : END DO
3248 :
3249 : !Preparing larger block sizes for efficient communication (less, bigger messages)
3250 : !we put 2 atoms per RI block
3251 190 : bfac = 2
3252 190 : natom = SIZE(ri_data%bsizes_RI)
3253 190 : nblks_RI = MAX(1, natom/bfac)
3254 570 : ALLOCATE (bsizes_tmp(nblks_RI))
3255 380 : DO i_blk = 1, nblks_RI
3256 190 : bo = get_limit(natom, nblks_RI, i_blk - 1)
3257 760 : bsizes_tmp(i_blk) = SUM(ri_data%bsizes_RI(bo(1):bo(2)))
3258 : END DO
3259 570 : ALLOCATE (bsizes_RI_ext(ri_data%ncell_RI*nblks_RI))
3260 1334 : DO i_RI = 1, ri_data%ncell_RI
3261 2478 : bsizes_RI_ext((i_RI - 1)*nblks_RI + 1:i_RI*nblks_RI) = bsizes_tmp(:)
3262 : END DO
3263 :
3264 190 : batch_size = ri_data%kp_stack_size
3265 190 : nblks_AO = SIZE(ri_data%bsizes_AO_split)
3266 570 : ALLOCATE (bsizes_stack(batch_size*nblks_AO))
3267 6270 : DO ib = 1, batch_size
3268 29886 : bsizes_stack((ib - 1)*nblks_AO + 1:ib*nblks_AO) = ri_data%bsizes_AO_split(:)
3269 : END DO
3270 :
3271 : !Create the pgrid for the configuration correspoinding to ri_data%t_3c_int_ctr_3
3272 190 : natom = SIZE(ri_data%bsizes_RI)
3273 190 : pdims = 0
3274 : CALL dbt_pgrid_create(para_env_sub, pdims, pgrid, &
3275 760 : tensor_dims=[SIZE(bsizes_RI_ext_split), 1, batch_size*SIZE(ri_data%bsizes_AO_split)])
3276 :
3277 : !Create all required 3c tensors in that configuration
3278 : CALL create_3c_tensor(t_3c_int(1), dist1, dist2, dist3, &
3279 : pgrid, bsizes_RI_ext_split, ri_data%bsizes_AO_split, &
3280 190 : ri_data%bsizes_AO_split, [1], [2, 3], name="(RI | AO AO)")
3281 190 : nimg = SIZE(t_3c_int)
3282 4804 : DO i_img = 2, nimg
3283 4804 : CALL dbt_create(t_3c_int(1), t_3c_int(i_img))
3284 : END DO
3285 :
3286 : !The stacked work tensors, in a distribution that matches that of t_3c_int
3287 380 : ALLOCATE (dist_stack(batch_size*nblks_AO))
3288 6270 : DO ib = 1, batch_size
3289 29886 : dist_stack((ib - 1)*nblks_AO + 1:ib*nblks_AO) = dist3(:)
3290 : END DO
3291 :
3292 190 : CALL dbt_distribution_new(t_dist, pgrid, dist1, dist2, dist_stack)
3293 : CALL dbt_create(t_3c_work_3(1), "work_3_stack", t_dist, [1], [2, 3], &
3294 190 : bsizes_RI_ext_split, ri_data%bsizes_AO_split, bsizes_stack)
3295 190 : CALL dbt_create(t_3c_work_3(1), t_3c_work_3(2))
3296 190 : CALL dbt_create(t_3c_work_3(1), t_3c_work_3(3))
3297 190 : CALL dbt_distribution_destroy(t_dist)
3298 190 : DEALLOCATE (dist1, dist2, dist3, dist_stack)
3299 :
3300 : !For more efficient communication, we use intermediate tensors with larger block size
3301 : CALL create_3c_tensor(work_atom_block_sub, dist1, dist2, dist3, &
3302 : pgrid, bsizes_RI_ext, ri_data%bsizes_AO, &
3303 190 : ri_data%bsizes_AO, [1], [2, 3], name="(RI | AO AO)")
3304 190 : DEALLOCATE (dist1, dist2, dist3)
3305 :
3306 : CALL create_3c_tensor(work_atom_block, dist1, dist2, dist3, &
3307 : ri_data%pgrid, bsizes_RI_ext, ri_data%bsizes_AO, &
3308 190 : ri_data%bsizes_AO, [1], [2, 3], name="(RI | AO AO)")
3309 190 : DEALLOCATE (dist1, dist2, dist3)
3310 :
3311 : !Finally copy the integrals into the subgroups (if not there already)
3312 190 : CALL timeset(routineN//"_ints", handle2)
3313 190 : IF (ALLOCATED(ri_data%kp_t_3c_int)) THEN
3314 3208 : DO i_img = 1, nimg
3315 3208 : CALL dbt_copy(ri_data%kp_t_3c_int(i_img), t_3c_int(i_img), move_data=.TRUE.)
3316 : END DO
3317 : ELSE
3318 2486 : ALLOCATE (ri_data%kp_t_3c_int(nimg))
3319 1786 : DO i_img = 1, nimg
3320 1716 : CALL dbt_create(t_3c_int(i_img), ri_data%kp_t_3c_int(i_img))
3321 1716 : CALL get_tensor_occupancy(ri_data%t_3c_int_ctr_1(1, i_img), nze, occ)
3322 1716 : IF (nze == 0) CYCLE
3323 1236 : CALL dbt_copy(ri_data%t_3c_int_ctr_1(1, i_img), work_atom_block, order=[2, 1, 3])
3324 1236 : CALL copy_3c_to_subgroup(work_atom_block_sub, work_atom_block, group_size, ngroups, para_env)
3325 3022 : CALL dbt_copy(work_atom_block_sub, t_3c_int(i_img), move_data=.TRUE.)
3326 : END DO
3327 : END IF
3328 190 : CALL timestop(handle2)
3329 190 : CALL dbt_pgrid_destroy(pgrid)
3330 190 : CALL dbt_destroy(work_atom_block)
3331 190 : CALL dbt_destroy(work_atom_block_sub)
3332 :
3333 : !Do the same for the t_3c_ctr_2 configuration
3334 190 : pdims = 0
3335 : CALL dbt_pgrid_create(para_env_sub, pdims, pgrid, &
3336 760 : tensor_dims=[1, SIZE(bsizes_RI_ext_split), batch_size*SIZE(ri_data%bsizes_AO_split)])
3337 :
3338 : !For more efficient communication, we use intermediate tensors with larger block size
3339 : CALL create_3c_tensor(work_atom_block_sub, dist1, dist2, dist3, &
3340 : pgrid, ri_data%bsizes_AO, bsizes_RI_ext, &
3341 190 : ri_data%bsizes_AO, [1], [2, 3], name="(AO RI | AO)")
3342 190 : DEALLOCATE (dist1, dist2, dist3)
3343 :
3344 : CALL create_3c_tensor(work_atom_block, dist1, dist2, dist3, &
3345 : ri_data%pgrid_1, ri_data%bsizes_AO, bsizes_RI_ext, &
3346 190 : ri_data%bsizes_AO, [1], [2, 3], name="(AO RI | AO)")
3347 190 : DEALLOCATE (dist1, dist2, dist3)
3348 :
3349 : !template for t_3c_apc_sub
3350 : CALL create_3c_tensor(tmp, dist1, dist2, dist3, &
3351 : pgrid, ri_data%bsizes_AO_split, bsizes_RI_ext_split, &
3352 190 : ri_data%bsizes_AO_split, [1], [2, 3], name="(AO RI | AO)")
3353 :
3354 : !create t_3c_work_2 tensors in a distribution that matches the above
3355 380 : ALLOCATE (dist_stack(batch_size*nblks_AO))
3356 6270 : DO ib = 1, batch_size
3357 29886 : dist_stack((ib - 1)*nblks_AO + 1:ib*nblks_AO) = dist3(:)
3358 : END DO
3359 :
3360 190 : CALL dbt_distribution_new(t_dist, pgrid, dist1, dist2, dist_stack)
3361 : CALL dbt_create(t_3c_work_2(1), "work_2_stack", t_dist, [1], [2, 3], &
3362 190 : ri_data%bsizes_AO_split, bsizes_RI_ext_split, bsizes_stack)
3363 190 : CALL dbt_create(t_3c_work_2(1), t_3c_work_2(2))
3364 190 : CALL dbt_create(t_3c_work_2(1), t_3c_work_2(3))
3365 190 : CALL dbt_distribution_destroy(t_dist)
3366 190 : DEALLOCATE (dist1, dist2, dist3, dist_stack)
3367 :
3368 : !Finally copy data from t_3c_apc to the subgroups
3369 570 : ALLOCATE (idx_to_at(SIZE(ri_data%bsizes_AO)))
3370 190 : CALL get_idx_to_atom(idx_to_at, ri_data%bsizes_AO, ri_data%bsizes_AO)
3371 190 : nspins = SIZE(t_3c_apc, 1)
3372 190 : CALL timeset(routineN//"_apc", handle2)
3373 4994 : DO i_img = 1, nimg
3374 11132 : DO i_spin = 1, nspins
3375 6328 : CALL dbt_create(tmp, t_3c_apc_sub(i_spin, i_img))
3376 6328 : CALL get_tensor_occupancy(t_3c_apc(i_spin, i_img), nze, occ)
3377 6328 : IF (nze == 0) CYCLE
3378 5462 : CALL dbt_copy(t_3c_apc(i_spin, i_img), work_atom_block, move_data=.TRUE.)
3379 : CALL copy_3c_to_subgroup(work_atom_block_sub, work_atom_block, group_size, &
3380 5462 : ngroups, para_env, ri_data%iatom_to_subgroup, 1, idx_to_at)
3381 16594 : CALL dbt_copy(work_atom_block_sub, t_3c_apc_sub(i_spin, i_img), move_data=.TRUE.)
3382 : END DO
3383 11322 : DO i_spin = 1, nspins
3384 11132 : CALL dbt_destroy(t_3c_apc(i_spin, i_img))
3385 : END DO
3386 : END DO
3387 190 : CALL timestop(handle2)
3388 190 : CALL dbt_pgrid_destroy(pgrid)
3389 190 : CALL dbt_destroy(tmp)
3390 190 : CALL dbt_destroy(work_atom_block)
3391 190 : CALL dbt_destroy(work_atom_block_sub)
3392 :
3393 190 : CALL timestop(handle)
3394 :
3395 760 : END SUBROUTINE get_subgroup_3c_tensors
3396 :
3397 : ! **************************************************************************************************
3398 : !> \brief copy all required 2c force tensors from the main MPI group to the subgroups
3399 : !> \param t_2c_inv ...
3400 : !> \param t_2c_bint ...
3401 : !> \param t_2c_metric ...
3402 : !> \param mat_2c_pot ...
3403 : !> \param t_2c_work ...
3404 : !> \param rho_ao_t ...
3405 : !> \param rho_ao_t_sub ...
3406 : !> \param t_2c_der_metric ...
3407 : !> \param t_2c_der_metric_sub ...
3408 : !> \param mat_der_pot ...
3409 : !> \param mat_der_pot_sub ...
3410 : !> \param group_size ...
3411 : !> \param ngroups ...
3412 : !> \param para_env ...
3413 : !> \param para_env_sub ...
3414 : !> \param ri_data ...
3415 : !> \note Main MPI group tensors are deleted within this routine, for memory optimization
3416 : ! **************************************************************************************************
3417 84 : SUBROUTINE get_subgroup_2c_derivs(t_2c_inv, t_2c_bint, t_2c_metric, mat_2c_pot, t_2c_work, rho_ao_t, &
3418 42 : rho_ao_t_sub, t_2c_der_metric, t_2c_der_metric_sub, mat_der_pot, &
3419 42 : mat_der_pot_sub, group_size, ngroups, para_env, para_env_sub, ri_data)
3420 : TYPE(dbt_type), DIMENSION(:), INTENT(INOUT) :: t_2c_inv, t_2c_bint, t_2c_metric
3421 : TYPE(dbcsr_type), DIMENSION(:), INTENT(INOUT) :: mat_2c_pot
3422 : TYPE(dbt_type), DIMENSION(:), INTENT(INOUT) :: t_2c_work
3423 : TYPE(dbt_type), DIMENSION(:, :), INTENT(INOUT) :: rho_ao_t, rho_ao_t_sub, t_2c_der_metric, &
3424 : t_2c_der_metric_sub
3425 : TYPE(dbcsr_type), DIMENSION(:, :), INTENT(INOUT) :: mat_der_pot, mat_der_pot_sub
3426 : INTEGER, INTENT(IN) :: group_size, ngroups
3427 : TYPE(mp_para_env_type), POINTER :: para_env, para_env_sub
3428 : TYPE(hfx_ri_type), INTENT(INOUT) :: ri_data
3429 :
3430 : CHARACTER(len=*), PARAMETER :: routineN = 'get_subgroup_2c_derivs'
3431 :
3432 : INTEGER :: handle, i, i_img, i_RI, i_spin, i_xyz, &
3433 : iatom, iproc, j, natom, nblks, nimg, &
3434 : nspins
3435 : INTEGER(int_8) :: nze
3436 : INTEGER, ALLOCATABLE, DIMENSION(:) :: bsizes_RI_ext, bsizes_RI_ext_split, &
3437 42 : dist1, dist2
3438 : INTEGER, DIMENSION(2) :: pdims_2d
3439 84 : INTEGER, DIMENSION(:), POINTER :: col_dist, RI_blk_size, row_dist
3440 42 : INTEGER, DIMENSION(:, :), POINTER :: dbcsr_pgrid
3441 : REAL(dp) :: occ
3442 : TYPE(dbcsr_distribution_type) :: dbcsr_dist_sub
3443 126 : TYPE(dbt_pgrid_type) :: pgrid_2d
3444 546 : TYPE(dbt_type) :: work, work_sub
3445 :
3446 42 : CALL timeset(routineN, handle)
3447 :
3448 : !Note: a fair portion of this routine is copied from the energy version of it
3449 : !Create the 2d pgrid
3450 42 : pdims_2d = 0
3451 42 : CALL dbt_pgrid_create(para_env_sub, pdims_2d, pgrid_2d)
3452 :
3453 42 : natom = SIZE(ri_data%bsizes_RI)
3454 42 : nblks = SIZE(ri_data%bsizes_RI_split)
3455 126 : ALLOCATE (bsizes_RI_ext(ri_data%ncell_RI*natom))
3456 126 : ALLOCATE (bsizes_RI_ext_split(ri_data%ncell_RI*nblks))
3457 294 : DO i_RI = 1, ri_data%ncell_RI
3458 756 : bsizes_RI_ext((i_RI - 1)*natom + 1:i_RI*natom) = ri_data%bsizes_RI(:)
3459 1334 : bsizes_RI_ext_split((i_RI - 1)*nblks + 1:i_RI*nblks) = ri_data%bsizes_RI_split(:)
3460 : END DO
3461 :
3462 : !nRI x nRI 2c tensors
3463 : CALL create_2c_tensor(t_2c_inv(1), dist1, dist2, pgrid_2d, &
3464 : bsizes_RI_ext, bsizes_RI_ext, &
3465 : name="(RI | RI)")
3466 42 : DEALLOCATE (dist1, dist2)
3467 :
3468 42 : CALL dbt_create(t_2c_inv(1), t_2c_bint(1))
3469 42 : CALL dbt_create(t_2c_inv(1), t_2c_metric(1))
3470 84 : DO iatom = 2, natom
3471 42 : CALL dbt_create(t_2c_inv(1), t_2c_inv(iatom))
3472 42 : CALL dbt_create(t_2c_inv(1), t_2c_bint(iatom))
3473 84 : CALL dbt_create(t_2c_inv(1), t_2c_metric(iatom))
3474 : END DO
3475 42 : CALL dbt_create(t_2c_inv(1), t_2c_work(1))
3476 42 : CALL dbt_create(t_2c_inv(1), t_2c_work(2))
3477 42 : CALL dbt_create(t_2c_inv(1), t_2c_work(3))
3478 42 : CALL dbt_create(t_2c_inv(1), t_2c_work(4))
3479 :
3480 : CALL create_2c_tensor(t_2c_work(5), dist1, dist2, pgrid_2d, &
3481 : bsizes_RI_ext_split, bsizes_RI_ext_split, &
3482 42 : name="(RI | RI)")
3483 42 : DEALLOCATE (dist1, dist2)
3484 :
3485 : !copy the data from the main group.
