Line data Source code
1 : !--------------------------------------------------------------------------------------------------!
2 : ! CP2K: A general program to perform molecular dynamics simulations !
3 : ! Copyright 2000-2026 CP2K developers group <https://cp2k.org> !
4 : ! !
5 : ! SPDX-License-Identifier: GPL-2.0-or-later !
6 : !--------------------------------------------------------------------------------------------------!
7 :
8 : ! **************************************************************************************************
9 : !> \brief
10 : !> \author Jan Wilhelm
11 : !> \date 07.2023
12 : ! **************************************************************************************************
13 : MODULE gw_utils
14 : USE atomic_kind_types, ONLY: atomic_kind_type,&
15 : get_atomic_kind_set
16 : USE basis_set_types, ONLY: get_gto_basis_set,&
17 : gto_basis_set_type
18 : USE bibliography, ONLY: Graml2024,&
19 : cite_reference
20 : USE cell_types, ONLY: cell_type,&
21 : pbc,&
22 : scaled_to_real
23 : USE cp_blacs_env, ONLY: cp_blacs_env_create,&
24 : cp_blacs_env_release,&
25 : cp_blacs_env_type
26 : USE cp_cfm_types, ONLY: cp_cfm_create,&
27 : cp_cfm_release,&
28 : cp_cfm_to_cfm,&
29 : cp_cfm_to_fm,&
30 : cp_cfm_type
31 : USE cp_control_types, ONLY: dft_control_type
32 : USE cp_dbcsr_api, ONLY: &
33 : dbcsr_create, dbcsr_distribution_release, dbcsr_distribution_type, dbcsr_p_type, &
34 : dbcsr_release, dbcsr_set, dbcsr_type, dbcsr_type_no_symmetry, dbcsr_type_symmetric
35 : USE cp_dbcsr_operations, ONLY: copy_dbcsr_to_fm,&
36 : copy_fm_to_dbcsr,&
37 : cp_dbcsr_dist2d_to_dist,&
38 : dbcsr_allocate_matrix_set,&
39 : dbcsr_deallocate_matrix_set
40 : USE cp_files, ONLY: close_file,&
41 : open_file
42 : USE cp_fm_basic_linalg, ONLY: cp_fm_scale_and_add
43 : USE cp_fm_struct, ONLY: cp_fm_struct_create,&
44 : cp_fm_struct_release,&
45 : cp_fm_struct_type
46 : USE cp_fm_types, ONLY: cp_fm_create,&
47 : cp_fm_get_diag,&
48 : cp_fm_release,&
49 : cp_fm_set_all,&
50 : cp_fm_type
51 : USE cp_log_handling, ONLY: cp_get_default_logger,&
52 : cp_logger_type
53 : USE cp_output_handling, ONLY: cp_print_key_generate_filename
54 : USE dbt_api, ONLY: &
55 : dbt_clear, dbt_create, dbt_destroy, dbt_filter, dbt_iterator_blocks_left, &
56 : dbt_iterator_next_block, dbt_iterator_start, dbt_iterator_stop, dbt_iterator_type, &
57 : dbt_mp_environ_pgrid, dbt_pgrid_create, dbt_pgrid_destroy, dbt_pgrid_type, dbt_type
58 : USE distribution_2d_types, ONLY: distribution_2d_type
59 : USE gw_communication, ONLY: fm_to_local_array
60 : USE gw_integrals, ONLY: build_3c_integral_block
61 : USE input_constants, ONLY: do_potential_truncated,&
62 : large_cell_Gamma,&
63 : ri_rpa_g0w0_crossing_newton,&
64 : rtp_method_bse,&
65 : small_cell_full_kp,&
66 : xc_none
67 : USE input_section_types, ONLY: section_vals_get,&
68 : section_vals_get_subs_vals,&
69 : section_vals_type,&
70 : section_vals_val_get,&
71 : section_vals_val_set
72 : USE kinds, ONLY: default_path_length,&
73 : default_string_length,&
74 : dp,&
75 : int_8
76 : USE kpoint_k_r_trafo_simple, ONLY: rs_to_kp
77 : USE kpoint_types, ONLY: get_kpoint_info,&
78 : kpoint_create,&
79 : kpoint_type
80 : USE libint_2c_3c, ONLY: libint_potential_type
81 : USE libint_wrapper, ONLY: cp_libint_static_cleanup,&
82 : cp_libint_static_init
83 : USE machine, ONLY: m_memory,&
84 : m_walltime
85 : USE mathconstants, ONLY: gaussi,&
86 : z_one,&
87 : z_zero
88 : USE mathlib, ONLY: diag_complex,&
89 : gcd
90 : USE message_passing, ONLY: mp_cart_type,&
91 : mp_para_env_type
92 : USE minimax_exp, ONLY: get_exp_minimax_coeff
93 : USE minimax_exp_gw, ONLY: get_exp_minimax_coeff_gw
94 : USE minimax_rpa, ONLY: get_rpa_minimax_coeff,&
95 : get_rpa_minimax_coeff_larger_grid
96 : USE mp2_gpw, ONLY: create_mat_munu
97 : USE mp2_grids, ONLY: get_l_sq_wghts_cos_tf_t_to_w,&
98 : get_l_sq_wghts_cos_tf_w_to_t,&
99 : get_l_sq_wghts_sin_tf_t_to_w
100 : USE mp2_ri_2c, ONLY: trunc_coulomb_for_exchange
101 : USE parallel_gemm_api, ONLY: parallel_gemm
102 : USE particle_methods, ONLY: get_particle_set
103 : USE particle_types, ONLY: particle_type
104 : USE physcon, ONLY: angstrom,&
105 : evolt
106 : USE post_scf_bandstructure_types, ONLY: post_scf_bandstructure_type
107 : USE post_scf_bandstructure_utils, ONLY: rsmat_to_kp
108 : USE qs_energy_types, ONLY: qs_energy_type
109 : USE qs_environment_types, ONLY: get_qs_env,&
110 : qs_env_part_release,&
111 : qs_environment_type
112 : USE qs_integral_utils, ONLY: basis_set_list_setup
113 : USE qs_interactions, ONLY: init_interaction_radii_orb_basis
114 : USE qs_kind_types, ONLY: get_qs_kind,&
115 : qs_kind_type
116 : USE qs_ks_methods, ONLY: qs_ks_build_kohn_sham_matrix
117 : USE qs_neighbor_list_types, ONLY: neighbor_list_set_p_type,&
118 : release_neighbor_list_sets
119 : USE qs_tensors, ONLY: build_2c_integrals,&
120 : build_2c_neighbor_lists,&
121 : build_3c_integrals,&
122 : build_3c_neighbor_lists,&
123 : get_tensor_occupancy,&
124 : neighbor_list_3c_destroy
125 : USE qs_tensors_types, ONLY: create_2c_tensor,&
126 : create_3c_tensor,&
127 : distribution_3d_create,&
128 : distribution_3d_type,&
129 : neighbor_list_3c_type
130 : USE rpa_gw, ONLY: continuation_pade
131 : #include "base/base_uses.f90"
132 :
133 : IMPLICIT NONE
134 :
135 : PRIVATE
136 :
137 : PUBLIC :: create_and_init_bs_env_for_gw, de_init_bs_env, get_i_j_atoms, &
138 : compute_xkp, time_to_freq, analyt_conti_and_print, &
139 : add_R, is_cell_in_index_to_cell, get_V_tr_R, power
140 :
141 : CHARACTER(len=*), PARAMETER, PRIVATE :: moduleN = 'gw_utils'
142 :
143 : CONTAINS
144 :
145 : ! **************************************************************************************************
146 : !> \brief ...
147 : !> \param qs_env ...
148 : !> \param bs_env ...
149 : !> \param bs_sec ...
150 : ! **************************************************************************************************
151 28 : SUBROUTINE create_and_init_bs_env_for_gw(qs_env, bs_env, bs_sec)
152 : TYPE(qs_environment_type), POINTER :: qs_env
153 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
154 : TYPE(section_vals_type), POINTER :: bs_sec
155 :
156 : CHARACTER(LEN=*), PARAMETER :: routineN = 'create_and_init_bs_env_for_gw'
157 :
158 : INTEGER :: handle
159 :
160 28 : CALL timeset(routineN, handle)
161 :
162 28 : CALL cite_reference(Graml2024)
163 :
164 28 : CALL read_gw_input_parameters(bs_env, bs_sec)
165 :
166 28 : CALL print_header_and_input_parameters(bs_env)
167 :
168 28 : CALL setup_AO_and_RI_basis_set(qs_env, bs_env)
169 :
170 28 : CALL get_RI_basis_and_basis_function_indices(qs_env, bs_env)
171 :
172 28 : CALL set_heuristic_parameters(bs_env, qs_env)
173 :
174 28 : CALL cp_libint_static_init()
175 :
176 28 : CALL setup_kpoints_chi_eps_W(bs_env, bs_env%kpoints_chi_eps_W)
177 :
178 28 : IF (bs_env%small_cell_full_kp_or_large_cell_Gamma == small_cell_full_kp) THEN
179 6 : CALL setup_cells_3c(qs_env, bs_env)
180 : END IF
181 :
182 28 : CALL set_parallelization_parameters(qs_env, bs_env)
183 :
184 28 : CALL allocate_matrices(qs_env, bs_env)
185 :
186 28 : CALL compute_V_xc(qs_env, bs_env)
187 :
188 28 : CALL create_tensors(qs_env, bs_env)
189 :
190 50 : SELECT CASE (bs_env%small_cell_full_kp_or_large_cell_Gamma)
191 : CASE (large_cell_Gamma)
192 :
193 22 : CALL allocate_GW_eigenvalues(bs_env)
194 :
195 22 : CALL check_sparsity_3c(qs_env, bs_env)
196 :
197 22 : CALL set_sparsity_parallelization_parameters(bs_env)
198 :
199 22 : CALL check_for_restart_files(qs_env, bs_env)
200 :
201 : CASE (small_cell_full_kp)
202 :
203 6 : CALL compute_3c_integrals(qs_env, bs_env)
204 :
205 6 : CALL setup_cells_Delta_R(bs_env)
206 :
207 6 : CALL setup_parallelization_Delta_R(bs_env)
208 :
209 6 : CALL allocate_matrices_small_cell_full_kp(qs_env, bs_env)
210 :
211 6 : CALL trafo_V_xc_R_to_kp(qs_env, bs_env)
212 :
213 34 : CALL heuristic_RI_regularization(qs_env, bs_env)
214 :
215 : END SELECT
216 :
217 28 : CALL setup_time_and_frequency_minimax_grid(bs_env)
218 :
219 : ! free memory in qs_env; only if one is not calculating the LDOS because
220 : ! we need real-space grid operations in pw_env, task_list for the LDOS
221 : ! Recommendation in case of memory issues: first perform GW calculation without calculating
222 : ! LDOS (to safe memor). Then, use GW restart files
223 : ! in a subsequent calculation to calculate the LDOS
224 : ! Marek : TODO - boolean that does not interfere with RTP init but sets this to correct value
225 : IF (.NOT. bs_env%do_ldos .AND. .FALSE.) THEN
226 : CALL qs_env_part_release(qs_env)
227 : END IF
228 :
229 28 : CALL timestop(handle)
230 :
231 28 : END SUBROUTINE create_and_init_bs_env_for_gw
232 :
233 : ! **************************************************************************************************
234 : !> \brief ...
235 : !> \param bs_env ...
236 : ! **************************************************************************************************
237 28 : SUBROUTINE de_init_bs_env(bs_env)
238 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
239 :
240 : CHARACTER(LEN=*), PARAMETER :: routineN = 'de_init_bs_env'
241 :
242 : INTEGER :: handle
243 :
244 28 : CALL timeset(routineN, handle)
245 : ! deallocate quantities here which:
246 : ! 1. cannot be deallocated in bs_env_release due to circular dependencies
247 : ! 2. consume a lot of memory and should not be kept until the quantity is
248 : ! deallocated in bs_env_release
249 :
250 28 : IF (ASSOCIATED(bs_env%nl_3c%ij_list) .AND. (bs_env%rtp_method == rtp_method_bse)) THEN
251 12 : IF (bs_env%unit_nr > 0) WRITE (bs_env%unit_nr, *) "Retaining nl_3c for RTBSE"
252 : ELSE
253 16 : CALL neighbor_list_3c_destroy(bs_env%nl_3c)
254 : END IF
255 :
256 28 : CALL cp_libint_static_cleanup()
257 :
258 28 : CALL timestop(handle)
259 :
260 28 : END SUBROUTINE de_init_bs_env
261 :
262 : ! **************************************************************************************************
263 : !> \brief ...