3486 126 : DO iatom = 1, natom
3487 84 : CALL copy_2c_to_subgroup(t_2c_inv(iatom), ri_data%t_2c_inv(1, iatom), group_size, ngroups, para_env)
3488 84 : CALL copy_2c_to_subgroup(t_2c_bint(iatom), ri_data%t_2c_int(1, iatom), group_size, ngroups, para_env)
3489 126 : CALL copy_2c_to_subgroup(t_2c_metric(iatom), ri_data%t_2c_pot(1, iatom), group_size, ngroups, para_env)
3490 : END DO
3491 :
3492 : !This includes the derivatives of the RI metric, for which there is one per atom
3493 168 : DO i_xyz = 1, 3
3494 420 : DO iatom = 1, natom
3495 252 : CALL dbt_create(t_2c_inv(1), t_2c_der_metric_sub(iatom, i_xyz))
3496 : CALL copy_2c_to_subgroup(t_2c_der_metric_sub(iatom, i_xyz), t_2c_der_metric(iatom, i_xyz), &
3497 252 : group_size, ngroups, para_env)
3498 378 : CALL dbt_destroy(t_2c_der_metric(iatom, i_xyz))
3499 : END DO
3500 : END DO
3501 :
3502 : !AO x AO 2c tensors
3503 : CALL create_2c_tensor(rho_ao_t_sub(1, 1), dist1, dist2, pgrid_2d, &
3504 : ri_data%bsizes_AO_split, ri_data%bsizes_AO_split, &
3505 : name="(AO | AO)")
3506 42 : DEALLOCATE (dist1, dist2)
3507 42 : nspins = SIZE(rho_ao_t, 1)
3508 42 : nimg = SIZE(rho_ao_t, 2)
3509 :
3510 1052 : DO i_img = 1, nimg
3511 2228 : DO i_spin = 1, nspins
3512 1176 : IF (.NOT. (i_img == 1 .AND. i_spin == 1)) &
3513 1134 : CALL dbt_create(rho_ao_t_sub(1, 1), rho_ao_t_sub(i_spin, i_img))
3514 : CALL copy_2c_to_subgroup(rho_ao_t_sub(i_spin, i_img), rho_ao_t(i_spin, i_img), &
3515 1176 : group_size, ngroups, para_env)
3516 2186 : CALL dbt_destroy(rho_ao_t(i_spin, i_img))
3517 : END DO
3518 : END DO
3519 :
3520 : !The RIxRI matrices, going through tensors
3521 : CALL create_2c_tensor(work_sub, dist1, dist2, pgrid_2d, &
3522 : ri_data%bsizes_RI, ri_data%bsizes_RI, &
3523 : name="(RI | RI)")
3524 42 : CALL dbt_create(ri_data%kp_mat_2c_pot(1, 1), work)
3525 :
3526 168 : ALLOCATE (dbcsr_pgrid(0:pdims_2d(1) - 1, 0:pdims_2d(2) - 1))
3527 42 : iproc = 0
3528 84 : DO i = 0, pdims_2d(1) - 1
3529 126 : DO j = 0, pdims_2d(2) - 1
3530 42 : dbcsr_pgrid(i, j) = iproc
3531 84 : iproc = iproc + 1
3532 : END DO
3533 : END DO
3534 :
3535 : !We need to have the same exact 2d block dist as the tensors
3536 168 : ALLOCATE (col_dist(natom), row_dist(natom))
3537 126 : row_dist(:) = dist1(:)
3538 126 : col_dist(:) = dist2(:)
3539 :
3540 84 : ALLOCATE (RI_blk_size(natom))
3541 126 : RI_blk_size(:) = ri_data%bsizes_RI(:)
3542 :
3543 : CALL dbcsr_distribution_new(dbcsr_dist_sub, group=para_env_sub%get_handle(), pgrid=dbcsr_pgrid, &
3544 42 : row_dist=row_dist, col_dist=col_dist)
3545 : CALL dbcsr_create(mat_2c_pot(1), dist=dbcsr_dist_sub, name="sub", matrix_type=dbcsr_type_no_symmetry, &
3546 42 : row_blk_size=RI_blk_size, col_blk_size=RI_blk_size)
3547 :
3548 : !The HFX potential
3549 1052 : DO i_img = 1, nimg
3550 1010 : IF (i_img > 1) CALL dbcsr_create(mat_2c_pot(i_img), template=mat_2c_pot(1))
3551 1010 : CALL dbt_copy_matrix_to_tensor(ri_data%kp_mat_2c_pot(1, i_img), work)
3552 1010 : CALL get_tensor_occupancy(work, nze, occ)
3553 1010 : IF (nze == 0) CYCLE
3554 :
3555 654 : CALL copy_2c_to_subgroup(work_sub, work, group_size, ngroups, para_env)
3556 654 : CALL dbt_copy_tensor_to_matrix(work_sub, mat_2c_pot(i_img))
3557 654 : CALL dbcsr_filter(mat_2c_pot(i_img), ri_data%filter_eps)
3558 1706 : CALL dbt_clear(work_sub)
3559 : END DO
3560 :
3561 : !The derivatives of the HFX potential
3562 168 : DO i_xyz = 1, 3
3563 3198 : DO i_img = 1, nimg
3564 3030 : CALL dbcsr_create(mat_der_pot_sub(i_img, i_xyz), template=mat_2c_pot(1))
3565 3030 : CALL dbt_copy_matrix_to_tensor(mat_der_pot(i_img, i_xyz), work)
3566 3030 : CALL dbcsr_release(mat_der_pot(i_img, i_xyz))
3567 3030 : CALL get_tensor_occupancy(work, nze, occ)
3568 3030 : IF (nze == 0) CYCLE
3569 :
3570 1958 : CALL copy_2c_to_subgroup(work_sub, work, group_size, ngroups, para_env)
3571 1958 : CALL dbt_copy_tensor_to_matrix(work_sub, mat_der_pot_sub(i_img, i_xyz))
3572 1958 : CALL dbcsr_filter(mat_der_pot_sub(i_img, i_xyz), ri_data%filter_eps)
3573 5114 : CALL dbt_clear(work_sub)
3574 : END DO
3575 : END DO
3576 :
3577 42 : CALL dbt_destroy(work)
3578 42 : CALL dbt_destroy(work_sub)
3579 42 : CALL dbt_pgrid_destroy(pgrid_2d)
3580 42 : CALL dbcsr_distribution_release(dbcsr_dist_sub)
3581 42 : DEALLOCATE (col_dist, row_dist, RI_blk_size, dbcsr_pgrid)
3582 :
3583 42 : CALL timestop(handle)
3584 :
3585 336 : END SUBROUTINE get_subgroup_2c_derivs
3586 :
3587 : ! **************************************************************************************************
3588 : !> \brief copy all required 3c derivative tensors from the main MPI group to the subgroups
3589 : !> \param t_3c_work_2 ...
3590 : !> \param t_3c_work_3 ...
3591 : !> \param t_3c_der_AO ...
3592 : !> \param t_3c_der_AO_sub ...
3593 : !> \param t_3c_der_RI ...
3594 : !> \param t_3c_der_RI_sub ...
3595 : !> \param t_3c_apc ...
3596 : !> \param t_3c_apc_sub ...
3597 : !> \param t_3c_der_stack ...
3598 : !> \param group_size ...
3599 : !> \param ngroups ...
3600 : !> \param para_env ...
3601 : !> \param para_env_sub ...
3602 : !> \param ri_data ...
3603 : !> \note the tensor containing the derivatives in the main MPI group are deleted for memory
3604 : ! **************************************************************************************************
3605 42 : SUBROUTINE get_subgroup_3c_derivs(t_3c_work_2, t_3c_work_3, t_3c_der_AO, t_3c_der_AO_sub, &
3606 42 : t_3c_der_RI, t_3c_der_RI_sub, t_3c_apc, t_3c_apc_sub, &
3607 42 : t_3c_der_stack, group_size, ngroups, para_env, para_env_sub, &
3608 : ri_data)
3609 : TYPE(dbt_type), DIMENSION(:), INTENT(INOUT) :: t_3c_work_2, t_3c_work_3
3610 : TYPE(dbt_type), DIMENSION(:, :), INTENT(INOUT) :: t_3c_der_AO, t_3c_der_AO_sub, &
3611 : t_3c_der_RI, t_3c_der_RI_sub, &
3612 : t_3c_apc, t_3c_apc_sub
3613 : TYPE(dbt_type), DIMENSION(:), INTENT(INOUT) :: t_3c_der_stack
3614 : INTEGER, INTENT(IN) :: group_size, ngroups
3615 : TYPE(mp_para_env_type), POINTER :: para_env, para_env_sub
3616 : TYPE(hfx_ri_type), INTENT(INOUT) :: ri_data
3617 :
3618 : CHARACTER(len=*), PARAMETER :: routineN = 'get_subgroup_3c_derivs'
3619 :
3620 : INTEGER :: batch_size, handle, i_img, i_RI, i_spin, &
3621 : i_xyz, ib, nblks_AO, nblks_RI, nimg, &
3622 : nspins, pdims(3)
3623 : INTEGER(int_8) :: nze
3624 42 : INTEGER, ALLOCATABLE, DIMENSION(:) :: bsizes_RI_ext, bsizes_RI_ext_split, &
3625 42 : bsizes_stack, dist1, dist2, dist3, &
3626 42 : dist_stack, idx_to_at
3627 : REAL(dp) :: occ
3628 378 : TYPE(dbt_distribution_type) :: t_dist
3629 126 : TYPE(dbt_pgrid_type) :: pgrid
3630 1050 : TYPE(dbt_type) :: tmp, work_atom_block, work_atom_block_sub
3631 :
3632 42 : CALL timeset(routineN, handle)
3633 :
3634 : !We use intermediate tensors with larger block size for more optimized communication
3635 42 : nblks_RI = SIZE(ri_data%bsizes_RI)
3636 126 : ALLOCATE (bsizes_RI_ext(ri_data%ncell_RI*nblks_RI))
3637 294 : DO i_RI = 1, ri_data%ncell_RI
3638 798 : bsizes_RI_ext((i_RI - 1)*nblks_RI + 1:i_RI*nblks_RI) = ri_data%bsizes_RI(:)
3639 : END DO
3640 :
3641 42 : CALL dbt_get_info(ri_data%kp_t_3c_int(1), pdims=pdims)
3642 42 : CALL dbt_pgrid_create(para_env_sub, pdims, pgrid)
3643 :
3644 : CALL create_3c_tensor(work_atom_block_sub, dist1, dist2, dist3, &
3645 : pgrid, bsizes_RI_ext, ri_data%bsizes_AO, &
3646 42 : ri_data%bsizes_AO, [1], [2, 3], name="(RI | AO AO)")
3647 42 : DEALLOCATE (dist1, dist2, dist3)
3648 :
3649 : CALL create_3c_tensor(work_atom_block, dist1, dist2, dist3, &
3650 : ri_data%pgrid_2, bsizes_RI_ext, ri_data%bsizes_AO, &
3651 42 : ri_data%bsizes_AO, [1], [2, 3], name="(RI | AO AO)")
3652 42 : DEALLOCATE (dist1, dist2, dist3)
3653 42 : CALL dbt_pgrid_destroy(pgrid)
3654 :
3655 : !We use the 3c integrals on the subgroup as template for the derivatives
3656 42 : nimg = ri_data%nimg
3657 168 : DO i_xyz = 1, 3
3658 3156 : DO i_img = 1, nimg
3659 3030 : CALL dbt_create(ri_data%kp_t_3c_int(1), t_3c_der_AO_sub(i_img, i_xyz))
3660 3030 : CALL get_tensor_occupancy(t_3c_der_AO(i_img, i_xyz), nze, occ)
3661 3030 : IF (nze == 0) CYCLE
3662 :
3663 1930 : CALL dbt_copy(t_3c_der_AO(i_img, i_xyz), work_atom_block, move_data=.TRUE.)
3664 : CALL copy_3c_to_subgroup(work_atom_block_sub, work_atom_block, &
3665 1930 : group_size, ngroups, para_env)
3666 5086 : CALL dbt_copy(work_atom_block_sub, t_3c_der_AO_sub(i_img, i_xyz), move_data=.TRUE.)
3667 : END DO
3668 :
3669 3156 : DO i_img = 1, nimg
3670 3030 : CALL dbt_create(ri_data%kp_t_3c_int(1), t_3c_der_RI_sub(i_img, i_xyz))
3671 3030 : CALL get_tensor_occupancy(t_3c_der_RI(i_img, i_xyz), nze, occ)
3672 3030 : IF (nze == 0) CYCLE
3673 :
3674 1910 : CALL dbt_copy(t_3c_der_RI(i_img, i_xyz), work_atom_block, move_data=.TRUE.)
3675 : CALL copy_3c_to_subgroup(work_atom_block_sub, work_atom_block, &
3676 1910 : group_size, ngroups, para_env)
3677 5066 : CALL dbt_copy(work_atom_block_sub, t_3c_der_RI_sub(i_img, i_xyz), move_data=.TRUE.)
3678 : END DO
3679 :
3680 3198 : DO i_img = 1, nimg
3681 3030 : CALL dbt_destroy(t_3c_der_RI(i_img, i_xyz))
3682 3156 : CALL dbt_destroy(t_3c_der_AO(i_img, i_xyz))
3683 : END DO
3684 : END DO
3685 42 : CALL dbt_destroy(work_atom_block_sub)
3686 42 : CALL dbt_destroy(work_atom_block)
3687 :
3688 : !Deal with t_3c_apc
3689 42 : nblks_RI = SIZE(ri_data%bsizes_RI_split)
3690 126 : ALLOCATE (bsizes_RI_ext_split(ri_data%ncell_RI*nblks_RI))
3691 294 : DO i_RI = 1, ri_data%ncell_RI
3692 1334 : bsizes_RI_ext_split((i_RI - 1)*nblks_RI + 1:i_RI*nblks_RI) = ri_data%bsizes_RI_split(:)
3693 : END DO
3694 :
3695 42 : pdims = 0
3696 : CALL dbt_pgrid_create(para_env_sub, pdims, pgrid, &
3697 168 : tensor_dims=[1, SIZE(bsizes_RI_ext_split), batch_size*SIZE(ri_data%bsizes_AO_split)])
3698 :
3699 : CALL create_3c_tensor(work_atom_block_sub, dist1, dist2, dist3, &
3700 : pgrid, ri_data%bsizes_AO, bsizes_RI_ext, &
3701 42 : ri_data%bsizes_AO, [1], [2, 3], name="(AO RI | AO)")
3702 42 : DEALLOCATE (dist1, dist2, dist3)
3703 :
3704 : CALL create_3c_tensor(work_atom_block, dist1, dist2, dist3, &
3705 : ri_data%pgrid_1, ri_data%bsizes_AO, bsizes_RI_ext, &
3706 42 : ri_data%bsizes_AO, [1], [2, 3], name="(AO RI | AO)")
3707 42 : DEALLOCATE (dist1, dist2, dist3)
3708 :
3709 : CALL create_3c_tensor(tmp, dist1, dist2, dist3, &
3710 : pgrid, ri_data%bsizes_AO_split, bsizes_RI_ext_split, &
3711 42 : ri_data%bsizes_AO_split, [1], [2, 3], name="(AO RI | AO)")
3712 42 : DEALLOCATE (dist1, dist2, dist3)
3713 :
3714 126 : ALLOCATE (idx_to_at(SIZE(ri_data%bsizes_AO)))
3715 42 : CALL get_idx_to_atom(idx_to_at, ri_data%bsizes_AO, ri_data%bsizes_AO)
3716 42 : nspins = SIZE(t_3c_apc, 1)
3717 1052 : DO i_img = 1, nimg
3718 2186 : DO i_spin = 1, nspins
3719 1176 : CALL dbt_create(tmp, t_3c_apc_sub(i_spin, i_img))
3720 1176 : CALL get_tensor_occupancy(t_3c_apc(i_spin, i_img), nze, occ)
3721 1176 : IF (nze == 0) CYCLE
3722 1152 : CALL dbt_copy(t_3c_apc(i_spin, i_img), work_atom_block, move_data=.TRUE.)