264 : !> \param bs_env ...
265 : !> \param bs_sec ...
266 : ! **************************************************************************************************
267 28 : SUBROUTINE read_gw_input_parameters(bs_env, bs_sec)
268 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
269 : TYPE(section_vals_type), POINTER :: bs_sec
270 :
271 : CHARACTER(LEN=*), PARAMETER :: routineN = 'read_gw_input_parameters'
272 :
273 : INTEGER :: handle
274 : TYPE(section_vals_type), POINTER :: gw_sec
275 :
276 28 : CALL timeset(routineN, handle)
277 :
278 28 : NULLIFY (gw_sec)
279 28 : gw_sec => section_vals_get_subs_vals(bs_sec, "GW")
280 :
281 28 : CALL section_vals_val_get(gw_sec, "NUM_TIME_FREQ_POINTS", i_val=bs_env%num_time_freq_points)
282 28 : CALL section_vals_val_get(gw_sec, "EPS_FILTER", r_val=bs_env%eps_filter)
283 28 : CALL section_vals_val_get(gw_sec, "REGULARIZATION_RI", r_val=bs_env%input_regularization_RI)
284 28 : CALL section_vals_val_get(gw_sec, "REGULARIZATION_MINIMAX", r_val=bs_env%input_regularization_minimax)
285 28 : CALL section_vals_val_get(gw_sec, "CUTOFF_RADIUS_RI", r_val=bs_env%ri_metric%cutoff_radius)
286 28 : CALL section_vals_val_get(gw_sec, "MEMORY_PER_PROC", r_val=bs_env%input_memory_per_proc_GB)
287 28 : CALL section_vals_val_get(gw_sec, "APPROX_KP_EXTRAPOL", l_val=bs_env%approx_kp_extrapol)
288 28 : CALL section_vals_val_get(gw_sec, "SIZE_LATTICE_SUM", i_val=bs_env%size_lattice_sum_V)
289 28 : CALL section_vals_val_get(gw_sec, "KPOINTS_W", i_vals=bs_env%nkp_grid_chi_eps_W_input)
290 28 : CALL section_vals_val_get(gw_sec, "HEDIN_SHIFT", l_val=bs_env%do_hedin_shift)
291 28 : CALL section_vals_val_get(gw_sec, "FREQ_MAX_FIT", r_val=bs_env%freq_max_fit)
292 28 : CALL section_vals_val_get(gw_sec, "PRINT%PRINT_DBT_CONTRACT", l_val=bs_env%print_contract)
293 28 : CALL section_vals_val_get(gw_sec, "PRINT%PRINT_DBT_CONTRACT_VERBOSE", l_val=bs_env%print_contract_verbose)
294 :
295 28 : IF (bs_env%print_contract) THEN
296 0 : bs_env%unit_nr_contract = bs_env%unit_nr
297 : ELSE
298 28 : bs_env%unit_nr_contract = 0
299 : END IF
300 28 : CALL timestop(handle)
301 :
302 28 : END SUBROUTINE read_gw_input_parameters
303 :
304 : ! **************************************************************************************************
305 : !> \brief ...
306 : !> \param qs_env ...
307 : !> \param bs_env ...
308 : ! **************************************************************************************************
309 28 : SUBROUTINE setup_AO_and_RI_basis_set(qs_env, bs_env)
310 : TYPE(qs_environment_type), POINTER :: qs_env
311 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
312 :
313 : CHARACTER(LEN=*), PARAMETER :: routineN = 'setup_AO_and_RI_basis_set'
314 :
315 : INTEGER :: handle, natom, nkind
316 28 : TYPE(particle_type), DIMENSION(:), POINTER :: particle_set
317 28 : TYPE(qs_kind_type), DIMENSION(:), POINTER :: qs_kind_set
318 :
319 28 : CALL timeset(routineN, handle)
320 :
321 : CALL get_qs_env(qs_env, &
322 : qs_kind_set=qs_kind_set, &
323 : particle_set=particle_set, &
324 28 : natom=natom, nkind=nkind)
325 :
326 : ! set up basis
327 140 : ALLOCATE (bs_env%sizes_RI(natom), bs_env%sizes_AO(natom))
328 228 : ALLOCATE (bs_env%basis_set_RI(nkind), bs_env%basis_set_AO(nkind))
329 :
330 28 : CALL basis_set_list_setup(bs_env%basis_set_RI, "RI_AUX", qs_kind_set)
331 28 : CALL basis_set_list_setup(bs_env%basis_set_AO, "ORB", qs_kind_set)
332 :
333 : CALL get_particle_set(particle_set, qs_kind_set, nsgf=bs_env%sizes_RI, &
334 28 : basis=bs_env%basis_set_RI)
335 : CALL get_particle_set(particle_set, qs_kind_set, nsgf=bs_env%sizes_AO, &
336 28 : basis=bs_env%basis_set_AO)
337 :
338 28 : CALL timestop(handle)
339 :
340 28 : END SUBROUTINE setup_AO_and_RI_basis_set
341 :
342 : ! **************************************************************************************************
343 : !> \brief ...
344 : !> \param qs_env ...
345 : !> \param bs_env ...
346 : ! **************************************************************************************************
347 28 : SUBROUTINE get_RI_basis_and_basis_function_indices(qs_env, bs_env)
348 : TYPE(qs_environment_type), POINTER :: qs_env
349 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
350 :
351 : CHARACTER(LEN=*), PARAMETER :: routineN = 'get_RI_basis_and_basis_function_indices'
352 :
353 : INTEGER :: handle, i_RI, iatom, ikind, iset, &
354 : max_AO_bf_per_atom, n_ao_test, n_atom, &
355 : n_kind, n_RI, nset, nsgf, u
356 28 : INTEGER, ALLOCATABLE, DIMENSION(:) :: kind_of
357 28 : INTEGER, DIMENSION(:), POINTER :: l_max, l_min, nsgf_set
358 28 : TYPE(atomic_kind_type), DIMENSION(:), POINTER :: atomic_kind_set
359 : TYPE(gto_basis_set_type), POINTER :: basis_set_a
360 28 : TYPE(qs_kind_type), DIMENSION(:), POINTER :: qs_kind_set
361 :
362 28 : CALL timeset(routineN, handle)
363 :
364 : ! determine RI basis set size
365 28 : CALL get_qs_env(qs_env, atomic_kind_set=atomic_kind_set, qs_kind_set=qs_kind_set)
366 :
367 28 : n_kind = SIZE(qs_kind_set)
368 28 : n_atom = bs_env%n_atom
369 :
370 28 : CALL get_atomic_kind_set(atomic_kind_set, kind_of=kind_of)
371 :
372 72 : DO ikind = 1, n_kind
373 : CALL get_qs_kind(qs_kind=qs_kind_set(ikind), basis_set=basis_set_a, &
374 44 : basis_type="RI_AUX")
375 72 : IF (.NOT. ASSOCIATED(basis_set_a)) THEN
376 : CALL cp_abort(__LOCATION__, &
377 0 : "At least one RI_AUX basis set was not explicitly invoked in &KIND-section.")
378 : END IF
379 : END DO
380 :
381 84 : ALLOCATE (bs_env%i_RI_start_from_atom(n_atom))
382 56 : ALLOCATE (bs_env%i_RI_end_from_atom(n_atom))
383 56 : ALLOCATE (bs_env%i_ao_start_from_atom(n_atom))
384 56 : ALLOCATE (bs_env%i_ao_end_from_atom(n_atom))
385 :
386 28 : n_RI = 0
387 92 : DO iatom = 1, n_atom
388 64 : bs_env%i_RI_start_from_atom(iatom) = n_RI + 1
389 64 : ikind = kind_of(iatom)
390 64 : CALL get_qs_kind(qs_kind=qs_kind_set(ikind), nsgf=nsgf, basis_type="RI_AUX")
391 64 : n_RI = n_RI + nsgf
392 92 : bs_env%i_RI_end_from_atom(iatom) = n_RI
393 : END DO
394 28 : bs_env%n_RI = n_RI
395 :
396 28 : max_AO_bf_per_atom = 0
397 28 : n_ao_test = 0
398 92 : DO iatom = 1, n_atom
399 64 : bs_env%i_ao_start_from_atom(iatom) = n_ao_test + 1
400 64 : ikind = kind_of(iatom)
401 64 : CALL get_qs_kind(qs_kind=qs_kind_set(ikind), nsgf=nsgf, basis_type="ORB")
402 64 : n_ao_test = n_ao_test + nsgf
403 64 : bs_env%i_ao_end_from_atom(iatom) = n_ao_test
404 92 : max_AO_bf_per_atom = MAX(max_AO_bf_per_atom, nsgf)
405 : END DO
406 28 : CPASSERT(n_ao_test == bs_env%n_ao)
407 28 : bs_env%max_AO_bf_per_atom = max_AO_bf_per_atom
408 :
409 84 : ALLOCATE (bs_env%l_RI(n_RI))
410 28 : i_RI = 0
411 92 : DO iatom = 1, n_atom
412 64 : ikind = kind_of(iatom)
413 :
414 64 : nset = bs_env%basis_set_RI(ikind)%gto_basis_set%nset
415 64 : l_max => bs_env%basis_set_RI(ikind)%gto_basis_set%lmax
416 64 : l_min => bs_env%basis_set_RI(ikind)%gto_basis_set%lmin
417 64 : nsgf_set => bs_env%basis_set_RI(ikind)%gto_basis_set%nsgf_set
418 :
419 268 : DO iset = 1, nset
420 176 : CPASSERT(l_max(iset) == l_min(iset))
421 536 : bs_env%l_RI(i_RI + 1:i_RI + nsgf_set(iset)) = l_max(iset)
422 240 : i_RI = i_RI + nsgf_set(iset)
423 : END DO
424 :
425 : END DO
426 28 : CPASSERT(i_RI == n_RI)
427 :
428 28 : u = bs_env%unit_nr
429 :
430 28 : IF (u > 0) THEN
431 14 : WRITE (u, FMT="(T2,A)") " "
432 14 : WRITE (u, FMT="(T2,2A,T75,I8)") "Number of auxiliary Gaussian basis functions ", &
433 28 : "for χ, ε, W", n_RI
434 : END IF
435 :
436 28 : CALL timestop(handle)
437 :
438 56 : END SUBROUTINE get_RI_basis_and_basis_function_indices
439 :
440 : ! **************************************************************************************************
441 : !> \brief ...
442 : !> \param bs_env ...
443 : !> \param kpoints ...
444 : ! **************************************************************************************************
445 28 : SUBROUTINE setup_kpoints_chi_eps_W(bs_env, kpoints)
446 :
447 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
448 : TYPE(kpoint_type), POINTER :: kpoints
449 :
450 : CHARACTER(LEN=*), PARAMETER :: routineN = 'setup_kpoints_chi_eps_W'
451 :
452 : INTEGER :: handle, i_dim, n_dim, nkp, nkp_extra, &
453 : nkp_orig, u
454 : INTEGER, DIMENSION(3) :: nkp_grid, nkp_grid_extra, periodic
455 : REAL(KIND=dp) :: exp_s_p, n_dim_inv
456 :
457 28 : CALL timeset(routineN, handle)
458 :
459 : ! routine adapted from mp2_integrals.F
460 28 : NULLIFY (kpoints)
461 28 : CALL kpoint_create(kpoints)
462 :
463 28 : kpoints%kp_scheme = "GENERAL"
464 :
465 112 : periodic(1:3) = bs_env%periodic(1:3)
466 :
467 28 : CPASSERT(SIZE(bs_env%nkp_grid_chi_eps_W_input) == 3)
468 :
469 : IF (bs_env%nkp_grid_chi_eps_W_input(1) > 0 .AND. &
470 28 : bs_env%nkp_grid_chi_eps_W_input(2) > 0 .AND. &
471 : bs_env%nkp_grid_chi_eps_W_input(3) > 0) THEN
472 : ! 1. k-point mesh for χ, ε, W from input
473 0 : DO i_dim = 1, 3
474 0 : SELECT CASE (periodic(i_dim))
475 : CASE (0)
476 0 : nkp_grid(i_dim) = 1
477 0 : nkp_grid_extra(i_dim) = 1
478 : CASE (1)
479 0 : nkp_grid(i_dim) = bs_env%nkp_grid_chi_eps_W_input(i_dim)
480 0 : nkp_grid_extra(i_dim) = nkp_grid(i_dim)*2
481 : CASE DEFAULT
482 0 : CPABORT("Error in periodicity.")