3723 : CALL copy_3c_to_subgroup(work_atom_block_sub, work_atom_block, group_size, &
3724 1152 : ngroups, para_env, ri_data%iatom_to_subgroup, 1, idx_to_at)
3725 3338 : CALL dbt_copy(work_atom_block_sub, t_3c_apc_sub(i_spin, i_img), move_data=.TRUE.)
3726 : END DO
3727 2228 : DO i_spin = 1, nspins
3728 2186 : CALL dbt_destroy(t_3c_apc(i_spin, i_img))
3729 : END DO
3730 : END DO
3731 42 : CALL dbt_destroy(tmp)
3732 42 : CALL dbt_destroy(work_atom_block)
3733 42 : CALL dbt_destroy(work_atom_block_sub)
3734 42 : CALL dbt_pgrid_destroy(pgrid)
3735 :
3736 : !t_3c_work_3 based on structure of 3c integrals/derivs
3737 42 : batch_size = ri_data%kp_stack_size
3738 42 : nblks_AO = SIZE(ri_data%bsizes_AO_split)
3739 126 : ALLOCATE (bsizes_stack(batch_size*nblks_AO))
3740 1386 : DO ib = 1, batch_size
3741 6826 : bsizes_stack((ib - 1)*nblks_AO + 1:ib*nblks_AO) = ri_data%bsizes_AO_split(:)
3742 : END DO
3743 :
3744 294 : ALLOCATE (dist1(ri_data%ncell_RI*nblks_RI), dist2(nblks_AO), dist3(nblks_AO))
3745 : CALL dbt_get_info(ri_data%kp_t_3c_int(1), proc_dist_1=dist1, proc_dist_2=dist2, &
3746 42 : proc_dist_3=dist3, pdims=pdims)
3747 :
3748 126 : ALLOCATE (dist_stack(batch_size*nblks_AO))
3749 1386 : DO ib = 1, batch_size
3750 6826 : dist_stack((ib - 1)*nblks_AO + 1:ib*nblks_AO) = dist3(:)
3751 : END DO
3752 :
3753 42 : CALL dbt_pgrid_create(para_env_sub, pdims, pgrid)
3754 42 : CALL dbt_distribution_new(t_dist, pgrid, dist1, dist2, dist_stack)
3755 : CALL dbt_create(t_3c_work_3(1), "work_3_stack", t_dist, [1], [2, 3], &
3756 42 : bsizes_RI_ext_split, ri_data%bsizes_AO_split, bsizes_stack)
3757 42 : CALL dbt_create(t_3c_work_3(1), t_3c_work_3(2))
3758 42 : CALL dbt_create(t_3c_work_3(1), t_3c_work_3(3))
3759 42 : CALL dbt_create(t_3c_work_3(1), t_3c_work_3(4))
3760 42 : CALL dbt_distribution_destroy(t_dist)
3761 42 : CALL dbt_pgrid_destroy(pgrid)
3762 42 : DEALLOCATE (dist1, dist2, dist3, dist_stack)
3763 :
3764 : !the derivatives are stacked in the same way
3765 42 : CALL dbt_create(t_3c_work_3(1), t_3c_der_stack(1))
3766 42 : CALL dbt_create(t_3c_work_3(1), t_3c_der_stack(2))
3767 42 : CALL dbt_create(t_3c_work_3(1), t_3c_der_stack(3))
3768 42 : CALL dbt_create(t_3c_work_3(1), t_3c_der_stack(4))
3769 42 : CALL dbt_create(t_3c_work_3(1), t_3c_der_stack(5))
3770 42 : CALL dbt_create(t_3c_work_3(1), t_3c_der_stack(6))
3771 :
3772 : !t_3c_work_2 based on structure of t_3c_apc
3773 294 : ALLOCATE (dist1(nblks_AO), dist2(ri_data%ncell_RI*nblks_RI), dist3(nblks_AO))
3774 : CALL dbt_get_info(t_3c_apc_sub(1, 1), proc_dist_1=dist1, proc_dist_2=dist2, &
3775 42 : proc_dist_3=dist3, pdims=pdims)
3776 :
3777 126 : ALLOCATE (dist_stack(batch_size*nblks_AO))
3778 1386 : DO ib = 1, batch_size
3779 6826 : dist_stack((ib - 1)*nblks_AO + 1:ib*nblks_AO) = dist3(:)
3780 : END DO
3781 :
3782 42 : CALL dbt_pgrid_create(para_env_sub, pdims, pgrid)
3783 42 : CALL dbt_distribution_new(t_dist, pgrid, dist1, dist2, dist_stack)
3784 : CALL dbt_create(t_3c_work_2(1), "work_3_stack", t_dist, [1], [2, 3], &
3785 42 : ri_data%bsizes_AO_split, bsizes_RI_ext_split, bsizes_stack)
3786 42 : CALL dbt_create(t_3c_work_2(1), t_3c_work_2(2))
3787 42 : CALL dbt_create(t_3c_work_2(1), t_3c_work_2(3))
3788 42 : CALL dbt_distribution_destroy(t_dist)
3789 42 : CALL dbt_pgrid_destroy(pgrid)
3790 42 : DEALLOCATE (dist1, dist2, dist3, dist_stack)
3791 :
3792 42 : CALL timestop(handle)
3793 :
3794 84 : END SUBROUTINE get_subgroup_3c_derivs
3795 :
3796 : ! **************************************************************************************************
3797 : !> \brief A routine that reorders the t_3c_int tensors such that all items which are fully empty
3798 : !> are bunched together. This way, we can get much more efficient screening based on NZE
3799 : !> \param t_3c_ints ...
3800 : !> \param ri_data ...
3801 : ! **************************************************************************************************
3802 70 : SUBROUTINE reorder_3c_ints(t_3c_ints, ri_data)
3803 : TYPE(dbt_type), DIMENSION(:), INTENT(INOUT) :: t_3c_ints
3804 : TYPE(hfx_ri_type), INTENT(INOUT) :: ri_data
3805 :
3806 : CHARACTER(LEN=*), PARAMETER :: routineN = 'reorder_3c_ints'
3807 :
3808 : INTEGER :: handle, i_img, idx, idx_empty, idx_full, &
3809 : nimg
3810 : INTEGER(int_8) :: nze
3811 : REAL(dp) :: occ
3812 70 : TYPE(dbt_type), ALLOCATABLE, DIMENSION(:) :: t_3c_tmp
3813 :
3814 70 : CALL timeset(routineN, handle)
3815 :
3816 70 : nimg = ri_data%nimg
3817 2486 : ALLOCATE (t_3c_tmp(nimg))
3818 1786 : DO i_img = 1, nimg
3819 1716 : CALL dbt_create(t_3c_ints(i_img), t_3c_tmp(i_img))
3820 1786 : CALL dbt_copy(t_3c_ints(i_img), t_3c_tmp(i_img), move_data=.TRUE.)
3821 : END DO
3822 :
3823 : !Loop over the images, check if ints have NZE == 0, and put them at the start or end of the
3824 : !initial tensor array. Keep the mapping in an array
3825 210 : ALLOCATE (ri_data%idx_to_img(nimg))
3826 70 : idx_full = 0
3827 70 : idx_empty = nimg + 1
3828 :
3829 1786 : DO i_img = 1, nimg
3830 1716 : CALL get_tensor_occupancy(t_3c_tmp(i_img), nze, occ)
3831 1716 : IF (nze == 0) THEN
3832 480 : idx_empty = idx_empty - 1
3833 480 : CALL dbt_copy(t_3c_tmp(i_img), t_3c_ints(idx_empty), move_data=.TRUE.)
3834 480 : ri_data%idx_to_img(idx_empty) = i_img
3835 : ELSE
3836 1236 : idx_full = idx_full + 1
3837 1236 : CALL dbt_copy(t_3c_tmp(i_img), t_3c_ints(idx_full), move_data=.TRUE.)
3838 1236 : ri_data%idx_to_img(idx_full) = i_img
3839 : END IF
3840 3502 : CALL dbt_destroy(t_3c_tmp(i_img))
3841 : END DO
3842 :
3843 : !store the highest image index with non-zero integrals
3844 70 : ri_data%nimg_nze = idx_full
3845 :
3846 140 : ALLOCATE (ri_data%img_to_idx(nimg))
3847 1786 : DO idx = 1, nimg
3848 1786 : ri_data%img_to_idx(ri_data%idx_to_img(idx)) = idx
3849 : END DO
3850 :
3851 70 : CALL timestop(handle)
3852 :
3853 1856 : END SUBROUTINE reorder_3c_ints
3854 :
3855 : ! **************************************************************************************************
3856 : !> \brief A routine that reorders the 3c derivatives, the same way that the integrals are, also to
3857 : !> increase efficiency of screening
3858 : !> \param t_3c_derivs ...
3859 : !> \param ri_data ...
3860 : ! **************************************************************************************************
3861 84 : SUBROUTINE reorder_3c_derivs(t_3c_derivs, ri_data)
3862 : TYPE(dbt_type), DIMENSION(:, :), INTENT(INOUT) :: t_3c_derivs
3863 : TYPE(hfx_ri_type), INTENT(INOUT) :: ri_data
3864 :
3865 : CHARACTER(LEN=*), PARAMETER :: routineN = 'reorder_3c_derivs'
3866 :
3867 : INTEGER :: handle, i_img, i_xyz, idx, nimg
3868 : INTEGER(int_8) :: nze
3869 : REAL(dp) :: occ
3870 84 : TYPE(dbt_type), ALLOCATABLE, DIMENSION(:) :: t_3c_tmp
3871 :
3872 84 : CALL timeset(routineN, handle)
3873 :
3874 84 : nimg = ri_data%nimg
3875 2944 : ALLOCATE (t_3c_tmp(nimg))
3876 2104 : DO i_img = 1, nimg
3877 2104 : CALL dbt_create(t_3c_derivs(1, 1), t_3c_tmp(i_img))
3878 : END DO
3879 :
3880 336 : DO i_xyz = 1, 3
3881 6312 : DO i_img = 1, nimg
3882 6312 : CALL dbt_copy(t_3c_derivs(i_img, i_xyz), t_3c_tmp(i_img), move_data=.TRUE.)
3883 : END DO
3884 6396 : DO i_img = 1, nimg
3885 6060 : idx = ri_data%img_to_idx(i_img)
3886 6060 : CALL dbt_copy(t_3c_tmp(i_img), t_3c_derivs(idx, i_xyz), move_data=.TRUE.)
3887 6060 : CALL get_tensor_occupancy(t_3c_derivs(idx, i_xyz), nze, occ)
3888 6312 : IF (nze > 0) ri_data%nimg_nze = MAX(idx, ri_data%nimg_nze)
3889 : END DO
3890 : END DO
3891 :
3892 2104 : DO i_img = 1, nimg
3893 2104 : CALL dbt_destroy(t_3c_tmp(i_img))
3894 : END DO
3895 :
3896 84 : CALL timestop(handle)
3897 :
3898 2188 : END SUBROUTINE reorder_3c_derivs
3899 :
3900 : ! **************************************************************************************************
3901 : !> \brief Get the sparsity pattern related to the non-symmetric AO basis overlap neighbor list
3902 : !> \param pattern ...
3903 : !> \param ri_data ...
3904 : !> \param qs_env ...
3905 : ! **************************************************************************************************
3906 232 : SUBROUTINE get_sparsity_pattern(pattern, ri_data, qs_env)
3907 : INTEGER, DIMENSION(:, :, :), INTENT(INOUT) :: pattern
3908 : TYPE(hfx_ri_type), INTENT(INOUT) :: ri_data
3909 : TYPE(qs_environment_type), POINTER :: qs_env
3910 :
3911 : INTEGER :: iatom, j_img, jatom, mj_img, natom, nimg
3912 232 : INTEGER, ALLOCATABLE, DIMENSION(:) :: bins
3913 232 : INTEGER, ALLOCATABLE, DIMENSION(:, :, :) :: tmp_pattern
3914 : INTEGER, DIMENSION(3) :: cell_j
3915 232 : INTEGER, DIMENSION(:, :), POINTER :: index_to_cell
3916 232 : INTEGER, DIMENSION(:, :, :), POINTER :: cell_to_index
3917 : TYPE(dft_control_type), POINTER :: dft_control
3918 : TYPE(kpoint_type), POINTER :: kpoints
3919 : TYPE(mp_para_env_type), POINTER :: para_env
3920 : TYPE(neighbor_list_iterator_p_type), &
3921 232 : DIMENSION(:), POINTER :: nl_iterator
3922 : TYPE(neighbor_list_set_p_type), DIMENSION(:), &
3923 232 : POINTER :: nl_2c
3924 :
3925 232 : NULLIFY (nl_2c, nl_iterator, kpoints, cell_to_index, dft_control, index_to_cell, para_env)
3926 :
3927 232 : CALL get_qs_env(qs_env, kpoints=kpoints, dft_control=dft_control, para_env=para_env, natom=natom)
3928 232 : CALL get_kpoint_info(kpoints, cell_to_index=cell_to_index, index_to_cell=index_to_cell, sab_nl=nl_2c)
3929 :
3930 232 : nimg = ri_data%nimg
3931 40930 : pattern(:, :, :) = 0
3932 :
3933 : !We use the symmetric nl for all images that have an opposite cell
3934 232 : CALL neighbor_list_iterator_create(nl_iterator, nl_2c)
3935 10017 : DO WHILE (neighbor_list_iterate(nl_iterator) == 0)
3936 9785 : CALL get_iterator_info(nl_iterator, iatom=iatom, jatom=jatom, cell=cell_j)
3937 :
3938 9785 : j_img = cell_to_index(cell_j(1), cell_j(2), cell_j(3))
3939 9785 : IF (j_img > nimg .OR. j_img < 1) CYCLE
3940 :
3941 6948 : mj_img = get_opp_index(j_img, qs_env)
3942 6948 : IF (mj_img > nimg .OR. mj_img < 1) CYCLE
3943 :
3944 6659 : IF (ri_data%present_images(j_img) == 0) CYCLE
3945 :
3946 9785 : pattern(iatom, jatom, j_img) = 1
3947 : END DO
3948 232 : CALL neighbor_list_iterator_release(nl_iterator)
3949 :
3950 : !If there is no opposite cell present, then we take into account the non-symmetric nl
3951 232 : CALL get_kpoint_info(kpoints, sab_nl_nosym=nl_2c)
3952 :
3953 232 : CALL neighbor_list_iterator_create(nl_iterator, nl_2c)
3954 13102 : DO WHILE (neighbor_list_iterate(nl_iterator) == 0)
3955 12870 : CALL get_iterator_info(nl_iterator, iatom=iatom, jatom=jatom, cell=cell_j)
3956 :
3957 12870 : j_img = cell_to_index(cell_j(1), cell_j(2), cell_j(3))
3958 12870 : IF (j_img > nimg .OR. j_img < 1) CYCLE
3959 :
3960 8872 : mj_img = get_opp_index(j_img, qs_env)
3961 8872 : IF (mj_img .LE. nimg .AND. mj_img > 0) CYCLE
3962 :
3963 298 : IF (ri_data%present_images(j_img) == 0) CYCLE
3964 :
3965 12870 : pattern(iatom, jatom, j_img) = 1
3966 : END DO
3967 232 : CALL neighbor_list_iterator_release(nl_iterator)
3968 :
3969 81628 : CALL para_env%sum(pattern)
3970 :
3971 : !If the opposite image is considered, then there is no need to compute diagonal twice
3972 5814 : DO j_img = 2, nimg
3973 16978 : DO iatom = 1, natom
3974 16746 : IF (pattern(iatom, iatom, j_img) .NE. 0) THEN
3975 3788 : mj_img = get_opp_index(j_img, qs_env)
3976 3788 : IF (mj_img > nimg .OR. mj_img < 1) CYCLE
3977 3788 : pattern(iatom, iatom, mj_img) = 0
3978 : END IF
3979 : END DO
3980 : END DO
3981 :
3982 : ! We want to equilibrate the sparsity pattern such that there are same amount of blocks
3983 : ! for each atom i of i,j pairs
3984 696 : ALLOCATE (bins(natom))
3985 696 : bins(:) = 0
3986 :
3987 1160 : ALLOCATE (tmp_pattern(natom, natom, nimg))
3988 40930 : tmp_pattern(:, :, :) = 0
3989 6046 : DO j_img = 1, nimg
3990 17674 : DO jatom = 1, natom
3991 40698 : DO iatom = 1, natom
3992 23256 : IF (pattern(iatom, jatom, j_img) == 0) CYCLE
3993 7622 : mj_img = get_opp_index(j_img, qs_env)
3994 :
3995 : !Should we take the i,j,b or th j,i,-b atomic block?
3996 19250 : IF (mj_img > nimg .OR. mj_img < 1) THEN
3997 : !No opposite image, no choice
3998 198 : bins(iatom) = bins(iatom) + 1
3999 198 : tmp_pattern(iatom, jatom, j_img) = 1
4000 : ELSE
4001 :
4002 7424 : IF (bins(iatom) > bins(jatom)) THEN
4003 1498 : bins(jatom) = bins(jatom) + 1
4004 1498 : tmp_pattern(jatom, iatom, mj_img) = 1
4005 : ELSE
4006 5926 : bins(iatom) = bins(iatom) + 1
4007 5926 : tmp_pattern(iatom, jatom, j_img) = 1
4008 : END IF
4009 : END IF
4010 : END DO
4011 : END DO
4012 : END DO
4013 :
4014 : ! -1 => unoccupied, 0 => occupied
4015 40930 : pattern(:, :, :) = tmp_pattern(:, :, :) - 1
4016 :
4017 464 : END SUBROUTINE get_sparsity_pattern
4018 :
4019 : ! **************************************************************************************************
4020 : !> \brief Distribute the iatom, jatom, b_img triplet over the subgroupd to spread the load
4021 : !> the group id for each triplet is passed as the value of sparsity_pattern(i, j, b),
4022 : !> with -1 being an unoccupied block
4023 : !> \param sparsity_pattern ...
4024 : !> \param ngroups ...
4025 : !> \param ri_data ...
4026 : ! **************************************************************************************************
4027 232 : SUBROUTINE get_sub_dist(sparsity_pattern, ngroups, ri_data)
4028 : INTEGER, DIMENSION(:, :, :), INTENT(INOUT) :: sparsity_pattern
4029 : INTEGER, INTENT(IN) :: ngroups
4030 : TYPE(hfx_ri_type), INTENT(INOUT) :: ri_data
4031 :
4032 : INTEGER :: b_img, ctr, iat, iatom, igroup, jatom, &
4033 : natom, nimg, ub
4034 232 : INTEGER, ALLOCATABLE, DIMENSION(:) :: max_at_per_group
4035 : REAL(dp) :: cost
4036 232 : REAL(dp), ALLOCATABLE, DIMENSION(:) :: bins
4037 :
4038 232 : natom = SIZE(sparsity_pattern, 2)
4039 232 : nimg = SIZE(sparsity_pattern, 3)
4040 :
4041 : !To avoid unnecessary data replication accross the subgroups, we want to have a limited number
4042 : !of subgroup with the data of a given iatom. At the minimum, all groups have 1 atom
4043 : !We assume that the cost associated to each iatom is roughly the same
4044 232 : IF (.NOT. ALLOCATED(ri_data%iatom_to_subgroup)) THEN
4045 350 : ALLOCATE (ri_data%iatom_to_subgroup(natom), max_at_per_group(ngroups))
4046 150 : DO iatom = 1, natom
4047 100 : NULLIFY (ri_data%iatom_to_subgroup(iatom)%array)
4048 200 : ALLOCATE (ri_data%iatom_to_subgroup(iatom)%array(ngroups))
4049 350 : ri_data%iatom_to_subgroup(iatom)%array(:) = .FALSE.
4050 : END DO
4051 :
4052 50 : ub = natom/ngroups
4053 50 : IF (ub*ngroups < natom) ub = ub + 1
4054 150 : max_at_per_group(:) = MAX(1, ub)
4055 :
4056 : !We want each atom to be present the same amount of times. Some groups might have more atoms
4057 : !than other to achieve this.
4058 : ctr = 0
4059 150 : DO WHILE (MODULO(SUM(max_at_per_group), natom) .NE. 0)
4060 0 : igroup = MODULO(ctr, ngroups) + 1
4061 0 : max_at_per_group(igroup) = max_at_per_group(igroup) + 1
4062 50 : ctr = ctr + 1
4063 : END DO
4064 :
4065 : ctr = 0
4066 150 : DO igroup = 1, ngroups
4067 250 : DO iat = 1, max_at_per_group(igroup)
4068 100 : iatom = MODULO(ctr, natom) + 1
4069 100 : ri_data%iatom_to_subgroup(iatom)%array(igroup) = .TRUE.
4070 200 : ctr = ctr + 1
4071 : END DO
4072 : END DO
4073 : END IF
4074 :
4075 696 : ALLOCATE (bins(ngroups))
4076 696 : bins = 0.0_dp
4077 6046 : DO b_img = 1, nimg
4078 17674 : DO jatom = 1, natom
4079 40698 : DO iatom = 1, natom
4080 23256 : IF (sparsity_pattern(iatom, jatom, b_img) == -1) CYCLE
4081 38110 : igroup = MINLOC(bins, 1, MASK=ri_data%iatom_to_subgroup(iatom)%array) - 1
4082 :
4083 : !Use cost information from previous SCF if available
4084 554266 : IF (ANY(ri_data%kp_cost > EPSILON(0.0_dp))) THEN
4085 5404 : cost = ri_data%kp_cost(iatom, jatom, b_img)
4086 : ELSE
4087 2218 : cost = REAL(ri_data%bsizes_AO(iatom)*ri_data%bsizes_AO(jatom), dp)
4088 : END IF
4089 7622 : bins(igroup + 1) = bins(igroup + 1) + cost
4090 34884 : sparsity_pattern(iatom, jatom, b_img) = igroup
4091 : END DO
4092 : END DO
4093 : END DO
4094 :
4095 232 : END SUBROUTINE get_sub_dist
4096 :
4097 : ! **************************************************************************************************
4098 : !> \brief A rouine that updates the sparsity pattern for force calculation, where all i,j,b combinations
4099 : !> are visited.
4100 : !> \param force_pattern ...
4101 : !> \param scf_pattern ...
4102 : !> \param ngroups ...
4103 : !> \param ri_data ...
4104 : !> \param qs_env ...
4105 : ! **************************************************************************************************
4106 42 : SUBROUTINE update_pattern_to_forces(force_pattern, scf_pattern, ngroups, ri_data, qs_env)
4107 : INTEGER, DIMENSION(:, :, :), INTENT(INOUT) :: force_pattern, scf_pattern
4108 : INTEGER, INTENT(IN) :: ngroups
4109 : TYPE(hfx_ri_type), INTENT(INOUT) :: ri_data
4110 : TYPE(qs_environment_type), POINTER :: qs_env
4111 :
4112 : INTEGER :: b_img, iatom, igroup, jatom, mb_img, &
4113 : natom, nimg
4114 42 : REAL(dp), ALLOCATABLE, DIMENSION(:) :: bins
4115 :
4116 42 : natom = SIZE(scf_pattern, 2)
4117 42 : nimg = SIZE(scf_pattern, 3)
4118 :
4119 126 : ALLOCATE (bins(ngroups))
4120 126 : bins = 0.0_dp
4121 :
4122 1052 : DO b_img = 1, nimg
4123 1010 : mb_img = get_opp_index(b_img, qs_env)
4124 3072 : DO jatom = 1, natom
4125 7070 : DO iatom = 1, natom
4126 : !Important: same distribution as KS matrix, because reuse t_3c_apc
4127 20200 : igroup = MINLOC(bins, 1, MASK=ri_data%iatom_to_subgroup(iatom)%array) - 1
4128 :
4129 : !check that block not already treated
4130 4040 : IF (scf_pattern(iatom, jatom, b_img) > -1) CYCLE
4131 :
4132 : !If not, take the cost of block j, i, -b (same energy contribution)
4133 4902 : IF (mb_img > 0 .AND. mb_img .LE. nimg) THEN
4134 2486 : IF (scf_pattern(jatom, iatom, mb_img) == -1) CYCLE
4135 1038 : bins(igroup + 1) = bins(igroup + 1) + ri_data%kp_cost(jatom, iatom, mb_img)
4136 1038 : force_pattern(iatom, jatom, b_img) = igroup
4137 : END IF
4138 : END DO
4139 : END DO
4140 : END DO
4141 :
4142 42 : END SUBROUTINE update_pattern_to_forces
4143 :
4144 : ! **************************************************************************************************
4145 : !> \brief A routine that determines the extend of the KP RI-HFX periodic images, including for the
4146 : !> extension of the RI basis
4147 : !> \param ri_data ...
4148 : !> \param qs_env ...
4149 : ! **************************************************************************************************
4150 70 : SUBROUTINE get_kp_and_ri_images(ri_data, qs_env)
4151 : TYPE(hfx_ri_type), INTENT(INOUT) :: ri_data
4152 : TYPE(qs_environment_type), POINTER :: qs_env
4153 :
4154 : CHARACTER(LEN=*), PARAMETER :: routineN = 'get_kp_and_ri_images'
4155 :
4156 : INTEGER :: cell_j(3), cell_k(3), handle, i_img, iatom, ikind, j_img, jatom, jcell, katom, &
4157 : kcell, kp_index_lbounds(3), kp_index_ubounds(3), natom, ngroups, nimg, nkind, pcoord(3), &
4158 : pdims(3)
4159 70 : INTEGER, ALLOCATABLE, DIMENSION(:) :: dist_AO_1, dist_AO_2, dist_RI, &
4160 70 : nRI_per_atom, present_img, RI_cells
4161 70 : INTEGER, DIMENSION(:, :, :), POINTER :: cell_to_index
4162 : REAL(dp) :: bump_fact, dij, dik, image_range, &
4163 : RI_range, rij(3), rik(3)
4164 490 : TYPE(dbt_type) :: t_dummy
4165 : TYPE(dft_control_type), POINTER :: dft_control
4166 : TYPE(distribution_2d_type), POINTER :: dist_2d
4167 : TYPE(distribution_3d_type) :: dist_3d
4168 : TYPE(gto_basis_set_p_type), ALLOCATABLE, &
4169 70 : DIMENSION(:), TARGET :: basis_set_AO, basis_set_RI
4170 : TYPE(kpoint_type), POINTER :: kpoints
4171 70 : TYPE(mp_cart_type) :: mp_comm_t3c
4172 : TYPE(mp_para_env_type), POINTER :: para_env
4173 : TYPE(neighbor_list_3c_iterator_type) :: nl_3c_iter
4174 : TYPE(neighbor_list_3c_type) :: nl_3c
4175 : TYPE(neighbor_list_iterator_p_type), &
4176 70 : DIMENSION(:), POINTER :: nl_iterator
4177 : TYPE(neighbor_list_set_p_type), DIMENSION(:), &
4178 70 : POINTER :: nl_2c
4179 70 : TYPE(particle_type), DIMENSION(:), POINTER :: particle_set
4180 70 : TYPE(qs_kind_type), DIMENSION(:), POINTER :: qs_kind_set
4181 : TYPE(section_vals_type), POINTER :: hfx_section
4182 :
4183 70 : NULLIFY (qs_kind_set, dist_2d, nl_2c, nl_iterator, dft_control, &
4184 70 : particle_set, kpoints, para_env, cell_to_index, hfx_section)
4185 :
4186 70 : CALL timeset(routineN, handle)
4187 :
4188 : CALL get_qs_env(qs_env, nkind=nkind, qs_kind_set=qs_kind_set, distribution_2d=dist_2d, &
4189 : dft_control=dft_control, particle_set=particle_set, kpoints=kpoints, &
4190 70 : para_env=para_env, natom=natom)
4191 70 : nimg = dft_control%nimages
4192 70 : CALL get_kpoint_info(kpoints, cell_to_index=cell_to_index)
4193 280 : kp_index_lbounds = LBOUND(cell_to_index)
4194 280 : kp_index_ubounds = UBOUND(cell_to_index)
4195 :
4196 70 : hfx_section => section_vals_get_subs_vals(qs_env%input, "DFT%XC%HF%RI")
4197 70 : CALL section_vals_val_get(hfx_section, "KP_NGROUPS", i_val=ngroups)
4198 :
4199 496 : ALLOCATE (basis_set_RI(nkind), basis_set_AO(nkind))
4200 70 : CALL basis_set_list_setup(basis_set_RI, ri_data%ri_basis_type, qs_kind_set)
4201 70 : CALL basis_set_list_setup(basis_set_AO, ri_data%orb_basis_type, qs_kind_set)
4202 :
4203 : !In case of shortrange HFX potential, it is imprtant to be consistent with the rest of the KP
4204 : !code, and use EPS_SCHWARZ to determine the range (rather than eps_filter_2c in normal RI-HFX)
4205 70 : IF (ri_data%hfx_pot%potential_type == do_potential_short) THEN
4206 0 : CALL erfc_cutoff(ri_data%eps_schwarz, ri_data%hfx_pot%omega, ri_data%hfx_pot%cutoff_radius)
4207 : END IF
4208 :
4209 : !Determine the range for contributing periodic images, and for the RI basis extension
4210 70 : ri_data%kp_RI_range = 0.0_dp
4211 70 : ri_data%kp_image_range = 0.0_dp
4212 178 : DO ikind = 1, nkind
4213 :
4214 108 : CALL init_interaction_radii_orb_basis(basis_set_AO(ikind)%gto_basis_set, ri_data%eps_pgf_orb)
4215 108 : CALL get_gto_basis_set(basis_set_AO(ikind)%gto_basis_set, kind_radius=RI_range)
4216 108 : ri_data%kp_RI_range = MAX(RI_range, ri_data%kp_RI_range)
4217 :
4218 108 : CALL init_interaction_radii_orb_basis(basis_set_AO(ikind)%gto_basis_set, ri_data%eps_pgf_orb)
4219 108 : CALL init_interaction_radii_orb_basis(basis_set_RI(ikind)%gto_basis_set, ri_data%eps_pgf_orb)
4220 108 : CALL get_gto_basis_set(basis_set_RI(ikind)%gto_basis_set, kind_radius=image_range)
4221 :
4222 108 : image_range = 2.0_dp*image_range + cutoff_screen_factor*ri_data%hfx_pot%cutoff_radius
4223 286 : ri_data%kp_image_range = MAX(image_range, ri_data%kp_image_range)
4224 : END DO
4225 :
4226 70 : CALL section_vals_val_get(hfx_section, "KP_RI_BUMP_FACTOR", r_val=bump_fact)
4227 70 : ri_data%kp_bump_rad = bump_fact*ri_data%kp_RI_range
4228 :
4229 : !For the extent of the KP RI-HFX images, we are limited by the RI-HFX potential in
4230 : !(mu^0 sigma^a|P^0) (P^0|Q^b) (Q^b|nu^b lambda^a+c), if there is no contact between
4231 : !any P^0 and Q^b, then image b does not contribute
4232 : CALL build_2c_neighbor_lists(nl_2c, basis_set_RI, basis_set_RI, ri_data%hfx_pot, &
4233 70 : "HFX_2c_nl_RI", qs_env, sym_ij=.FALSE., dist_2d=dist_2d)
4234 :
4235 210 : ALLOCATE (present_img(nimg))
4236 3120 : present_img = 0
4237 70 : ri_data%nimg = 0
4238 70 : CALL neighbor_list_iterator_create(nl_iterator, nl_2c)
4239 1568 : DO WHILE (neighbor_list_iterate(nl_iterator) == 0)
4240 1498 : CALL get_iterator_info(nl_iterator, r=rij, cell=cell_j)
4241 :
4242 5992 : dij = NORM2(rij)
4243 :
4244 1498 : j_img = cell_to_index(cell_j(1), cell_j(2), cell_j(3))
4245 1498 : IF (j_img > nimg .OR. j_img < 1) CYCLE
4246 :
4247 1466 : IF (dij > ri_data%kp_image_range) CYCLE
4248 :
4249 1466 : ri_data%nimg = MAX(j_img, ri_data%nimg)
4250 1498 : present_img(j_img) = 1
4251 :
4252 : END DO
4253 70 : CALL neighbor_list_iterator_release(nl_iterator)
4254 70 : CALL release_neighbor_list_sets(nl_2c)
4255 70 : CALL para_env%max(ri_data%nimg)
4256 70 : IF (ri_data%nimg > nimg) &
4257 0 : CPABORT("Make sure the smallest exponent of the RI-HFX basis is larger than that of the ORB basis.")