483 : END SELECT
484 : END DO
485 :
486 : ELSE IF (bs_env%nkp_grid_chi_eps_W_input(1) == -1 .AND. &
487 28 : bs_env%nkp_grid_chi_eps_W_input(2) == -1 .AND. &
488 : bs_env%nkp_grid_chi_eps_W_input(3) == -1) THEN
489 : ! 2. automatic k-point mesh for χ, ε, W
490 :
491 112 : DO i_dim = 1, 3
492 :
493 84 : CPASSERT(periodic(i_dim) == 0 .OR. periodic(i_dim) == 1)
494 :
495 28 : SELECT CASE (periodic(i_dim))
496 : CASE (0)
497 52 : nkp_grid(i_dim) = 1
498 52 : nkp_grid_extra(i_dim) = 1
499 : CASE (1)
500 52 : SELECT CASE (bs_env%small_cell_full_kp_or_large_cell_Gamma)
501 : CASE (large_cell_Gamma)
502 20 : nkp_grid(i_dim) = 4
503 20 : nkp_grid_extra(i_dim) = 6
504 : CASE (small_cell_full_kp)
505 12 : nkp_grid(i_dim) = bs_env%kpoints_scf_desymm%nkp_grid(i_dim)*4
506 32 : nkp_grid_extra(i_dim) = bs_env%kpoints_scf_desymm%nkp_grid(i_dim)*8
507 : END SELECT
508 : CASE DEFAULT
509 84 : CPABORT("Error in periodicity.")
510 : END SELECT
511 :
512 : END DO
513 :
514 : ELSE
515 :
516 0 : CPABORT("An error occured when setting up the k-mesh for W.")
517 :
518 : END IF
519 :
520 28 : nkp_orig = MAX(nkp_grid(1)*nkp_grid(2)*nkp_grid(3)/2, 1)
521 :
522 28 : nkp_extra = nkp_grid_extra(1)*nkp_grid_extra(2)*nkp_grid_extra(3)/2
523 :
524 28 : nkp = nkp_orig + nkp_extra
525 :
526 112 : kpoints%nkp_grid(1:3) = nkp_grid(1:3)
527 28 : kpoints%nkp = nkp
528 :
529 112 : bs_env%nkp_grid_chi_eps_W_orig(1:3) = nkp_grid(1:3)
530 112 : bs_env%nkp_grid_chi_eps_W_extra(1:3) = nkp_grid_extra(1:3)
531 28 : bs_env%nkp_chi_eps_W_orig = nkp_orig
532 28 : bs_env%nkp_chi_eps_W_extra = nkp_extra
533 28 : bs_env%nkp_chi_eps_W_orig_plus_extra = nkp
534 :
535 140 : ALLOCATE (kpoints%xkp(3, nkp), kpoints%wkp(nkp))
536 140 : ALLOCATE (bs_env%wkp_no_extra(nkp), bs_env%wkp_s_p(nkp))
537 :
538 28 : CALL compute_xkp(kpoints%xkp, 1, nkp_orig, nkp_grid)
539 28 : CALL compute_xkp(kpoints%xkp, nkp_orig + 1, nkp, nkp_grid_extra)
540 :
541 112 : n_dim = SUM(periodic)
542 28 : IF (n_dim == 0) THEN
543 : ! molecules
544 12 : kpoints%wkp(1) = 1.0_dp
545 12 : bs_env%wkp_s_p(1) = 1.0_dp
546 12 : bs_env%wkp_no_extra(1) = 1.0_dp
547 : ELSE
548 :
549 16 : n_dim_inv = 1.0_dp/REAL(n_dim, KIND=dp)
550 :
551 : ! k-point weights are chosen to automatically extrapolate the k-point mesh
552 16 : CALL compute_wkp(kpoints%wkp(1:nkp_orig), nkp_orig, nkp_extra, n_dim_inv)
553 16 : CALL compute_wkp(kpoints%wkp(nkp_orig + 1:nkp), nkp_extra, nkp_orig, n_dim_inv)
554 :
555 864 : bs_env%wkp_no_extra(1:nkp_orig) = 0.0_dp
556 3268 : bs_env%wkp_no_extra(nkp_orig + 1:nkp) = 1.0_dp/REAL(nkp_extra, KIND=dp)
557 :
558 16 : IF (n_dim == 3) THEN
559 : ! W_PQ(k) for an s-function P and a p-function Q diverges as 1/k at k=0
560 : ! (instead of 1/k^2 for P and Q both being s-functions).
561 0 : exp_s_p = 2.0_dp*n_dim_inv
562 0 : CALL compute_wkp(bs_env%wkp_s_p(1:nkp_orig), nkp_orig, nkp_extra, exp_s_p)
563 0 : CALL compute_wkp(bs_env%wkp_s_p(nkp_orig + 1:nkp), nkp_extra, nkp_orig, exp_s_p)
564 : ELSE
565 4116 : bs_env%wkp_s_p(1:nkp) = bs_env%wkp_no_extra(1:nkp)
566 : END IF
567 :
568 : END IF
569 :
570 28 : IF (bs_env%approx_kp_extrapol) THEN
571 2 : bs_env%wkp_orig = 1.0_dp/REAL(nkp_orig, KIND=dp)
572 : END IF
573 :
574 : ! heuristic parameter: how many k-points for χ, ε, and W are used simultaneously
575 : ! (less simultaneous k-points: less memory, but more computational effort because of
576 : ! recomputation of V(k))
577 28 : bs_env%nkp_chi_eps_W_batch = 4
578 :
579 : bs_env%num_chi_eps_W_batches = (bs_env%nkp_chi_eps_W_orig_plus_extra - 1)/ &
580 28 : bs_env%nkp_chi_eps_W_batch + 1
581 :
582 28 : u = bs_env%unit_nr
583 :
584 28 : IF (u > 0) THEN
585 14 : WRITE (u, FMT="(T2,A)") " "
586 14 : WRITE (u, FMT="(T2,1A,T71,3I4)") "K-point mesh 1 for χ, ε, W", nkp_grid(1:3)
587 14 : WRITE (u, FMT="(T2,2A,T71,3I4)") "K-point mesh 2 for χ, ε, W ", &
588 28 : "(for k-point extrapolation of W)", nkp_grid_extra(1:3)
589 14 : WRITE (u, FMT="(T2,A,T80,L)") "Approximate the k-point extrapolation", &
590 28 : bs_env%approx_kp_extrapol
591 : END IF
592 :
593 28 : CALL timestop(handle)
594 :
595 28 : END SUBROUTINE setup_kpoints_chi_eps_W
596 :
597 : ! **************************************************************************************************
598 : !> \brief ...
599 : !> \param xkp ...
600 : !> \param ikp_start ...
601 : !> \param ikp_end ...
602 : !> \param grid ...
603 : ! **************************************************************************************************
604 56 : SUBROUTINE compute_xkp(xkp, ikp_start, ikp_end, grid)
605 :
606 : REAL(KIND=dp), DIMENSION(:, :), POINTER :: xkp
607 : INTEGER :: ikp_start, ikp_end
608 : INTEGER, DIMENSION(3) :: grid
609 :
610 : CHARACTER(LEN=*), PARAMETER :: routineN = 'compute_xkp'
611 :
612 : INTEGER :: handle, i, ix, iy, iz
613 :
614 56 : CALL timeset(routineN, handle)
615 :
616 56 : i = ikp_start
617 236 : DO ix = 1, grid(1)
618 3280 : DO iy = 1, grid(2)
619 11448 : DO iz = 1, grid(3)
620 :
621 8224 : IF (i > ikp_end) CYCLE
622 :
623 4112 : xkp(1, i) = REAL(2*ix - grid(1) - 1, KIND=dp)/(2._dp*REAL(grid(1), KIND=dp))
624 4112 : xkp(2, i) = REAL(2*iy - grid(2) - 1, KIND=dp)/(2._dp*REAL(grid(2), KIND=dp))
625 4112 : xkp(3, i) = REAL(2*iz - grid(3) - 1, KIND=dp)/(2._dp*REAL(grid(3), KIND=dp))
626 11268 : i = i + 1
627 :
628 : END DO
629 : END DO
630 : END DO
631 :
632 56 : CALL timestop(handle)
633 :
634 56 : END SUBROUTINE compute_xkp
635 :
636 : ! **************************************************************************************************
637 : !> \brief ...
638 : !> \param wkp ...
639 : !> \param nkp_1 ...
640 : !> \param nkp_2 ...
641 : !> \param exponent ...
642 : ! **************************************************************************************************
643 32 : SUBROUTINE compute_wkp(wkp, nkp_1, nkp_2, exponent)
644 : REAL(KIND=dp), DIMENSION(:) :: wkp
645 : INTEGER :: nkp_1, nkp_2
646 : REAL(KIND=dp) :: exponent
647 :
648 : CHARACTER(LEN=*), PARAMETER :: routineN = 'compute_wkp'
649 :
650 : INTEGER :: handle
651 : REAL(KIND=dp) :: nkp_ratio
652 :
653 32 : CALL timeset(routineN, handle)
654 :
655 32 : nkp_ratio = REAL(nkp_2, KIND=dp)/REAL(nkp_1, KIND=dp)
656 :
657 4132 : wkp(:) = 1.0_dp/REAL(nkp_1, KIND=dp)/(1.0_dp - nkp_ratio**exponent)
658 :
659 32 : CALL timestop(handle)
660 :
661 32 : END SUBROUTINE compute_wkp
662 :
663 : ! **************************************************************************************************
664 : !> \brief ...
665 : !> \param qs_env ...
666 : !> \param bs_env ...
667 : ! **************************************************************************************************
668 28 : SUBROUTINE allocate_matrices(qs_env, bs_env)
669 : TYPE(qs_environment_type), POINTER :: qs_env
670 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
671 :
672 : CHARACTER(LEN=*), PARAMETER :: routineN = 'allocate_matrices'
673 :
674 : INTEGER :: handle, i_t
675 : TYPE(cp_blacs_env_type), POINTER :: blacs_env, blacs_env_tensor
676 : TYPE(cp_fm_struct_type), POINTER :: fm_struct, fm_struct_RI_global
677 : TYPE(mp_para_env_type), POINTER :: para_env
678 :
679 28 : CALL timeset(routineN, handle)
680 :
681 28 : CALL get_qs_env(qs_env, para_env=para_env, blacs_env=blacs_env)
682 :
683 28 : fm_struct => bs_env%fm_ks_Gamma(1)%matrix_struct
684 :
685 28 : CALL cp_fm_create(bs_env%fm_Gocc, fm_struct)
686 28 : CALL cp_fm_create(bs_env%fm_Gvir, fm_struct)
687 :
688 28 : NULLIFY (fm_struct_RI_global)
689 : CALL cp_fm_struct_create(fm_struct_RI_global, context=blacs_env, nrow_global=bs_env%n_RI, &
690 28 : ncol_global=bs_env%n_RI, para_env=para_env)
691 28 : CALL cp_fm_create(bs_env%fm_RI_RI, fm_struct_RI_global)
692 28 : CALL cp_fm_create(bs_env%fm_chi_Gamma_freq, fm_struct_RI_global)
693 28 : CALL cp_fm_create(bs_env%fm_W_MIC_freq, fm_struct_RI_global)
694 28 : IF (bs_env%approx_kp_extrapol) THEN
695 2 : CALL cp_fm_create(bs_env%fm_W_MIC_freq_1_extra, fm_struct_RI_global)
696 2 : CALL cp_fm_create(bs_env%fm_W_MIC_freq_1_no_extra, fm_struct_RI_global)
697 2 : CALL cp_fm_set_all(bs_env%fm_W_MIC_freq_1_extra, 0.0_dp)
698 2 : CALL cp_fm_set_all(bs_env%fm_W_MIC_freq_1_no_extra, 0.0_dp)
699 : END IF
700 28 : CALL cp_fm_struct_release(fm_struct_RI_global)
701 :
702 : ! create blacs_env for subgroups of tensor operations
703 28 : NULLIFY (blacs_env_tensor)
704 28 : CALL cp_blacs_env_create(blacs_env=blacs_env_tensor, para_env=bs_env%para_env_tensor)
705 :
706 : ! allocate dbcsr matrices in the tensor subgroup; actually, one only needs a small
707 : ! subset of blocks in the tensor subgroup, however, all atomic blocks are allocated.
708 : ! One might think of creating a dbcsr matrix with only the blocks that are needed
709 : ! in the tensor subgroup
710 : CALL create_mat_munu(bs_env%mat_ao_ao_tensor, qs_env, bs_env%eps_atom_grid_2d_mat, &
711 28 : blacs_env_tensor, do_ri_aux_basis=.FALSE.)