4258 :
4259 : !Keep track of which images will not contribute, so that can be ignored before calculation
4260 70 : CALL para_env%sum(present_img)
4261 210 : ALLOCATE (ri_data%present_images(ri_data%nimg))
4262 1786 : ri_data%present_images = 0
4263 1786 : DO i_img = 1, ri_data%nimg
4264 1786 : IF (present_img(i_img) > 0) ri_data%present_images(i_img) = 1
4265 : END DO
4266 :
4267 : CALL create_3c_tensor(t_dummy, dist_AO_1, dist_AO_2, dist_RI, &
4268 : ri_data%pgrid, ri_data%bsizes_AO, ri_data%bsizes_AO, ri_data%bsizes_RI, &
4269 70 : map1=[1, 2], map2=[3], name="(AO AO | RI)")
4270 :
4271 70 : CALL dbt_mp_environ_pgrid(ri_data%pgrid, pdims, pcoord)
4272 70 : CALL mp_comm_t3c%create(ri_data%pgrid%mp_comm_2d, 3, pdims)
4273 : CALL distribution_3d_create(dist_3d, dist_AO_1, dist_AO_2, dist_RI, &
4274 70 : nkind, particle_set, mp_comm_t3c, own_comm=.TRUE.)
4275 70 : DEALLOCATE (dist_RI, dist_AO_1, dist_AO_2)
4276 70 : CALL dbt_destroy(t_dummy)
4277 :
4278 : !For the extension of the RI basis P in (mu^0 sigma^a |P^i), we consider an atom if the distance,
4279 : !between mu^0 and P^i if smaller or equal to the kind radius of mu^0
4280 : CALL build_3c_neighbor_lists(nl_3c, basis_set_AO, basis_set_AO, basis_set_RI, dist_3d, &
4281 : ri_data%ri_metric, "HFX_3c_nl", qs_env, op_pos=2, sym_ij=.FALSE., &
4282 70 : own_dist=.TRUE.)
4283 :
4284 140 : ALLOCATE (RI_cells(nimg))
4285 3120 : RI_cells = 0
4286 :
4287 210 : ALLOCATE (nRI_per_atom(natom))
4288 210 : nRI_per_atom = 0
4289 :
4290 70 : CALL neighbor_list_3c_iterator_create(nl_3c_iter, nl_3c)
4291 58508 : DO WHILE (neighbor_list_3c_iterate(nl_3c_iter) == 0)
4292 : CALL get_3c_iterator_info(nl_3c_iter, cell_k=cell_k, rik=rik, cell_j=cell_j, &
4293 58438 : iatom=iatom, jatom=jatom, katom=katom)
4294 233752 : dik = NORM2(rik)
4295 :
4296 409066 : IF (ANY([cell_j(1), cell_j(2), cell_j(3)] < kp_index_lbounds) .OR. &
4297 : ANY([cell_j(1), cell_j(2), cell_j(3)] > kp_index_ubounds)) CYCLE
4298 :
4299 58438 : jcell = cell_to_index(cell_j(1), cell_j(2), cell_j(3))
4300 58438 : IF (jcell > nimg .OR. jcell < 1) CYCLE
4301 :
4302 385667 : IF (ANY([cell_k(1), cell_k(2), cell_k(3)] < kp_index_lbounds) .OR. &
4303 : ANY([cell_k(1), cell_k(2), cell_k(3)] > kp_index_ubounds)) CYCLE
4304 :
4305 51169 : kcell = cell_to_index(cell_k(1), cell_k(2), cell_k(3))
4306 51169 : IF (kcell > nimg .OR. kcell < 1) CYCLE
4307 :
4308 43523 : IF (dik > ri_data%kp_RI_range) CYCLE
4309 5791 : RI_cells(kcell) = 1
4310 :
4311 5861 : IF (jcell == 1 .AND. iatom == jatom) nRI_per_atom(iatom) = nRI_per_atom(iatom) + ri_data%bsizes_RI(katom)
4312 : END DO
4313 70 : CALL neighbor_list_3c_iterator_destroy(nl_3c_iter)
4314 70 : CALL neighbor_list_3c_destroy(nl_3c)
4315 70 : CALL para_env%sum(RI_cells)
4316 70 : CALL para_env%sum(nRI_per_atom)
4317 :
4318 140 : ALLOCATE (ri_data%img_to_RI_cell(nimg))
4319 70 : ri_data%ncell_RI = 0
4320 3120 : ri_data%img_to_RI_cell = 0
4321 3120 : DO i_img = 1, nimg
4322 3120 : IF (RI_cells(i_img) > 0) THEN
4323 436 : ri_data%ncell_RI = ri_data%ncell_RI + 1
4324 436 : ri_data%img_to_RI_cell(i_img) = ri_data%ncell_RI
4325 : END IF
4326 : END DO
4327 :
4328 210 : ALLOCATE (ri_data%RI_cell_to_img(ri_data%ncell_RI))
4329 3120 : DO i_img = 1, nimg
4330 3120 : IF (ri_data%img_to_RI_cell(i_img) > 0) ri_data%RI_cell_to_img(ri_data%img_to_RI_cell(i_img)) = i_img
4331 : END DO
4332 :
4333 : !Print some info
4334 70 : IF (ri_data%unit_nr > 0) THEN
4335 : WRITE (ri_data%unit_nr, FMT="(/T3,A,I29)") &
4336 35 : "KP-HFX_RI_INFO| Number of RI-KP parallel groups:", ngroups
4337 : WRITE (ri_data%unit_nr, FMT="(T3,A,F31.3,A)") &
4338 35 : "KP-HFX_RI_INFO| RI basis extension radius:", ri_data%kp_RI_range*angstrom, " Ang"
4339 : WRITE (ri_data%unit_nr, FMT="(T3,A,F12.3,A, F6.3, A)") &
4340 35 : "KP-HFX_RI_INFO| RI basis bump factor and bump radius:", bump_fact, " /", &
4341 70 : ri_data%kp_bump_rad*angstrom, " Ang"
4342 : WRITE (ri_data%unit_nr, FMT="(T3,A,I16,A)") &
4343 35 : "KP-HFX_RI_INFO| The extended RI bases cover up to ", ri_data%ncell_RI, " unit cells"
4344 : WRITE (ri_data%unit_nr, FMT="(T3,A,I18)") &
4345 105 : "KP-HFX_RI_INFO| Average number of sgf in extended RI bases:", SUM(nRI_per_atom)/natom
4346 : WRITE (ri_data%unit_nr, FMT="(T3,A,F13.3,A)") &
4347 35 : "KP-HFX_RI_INFO| Consider all image cells within a radius of ", ri_data%kp_image_range*angstrom, " Ang"
4348 : WRITE (ri_data%unit_nr, FMT="(T3,A,I27/)") &
4349 35 : "KP-HFX_RI_INFO| Number of image cells considered: ", ri_data%nimg
4350 35 : CALL m_flush(ri_data%unit_nr)
4351 : END IF
4352 :
4353 70 : CALL timestop(handle)
4354 :
4355 840 : END SUBROUTINE get_kp_and_ri_images
4356 :
4357 : ! **************************************************************************************************
4358 : !> \brief A routine that creates tensors structure for rho_ao and 3c_ints in a stacked format for
4359 : !> the efficient contractions of rho_sigma^0,lambda^c * (mu^0 sigam^a | P) => TAS tensors
4360 : !> \param res_stack ...
4361 : !> \param rho_stack ...
4362 : !> \param ints_stack ...
4363 : !> \param rho_template ...
4364 : !> \param ints_template ...
4365 : !> \param stack_size ...
4366 : !> \param ri_data ...
4367 : !> \param qs_env ...
4368 : !> \note The result tensor has the exact same shape and distribution as the integral tensor
4369 : ! **************************************************************************************************
4370 232 : SUBROUTINE get_stack_tensors(res_stack, rho_stack, ints_stack, rho_template, ints_template, &
4371 : stack_size, ri_data, qs_env)
4372 : TYPE(dbt_type), DIMENSION(:), INTENT(INOUT) :: res_stack, rho_stack, ints_stack
4373 : TYPE(dbt_type), INTENT(INOUT) :: rho_template, ints_template
4374 : INTEGER, INTENT(IN) :: stack_size
4375 : TYPE(hfx_ri_type), INTENT(INOUT) :: ri_data
4376 : TYPE(qs_environment_type), POINTER :: qs_env
4377 :
4378 : INTEGER :: is, nblks, nblks_3c(3), pdims_3d(3)
4379 232 : INTEGER, ALLOCATABLE, DIMENSION(:) :: bsizes_RI_ext, bsizes_stack, dist1, &
4380 232 : dist2, dist3, dist_stack1, &
4381 232 : dist_stack2, dist_stack3
4382 2088 : TYPE(dbt_distribution_type) :: t_dist
4383 696 : TYPE(dbt_pgrid_type) :: pgrid
4384 : TYPE(mp_para_env_type), POINTER :: para_env
4385 :
4386 232 : NULLIFY (para_env)
4387 :
4388 232 : CALL get_qs_env(qs_env, para_env=para_env)
4389 :
4390 232 : nblks = SIZE(ri_data%bsizes_AO_split)
4391 696 : ALLOCATE (bsizes_stack(stack_size*nblks))
4392 3082 : DO is = 1, stack_size
4393 11582 : bsizes_stack((is - 1)*nblks + 1:is*nblks) = ri_data%bsizes_AO_split(:)
4394 : END DO
4395 :
4396 2088 : ALLOCATE (dist1(nblks), dist2(nblks), dist_stack1(stack_size*nblks), dist_stack2(stack_size*nblks))
4397 232 : CALL dbt_get_info(rho_template, proc_dist_1=dist1, proc_dist_2=dist2)
4398 3082 : DO is = 1, stack_size
4399 11350 : dist_stack1((is - 1)*nblks + 1:is*nblks) = dist1(:)
4400 11582 : dist_stack2((is - 1)*nblks + 1:is*nblks) = dist2(:)
4401 : END DO
4402 :
4403 : !First 2c tensor matches the distribution of template
4404 : !It is stacked in both directions
4405 232 : CALL dbt_distribution_new(t_dist, ri_data%pgrid_2d, dist_stack1, dist_stack2)
4406 232 : CALL dbt_create(rho_stack(1), "RHO_stack", t_dist, [1], [2], bsizes_stack, bsizes_stack)
4407 232 : CALL dbt_distribution_destroy(t_dist)
4408 232 : DEALLOCATE (dist1, dist2, dist_stack1, dist_stack2)
4409 :
4410 : !Second 2c tensor has optimal distribution on the 2d pgrid
4411 232 : CALL create_2c_tensor(rho_stack(2), dist1, dist2, ri_data%pgrid_2d, bsizes_stack, bsizes_stack, name="RHO_stack")
4412 232 : DEALLOCATE (dist1, dist2)
4413 :
4414 232 : CALL dbt_get_info(ints_template, nblks_total=nblks_3c)
4415 1624 : ALLOCATE (dist1(nblks_3c(1)), dist2(nblks_3c(2)), dist3(nblks_3c(3)))
4416 1160 : ALLOCATE (dist_stack3(stack_size*nblks_3c(3)), bsizes_RI_ext(nblks_3c(2)))
4417 : CALL dbt_get_info(ints_template, proc_dist_1=dist1, proc_dist_2=dist2, &
4418 232 : proc_dist_3=dist3, blk_size_2=bsizes_RI_ext)
4419 3082 : DO is = 1, stack_size
4420 11582 : dist_stack3((is - 1)*nblks_3c(3) + 1:is*nblks_3c(3)) = dist3(:)
4421 : END DO
4422 :
4423 : !First 3c tensor matches the distribution of template
4424 232 : CALL dbt_distribution_new(t_dist, ri_data%pgrid_1, dist1, dist2, dist_stack3)
4425 : CALL dbt_create(ints_stack(1), "ints_stack", t_dist, [1, 2], [3], ri_data%bsizes_AO_split, &
4426 232 : bsizes_RI_ext, bsizes_stack)
4427 232 : CALL dbt_distribution_destroy(t_dist)
4428 232 : DEALLOCATE (dist1, dist2, dist3, dist_stack3)
4429 :
4430 : !Second 3c tensor has optimal pgrid
4431 232 : pdims_3d = 0
4432 928 : CALL dbt_pgrid_create(para_env, pdims_3d, pgrid, tensor_dims=[nblks_3c(1), nblks_3c(2), stack_size*nblks_3c(3)])
4433 : CALL create_3c_tensor(ints_stack(2), dist1, dist2, dist3, pgrid, ri_data%bsizes_AO_split, &
4434 232 : bsizes_RI_ext, bsizes_stack, [1, 2], [3], name="ints_stack")
4435 232 : DEALLOCATE (dist1, dist2, dist3)
4436 232 : CALL dbt_pgrid_destroy(pgrid)
4437 :
4438 : !The result tensor has the same shape and dist as the integral tensor
4439 232 : CALL dbt_create(ints_stack(1), res_stack(1))
4440 232 : CALL dbt_create(ints_stack(2), res_stack(2))
4441 :
4442 464 : END SUBROUTINE get_stack_tensors
4443 :
4444 : ! **************************************************************************************************
4445 : !> \brief Fill the stack of 3c tensors accrding to the order in the images input
4446 : !> \param t_3c_stack ...
4447 : !> \param t_3c_in ...
4448 : !> \param images ...
4449 : !> \param stack_dim ...
4450 : !> \param ri_data ...
4451 : !> \param filter_at ...
4452 : !> \param filter_dim ...
4453 : !> \param idx_to_at ...
4454 : !> \param img_bounds ...