712 :
713 : CALL create_mat_munu(bs_env%mat_RI_RI_tensor, qs_env, bs_env%eps_atom_grid_2d_mat, &
714 28 : blacs_env_tensor, do_ri_aux_basis=.TRUE.)
715 :
716 : CALL create_mat_munu(bs_env%mat_RI_RI, qs_env, bs_env%eps_atom_grid_2d_mat, &
717 28 : blacs_env, do_ri_aux_basis=.TRUE.)
718 :
719 28 : CALL cp_blacs_env_release(blacs_env_tensor)
720 :
721 28 : NULLIFY (bs_env%mat_chi_Gamma_tau)
722 28 : CALL dbcsr_allocate_matrix_set(bs_env%mat_chi_Gamma_tau, bs_env%num_time_freq_points)
723 :
724 396 : DO i_t = 1, bs_env%num_time_freq_points
725 368 : ALLOCATE (bs_env%mat_chi_Gamma_tau(i_t)%matrix)
726 396 : CALL dbcsr_create(bs_env%mat_chi_Gamma_tau(i_t)%matrix, template=bs_env%mat_RI_RI%matrix)
727 : END DO
728 :
729 28 : CALL timestop(handle)
730 :
731 28 : END SUBROUTINE allocate_matrices
732 :
733 : ! **************************************************************************************************
734 : !> \brief ...
735 : !> \param bs_env ...
736 : ! **************************************************************************************************
737 22 : SUBROUTINE allocate_GW_eigenvalues(bs_env)
738 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
739 :
740 : CHARACTER(LEN=*), PARAMETER :: routineN = 'allocate_GW_eigenvalues'
741 :
742 : INTEGER :: handle
743 :
744 22 : CALL timeset(routineN, handle)
745 :
746 110 : ALLOCATE (bs_env%eigenval_G0W0(bs_env%n_ao, bs_env%nkp_bs_and_DOS, bs_env%n_spin))
747 110 : ALLOCATE (bs_env%eigenval_HF(bs_env%n_ao, bs_env%nkp_bs_and_DOS, bs_env%n_spin))
748 :
749 22 : CALL timestop(handle)
750 :
751 22 : END SUBROUTINE allocate_GW_eigenvalues
752 :
753 : ! **************************************************************************************************
754 : !> \brief ...
755 : !> \param qs_env ...
756 : !> \param bs_env ...
757 : ! **************************************************************************************************
758 28 : SUBROUTINE create_tensors(qs_env, bs_env)
759 : TYPE(qs_environment_type), POINTER :: qs_env
760 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
761 :
762 : CHARACTER(LEN=*), PARAMETER :: routineN = 'create_tensors'
763 :
764 : INTEGER :: handle
765 :
766 28 : CALL timeset(routineN, handle)
767 :
768 28 : CALL init_interaction_radii(bs_env)
769 :
770 : ! split blocks does not improve load balancing/efficienfy for tensor contraction, so we go
771 : ! with the standard atomic blocks
772 : CALL create_3c_t(bs_env%t_RI_AO__AO, bs_env%para_env_tensor, "(RI AO | AO)", [1, 2], [3], &
773 : bs_env%sizes_RI, bs_env%sizes_AO, &
774 28 : create_nl_3c=.TRUE., nl_3c=bs_env%nl_3c, qs_env=qs_env)
775 : CALL create_3c_t(bs_env%t_RI__AO_AO, bs_env%para_env_tensor, "(RI | AO AO)", [1], [2, 3], &
776 28 : bs_env%sizes_RI, bs_env%sizes_AO)
777 :
778 28 : CALL create_2c_t(bs_env, bs_env%sizes_RI, bs_env%sizes_AO)
779 :
780 28 : CALL timestop(handle)
781 :
782 28 : END SUBROUTINE create_tensors
783 :
784 : ! **************************************************************************************************
785 : !> \brief ...
786 : !> \param qs_env ...
787 : !> \param bs_env ...
788 : ! **************************************************************************************************
789 22 : SUBROUTINE check_sparsity_3c(qs_env, bs_env)
790 : TYPE(qs_environment_type), POINTER :: qs_env
791 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
792 :
793 : CHARACTER(LEN=*), PARAMETER :: routineN = 'check_sparsity_3c'
794 :
795 : INTEGER :: handle, n_atom_step, RI_atom
796 : INTEGER(int_8) :: non_zero_elements_sum, nze
797 : REAL(dp) :: max_dist_AO_atoms, occ, occupation_sum
798 : REAL(KIND=dp) :: t1, t2
799 22 : TYPE(dbt_type), ALLOCATABLE, DIMENSION(:, :) :: t_3c_global_array
800 :
801 : !TYPE(dbt_type) :: t_3c_global
802 :
803 : !TYPE(neighbor_list_3c_type) :: nl_3c_global
804 :
805 22 : CALL timeset(routineN, handle)
806 :
807 : ! check the sparsity of 3c integral tensor (µν|P); calculate maximum distance between
808 : ! AO atoms µ, ν where at least a single integral (µν|P) is larger than the filter threshold
809 :
810 198 : ALLOCATE (t_3c_global_array(1, 1))
811 22 : CALL dbt_create(bs_env%t_RI_AO__AO, t_3c_global_array(1, 1))
812 :
813 : ! Allocate arrays to store min/max indices for overlap with other AO/RI functions on each atom
814 : ! (Filled during loop)
815 88 : ALLOCATE (bs_env%min_RI_idx_from_AO_AO_atom(bs_env%n_atom, bs_env%n_atom))
816 88 : ALLOCATE (bs_env%max_RI_idx_from_AO_AO_atom(bs_env%n_atom, bs_env%n_atom))
817 88 : ALLOCATE (bs_env%min_AO_idx_from_RI_AO_atom(bs_env%n_atom, bs_env%n_atom))
818 88 : ALLOCATE (bs_env%max_AO_idx_from_RI_AO_atom(bs_env%n_atom, bs_env%n_atom))
819 178 : bs_env%min_RI_idx_from_AO_AO_atom(:, :) = bs_env%n_RI
820 178 : bs_env%max_RI_idx_from_AO_AO_atom(:, :) = 1
821 178 : bs_env%min_AO_idx_from_RI_AO_atom(:, :) = bs_env%n_AO
822 178 : bs_env%max_AO_idx_from_RI_AO_atom(:, :) = 1
823 :
824 22 : CALL bs_env%para_env%sync()
825 22 : t1 = m_walltime()
826 :
827 22 : occupation_sum = 0.0_dp
828 22 : non_zero_elements_sum = 0
829 22 : max_dist_AO_atoms = 0.0_dp
830 22 : n_atom_step = INT(SQRT(REAL(bs_env%n_atom, KIND=dp)))
831 : ! do not compute full 3c integrals at once because it may cause out of memory
832 70 : DO RI_atom = 1, bs_env%n_atom, n_atom_step
833 :
834 : CALL build_3c_integrals(t_3c_global_array, &
835 : bs_env%eps_filter, &
836 : qs_env, &
837 : bs_env%nl_3c, &
838 : int_eps=bs_env%eps_filter, &
839 : basis_i=bs_env%basis_set_RI, &
840 : basis_j=bs_env%basis_set_AO, &
841 : basis_k=bs_env%basis_set_AO, &
842 : bounds_i=[RI_atom, MIN(RI_atom + n_atom_step - 1, bs_env%n_atom)], &
843 : potential_parameter=bs_env%ri_metric, &
844 144 : desymmetrize=.FALSE.)
845 :
846 48 : CALL dbt_filter(t_3c_global_array(1, 1), bs_env%eps_filter)
847 :
848 48 : CALL bs_env%para_env%sync()
849 :
850 48 : CALL get_tensor_occupancy(t_3c_global_array(1, 1), nze, occ)
851 48 : non_zero_elements_sum = non_zero_elements_sum + nze
852 48 : occupation_sum = occupation_sum + occ
853 :
854 48 : CALL get_max_dist_AO_atoms(t_3c_global_array(1, 1), max_dist_AO_atoms, qs_env)
855 :
856 : ! Extract indices per block
857 48 : CALL get_i_j_atom_ranges(t_3c_global_array(1, 1), bs_env)
858 :
859 118 : CALL dbt_clear(t_3c_global_array(1, 1))
860 :
861 : END DO
862 :
863 22 : t2 = m_walltime()
864 :
865 22 : CALL bs_env%para_env%min(bs_env%min_RI_idx_from_AO_AO_atom)
866 22 : CALL bs_env%para_env%max(bs_env%max_RI_idx_from_AO_AO_atom)
867 22 : CALL bs_env%para_env%min(bs_env%min_AO_idx_from_RI_AO_atom)
868 22 : CALL bs_env%para_env%max(bs_env%max_AO_idx_from_RI_AO_atom)
869 :
870 22 : bs_env%occupation_3c_int = occupation_sum
871 22 : bs_env%max_dist_AO_atoms = max_dist_AO_atoms
872 :
873 22 : CALL dbt_destroy(t_3c_global_array(1, 1))
874 44 : DEALLOCATE (t_3c_global_array)
875 :
876 22 : IF (bs_env%unit_nr > 0) THEN
877 11 : WRITE (bs_env%unit_nr, '(T2,A)') ''
878 : WRITE (bs_env%unit_nr, '(T2,A,F27.1,A)') &
879 11 : 'Computed 3-center integrals (µν|P), execution time', t2 - t1, ' s'
880 11 : WRITE (bs_env%unit_nr, '(T2,A,F48.3,A)') 'Percentage of non-zero (µν|P)', &
881 22 : occupation_sum*100, ' %'
882 11 : WRITE (bs_env%unit_nr, '(T2,A,F33.1,A)') 'Max. distance between µ,ν in non-zero (µν|P)', &
883 22 : max_dist_AO_atoms*angstrom, ' A'
884 11 : WRITE (bs_env%unit_nr, '(T2,2A,I20,A)') 'Required memory if storing all 3-center ', &
885 22 : 'integrals (µν|P)', INT(REAL(non_zero_elements_sum, KIND=dp)*8.0E-9_dp), ' GB'
886 : END IF
887 :
888 22 : CALL timestop(handle)
889 :
890 44 : END SUBROUTINE check_sparsity_3c
891 :
892 : ! **************************************************************************************************
893 : !> \brief ...
894 : !> \param t_3c ...
895 : !> \param bs_env ...
896 : ! **************************************************************************************************
897 48 : SUBROUTINE get_i_j_atom_ranges(t_3c, bs_env)
898 : TYPE(dbt_type) :: t_3c
899 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
900 :
901 : CHARACTER(LEN=*), PARAMETER :: routineN = 'get_i_j_atom_ranges'
902 :
903 : INTEGER :: handle, idx_AO_end, idx_AO_start, &
904 : idx_RI_end, idx_RI_start
905 : INTEGER, DIMENSION(3) :: atom_ind
906 : TYPE(dbt_iterator_type) :: iter
907 :
908 48 : CALL timeset(routineN, handle)
909 :
910 : ! Loop over blocks in 3c, for given min_atom: RI_min/max index from min_atom
911 : !$OMP PARALLEL DEFAULT(NONE) &
912 : !$OMP SHARED(t_3c, bs_env) &
913 : !$OMP PRIVATE(iter, atom_ind, &
914 48 : !$OMP idx_RI_start, idx_RI_end, idx_AO_start, idx_AO_end)
915 :
916 : CALL dbt_iterator_start(iter, t_3c)
917 : DO WHILE (dbt_iterator_blocks_left(iter))
918 : CALL dbt_iterator_next_block(iter, atom_ind)
919 :
920 : ! Pre-fetch indices to avoid referencing 'bs_env' twice inside the ATOMIC blocks
921 : idx_RI_start = bs_env%i_RI_start_from_atom(atom_ind(1))
922 : idx_RI_end = bs_env%i_RI_end_from_atom(atom_ind(1))
923 :
924 : idx_AO_start = bs_env%i_ao_start_from_atom(atom_ind(2))
925 : idx_AO_end = bs_env%i_ao_end_from_atom(atom_ind(2))
926 :
927 : ! Update values safely inside ATOMIC blocks, otherwise race conditions occur
928 : !$OMP ATOMIC UPDATE
929 : bs_env%min_RI_idx_from_AO_AO_atom(atom_ind(2), atom_ind(3)) = &
930 : MIN(bs_env%min_RI_idx_from_AO_AO_atom(atom_ind(2), atom_ind(3)), idx_RI_start)
931 : !$OMP ATOMIC UPDATE
932 : bs_env%max_RI_idx_from_AO_AO_atom(atom_ind(2), atom_ind(3)) = &
933 : MAX(bs_env%max_RI_idx_from_AO_AO_atom(atom_ind(2), atom_ind(3)), idx_RI_end)
934 :
935 : !$OMP ATOMIC UPDATE
936 : bs_env%min_AO_idx_from_RI_AO_atom(atom_ind(1), atom_ind(3)) = &
937 : MIN(bs_env%min_AO_idx_from_RI_AO_atom(atom_ind(1), atom_ind(3)), idx_AO_start)
938 : !$OMP ATOMIC UPDATE
939 : bs_env%max_AO_idx_from_RI_AO_atom(atom_ind(1), atom_ind(3)) = &
940 : MAX(bs_env%max_AO_idx_from_RI_AO_atom(atom_ind(1), atom_ind(3)), idx_AO_end)
941 :
942 : END DO
943 : CALL dbt_iterator_stop(iter)
944 : !$OMP END PARALLEL
945 :
946 48 : CALL timestop(handle)
947 :
948 48 : END SUBROUTINE get_i_j_atom_ranges
949 :
950 : ! **************************************************************************************************
951 : !> \brief ...
952 : !> \param bs_env ...
953 : !> \param sizes_RI ...
954 : !> \param sizes_AO ...
955 : ! **************************************************************************************************
956 28 : SUBROUTINE create_2c_t(bs_env, sizes_RI, sizes_AO)
957 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
958 : INTEGER, ALLOCATABLE, DIMENSION(:) :: sizes_RI, sizes_AO
959 :
960 : CHARACTER(LEN=*), PARAMETER :: routineN = 'create_2c_t'
961 :
962 : INTEGER :: handle
963 28 : INTEGER, ALLOCATABLE, DIMENSION(:) :: dist_1, dist_2
964 : INTEGER, DIMENSION(2) :: pdims_2d
965 84 : TYPE(dbt_pgrid_type) :: pgrid_2d
966 :
967 28 : CALL timeset(routineN, handle)
968 :
969 : ! inspired from rpa_im_time.F / hfx_types.F
970 :
971 28 : pdims_2d = 0
972 28 : CALL dbt_pgrid_create(bs_env%para_env_tensor, pdims_2d, pgrid_2d)
973 :
974 : CALL create_2c_tensor(bs_env%t_G, dist_1, dist_2, pgrid_2d, sizes_AO, sizes_AO, &
975 28 : name="(AO | AO)")
976 28 : DEALLOCATE (dist_1, dist_2)
977 : CALL create_2c_tensor(bs_env%t_chi, dist_1, dist_2, pgrid_2d, sizes_RI, sizes_RI, &
978 28 : name="(RI | RI)")
979 28 : DEALLOCATE (dist_1, dist_2)
980 : CALL create_2c_tensor(bs_env%t_W, dist_1, dist_2, pgrid_2d, sizes_RI, sizes_RI, &
981 28 : name="(RI | RI)")
982 28 : DEALLOCATE (dist_1, dist_2)
983 28 : CALL dbt_pgrid_destroy(pgrid_2d)
984 :
985 28 : CALL timestop(handle)
986 :
987 28 : END SUBROUTINE create_2c_t
988 :
989 : ! **************************************************************************************************
990 : !> \brief ...
991 : !> \param tensor ...
992 : !> \param para_env ...
993 : !> \param tensor_name ...
994 : !> \param map1 ...
995 : !> \param map2 ...
996 : !> \param sizes_RI ...
997 : !> \param sizes_AO ...
998 : !> \param create_nl_3c ...
999 : !> \param nl_3c ...
1000 : !> \param qs_env ...
1001 : ! **************************************************************************************************
1002 56 : SUBROUTINE create_3c_t(tensor, para_env, tensor_name, map1, map2, sizes_RI, sizes_AO, &
1003 : create_nl_3c, nl_3c, qs_env)
1004 : TYPE(dbt_type) :: tensor
1005 : TYPE(mp_para_env_type), POINTER :: para_env
1006 : CHARACTER(LEN=12) :: tensor_name
1007 : INTEGER, DIMENSION(:) :: map1, map2
1008 : INTEGER, ALLOCATABLE, DIMENSION(:) :: sizes_RI, sizes_AO
1009 : LOGICAL, OPTIONAL :: create_nl_3c
1010 : TYPE(neighbor_list_3c_type), OPTIONAL :: nl_3c
1011 : TYPE(qs_environment_type), OPTIONAL, POINTER :: qs_env
1012 :
1013 : CHARACTER(LEN=*), PARAMETER :: routineN = 'create_3c_t'
1014 :
1015 : INTEGER :: handle, nkind
1016 56 : INTEGER, ALLOCATABLE, DIMENSION(:) :: dist_AO_1, dist_AO_2, dist_RI
1017 : INTEGER, DIMENSION(3) :: pcoord, pdims, pdims_3d
1018 : LOGICAL :: my_create_nl_3c
1019 168 : TYPE(dbt_pgrid_type) :: pgrid_3d
1020 : TYPE(distribution_3d_type) :: dist_3d
1021 56 : TYPE(mp_cart_type) :: mp_comm_t3c_2
1022 56 : TYPE(particle_type), DIMENSION(:), POINTER :: particle_set
1023 :
1024 56 : CALL timeset(routineN, handle)
1025 :
1026 56 : pdims_3d = 0
1027 56 : CALL dbt_pgrid_create(para_env, pdims_3d, pgrid_3d)
1028 : CALL create_3c_tensor(tensor, dist_RI, dist_AO_1, dist_AO_2, &
1029 : pgrid_3d, sizes_RI, sizes_AO, sizes_AO, &
1030 56 : map1=map1, map2=map2, name=tensor_name)
1031 :
1032 56 : IF (PRESENT(create_nl_3c)) THEN
1033 28 : my_create_nl_3c = create_nl_3c
1034 : ELSE
1035 : my_create_nl_3c = .FALSE.
1036 : END IF
1037 :
1038 28 : IF (my_create_nl_3c) THEN
1039 28 : CALL get_qs_env(qs_env, nkind=nkind, particle_set=particle_set)
1040 28 : CALL dbt_mp_environ_pgrid(pgrid_3d, pdims, pcoord)
1041 28 : CALL mp_comm_t3c_2%create(pgrid_3d%mp_comm_2d, 3, pdims)
1042 : CALL distribution_3d_create(dist_3d, dist_RI, dist_AO_1, dist_AO_2, &
1043 28 : nkind, particle_set, mp_comm_t3c_2, own_comm=.TRUE.)
1044 :
1045 : CALL build_3c_neighbor_lists(nl_3c, &
1046 : qs_env%bs_env%basis_set_RI, &
1047 : qs_env%bs_env%basis_set_AO, &
1048 : qs_env%bs_env%basis_set_AO, &
1049 : dist_3d, qs_env%bs_env%ri_metric, &
1050 28 : "GW_3c_nl", qs_env, own_dist=.TRUE.)
1051 : END IF
1052 :
1053 56 : DEALLOCATE (dist_RI, dist_AO_1, dist_AO_2)
1054 56 : CALL dbt_pgrid_destroy(pgrid_3d)
1055 :
1056 56 : CALL timestop(handle)
1057 :
1058 112 : END SUBROUTINE create_3c_t
1059 :
1060 : ! **************************************************************************************************
1061 : !> \brief ...
1062 : !> \param bs_env ...
1063 : ! **************************************************************************************************
1064 28 : SUBROUTINE init_interaction_radii(bs_env)
1065 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
1066 :
1067 : CHARACTER(LEN=*), PARAMETER :: routineN = 'init_interaction_radii'
1068 :
1069 : INTEGER :: handle, ibasis
1070 : TYPE(gto_basis_set_type), POINTER :: orb_basis, ri_basis
1071 :
1072 28 : CALL timeset(routineN, handle)
1073 :
1074 72 : DO ibasis = 1, SIZE(bs_env%basis_set_AO)
1075 :
1076 44 : orb_basis => bs_env%basis_set_AO(ibasis)%gto_basis_set
1077 44 : CALL init_interaction_radii_orb_basis(orb_basis, bs_env%eps_filter)
1078 :
1079 44 : ri_basis => bs_env%basis_set_RI(ibasis)%gto_basis_set
1080 72 : CALL init_interaction_radii_orb_basis(ri_basis, bs_env%eps_filter)
1081 :
1082 : END DO
1083 :
1084 28 : CALL timestop(handle)
1085 :
1086 28 : END SUBROUTINE init_interaction_radii
1087 :
1088 : ! **************************************************************************************************
1089 : !> \brief ...
1090 : !> \param t_3c_int ...
1091 : !> \param max_dist_AO_atoms ...
1092 : !> \param qs_env ...
1093 : ! **************************************************************************************************
1094 48 : SUBROUTINE get_max_dist_AO_atoms(t_3c_int, max_dist_AO_atoms, qs_env)
1095 : TYPE(dbt_type) :: t_3c_int
1096 : REAL(KIND=dp) :: max_dist_AO_atoms
1097 : TYPE(qs_environment_type), POINTER :: qs_env
1098 :
1099 : CHARACTER(LEN=*), PARAMETER :: routineN = 'get_max_dist_AO_atoms'
1100 :
1101 : INTEGER :: atom_1, atom_2, handle, num_cells
1102 : INTEGER, DIMENSION(3) :: atom_ind
1103 48 : INTEGER, DIMENSION(:, :), POINTER :: index_to_cell
1104 : REAL(KIND=dp) :: abs_rab
1105 : REAL(KIND=dp), DIMENSION(3) :: rab
1106 : TYPE(cell_type), POINTER :: cell
1107 : TYPE(dbt_iterator_type) :: iter
1108 : TYPE(mp_para_env_type), POINTER :: para_env
1109 48 : TYPE(particle_type), DIMENSION(:), POINTER :: particle_set
1110 :
1111 48 : CALL timeset(routineN, handle)
1112 :
1113 48 : NULLIFY (cell, particle_set, para_env)
1114 48 : CALL get_qs_env(qs_env, cell=cell, particle_set=particle_set, para_env=para_env)
1115 :
1116 : ! IMPORTANT: Use thread-local copy for max_dist_AO_atoms via REDUCTION to avoid race conditions
1117 : !$OMP PARALLEL DEFAULT(NONE) &
1118 : !$OMP SHARED(t_3c_int, num_cells, index_to_cell, particle_set, cell) &
1119 : !$OMP PRIVATE(iter, atom_ind, rab, abs_rab, atom_1, atom_2) &
1120 48 : !$OMP REDUCTION(MAX:max_dist_AO_atoms)
1121 :
1122 : CALL dbt_iterator_start(iter, t_3c_int)
1123 : DO WHILE (dbt_iterator_blocks_left(iter))
1124 : CALL dbt_iterator_next_block(iter, atom_ind)
1125 :
1126 : atom_1 = atom_ind(2)
1127 : atom_2 = atom_ind(3)
1128 : rab = pbc(particle_set(atom_1)%r(1:3), particle_set(atom_2)%r(1:3), cell)
1129 : abs_rab = SQRT(rab(1)**2 + rab(2)**2 + rab(3)**2)
1130 :
1131 : ! Reduction takes care of using a thread-local copy
1132 : max_dist_AO_atoms = MAX(max_dist_AO_atoms, abs_rab)
1133 :
1134 : END DO
1135 : CALL dbt_iterator_stop(iter)
1136 : !$OMP END PARALLEL
1137 :
1138 48 : CALL para_env%max(max_dist_AO_atoms)
1139 :
1140 48 : CALL timestop(handle)
1141 :
1142 48 : END SUBROUTINE get_max_dist_AO_atoms
1143 :
1144 : ! **************************************************************************************************
1145 : !> \brief ...
1146 : !> \param bs_env ...