4455 : ! **************************************************************************************************
4456 22326 : SUBROUTINE fill_3c_stack(t_3c_stack, t_3c_in, images, stack_dim, ri_data, filter_at, filter_dim, &
4457 22326 : idx_to_at, img_bounds)
4458 : TYPE(dbt_type), INTENT(INOUT) :: t_3c_stack
4459 : TYPE(dbt_type), DIMENSION(:), INTENT(INOUT) :: t_3c_in
4460 : INTEGER, DIMENSION(:), INTENT(INOUT) :: images
4461 : INTEGER, INTENT(IN) :: stack_dim
4462 : TYPE(hfx_ri_type), INTENT(INOUT) :: ri_data
4463 : INTEGER, INTENT(IN), OPTIONAL :: filter_at, filter_dim
4464 : INTEGER, DIMENSION(:), INTENT(INOUT), OPTIONAL :: idx_to_at
4465 : INTEGER, INTENT(IN), OPTIONAL :: img_bounds(2)
4466 :
4467 : INTEGER :: dest(3), i_img, idx, ind(3), lb, nblks, &
4468 : nimg, offset, ub
4469 : LOGICAL :: do_filter, found
4470 22326 : REAL(dp), ALLOCATABLE, DIMENSION(:, :, :) :: blk
4471 : TYPE(dbt_iterator_type) :: iter
4472 :
4473 : !We loop over the a images from the ac_pairs, then copy the 3c ints to the correct spot in
4474 : !in the stack tensor (corresponding to pair index). Distributions match by construction
4475 22326 : nimg = ri_data%nimg
4476 22326 : nblks = SIZE(ri_data%bsizes_AO_split)
4477 :
4478 22326 : do_filter = .FALSE.
4479 21806 : IF (PRESENT(filter_at) .AND. PRESENT(filter_dim) .AND. PRESENT(idx_to_at)) do_filter = .TRUE.
4480 :
4481 22326 : lb = 1
4482 22326 : ub = nimg
4483 22326 : offset = 0
4484 22326 : IF (PRESENT(img_bounds)) THEN
4485 22326 : lb = img_bounds(1)
4486 22326 : ub = img_bounds(2) - 1
4487 22326 : offset = lb - 1
4488 : END IF
4489 :
4490 449976 : DO idx = lb, ub
4491 427650 : i_img = images(idx)
4492 427650 : IF (i_img == 0 .OR. i_img > nimg) CYCLE
4493 :
4494 : !$OMP PARALLEL DEFAULT(NONE) &
4495 : !$OMP SHARED(idx,i_img,t_3c_in,t_3c_stack,nblks,stack_dim,filter_at,filter_dim,idx_to_at,do_filter,offset) &
4496 449976 : !$OMP PRIVATE(iter,ind,blk,found,dest)
4497 : CALL dbt_iterator_start(iter, t_3c_in(i_img))
4498 : DO WHILE (dbt_iterator_blocks_left(iter))
4499 : CALL dbt_iterator_next_block(iter, ind)
4500 : CALL dbt_get_block(t_3c_in(i_img), ind, blk, found)
4501 : IF (.NOT. found) CYCLE
4502 :
4503 : IF (do_filter) THEN
4504 : IF (.NOT. idx_to_at(ind(filter_dim)) == filter_at) CYCLE
4505 : END IF
4506 :
4507 : IF (stack_dim == 1) THEN
4508 : dest = [(idx - offset - 1)*nblks + ind(1), ind(2), ind(3)]
4509 : ELSE IF (stack_dim == 2) THEN
4510 : dest = [ind(1), (idx - offset - 1)*nblks + ind(2), ind(3)]
4511 : ELSE
4512 : dest = [ind(1), ind(2), (idx - offset - 1)*nblks + ind(3)]
4513 : END IF
4514 :
4515 : CALL dbt_put_block(t_3c_stack, dest, SHAPE(blk), blk)
4516 : DEALLOCATE (blk)
4517 : END DO
4518 : CALL dbt_iterator_stop(iter)
4519 : !$OMP END PARALLEL
4520 : END DO !i_img
4521 22326 : CALL dbt_finalize(t_3c_stack)
4522 :
4523 44652 : END SUBROUTINE fill_3c_stack
4524 :
4525 : ! **************************************************************************************************
4526 : !> \brief Fill the stack of 2c tensors based on the content of images input
4527 : !> \param t_2c_stack ...
4528 : !> \param t_2c_in ...
4529 : !> \param images ...
4530 : !> \param stack_dim ...
4531 : !> \param ri_data ...
4532 : !> \param img_bounds ...
4533 : !> \param shift ...
4534 : ! **************************************************************************************************
4535 15228 : SUBROUTINE fill_2c_stack(t_2c_stack, t_2c_in, images, stack_dim, ri_data, img_bounds, shift)
4536 : TYPE(dbt_type), INTENT(INOUT) :: t_2c_stack
4537 : TYPE(dbt_type), DIMENSION(:), INTENT(INOUT) :: t_2c_in
4538 : INTEGER, DIMENSION(:), INTENT(INOUT) :: images
4539 : INTEGER, INTENT(IN) :: stack_dim
4540 : TYPE(hfx_ri_type), INTENT(INOUT) :: ri_data
4541 : INTEGER, INTENT(IN), OPTIONAL :: img_bounds(2), shift
4542 :
4543 : INTEGER :: dest(2), i_img, idx, ind(2), lb, &
4544 : my_shift, nblks, nimg, offset, ub
4545 : LOGICAL :: found
4546 15228 : REAL(dp), ALLOCATABLE, DIMENSION(:, :) :: blk
4547 : TYPE(dbt_iterator_type) :: iter
4548 :
4549 : !We loop over the a images from the ac_pairs, then copy the 3c ints to the correct spot in
4550 : !in the stack tensor (corresponding to pair index). Distributions match by construction
4551 15228 : nimg = ri_data%nimg
4552 15228 : nblks = SIZE(ri_data%bsizes_AO_split)
4553 :
4554 15228 : lb = 1
4555 15228 : ub = nimg
4556 15228 : offset = 0
4557 15228 : IF (PRESENT(img_bounds)) THEN
4558 15228 : lb = img_bounds(1)
4559 15228 : ub = img_bounds(2) - 1
4560 15228 : offset = lb - 1
4561 : END IF
4562 :
4563 15228 : my_shift = 1
4564 15228 : IF (PRESENT(shift)) my_shift = shift
4565 :
4566 202454 : DO idx = lb, ub
4567 187226 : i_img = images(idx)
4568 187226 : IF (i_img == 0 .OR. i_img > nimg) CYCLE
4569 :
4570 : !$OMP PARALLEL DEFAULT(NONE) SHARED(idx,i_img,t_2c_in,t_2c_stack,nblks,stack_dim,offset,my_shift) &
4571 202454 : !$OMP PRIVATE(iter,ind,blk,found,dest)
4572 : CALL dbt_iterator_start(iter, t_2c_in(i_img))
4573 : DO WHILE (dbt_iterator_blocks_left(iter))
4574 : CALL dbt_iterator_next_block(iter, ind)
4575 : CALL dbt_get_block(t_2c_in(i_img), ind, blk, found)
4576 : IF (.NOT. found) CYCLE
4577 :
4578 : IF (stack_dim == 1) THEN
4579 : dest = [(idx - offset - 1)*nblks + ind(1), (my_shift - 1)*nblks + ind(2)]
4580 : ELSE
4581 : dest = [(my_shift - 1)*nblks + ind(1), (idx - offset - 1)*nblks + ind(2)]
4582 : END IF
4583 :
4584 : CALL dbt_put_block(t_2c_stack, dest, SHAPE(blk), blk)
4585 : DEALLOCATE (blk)
4586 : END DO
4587 : CALL dbt_iterator_stop(iter)
4588 : !$OMP END PARALLEL
4589 : END DO !idx
4590 15228 : CALL dbt_finalize(t_2c_stack)
4591 :
4592 30456 : END SUBROUTINE fill_2c_stack
4593 :
4594 : ! **************************************************************************************************
4595 : !> \brief Unstacks a stacked 3c tensor containing t_3c_apc
4596 : !> \param t_3c_apc ...
4597 : !> \param t_stacked ...
4598 : !> \param idx ...
4599 : ! **************************************************************************************************
4600 17830 : SUBROUTINE unstack_t_3c_apc(t_3c_apc, t_stacked, idx)
4601 : TYPE(dbt_type), INTENT(INOUT) :: t_3c_apc, t_stacked
4602 : INTEGER, INTENT(IN) :: idx
4603 :
4604 : INTEGER :: current_idx
4605 : INTEGER, DIMENSION(3) :: ind, nblks_3c
4606 : LOGICAL :: found
4607 17830 : REAL(dp), ALLOCATABLE, DIMENSION(:, :, :) :: blk
4608 : TYPE(dbt_iterator_type) :: iter
4609 :
4610 : !Note: t_3c_apc and t_stacked must have the same ditribution
4611 17830 : CALL dbt_get_info(t_3c_apc, nblks_total=nblks_3c)
4612 :
4613 17830 : !$OMP PARALLEL DEFAULT(NONE) SHARED(t_3c_apc,t_stacked,idx,nblks_3c) PRIVATE(iter,ind,blk,found,current_idx)
4614 : CALL dbt_iterator_start(iter, t_stacked)
4615 : DO WHILE (dbt_iterator_blocks_left(iter))
4616 : CALL dbt_iterator_next_block(iter, ind)
4617 :
4618 : !tensor is stacked along the 3rd dimension
4619 : current_idx = (ind(3) - 1)/nblks_3c(3) + 1
4620 : IF (.NOT. idx == current_idx) CYCLE
4621 :
4622 : CALL dbt_get_block(t_stacked, ind, blk, found)
4623 : IF (.NOT. found) CYCLE
4624 :
4625 : CALL dbt_put_block(t_3c_apc, [ind(1), ind(2), ind(3) - (idx - 1)*nblks_3c(3)], SHAPE(blk), blk)
4626 : DEALLOCATE (blk)
4627 : END DO
4628 : CALL dbt_iterator_stop(iter)
4629 : !$OMP END PARALLEL
4630 :
4631 17830 : END SUBROUTINE unstack_t_3c_apc
4632 :
4633 : ! **************************************************************************************************
4634 : !> \brief copies the 3c integrals correspoinding to a single atom mu from the general (P^0| mu^0 sigam^a)
4635 : !> \param t_3c_at ...
4636 : !> \param t_3c_ints ...
4637 : !> \param iatom ...
4638 : !> \param dim_at ...
4639 : !> \param idx_to_at ...
4640 : ! **************************************************************************************************
4641 0 : SUBROUTINE get_atom_3c_ints(t_3c_at, t_3c_ints, iatom, dim_at, idx_to_at)
4642 : TYPE(dbt_type), INTENT(INOUT) :: t_3c_at, t_3c_ints
4643 : INTEGER, INTENT(IN) :: iatom, dim_at
4644 : INTEGER, DIMENSION(:), INTENT(IN) :: idx_to_at
4645 :
4646 : INTEGER, DIMENSION(3) :: ind
4647 : LOGICAL :: found
4648 0 : REAL(dp), ALLOCATABLE, DIMENSION(:, :, :) :: blk
4649 : TYPE(dbt_iterator_type) :: iter
4650 :
4651 0 : !$OMP PARALLEL DEFAULT(NONE) SHARED(t_3c_ints,t_3c_at,iatom,idx_to_at,dim_at) PRIVATE(iter,ind,blk,found)
4652 : CALL dbt_iterator_start(iter, t_3c_ints)
4653 : DO WHILE (dbt_iterator_blocks_left(iter))
4654 : CALL dbt_iterator_next_block(iter, ind)
4655 : IF (.NOT. idx_to_at(ind(dim_at)) == iatom) CYCLE
4656 :
4657 : CALL dbt_get_block(t_3c_ints, ind, blk, found)
4658 : IF (.NOT. found) CYCLE
4659 :
4660 : CALL dbt_put_block(t_3c_at, ind, SHAPE(blk), blk)
4661 : DEALLOCATE (blk)
4662 : END DO
4663 : CALL dbt_iterator_stop(iter)
4664 : !$OMP END PARALLEL
4665 0 : CALL dbt_finalize(t_3c_at)
4666 :
4667 0 : END SUBROUTINE get_atom_3c_ints
4668 :
4669 : ! **************************************************************************************************
4670 : !> \brief Precalculate the 3c and 2c derivatives tensors
4671 : !> \param t_3c_der_RI ...
4672 : !> \param t_3c_der_AO ...
4673 : !> \param mat_der_pot ...
4674 : !> \param t_2c_der_metric ...
4675 : !> \param ri_data ...
4676 : !> \param qs_env ...
4677 : ! **************************************************************************************************
4678 42 : SUBROUTINE precalc_derivatives(t_3c_der_RI, t_3c_der_AO, mat_der_pot, t_2c_der_metric, ri_data, qs_env)
4679 : TYPE(dbt_type), DIMENSION(:, :), INTENT(INOUT) :: t_3c_der_RI, t_3c_der_AO
4680 : TYPE(dbcsr_type), DIMENSION(:, :), INTENT(INOUT) :: mat_der_pot
4681 : TYPE(dbt_type), DIMENSION(:, :), INTENT(INOUT) :: t_2c_der_metric
4682 : TYPE(hfx_ri_type), INTENT(INOUT) :: ri_data
4683 : TYPE(qs_environment_type), POINTER :: qs_env
4684 :
4685 : CHARACTER(LEN=*), PARAMETER :: routineN = 'precalc_derivatives'
4686 :
4687 : INTEGER :: handle, handle2, i_img, i_mem, i_RI, &
4688 : i_xyz, iatom, n_mem, natom, nblks_RI, &
4689 : ncell_RI, nimg, nkind, nthreads
4690 : INTEGER(int_8) :: nze
4691 42 : INTEGER, ALLOCATABLE, DIMENSION(:) :: bsizes_RI_ext, bsizes_RI_ext_split, dist_AO_1, &
4692 84 : dist_AO_2, dist_RI, dist_RI_ext, dummy_end, dummy_start, end_blocks, start_blocks
4693 : INTEGER, DIMENSION(3) :: pcoord, pdims
4694 84 : INTEGER, DIMENSION(:), POINTER :: col_bsize, row_bsize
4695 : REAL(dp) :: occ
4696 : TYPE(dbcsr_distribution_type) :: dbcsr_dist
4697 : TYPE(dbcsr_type) :: dbcsr_template
4698 42 : TYPE(dbcsr_type), ALLOCATABLE, DIMENSION(:, :) :: mat_der_metric
4699 378 : TYPE(dbt_distribution_type) :: t_dist
4700 126 : TYPE(dbt_pgrid_type) :: pgrid
4701 378 : TYPE(dbt_type) :: t_3c_template
4702 42 : TYPE(dbt_type), ALLOCATABLE, DIMENSION(:, :, :) :: t_3c_der_AO_prv, t_3c_der_RI_prv
4703 : TYPE(dft_control_type), POINTER :: dft_control
4704 : TYPE(distribution_2d_type), POINTER :: dist_2d
4705 : TYPE(distribution_3d_type) :: dist_3d
4706 : TYPE(gto_basis_set_p_type), ALLOCATABLE, &
4707 42 : DIMENSION(:), TARGET :: basis_set_AO, basis_set_RI
4708 42 : TYPE(mp_cart_type) :: mp_comm_t3c
4709 : TYPE(mp_para_env_type), POINTER :: para_env
4710 : TYPE(neighbor_list_3c_type) :: nl_3c
4711 : TYPE(neighbor_list_set_p_type), DIMENSION(:), &
4712 42 : POINTER :: nl_2c
4713 42 : TYPE(particle_type), DIMENSION(:), POINTER :: particle_set
4714 42 : TYPE(qs_kind_type), DIMENSION(:), POINTER :: qs_kind_set
4715 :
4716 42 : NULLIFY (qs_kind_set, dist_2d, nl_2c, particle_set, dft_control, para_env, row_bsize, col_bsize)
4717 :
4718 42 : CALL timeset(routineN, handle)
4719 :
4720 : CALL get_qs_env(qs_env, nkind=nkind, qs_kind_set=qs_kind_set, distribution_2d=dist_2d, natom=natom, &
4721 42 : particle_set=particle_set, dft_control=dft_control, para_env=para_env)
4722 :
4723 42 : nimg = ri_data%nimg
4724 42 : ncell_RI = ri_data%ncell_RI
4725 :
4726 300 : ALLOCATE (basis_set_RI(nkind), basis_set_AO(nkind))
4727 42 : CALL basis_set_list_setup(basis_set_RI, ri_data%ri_basis_type, qs_kind_set)
4728 42 : CALL get_particle_set(particle_set, qs_kind_set, basis=basis_set_RI)
4729 42 : CALL basis_set_list_setup(basis_set_AO, ri_data%orb_basis_type, qs_kind_set)
4730 42 : CALL get_particle_set(particle_set, qs_kind_set, basis=basis_set_AO)
4731 :
4732 : !Dealing with the 3c derivatives
4733 42 : nthreads = 1
4734 42 : !$ nthreads = omp_get_num_threads()
4735 42 : pdims = 0
4736 168 : CALL dbt_pgrid_create(para_env, pdims, pgrid, tensor_dims=[MAX(1, natom/(ri_data%n_mem*nthreads)), natom, natom])
4737 :
4738 : CALL create_3c_tensor(t_3c_template, dist_AO_1, dist_AO_2, dist_RI, pgrid, &
4739 : ri_data%bsizes_AO, ri_data%bsizes_AO, ri_data%bsizes_RI, &
4740 42 : map1=[1, 2], map2=[3], name="tmp")
4741 42 : CALL dbt_destroy(t_3c_template)
4742 :
4743 : !We stack the RI basis images. Keep consistent distribution
4744 42 : nblks_RI = SIZE(ri_data%bsizes_RI_split)
4745 126 : ALLOCATE (dist_RI_ext(natom*ncell_RI))
4746 84 : ALLOCATE (bsizes_RI_ext(natom*ncell_RI))
4747 126 : ALLOCATE (bsizes_RI_ext_split(nblks_RI*ncell_RI))
4748 294 : DO i_RI = 1, ncell_RI
4749 756 : bsizes_RI_ext((i_RI - 1)*natom + 1:i_RI*natom) = ri_data%bsizes_RI(:)
4750 756 : dist_RI_ext((i_RI - 1)*natom + 1:i_RI*natom) = dist_RI(:)
4751 1334 : bsizes_RI_ext_split((i_RI - 1)*nblks_RI + 1:i_RI*nblks_RI) = ri_data%bsizes_RI_split(:)
4752 : END DO
4753 :
4754 42 : CALL dbt_distribution_new(t_dist, pgrid, dist_AO_1, dist_AO_2, dist_RI_ext)
4755 : CALL dbt_create(t_3c_template, "KP_3c_der", t_dist, [1, 2], [3], &
4756 42 : ri_data%bsizes_AO, ri_data%bsizes_AO, bsizes_RI_ext)
4757 42 : CALL dbt_distribution_destroy(t_dist)
4758 :
4759 7404 : ALLOCATE (t_3c_der_RI_prv(nimg, 1, 3), t_3c_der_AO_prv(nimg, 1, 3))
4760 168 : DO i_xyz = 1, 3
4761 3198 : DO i_img = 1, nimg
4762 3030 : CALL dbt_create(t_3c_template, t_3c_der_RI_prv(i_img, 1, i_xyz))
4763 3156 : CALL dbt_create(t_3c_template, t_3c_der_AO_prv(i_img, 1, i_xyz))
4764 : END DO
4765 : END DO
4766 42 : CALL dbt_destroy(t_3c_template)
4767 :
4768 42 : CALL dbt_mp_environ_pgrid(pgrid, pdims, pcoord)
4769 42 : CALL mp_comm_t3c%create(pgrid%mp_comm_2d, 3, pdims)
4770 : CALL distribution_3d_create(dist_3d, dist_AO_1, dist_AO_2, dist_RI, &
4771 42 : nkind, particle_set, mp_comm_t3c, own_comm=.TRUE.)