1147 : ! **************************************************************************************************
1148 22 : SUBROUTINE set_sparsity_parallelization_parameters(bs_env)
1149 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
1150 :
1151 : CHARACTER(LEN=*), PARAMETER :: routineN = 'set_sparsity_parallelization_parameters'
1152 :
1153 : INTEGER :: handle, i_ivl, IL_ivl, j_ivl, n_atom_per_IL_ivl, n_atom_per_ivl, n_intervals_i, &
1154 : n_intervals_inner_loop_atoms, n_intervals_j, u
1155 : INTEGER(KIND=int_8) :: input_memory_per_proc
1156 :
1157 22 : CALL timeset(routineN, handle)
1158 :
1159 : ! heuristic parameter to prevent out of memory
1160 22 : bs_env%safety_factor_memory = 0.10_dp
1161 :
1162 22 : input_memory_per_proc = INT(bs_env%input_memory_per_proc_GB*1.0E9_dp, KIND=int_8)
1163 :
1164 : ! choose atomic range for λ ("i_atom"), ν ("j_atom") in
1165 : ! M_λνP(iτ) = sum_µ (µν|P) G^occ_µλ(i|τ|,k=0)
1166 : ! N_νλQ(iτ) = sum_σ (σλ|Q) G^vir_σν(i|τ|,k=0)
1167 : ! such that M and N fit into the memory
1168 : n_atom_per_ivl = INT(SQRT(bs_env%safety_factor_memory*input_memory_per_proc &
1169 : *bs_env%group_size_tensor/24/bs_env%n_RI &
1170 22 : /SQRT(bs_env%occupation_3c_int)))/bs_env%max_AO_bf_per_atom
1171 :
1172 22 : n_intervals_i = (bs_env%n_atom_i - 1)/n_atom_per_ivl + 1
1173 22 : n_intervals_j = (bs_env%n_atom_j - 1)/n_atom_per_ivl + 1
1174 :
1175 22 : bs_env%n_atom_per_interval_ij = n_atom_per_ivl
1176 22 : bs_env%n_intervals_i = n_intervals_i
1177 22 : bs_env%n_intervals_j = n_intervals_j
1178 :
1179 66 : ALLOCATE (bs_env%i_atom_intervals(2, n_intervals_i))
1180 66 : ALLOCATE (bs_env%j_atom_intervals(2, n_intervals_j))
1181 :
1182 44 : DO i_ivl = 1, n_intervals_i
1183 22 : bs_env%i_atom_intervals(1, i_ivl) = (i_ivl - 1)*n_atom_per_ivl + bs_env%atoms_i(1)
1184 : bs_env%i_atom_intervals(2, i_ivl) = MIN(i_ivl*n_atom_per_ivl + bs_env%atoms_i(1) - 1, &
1185 44 : bs_env%atoms_i(2))
1186 : END DO
1187 :
1188 44 : DO j_ivl = 1, n_intervals_j
1189 22 : bs_env%j_atom_intervals(1, j_ivl) = (j_ivl - 1)*n_atom_per_ivl + bs_env%atoms_j(1)
1190 : bs_env%j_atom_intervals(2, j_ivl) = MIN(j_ivl*n_atom_per_ivl + bs_env%atoms_j(1) - 1, &
1191 44 : bs_env%atoms_j(2))
1192 : END DO
1193 :
1194 88 : ALLOCATE (bs_env%skip_Sigma_occ(n_intervals_i, n_intervals_j))
1195 66 : ALLOCATE (bs_env%skip_Sigma_vir(n_intervals_i, n_intervals_j))
1196 66 : bs_env%skip_Sigma_occ(:, :) = .FALSE.
1197 66 : bs_env%skip_Sigma_vir(:, :) = .FALSE.
1198 22 : bs_env%n_skip_chi = 0
1199 :
1200 66 : ALLOCATE (bs_env%skip_chi(n_intervals_i, n_intervals_j))
1201 66 : bs_env%skip_chi(:, :) = .FALSE.
1202 22 : bs_env%n_skip_sigma = 0
1203 :
1204 : ! choose atomic range for µ and σ ("inner loop (IL) atom") in
1205 : ! M_λνP(iτ) = sum_µ (µν|P) G^occ_µλ(i|τ|,k=0)
1206 : ! N_νλQ(iτ) = sum_σ (σλ|Q) G^vir_σν(i|τ|,k=0)
1207 : n_atom_per_IL_ivl = MIN(INT(bs_env%safety_factor_memory*input_memory_per_proc &
1208 : *bs_env%group_size_tensor/n_atom_per_ivl &
1209 : /bs_env%max_AO_bf_per_atom &
1210 : /bs_env%n_RI/8/SQRT(bs_env%occupation_3c_int) &
1211 22 : /bs_env%max_AO_bf_per_atom), bs_env%n_atom)
1212 :
1213 22 : n_intervals_inner_loop_atoms = (bs_env%n_atom - 1)/n_atom_per_IL_ivl + 1
1214 :
1215 22 : bs_env%n_atom_per_IL_interval = n_atom_per_IL_ivl
1216 22 : bs_env%n_intervals_inner_loop_atoms = n_intervals_inner_loop_atoms
1217 :
1218 66 : ALLOCATE (bs_env%inner_loop_atom_intervals(2, n_intervals_inner_loop_atoms))
1219 44 : DO IL_ivl = 1, n_intervals_inner_loop_atoms
1220 22 : bs_env%inner_loop_atom_intervals(1, IL_ivl) = (IL_ivl - 1)*n_atom_per_IL_ivl + 1
1221 44 : bs_env%inner_loop_atom_intervals(2, IL_ivl) = MIN(IL_ivl*n_atom_per_IL_ivl, bs_env%n_atom)
1222 : END DO
1223 :
1224 22 : u = bs_env%unit_nr
1225 22 : IF (u > 0) THEN
1226 11 : WRITE (u, '(T2,A)') ''
1227 11 : WRITE (u, '(T2,A,I33)') 'Number of i and j atoms in M_λνP(τ), N_νλQ(τ):', n_atom_per_ivl
1228 11 : WRITE (u, '(T2,A,I18)') 'Number of inner loop atoms for µ in M_λνP = sum_µ (µν|P) G_µλ', &
1229 22 : n_atom_per_IL_ivl
1230 : END IF
1231 :
1232 22 : CALL timestop(handle)
1233 :
1234 22 : END SUBROUTINE set_sparsity_parallelization_parameters
1235 :
1236 : ! **************************************************************************************************
1237 : !> \brief ...
1238 : !> \param qs_env ...
1239 : !> \param bs_env ...
1240 : ! **************************************************************************************************
1241 22 : SUBROUTINE check_for_restart_files(qs_env, bs_env)
1242 : TYPE(qs_environment_type), POINTER :: qs_env
1243 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
1244 :
1245 : CHARACTER(LEN=*), PARAMETER :: routineN = 'check_for_restart_files'
1246 :
1247 : CHARACTER(LEN=9) :: frmt
1248 : CHARACTER(len=default_path_length) :: project_name
1249 : CHARACTER(len=default_string_length) :: f_chi, f_s_n, f_s_p, f_s_x, f_w_t, prefix
1250 : INTEGER :: handle, i_spin, i_t_or_w, ind, n_spin, &
1251 : num_time_freq_points
1252 : LOGICAL :: chi_exists, Sigma_neg_time_exists, &
1253 : Sigma_pos_time_exists, &
1254 : Sigma_x_spin_exists, W_time_exists
1255 : TYPE(cp_logger_type), POINTER :: logger
1256 : TYPE(section_vals_type), POINTER :: input, print_key
1257 :
1258 22 : CALL timeset(routineN, handle)
1259 :
1260 22 : num_time_freq_points = bs_env%num_time_freq_points
1261 22 : n_spin = bs_env%n_spin
1262 :
1263 66 : ALLOCATE (bs_env%read_chi(num_time_freq_points))
1264 44 : ALLOCATE (bs_env%calc_chi(num_time_freq_points))
1265 88 : ALLOCATE (bs_env%Sigma_c_exists(num_time_freq_points, n_spin))
1266 :
1267 22 : CALL get_qs_env(qs_env, input=input)
1268 :
1269 22 : logger => cp_get_default_logger()
1270 22 : print_key => section_vals_get_subs_vals(input, 'PROPERTIES%BANDSTRUCTURE%GW%PRINT%RESTART')
1271 : project_name = cp_print_key_generate_filename(logger, print_key, extension="", &
1272 22 : my_local=.FALSE.)
1273 22 : WRITE (prefix, '(2A)') TRIM(project_name), "-RESTART_"
1274 22 : bs_env%prefix = prefix
1275 :
1276 22 : bs_env%all_W_exist = .TRUE.
1277 :
1278 346 : DO i_t_or_w = 1, num_time_freq_points
1279 :
1280 324 : IF (i_t_or_w < 10) THEN
1281 186 : WRITE (frmt, '(A)') '(3A,I1,A)'
1282 186 : WRITE (f_chi, frmt) TRIM(prefix), bs_env%chi_name, "_0", i_t_or_w, ".matrix"
1283 186 : WRITE (f_W_t, frmt) TRIM(prefix), bs_env%W_time_name, "_0", i_t_or_w, ".matrix"
1284 138 : ELSE IF (i_t_or_w < 100) THEN
1285 138 : WRITE (frmt, '(A)') '(3A,I2,A)'
1286 138 : WRITE (f_chi, frmt) TRIM(prefix), bs_env%chi_name, "_", i_t_or_w, ".matrix"
1287 138 : WRITE (f_W_t, frmt) TRIM(prefix), bs_env%W_time_name, "_", i_t_or_w, ".matrix"
1288 : ELSE
1289 0 : CPABORT('Please implement more than 99 time/frequency points.')
1290 : END IF
1291 :
1292 324 : INQUIRE (file=TRIM(f_chi), exist=chi_exists)
1293 324 : INQUIRE (file=TRIM(f_W_t), exist=W_time_exists)
1294 :
1295 324 : bs_env%read_chi(i_t_or_w) = chi_exists
1296 324 : bs_env%calc_chi(i_t_or_w) = .NOT. chi_exists
1297 :
1298 324 : bs_env%all_W_exist = bs_env%all_W_exist .AND. W_time_exists
1299 :
1300 : ! the self-energy is spin-dependent
1301 710 : DO i_spin = 1, n_spin
1302 :
1303 364 : ind = i_t_or_w + (i_spin - 1)*num_time_freq_points
1304 :
1305 364 : IF (ind < 10) THEN
1306 186 : WRITE (frmt, '(A)') '(3A,I1,A)'
1307 186 : WRITE (f_S_p, frmt) TRIM(prefix), bs_env%Sigma_p_name, "_0", ind, ".matrix"
1308 186 : WRITE (f_S_n, frmt) TRIM(prefix), bs_env%Sigma_n_name, "_0", ind, ".matrix"
1309 178 : ELSE IF (i_t_or_w < 100) THEN
1310 178 : WRITE (frmt, '(A)') '(3A,I2,A)'
1311 178 : WRITE (f_S_p, frmt) TRIM(prefix), bs_env%Sigma_p_name, "_", ind, ".matrix"
1312 178 : WRITE (f_S_n, frmt) TRIM(prefix), bs_env%Sigma_n_name, "_", ind, ".matrix"
1313 : END IF
1314 :
1315 364 : INQUIRE (file=TRIM(f_S_p), exist=Sigma_pos_time_exists)
1316 364 : INQUIRE (file=TRIM(f_S_n), exist=Sigma_neg_time_exists)
1317 :
1318 : bs_env%Sigma_c_exists(i_t_or_w, i_spin) = Sigma_pos_time_exists .AND. &
1319 932 : Sigma_neg_time_exists
1320 :
1321 : END DO
1322 :
1323 : END DO
1324 :
1325 : ! Marek : In the RTBSE run, check also for zero frequency W
1326 22 : IF (bs_env%rtp_method == rtp_method_bse) THEN
1327 12 : WRITE (f_W_t, '(3A,I1,A)') TRIM(prefix), "W_freq_rtp", "_0", 0, ".matrix"
1328 12 : INQUIRE (file=TRIM(f_W_t), exist=W_time_exists)
1329 20 : bs_env%all_W_exist = bs_env%all_W_exist .AND. W_time_exists
1330 : END IF
1331 :
1332 22 : IF (bs_env%all_W_exist) THEN
1333 106 : bs_env%read_chi(:) = .FALSE.
1334 106 : bs_env%calc_chi(:) = .FALSE.
1335 : END IF
1336 :
1337 22 : bs_env%Sigma_x_exists = .TRUE.
1338 48 : DO i_spin = 1, n_spin
1339 26 : WRITE (f_S_x, '(3A,I1,A)') TRIM(prefix), bs_env%Sigma_x_name, "_0", i_spin, ".matrix"
1340 26 : INQUIRE (file=TRIM(f_S_x), exist=Sigma_x_spin_exists)
1341 66 : bs_env%Sigma_x_exists = bs_env%Sigma_x_exists .AND. Sigma_x_spin_exists
1342 : END DO
1343 :
1344 : ! If any restart files are read, check if the SCF converged in 1 step.
1345 : ! This is important because a re-iterated SCF can lead to spurious GW results
1346 : IF (ANY(bs_env%read_chi(:)) &
1347 : .OR. ANY(bs_env%Sigma_c_exists) &
1348 : .OR. bs_env%all_W_exist &
1349 608 : .OR. bs_env%Sigma_x_exists &
1350 : ) THEN
1351 :
1352 6 : IF (qs_env%scf_env%iter_count /= 1) THEN
1353 : CALL cp_warn(__LOCATION__, "SCF needed more than 1 step, "// &
1354 6 : "which might lead to spurious GW results when using GW restart files. ")
1355 : END IF
1356 : END IF
1357 :
1358 22 : CALL timestop(handle)
1359 :
1360 22 : END SUBROUTINE check_for_restart_files
1361 :
1362 : ! **************************************************************************************************
1363 : !> \brief ...
1364 : !> \param qs_env ...
1365 : !> \param bs_env ...
1366 : ! **************************************************************************************************
1367 28 : SUBROUTINE set_parallelization_parameters(qs_env, bs_env)
1368 : TYPE(qs_environment_type), POINTER :: qs_env
1369 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
1370 :
1371 : CHARACTER(LEN=*), PARAMETER :: routineN = 'set_parallelization_parameters'
1372 :
1373 : INTEGER :: color_sub, dummy_1, dummy_2, handle, &
1374 : num_pe, num_t_groups, u
1375 : TYPE(mp_para_env_type), POINTER :: para_env
1376 :
1377 28 : CALL timeset(routineN, handle)
1378 :
1379 28 : CALL get_qs_env(qs_env, para_env=para_env)
1380 :
1381 28 : num_pe = para_env%num_pe
1382 : ! if not already set, use all processors for the group (for large-cell GW, performance
1383 : ! seems to be best for a single group with all MPI processes per group)
1384 28 : IF (bs_env%group_size_tensor < 0 .OR. bs_env%group_size_tensor > num_pe) &
1385 22 : bs_env%group_size_tensor = num_pe
1386 :
1387 : ! group_size_tensor must divide num_pe without rest; otherwise everything will be complicated
1388 28 : IF (MODULO(num_pe, bs_env%group_size_tensor) /= 0) THEN
1389 0 : CALL find_good_group_size(num_pe, bs_env%group_size_tensor)
1390 : END IF
1391 :
1392 : ! para_env_tensor for tensor subgroups
1393 28 : color_sub = para_env%mepos/bs_env%group_size_tensor
1394 28 : bs_env%tensor_group_color = color_sub
1395 :
1396 28 : ALLOCATE (bs_env%para_env_tensor)
1397 28 : CALL bs_env%para_env_tensor%from_split(para_env, color_sub)
1398 :
1399 28 : num_t_groups = para_env%num_pe/bs_env%group_size_tensor
1400 28 : bs_env%num_tensor_groups = num_t_groups
1401 :
1402 : CALL get_i_j_atoms(bs_env%atoms_i, bs_env%atoms_j, bs_env%n_atom_i, bs_env%n_atom_j, &
1403 28 : color_sub, bs_env)
1404 :
1405 84 : ALLOCATE (bs_env%atoms_i_t_group(2, num_t_groups))
1406 84 : ALLOCATE (bs_env%atoms_j_t_group(2, num_t_groups))
1407 62 : DO color_sub = 0, num_t_groups - 1
1408 : CALL get_i_j_atoms(bs_env%atoms_i_t_group(1:2, color_sub + 1), &
1409 : bs_env%atoms_j_t_group(1:2, color_sub + 1), &
1410 62 : dummy_1, dummy_2, color_sub, bs_env)
1411 : END DO
1412 :
1413 28 : u = bs_env%unit_nr
1414 28 : IF (u > 0) THEN
1415 14 : WRITE (u, '(T2,A,I47)') 'Group size for tensor operations', bs_env%group_size_tensor
1416 14 : IF (bs_env%group_size_tensor > 1 .AND. bs_env%n_atom < 5) THEN
1417 11 : WRITE (u, '(T2,A)') 'The requested group size is > 1 which can lead to bad performance.'
1418 11 : WRITE (u, '(T2,A)') 'Using more memory per MPI process might improve performance.'
1419 11 : WRITE (u, '(T2,A)') '(Also increase MEMORY_PER_PROC when using more memory per process.)'
1420 : END IF
1421 : END IF
1422 :
1423 28 : CALL timestop(handle)
1424 :
1425 28 : END SUBROUTINE set_parallelization_parameters
1426 :
1427 : ! **************************************************************************************************
1428 : !> \brief ...
1429 : !> \param num_pe ...
1430 : !> \param group_size ...
1431 : ! **************************************************************************************************
1432 0 : SUBROUTINE find_good_group_size(num_pe, group_size)
1433 :
1434 : INTEGER :: num_pe, group_size
1435 :
1436 : CHARACTER(LEN=*), PARAMETER :: routineN = 'find_good_group_size'
1437 :
1438 : INTEGER :: group_size_minus, group_size_orig, &
1439 : group_size_plus, handle, i_diff
1440 :
1441 0 : CALL timeset(routineN, handle)
1442 :
1443 0 : group_size_orig = group_size
1444 :
1445 0 : DO i_diff = 1, num_pe
1446 :
1447 0 : group_size_minus = group_size - i_diff
1448 :
1449 0 : IF (MODULO(num_pe, group_size_minus) == 0 .AND. group_size_minus > 0) THEN
1450 0 : group_size = group_size_minus
1451 0 : EXIT
1452 : END IF
1453 :
1454 0 : group_size_plus = group_size + i_diff
1455 :
1456 0 : IF (MODULO(num_pe, group_size_plus) == 0 .AND. group_size_plus <= num_pe) THEN
1457 0 : group_size = group_size_plus
1458 0 : EXIT
1459 : END IF
1460 :
1461 : END DO
1462 :
1463 0 : IF (group_size_orig == group_size) CPABORT("Group size error")
1464 :
1465 0 : CALL timestop(handle)
1466 :
1467 0 : END SUBROUTINE find_good_group_size
1468 :
1469 : ! **************************************************************************************************
1470 : !> \brief ...
1471 : !> \param atoms_i ...
1472 : !> \param atoms_j ...
1473 : !> \param n_atom_i ...
1474 : !> \param n_atom_j ...
1475 : !> \param color_sub ...
1476 : !> \param bs_env ...
1477 : ! **************************************************************************************************
1478 62 : SUBROUTINE get_i_j_atoms(atoms_i, atoms_j, n_atom_i, n_atom_j, color_sub, bs_env)
1479 :
1480 : INTEGER, DIMENSION(2) :: atoms_i, atoms_j
1481 : INTEGER :: n_atom_i, n_atom_j, color_sub
1482 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
1483 :
1484 : CHARACTER(LEN=*), PARAMETER :: routineN = 'get_i_j_atoms'
1485 :
1486 : INTEGER :: handle, i_atoms_per_group, i_group, &
1487 : ipcol, ipcol_loop, iprow, iprow_loop, &
1488 : j_atoms_per_group, npcol, nprow
1489 :
1490 62 : CALL timeset(routineN, handle)
1491 :
1492 : ! create a square mesh of tensor groups for iatom and jatom; code from blacs_env_create
1493 62 : CALL square_mesh(nprow, npcol, bs_env%num_tensor_groups)
1494 :
1495 62 : i_group = 0
1496 124 : DO ipcol_loop = 0, npcol - 1
1497 204 : DO iprow_loop = 0, nprow - 1
1498 80 : IF (i_group == color_sub) THEN
1499 62 : iprow = iprow_loop
1500 62 : ipcol = ipcol_loop
1501 : END IF
1502 142 : i_group = i_group + 1
1503 : END DO
1504 : END DO
1505 :
1506 62 : IF (MODULO(bs_env%n_atom, nprow) == 0) THEN
1507 50 : i_atoms_per_group = bs_env%n_atom/nprow
1508 : ELSE
1509 12 : i_atoms_per_group = bs_env%n_atom/nprow + 1
1510 : END IF
1511 :
1512 62 : IF (MODULO(bs_env%n_atom, npcol) == 0) THEN
1513 62 : j_atoms_per_group = bs_env%n_atom/npcol
1514 : ELSE
1515 0 : j_atoms_per_group = bs_env%n_atom/npcol + 1
1516 : END IF
1517 :
1518 62 : atoms_i(1) = iprow*i_atoms_per_group + 1
1519 62 : atoms_i(2) = MIN((iprow + 1)*i_atoms_per_group, bs_env%n_atom)
1520 62 : n_atom_i = atoms_i(2) - atoms_i(1) + 1
1521 :
1522 62 : atoms_j(1) = ipcol*j_atoms_per_group + 1
1523 62 : atoms_j(2) = MIN((ipcol + 1)*j_atoms_per_group, bs_env%n_atom)
1524 62 : n_atom_j = atoms_j(2) - atoms_j(1) + 1
1525 :
1526 62 : CALL timestop(handle)
1527 :
1528 62 : END SUBROUTINE get_i_j_atoms
1529 :
1530 : ! **************************************************************************************************
1531 : !> \brief ...
1532 : !> \param nprow ...
1533 : !> \param npcol ...
1534 : !> \param nproc ...
1535 : ! **************************************************************************************************
1536 62 : SUBROUTINE square_mesh(nprow, npcol, nproc)
1537 : INTEGER :: nprow, npcol, nproc
1538 :
1539 : CHARACTER(LEN=*), PARAMETER :: routineN = 'square_mesh'
1540 :
1541 : INTEGER :: gcd_max, handle, ipe, jpe
1542 :
1543 62 : CALL timeset(routineN, handle)
1544 :
1545 62 : gcd_max = -1
1546 142 : DO ipe = 1, CEILING(SQRT(REAL(nproc, dp)))
1547 80 : jpe = nproc/ipe
1548 80 : IF (ipe*jpe /= nproc) CYCLE
1549 142 : IF (gcd(ipe, jpe) >= gcd_max) THEN
1550 80 : nprow = ipe
1551 80 : npcol = jpe
1552 80 : gcd_max = gcd(ipe, jpe)
1553 : END IF
1554 : END DO
1555 :
1556 62 : CALL timestop(handle)
1557 :
1558 62 : END SUBROUTINE square_mesh
1559 :
1560 : ! **************************************************************************************************
1561 : !> \brief ...
1562 : !> \param bs_env ...
1563 : !> \param qs_env ...