4772 42 : DEALLOCATE (dist_RI, dist_AO_1, dist_AO_2)
4773 42 : CALL dbt_pgrid_destroy(pgrid)
4774 :
4775 : CALL build_3c_neighbor_lists(nl_3c, basis_set_AO, basis_set_AO, basis_set_RI, dist_3d, ri_data%ri_metric, &
4776 42 : "HFX_3c_nl", qs_env, op_pos=2, sym_jk=.FALSE., own_dist=.TRUE.)
4777 :
4778 42 : n_mem = ri_data%n_mem
4779 : CALL create_tensor_batches(ri_data%bsizes_RI, n_mem, dummy_start, dummy_end, &
4780 : start_blocks, end_blocks)
4781 42 : DEALLOCATE (dummy_start, dummy_end)
4782 :
4783 : CALL create_3c_tensor(t_3c_template, dist_RI, dist_AO_1, dist_AO_2, ri_data%pgrid_2, &
4784 : bsizes_RI_ext_split, ri_data%bsizes_AO_split, ri_data%bsizes_AO_split, &
4785 42 : map1=[1], map2=[2, 3], name="der (RI | AO AO)")
4786 168 : DO i_xyz = 1, 3
4787 3198 : DO i_img = 1, nimg
4788 3030 : CALL dbt_create(t_3c_template, t_3c_der_RI(i_img, i_xyz))
4789 3156 : CALL dbt_create(t_3c_template, t_3c_der_AO(i_img, i_xyz))
4790 : END DO
4791 : END DO
4792 :
4793 116 : DO i_mem = 1, n_mem
4794 : CALL build_3c_derivatives(t_3c_der_AO_prv, t_3c_der_RI_prv, ri_data%filter_eps, qs_env, &
4795 : nl_3c, basis_set_AO, basis_set_AO, basis_set_RI, &
4796 : ri_data%ri_metric, der_eps=ri_data%eps_schwarz_forces, op_pos=2, &
4797 : do_kpoints=.TRUE., do_hfx_kpoints=.TRUE., &
4798 : bounds_k=[start_blocks(i_mem), end_blocks(i_mem)], &
4799 222 : RI_range=ri_data%kp_RI_range, img_to_RI_cell=ri_data%img_to_RI_cell)
4800 :
4801 74 : CALL timeset(routineN//"_cpy", handle2)
4802 : !We go from (mu^0 sigma^i | P^j) to (P^i| sigma^j mu^0) and finally to (P^i| mu^0 sigma^j)
4803 1994 : DO i_img = 1, nimg
4804 7754 : DO i_xyz = 1, 3
4805 : !derivative wrt to mu^0
4806 5760 : CALL get_tensor_occupancy(t_3c_der_AO_prv(i_img, 1, i_xyz), nze, occ)
4807 5760 : IF (nze > 0) THEN
4808 : CALL dbt_copy(t_3c_der_AO_prv(i_img, 1, i_xyz), t_3c_template, &
4809 3434 : order=[3, 2, 1], move_data=.TRUE.)
4810 3434 : CALL dbt_filter(t_3c_template, ri_data%filter_eps)
4811 : CALL dbt_copy(t_3c_template, t_3c_der_AO(i_img, i_xyz), &
4812 3434 : order=[1, 3, 2], move_data=.TRUE., summation=.TRUE.)
4813 : END IF
4814 :
4815 : !derivative wrt to P^i
4816 5760 : CALL get_tensor_occupancy(t_3c_der_RI_prv(i_img, 1, i_xyz), nze, occ)
4817 13440 : IF (nze > 0) THEN
4818 : CALL dbt_copy(t_3c_der_RI_prv(i_img, 1, i_xyz), t_3c_template, &
4819 3424 : order=[3, 2, 1], move_data=.TRUE.)
4820 3424 : CALL dbt_filter(t_3c_template, ri_data%filter_eps)
4821 : CALL dbt_copy(t_3c_template, t_3c_der_RI(i_img, i_xyz), &
4822 3424 : order=[1, 3, 2], move_data=.TRUE., summation=.TRUE.)
4823 : END IF
4824 : END DO
4825 : END DO
4826 190 : CALL timestop(handle2)
4827 : END DO
4828 42 : CALL dbt_destroy(t_3c_template)
4829 :
4830 42 : CALL neighbor_list_3c_destroy(nl_3c)
4831 168 : DO i_xyz = 1, 3
4832 3198 : DO i_img = 1, nimg
4833 3030 : CALL dbt_destroy(t_3c_der_RI_prv(i_img, 1, i_xyz))
4834 3156 : CALL dbt_destroy(t_3c_der_AO_prv(i_img, 1, i_xyz))
4835 : END DO
4836 : END DO
4837 6102 : DEALLOCATE (t_3c_der_RI_prv, t_3c_der_AO_prv)
4838 :
4839 : !Reorder 3c derivatives to be consistant with ints
4840 42 : CALL reorder_3c_derivs(t_3c_der_RI, ri_data)
4841 42 : CALL reorder_3c_derivs(t_3c_der_AO, ri_data)
4842 :
4843 42 : CALL timeset(routineN//"_2c", handle2)
4844 : !The 2-center derivatives
4845 42 : CALL cp_dbcsr_dist2d_to_dist(dist_2d, dbcsr_dist)
4846 126 : ALLOCATE (row_bsize(SIZE(ri_data%bsizes_RI)))
4847 84 : ALLOCATE (col_bsize(SIZE(ri_data%bsizes_RI)))
4848 126 : row_bsize(:) = ri_data%bsizes_RI
4849 126 : col_bsize(:) = ri_data%bsizes_RI
4850 :
4851 : CALL dbcsr_create(dbcsr_template, "2c_der", dbcsr_dist, dbcsr_type_no_symmetry, &
4852 42 : row_bsize, col_bsize)
4853 42 : CALL dbcsr_distribution_release(dbcsr_dist)
4854 42 : DEALLOCATE (col_bsize, row_bsize)
4855 :
4856 3282 : ALLOCATE (mat_der_metric(nimg, 3))
4857 168 : DO i_xyz = 1, 3
4858 3198 : DO i_img = 1, nimg
4859 3030 : CALL dbcsr_create(mat_der_pot(i_img, i_xyz), template=dbcsr_template)
4860 3156 : CALL dbcsr_create(mat_der_metric(i_img, i_xyz), template=dbcsr_template)
4861 : END DO
4862 : END DO
4863 42 : CALL dbcsr_release(dbcsr_template)
4864 :
4865 : !HFX potential derivatives
4866 : CALL build_2c_neighbor_lists(nl_2c, basis_set_RI, basis_set_RI, ri_data%hfx_pot, &
4867 42 : "HFX_2c_nl_pot", qs_env, sym_ij=.FALSE., dist_2d=dist_2d)
4868 : CALL build_2c_derivatives(mat_der_pot, ri_data%filter_eps_2c, qs_env, nl_2c, &
4869 42 : basis_set_RI, basis_set_RI, ri_data%hfx_pot, do_kpoints=.TRUE.)
4870 42 : CALL release_neighbor_list_sets(nl_2c)
4871 :
4872 : !RI metric derivatives
4873 : CALL build_2c_neighbor_lists(nl_2c, basis_set_RI, basis_set_RI, ri_data%ri_metric, &
4874 42 : "HFX_2c_nl_pot", qs_env, sym_ij=.FALSE., dist_2d=dist_2d)
4875 : CALL build_2c_derivatives(mat_der_metric, ri_data%filter_eps_2c, qs_env, nl_2c, &
4876 42 : basis_set_RI, basis_set_RI, ri_data%ri_metric, do_kpoints=.TRUE.)
4877 42 : CALL release_neighbor_list_sets(nl_2c)
4878 :
4879 : !Get into extended RI basis and tensor format
4880 168 : DO i_xyz = 1, 3
4881 378 : DO iatom = 1, natom
4882 252 : CALL dbt_create(ri_data%t_2c_inv(1, 1), t_2c_der_metric(iatom, i_xyz))
4883 : CALL get_ext_2c_int(t_2c_der_metric(iatom, i_xyz), mat_der_metric(:, i_xyz), &
4884 378 : iatom, iatom, 1, ri_data, qs_env)
4885 : END DO
4886 3198 : DO i_img = 1, nimg
4887 3156 : CALL dbcsr_release(mat_der_metric(i_img, i_xyz))
4888 : END DO
4889 : END DO
4890 42 : CALL timestop(handle2)
4891 :
4892 42 : CALL timestop(handle)
4893 :
4894 252 : END SUBROUTINE precalc_derivatives
4895 :
4896 : ! **************************************************************************************************
4897 : !> \brief Update the forces due to the derivative of the a 2-center product d/dR (Q|R)
4898 : !> \param force ...
4899 : !> \param t_2c_contr A precontracted tensor containing sum_abcdPS (ab|P)(P|Q)^-1 (R|S)^-1 (S|cd) P_ac P_bd
4900 : !> \param t_2c_der the d/dR (Q|R) tensor, in all 3 cartesian directions
4901 : !> \param atom_of_kind ...
4902 : !> \param kind_of ...
4903 : !> \param img in which periodic image the second center of the tensor is
4904 : !> \param pref ...
4905 : !> \param ri_data ...
4906 : !> \param qs_env ...
4907 : !> \param work_virial ...
4908 : !> \param cell ...
4909 : !> \param particle_set ...
4910 : !> \param diag ...
4911 : !> \param offdiag ...
4912 : !> \note IMPORTANT: t_tc_contr and t_2c_der need to have the same distribution. Atomic block sizes are
4913 : !> assumed
4914 : ! **************************************************************************************************
4915 2895 : SUBROUTINE get_2c_der_force(force, t_2c_contr, t_2c_der, atom_of_kind, kind_of, img, pref, &
4916 : ri_data, qs_env, work_virial, cell, particle_set, diag, offdiag)
4917 :
4918 : TYPE(qs_force_type), DIMENSION(:), POINTER :: force
4919 : TYPE(dbt_type), INTENT(INOUT) :: t_2c_contr
4920 : TYPE(dbt_type), DIMENSION(:), INTENT(INOUT) :: t_2c_der
4921 : INTEGER, DIMENSION(:), INTENT(IN) :: atom_of_kind, kind_of
4922 : INTEGER, INTENT(IN) :: img
4923 : REAL(dp), INTENT(IN) :: pref
4924 : TYPE(hfx_ri_type), INTENT(INOUT) :: ri_data
4925 : TYPE(qs_environment_type), POINTER :: qs_env
4926 : REAL(dp), DIMENSION(3, 3), INTENT(INOUT), OPTIONAL :: work_virial
4927 : TYPE(cell_type), OPTIONAL, POINTER :: cell
4928 : TYPE(particle_type), DIMENSION(:), OPTIONAL, &
4929 : POINTER :: particle_set
4930 : LOGICAL, INTENT(IN), OPTIONAL :: diag, offdiag
4931 :
4932 : CHARACTER(LEN=*), PARAMETER :: routineN = 'get_2c_der_force'
4933 :
4934 : INTEGER :: handle, i_img, i_RI, i_xyz, iat, &
4935 : iat_of_kind, ikind, j_img, j_RI, &
4936 : j_xyz, jat, jat_of_kind, jkind, natom
4937 : INTEGER, DIMENSION(2) :: ind
4938 2895 : INTEGER, DIMENSION(:, :), POINTER :: index_to_cell
4939 : LOGICAL :: found, my_diag, my_offdiag, use_virial
4940 : REAL(dp) :: new_force
4941 2895 : REAL(dp), ALLOCATABLE, DIMENSION(:, :), TARGET :: contr_blk, der_blk
4942 : REAL(dp), DIMENSION(3) :: scoord
4943 : TYPE(dbt_iterator_type) :: iter
4944 : TYPE(kpoint_type), POINTER :: kpoints
4945 :
4946 2895 : NULLIFY (kpoints, index_to_cell)
4947 :
4948 : !Loop over the blocks of d/dR (Q|R), contract with the corresponding block of t_2c_contr and
4949 : !update the relevant force
4950 :
4951 2895 : CALL timeset(routineN, handle)
4952 :
4953 2895 : use_virial = .FALSE.
4954 2895 : IF (PRESENT(work_virial) .AND. PRESENT(cell) .AND. PRESENT(particle_set)) use_virial = .TRUE.
4955 :
4956 2895 : my_diag = .FALSE.
4957 2895 : IF (PRESENT(diag)) my_diag = diag
4958 :
4959 2316 : my_offdiag = .FALSE.
4960 2316 : IF (PRESENT(diag)) my_offdiag = offdiag
4961 :
4962 2895 : CALL get_qs_env(qs_env, kpoints=kpoints, natom=natom)
4963 2895 : CALL get_kpoint_info(kpoints, index_to_cell=index_to_cell)
4964 :
4965 : !$OMP PARALLEL DEFAULT(NONE) &
4966 : !$OMP SHARED(t_2c_der,t_2c_contr,work_virial,force,use_virial,natom,index_to_cell,ri_data,img) &
4967 : !$OMP SHARED(pref,atom_of_kind,kind_of,particle_set,cell,my_diag,my_offdiag) &
4968 : !$OMP PRIVATE(i_xyz,j_xyz,iter,ind,der_blk,contr_blk,found,new_force,i_RI,i_img,j_RI,j_img) &
4969 2895 : !$OMP PRIVATE(iat,jat,iat_of_kind,jat_of_kind,ikind,jkind,scoord)
4970 : DO i_xyz = 1, 3
4971 : CALL dbt_iterator_start(iter, t_2c_der(i_xyz))
4972 : DO WHILE (dbt_iterator_blocks_left(iter))
4973 : CALL dbt_iterator_next_block(iter, ind)
4974 :
4975 : !Only take forecs due to block diagonal or block off-diagonal, depending on arguments
4976 : IF ((my_diag .AND. .NOT. my_offdiag) .OR. (.NOT. my_diag .AND. my_offdiag)) THEN
4977 : IF (my_diag .AND. (ind(1) .NE. ind(2))) CYCLE
4978 : IF (my_offdiag .AND. (ind(1) == ind(2))) CYCLE
4979 : END IF
4980 :
4981 : CALL dbt_get_block(t_2c_der(i_xyz), ind, der_blk, found)
4982 : CPASSERT(found)
4983 : CALL dbt_get_block(t_2c_contr, ind, contr_blk, found)
4984 :
4985 : IF (found) THEN
4986 :
4987 : !an element of d/dR (Q|R) corresponds to 2 things because of translational invariance
4988 : !(Q'| R) = - (Q| R'), once wrt the center on Q, and once on R
4989 : new_force = pref*SUM(der_blk(:, :)*contr_blk(:, :))
4990 :
4991 : i_RI = (ind(1) - 1)/natom + 1
4992 : i_img = ri_data%RI_cell_to_img(i_RI)
4993 : iat = ind(1) - (i_RI - 1)*natom
4994 : iat_of_kind = atom_of_kind(iat)
4995 : ikind = kind_of(iat)
4996 :
4997 : j_RI = (ind(2) - 1)/natom + 1
4998 : j_img = ri_data%RI_cell_to_img(j_RI)
4999 : jat = ind(2) - (j_RI - 1)*natom
5000 : jat_of_kind = atom_of_kind(jat)
5001 : jkind = kind_of(jat)
5002 :
5003 : !Force on iatom (first center)
5004 : !$OMP ATOMIC
5005 : force(ikind)%fock_4c(i_xyz, iat_of_kind) = force(ikind)%fock_4c(i_xyz, iat_of_kind) &
5006 : + new_force
5007 :
5008 : IF (use_virial) THEN
5009 :
5010 : CALL real_to_scaled(scoord, pbc(particle_set(iat)%r, cell), cell)
5011 : scoord(:) = scoord(:) + REAL(index_to_cell(:, i_img), dp)
5012 :
5013 : DO j_xyz = 1, 3
5014 : !$OMP ATOMIC
5015 : work_virial(i_xyz, j_xyz) = work_virial(i_xyz, j_xyz) + new_force*scoord(j_xyz)
5016 : END DO
5017 : END IF
5018 :
5019 : !Force on jatom (second center)
5020 : !$OMP ATOMIC
5021 : force(jkind)%fock_4c(i_xyz, jat_of_kind) = force(jkind)%fock_4c(i_xyz, jat_of_kind) &
5022 : - new_force
5023 :
5024 : IF (use_virial) THEN
5025 :
5026 : CALL real_to_scaled(scoord, pbc(particle_set(jat)%r, cell), cell)
5027 : scoord(:) = scoord(:) + REAL(index_to_cell(:, j_img) + index_to_cell(:, img), dp)
5028 :
5029 : DO j_xyz = 1, 3
5030 : !$OMP ATOMIC
5031 : work_virial(i_xyz, j_xyz) = work_virial(i_xyz, j_xyz) - new_force*scoord(j_xyz)
5032 : END DO
5033 : END IF
5034 :
5035 : DEALLOCATE (contr_blk)
5036 : END IF
5037 :
5038 : DEALLOCATE (der_blk)
5039 : END DO !iter
5040 : CALL dbt_iterator_stop(iter)
5041 :
5042 : END DO !i_xyz
5043 : !$OMP END PARALLEL
5044 2895 : CALL timestop(handle)
5045 :
5046 5790 : END SUBROUTINE get_2c_der_force
5047 :
5048 : ! **************************************************************************************************
5049 : !> \brief This routines calculates the force contribution from a trace over 3D tensors, i.e.
5050 : !> force = sum_ijk A_ijk B_ijk., the B tensor is (P^0| sigma^0 lambda^img), with P in the
5051 : !> extended RI basis. Note that all tensors are stacked along the 3rd dimension
5052 : !> \param force ...
5053 : !> \param t_3c_contr ...
5054 : !> \param t_3c_der_1 ...
5055 : !> \param t_3c_der_2 ...
5056 : !> \param atom_of_kind ...
5057 : !> \param kind_of ...
5058 : !> \param idx_to_at_RI ...
5059 : !> \param idx_to_at_AO ...
5060 : !> \param i_images ...
5061 : !> \param lb_img ...
5062 : !> \param pref ...
5063 : !> \param ri_data ...
5064 : !> \param qs_env ...
5065 : !> \param work_virial ...
5066 : !> \param cell ...
5067 : !> \param particle_set ...
5068 : ! **************************************************************************************************
5069 1666 : SUBROUTINE get_force_from_3c_trace(force, t_3c_contr, t_3c_der_1, t_3c_der_2, atom_of_kind, kind_of, &
5070 3332 : idx_to_at_RI, idx_to_at_AO, i_images, lb_img, pref, &
5071 : ri_data, qs_env, work_virial, cell, particle_set)
5072 :
5073 : TYPE(qs_force_type), DIMENSION(:), POINTER :: force
5074 : TYPE(dbt_type), INTENT(INOUT) :: t_3c_contr
5075 : TYPE(dbt_type), DIMENSION(3), INTENT(INOUT) :: t_3c_der_1, t_3c_der_2
5076 : INTEGER, DIMENSION(:), INTENT(IN) :: atom_of_kind, kind_of, idx_to_at_RI, &
5077 : idx_to_at_AO, i_images
5078 : INTEGER, INTENT(IN) :: lb_img
5079 : REAL(dp), INTENT(IN) :: pref
5080 : TYPE(hfx_ri_type), INTENT(INOUT) :: ri_data
5081 : TYPE(qs_environment_type), POINTER :: qs_env
5082 : REAL(dp), DIMENSION(3, 3), INTENT(INOUT), OPTIONAL :: work_virial
5083 : TYPE(cell_type), OPTIONAL, POINTER :: cell
5084 : TYPE(particle_type), DIMENSION(:), OPTIONAL, &
5085 : POINTER :: particle_set
5086 :
5087 : CHARACTER(LEN=*), PARAMETER :: routineN = 'get_force_from_3c_trace'
5088 :
5089 : INTEGER :: handle, i_RI, i_xyz, iat, iat_of_kind, idx, ikind, j_xyz, jat, jat_of_kind, &
5090 : jkind, kat, kat_of_kind, kkind, nblks_AO, nblks_RI, RI_img
5091 : INTEGER, DIMENSION(3) :: ind
5092 1666 : INTEGER, DIMENSION(:, :), POINTER :: index_to_cell
5093 : LOGICAL :: found, found_1, found_2, use_virial
5094 : REAL(dp) :: new_force
5095 1666 : REAL(dp), ALLOCATABLE, DIMENSION(:, :, :), TARGET :: contr_blk, der_blk_1, der_blk_2, &
5096 1666 : der_blk_3
5097 : REAL(dp), DIMENSION(3) :: scoord
5098 : TYPE(dbt_iterator_type) :: iter
5099 : TYPE(kpoint_type), POINTER :: kpoints
5100 :
5101 1666 : NULLIFY (kpoints, index_to_cell)
5102 :
5103 1666 : CALL timeset(routineN, handle)
5104 :
5105 1666 : CALL get_qs_env(qs_env, kpoints=kpoints)
5106 1666 : CALL get_kpoint_info(kpoints, index_to_cell=index_to_cell)
5107 :
5108 1666 : nblks_RI = SIZE(ri_data%bsizes_RI_split)
5109 1666 : nblks_AO = SIZE(ri_data%bsizes_AO_split)
5110 :
5111 1666 : use_virial = .FALSE.
5112 1666 : IF (PRESENT(work_virial) .AND. PRESENT(cell) .AND. PRESENT(particle_set)) use_virial = .TRUE.
5113 :
5114 : !$OMP PARALLEL DEFAULT(NONE) &
5115 : !$OMP SHARED(t_3c_der_1, t_3c_der_2,t_3c_contr,work_virial,force,use_virial,index_to_cell,i_images,lb_img) &
5116 : !$OMP SHARED(pref,idx_to_at_AO,atom_of_kind,kind_of,particle_set,cell,idx_to_at_RI,ri_data,nblks_RI,nblks_AO) &
5117 : !$OMP PRIVATE(i_xyz,j_xyz,iter,ind,der_blk_1,contr_blk,found,new_force,iat,iat_of_kind,ikind,scoord) &
5118 1666 : !$OMP PRIVATE(jat,kat,jat_of_kind,kat_of_kind,jkind,kkind,i_RI,RI_img,der_blk_2,der_blk_3,found_1,found_2,idx)
5119 : CALL dbt_iterator_start(iter, t_3c_contr)
5120 : DO WHILE (dbt_iterator_blocks_left(iter))
5121 : CALL dbt_iterator_next_block(iter, ind)
5122 :
5123 : CALL dbt_get_block(t_3c_contr, ind, contr_blk, found)
5124 : IF (found) THEN
5125 :
5126 : DO i_xyz = 1, 3
5127 : CALL dbt_get_block(t_3c_der_1(i_xyz), ind, der_blk_1, found_1)
5128 : IF (.NOT. found_1) THEN
5129 : DEALLOCATE (der_blk_1)
5130 : ALLOCATE (der_blk_1(SIZE(contr_blk, 1), SIZE(contr_blk, 2), SIZE(contr_blk, 3)))
5131 : der_blk_1(:, :, :) = 0.0_dp
5132 : END IF
5133 : CALL dbt_get_block(t_3c_der_2(i_xyz), ind, der_blk_2, found_2)
5134 : IF (.NOT. found_2) THEN
5135 : DEALLOCATE (der_blk_2)
5136 : ALLOCATE (der_blk_2(SIZE(contr_blk, 1), SIZE(contr_blk, 2), SIZE(contr_blk, 3)))
5137 : der_blk_2(:, :, :) = 0.0_dp
5138 : END IF
5139 :
5140 : ALLOCATE (der_blk_3(SIZE(contr_blk, 1), SIZE(contr_blk, 2), SIZE(contr_blk, 3)))
5141 : der_blk_3(:, :, :) = -(der_blk_1(:, :, :) + der_blk_2(:, :, :))
5142 :
5143 : !We assume the tensors are in the format (P^0| sigma^0 mu^a+c-b), with P a member of the
5144 : !extended RI basis set
5145 :
5146 : !Force for the first center (RI extended basis, zero cell)
5147 : new_force = pref*SUM(der_blk_1(:, :, :)*contr_blk(:, :, :))
5148 :
5149 : i_RI = (ind(1) - 1)/nblks_RI + 1
5150 : RI_img = ri_data%RI_cell_to_img(i_RI)
5151 : iat = idx_to_at_RI(ind(1) - (i_RI - 1)*nblks_RI)
5152 : iat_of_kind = atom_of_kind(iat)
5153 : ikind = kind_of(iat)
5154 :
5155 : !$OMP ATOMIC
5156 : force(ikind)%fock_4c(i_xyz, iat_of_kind) = force(ikind)%fock_4c(i_xyz, iat_of_kind) &
5157 : + new_force
5158 :
5159 : IF (use_virial) THEN
5160 :
5161 : CALL real_to_scaled(scoord, pbc(particle_set(iat)%r, cell), cell)
5162 : scoord(:) = scoord(:) + REAL(index_to_cell(:, RI_img), dp)
5163 :
5164 : DO j_xyz = 1, 3
5165 : !$OMP ATOMIC
5166 : work_virial(i_xyz, j_xyz) = work_virial(i_xyz, j_xyz) + new_force*scoord(j_xyz)
5167 : END DO
5168 : END IF
5169 :
5170 : !Force with respect to the second center (AO basis, zero cell)
5171 : new_force = pref*SUM(der_blk_2(:, :, :)*contr_blk(:, :, :))
5172 : jat = idx_to_at_AO(ind(2))
5173 : jat_of_kind = atom_of_kind(jat)
5174 : jkind = kind_of(jat)
5175 :
5176 : !$OMP ATOMIC
5177 : force(jkind)%fock_4c(i_xyz, jat_of_kind) = force(jkind)%fock_4c(i_xyz, jat_of_kind) &
5178 : + new_force
5179 :
5180 : IF (use_virial) THEN
5181 :
5182 : CALL real_to_scaled(scoord, pbc(particle_set(jat)%r, cell), cell)
5183 :
5184 : DO j_xyz = 1, 3
5185 : !$OMP ATOMIC
5186 : work_virial(i_xyz, j_xyz) = work_virial(i_xyz, j_xyz) + new_force*scoord(j_xyz)
5187 : END DO
5188 : END IF
5189 :
5190 : !Force with respect to the third center (AO basis, apc_img - b_img)
5191 : !Note: tensors are stacked along the 3rd direction
5192 : new_force = pref*SUM(der_blk_3(:, :, :)*contr_blk(:, :, :))
5193 : idx = (ind(3) - 1)/nblks_AO + 1
5194 : kat = idx_to_at_AO(ind(3) - (idx - 1)*nblks_AO)
5195 : kat_of_kind = atom_of_kind(kat)
5196 : kkind = kind_of(kat)
5197 :
5198 : !$OMP ATOMIC
5199 : force(kkind)%fock_4c(i_xyz, kat_of_kind) = force(kkind)%fock_4c(i_xyz, kat_of_kind) &
5200 : + new_force
5201 :
5202 : IF (use_virial) THEN
5203 : CALL real_to_scaled(scoord, pbc(particle_set(kat)%r, cell), cell)
5204 : scoord(:) = scoord(:) + REAL(index_to_cell(:, i_images(lb_img - 1 + idx)), dp)
5205 :
5206 : DO j_xyz = 1, 3
5207 : !$OMP ATOMIC
5208 : work_virial(i_xyz, j_xyz) = work_virial(i_xyz, j_xyz) + new_force*scoord(j_xyz)
5209 : END DO
5210 : END IF
5211 :
5212 : DEALLOCATE (der_blk_1, der_blk_2, der_blk_3)
5213 : END DO !i_xyz
5214 : DEALLOCATE (contr_blk)
5215 : END IF !found
5216 : END DO !iter
5217 : CALL dbt_iterator_stop(iter)
5218 : !$OMP END PARALLEL
5219 1666 : CALL timestop(handle)
5220 :
5221 3332 : END SUBROUTINE get_force_from_3c_trace
5222 :
5223 : END MODULE
|