1564 : ! **************************************************************************************************
1565 28 : SUBROUTINE set_heuristic_parameters(bs_env, qs_env)
1566 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
1567 : TYPE(qs_environment_type), OPTIONAL, POINTER :: qs_env
1568 :
1569 : CHARACTER(LEN=*), PARAMETER :: routineN = 'set_heuristic_parameters'
1570 :
1571 : INTEGER :: handle, u
1572 : LOGICAL :: do_BvK_cell
1573 :
1574 28 : CALL timeset(routineN, handle)
1575 :
1576 : ! for generating numerically stable minimax Fourier integration weights
1577 28 : bs_env%num_points_per_magnitude = 200
1578 :
1579 28 : IF (bs_env%input_regularization_minimax > -1.0E-12_dp) THEN
1580 0 : bs_env%regularization_minimax = bs_env%input_regularization_minimax
1581 : ELSE
1582 : ! for periodic systems and for 20 minimax points, we use a regularized minimax mesh
1583 : ! (from experience: regularized minimax meshes converges faster for periodic systems
1584 : ! and for 20 pts)
1585 112 : IF (SUM(bs_env%periodic) /= 0 .OR. bs_env%num_time_freq_points >= 20) THEN
1586 28 : bs_env%regularization_minimax = 1.0E-6_dp
1587 : ELSE
1588 0 : bs_env%regularization_minimax = 0.0_dp
1589 : END IF
1590 : END IF
1591 :
1592 28 : bs_env%stabilize_exp = 70.0_dp
1593 28 : bs_env%eps_atom_grid_2d_mat = 1.0E-50_dp
1594 :
1595 : ! use a 16-parameter Padé fit
1596 28 : bs_env%nparam_pade = 16
1597 :
1598 : ! resolution of the identity with the truncated Coulomb metric, cutoff radius 3 Angström
1599 28 : bs_env%ri_metric%potential_type = do_potential_truncated
1600 28 : bs_env%ri_metric%omega = 0.0_dp
1601 : ! cutoff radius is specified in the input
1602 28 : bs_env%ri_metric%filename = "t_c_g.dat"
1603 :
1604 28 : bs_env%eps_eigval_mat_RI = 0.0_dp
1605 :
1606 28 : IF (bs_env%input_regularization_RI > -1.0E-12_dp) THEN
1607 0 : bs_env%regularization_RI = bs_env%input_regularization_RI
1608 : ELSE
1609 : ! default case:
1610 :
1611 : ! 1. for periodic systems, we use the regularized resolution of the identity per default
1612 28 : bs_env%regularization_RI = 1.0E-2_dp
1613 :
1614 : ! 2. for molecules, no regularization is necessary
1615 112 : IF (SUM(bs_env%periodic) == 0) bs_env%regularization_RI = 0.0_dp
1616 :
1617 : END IF
1618 :
1619 : ! truncated Coulomb operator for exchange self-energy
1620 : ! (see details in Guidon, VandeVondele, Hutter, JCTC 5, 3010 (2009) and references therein)
1621 28 : do_BvK_cell = bs_env%small_cell_full_kp_or_large_cell_Gamma == small_cell_full_kp
1622 : CALL trunc_coulomb_for_exchange(qs_env, bs_env%trunc_coulomb, &
1623 : rel_cutoff_trunc_coulomb_ri_x=0.5_dp, &
1624 : cell_grid=bs_env%cell_grid_scf_desymm, &
1625 28 : do_BvK_cell=do_BvK_cell)
1626 :
1627 : ! for small-cell GW, we need more cells than normally used by the filter bs_env%eps_filter
1628 : ! (in particular for computing the self-energy because of higher number of cells needed)
1629 28 : bs_env%heuristic_filter_factor = 1.0E-4
1630 :
1631 28 : u = bs_env%unit_nr
1632 28 : IF (u > 0) THEN
1633 14 : WRITE (u, FMT="(T2,2A,F21.1,A)") "Cutoff radius for the truncated Coulomb ", &
1634 28 : "operator in Σ^x:", bs_env%trunc_coulomb%cutoff_radius*angstrom, " Å"
1635 14 : WRITE (u, FMT="(T2,2A,F15.1,A)") "Cutoff radius for the truncated Coulomb ", &
1636 28 : "operator in RI metric:", bs_env%ri_metric%cutoff_radius*angstrom, " Å"
1637 14 : WRITE (u, FMT="(T2,A,ES48.1)") "Regularization parameter of RI ", bs_env%regularization_RI
1638 14 : WRITE (u, FMT="(T2,A,ES38.1)") "Regularization parameter of minimax grids", &
1639 28 : bs_env%regularization_minimax
1640 14 : WRITE (u, FMT="(T2,A,I53)") "Lattice sum size for V(k):", bs_env%size_lattice_sum_V
1641 : END IF
1642 :
1643 28 : CALL timestop(handle)
1644 :
1645 28 : END SUBROUTINE set_heuristic_parameters
1646 :
1647 : ! **************************************************************************************************
1648 : !> \brief ...
1649 : !> \param bs_env ...
1650 : ! **************************************************************************************************
1651 28 : SUBROUTINE print_header_and_input_parameters(bs_env)
1652 :
1653 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
1654 :
1655 : CHARACTER(LEN=*), PARAMETER :: routineN = 'print_header_and_input_parameters'
1656 :
1657 : INTEGER :: handle, u
1658 :
1659 28 : CALL timeset(routineN, handle)
1660 :
1661 28 : u = bs_env%unit_nr
1662 :
1663 28 : IF (u > 0) THEN
1664 14 : WRITE (u, '(T2,A)') ' '
1665 14 : WRITE (u, '(T2,A)') REPEAT('-', 79)
1666 14 : WRITE (u, '(T2,A,A78)') '-', '-'
1667 14 : WRITE (u, '(T2,A,A46,A32)') '-', 'GW CALCULATION', '-'
1668 14 : WRITE (u, '(T2,A,A78)') '-', '-'
1669 14 : WRITE (u, '(T2,A)') REPEAT('-', 79)
1670 14 : WRITE (u, '(T2,A)') ' '
1671 14 : WRITE (u, '(T2,A,I45)') 'Input: Number of time/freq. points', bs_env%num_time_freq_points
1672 14 : WRITE (u, "(T2,A,F44.1,A)") 'Input: ω_max for fitting Σ(iω) (eV)', bs_env%freq_max_fit*evolt
1673 14 : WRITE (u, '(T2,A,ES27.1)') 'Input: Filter threshold for sparse tensor operations', &
1674 28 : bs_env%eps_filter
1675 14 : WRITE (u, "(T2,A,L55)") 'Input: Apply Hedin shift', bs_env%do_hedin_shift
1676 14 : WRITE (u, '(T2,A,F37.1,A)') 'Input: Available memory per MPI process', &
1677 28 : bs_env%input_memory_per_proc_GB, ' GB'
1678 : END IF
1679 :
1680 28 : CALL timestop(handle)
1681 :
1682 28 : END SUBROUTINE print_header_and_input_parameters
1683 :
1684 : ! **************************************************************************************************
1685 : !> \brief ...
1686 : !> \param qs_env ...
1687 : !> \param bs_env ...
1688 : ! **************************************************************************************************
1689 56 : SUBROUTINE compute_V_xc(qs_env, bs_env)
1690 : TYPE(qs_environment_type), POINTER :: qs_env
1691 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
1692 :
1693 : CHARACTER(LEN=*), PARAMETER :: routineN = 'compute_V_xc'
1694 :
1695 : INTEGER :: handle, img, ispin, myfun, nimages
1696 : LOGICAL :: hf_present
1697 : REAL(KIND=dp) :: energy_ex, energy_exc, energy_total, &
1698 : myfraction
1699 28 : TYPE(dbcsr_p_type), DIMENSION(:), POINTER :: mat_ks_without_v_xc
1700 28 : TYPE(dbcsr_p_type), DIMENSION(:, :), POINTER :: matrix_ks_kp
1701 : TYPE(dft_control_type), POINTER :: dft_control
1702 : TYPE(qs_energy_type), POINTER :: energy
1703 : TYPE(section_vals_type), POINTER :: hf_section, input, xc_section
1704 :
1705 28 : CALL timeset(routineN, handle)
1706 :
1707 28 : CALL get_qs_env(qs_env, input=input, energy=energy, dft_control=dft_control)
1708 :
1709 : ! previously, dft_control%nimages set to # neighbor cells, revert for Γ-only KS matrix
1710 28 : nimages = dft_control%nimages
1711 28 : dft_control%nimages = bs_env%nimages_scf
1712 :
1713 : ! we need to reset XC functional, therefore, get XC input
1714 28 : xc_section => section_vals_get_subs_vals(input, "DFT%XC")
1715 28 : CALL section_vals_val_get(xc_section, "XC_FUNCTIONAL%_SECTION_PARAMETERS_", i_val=myfun)
1716 28 : CALL section_vals_val_set(xc_section, "XC_FUNCTIONAL%_SECTION_PARAMETERS_", i_val=xc_none)
1717 : ! IF (ASSOCIATED(section_vals_get_subs_vals(xc_section, "HF", can_return_null=.TRUE.))) THEN
1718 28 : hf_section => section_vals_get_subs_vals(input, "DFT%XC%HF", can_return_null=.TRUE.)
1719 28 : hf_present = .FALSE.
1720 28 : IF (ASSOCIATED(hf_section)) THEN
1721 28 : CALL section_vals_get(hf_section, explicit=hf_present)
1722 : END IF
1723 28 : IF (hf_present) THEN
1724 : ! Special case for handling hfx
1725 0 : CALL section_vals_val_get(xc_section, "HF%FRACTION", r_val=myfraction)
1726 0 : CALL section_vals_val_set(xc_section, "HF%FRACTION", r_val=0.0_dp)
1727 : END IF
1728 :
1729 : ! save the energy before the energy gets updated
1730 28 : energy_total = energy%total
1731 28 : energy_exc = energy%exc
1732 28 : energy_ex = energy%ex
1733 :
1734 50 : SELECT CASE (bs_env%small_cell_full_kp_or_large_cell_Gamma)
1735 : CASE (large_cell_Gamma)
1736 :
1737 22 : NULLIFY (mat_ks_without_v_xc)
1738 22 : CALL dbcsr_allocate_matrix_set(mat_ks_without_v_xc, bs_env%n_spin)
1739 :
1740 48 : DO ispin = 1, bs_env%n_spin
1741 26 : ALLOCATE (mat_ks_without_v_xc(ispin)%matrix)
1742 48 : IF (hf_present) THEN
1743 : CALL dbcsr_create(mat_ks_without_v_xc(ispin)%matrix, template=bs_env%mat_ao_ao%matrix, &
1744 0 : matrix_type=dbcsr_type_symmetric)
1745 : ELSE
1746 26 : CALL dbcsr_create(mat_ks_without_v_xc(ispin)%matrix, template=bs_env%mat_ao_ao%matrix)
1747 : END IF
1748 : END DO
1749 :
1750 : ! calculate KS-matrix without XC
1751 : CALL qs_ks_build_kohn_sham_matrix(qs_env, calculate_forces=.FALSE., just_energy=.FALSE., &
1752 22 : ext_ks_matrix=mat_ks_without_v_xc)
1753 :
1754 48 : DO ispin = 1, bs_env%n_spin
1755 : ! transfer dbcsr matrix to fm
1756 26 : CALL cp_fm_create(bs_env%fm_V_xc_Gamma(ispin), bs_env%fm_s_Gamma%matrix_struct)
1757 26 : CALL copy_dbcsr_to_fm(mat_ks_without_v_xc(ispin)%matrix, bs_env%fm_V_xc_Gamma(ispin))
1758 :
1759 : ! v_xc = h_ks - h_ks(v_xc = 0)
1760 : CALL cp_fm_scale_and_add(alpha=-1.0_dp, matrix_a=bs_env%fm_V_xc_Gamma(ispin), &
1761 48 : beta=1.0_dp, matrix_b=bs_env%fm_ks_Gamma(ispin))
1762 : END DO
1763 :
1764 22 : CALL dbcsr_deallocate_matrix_set(mat_ks_without_v_xc)
1765 :
1766 : CASE (small_cell_full_kp)
1767 :
1768 : ! calculate KS-matrix without XC
1769 6 : CALL qs_ks_build_kohn_sham_matrix(qs_env, calculate_forces=.FALSE., just_energy=.FALSE.)
1770 6 : CALL get_qs_env(qs_env=qs_env, matrix_ks_kp=matrix_ks_kp)
1771 :
1772 208 : ALLOCATE (bs_env%fm_V_xc_R(dft_control%nimages, bs_env%n_spin))
1773 40 : DO ispin = 1, bs_env%n_spin
1774 190 : DO img = 1, dft_control%nimages
1775 : ! safe fm_V_xc_R in fm_matrix because saving in dbcsr matrix caused trouble...
1776 178 : CALL copy_dbcsr_to_fm(matrix_ks_kp(ispin, img)%matrix, bs_env%fm_work_mo(1))
1777 : CALL cp_fm_create(bs_env%fm_V_xc_R(img, ispin), bs_env%fm_work_mo(1)%matrix_struct, &
1778 178 : set_zero=.TRUE.)
1779 : ! store h_ks(v_xc = 0) in fm_V_xc_R
1780 : CALL cp_fm_scale_and_add(alpha=1.0_dp, matrix_a=bs_env%fm_V_xc_R(img, ispin), &
1781 184 : beta=1.0_dp, matrix_b=bs_env%fm_work_mo(1))
1782 : END DO
1783 : END DO
1784 :
1785 : END SELECT
1786 :
1787 : ! set back the energy
1788 28 : energy%total = energy_total
1789 28 : energy%exc = energy_exc
1790 28 : energy%ex = energy_ex
1791 :
1792 : ! set back nimages
1793 28 : dft_control%nimages = nimages
1794 :
1795 : ! set the DFT functional and HF fraction back
1796 : CALL section_vals_val_set(xc_section, "XC_FUNCTIONAL%_SECTION_PARAMETERS_", &
1797 28 : i_val=myfun)
1798 28 : IF (hf_present) THEN
1799 : CALL section_vals_val_set(xc_section, "HF%FRACTION", &
1800 0 : r_val=myfraction)
1801 : END IF
1802 :
1803 28 : IF (bs_env%small_cell_full_kp_or_large_cell_Gamma == small_cell_full_kp) THEN
1804 : ! calculate KS-matrix again with XC
1805 6 : CALL qs_ks_build_kohn_sham_matrix(qs_env, calculate_forces=.FALSE., just_energy=.FALSE.)
1806 12 : DO ispin = 1, bs_env%n_spin
1807 190 : DO img = 1, dft_control%nimages
1808 : ! store h_ks in fm_work_mo
1809 178 : CALL copy_dbcsr_to_fm(matrix_ks_kp(ispin, img)%matrix, bs_env%fm_work_mo(1))
1810 : ! v_xc = h_ks - h_ks(v_xc = 0)
1811 : CALL cp_fm_scale_and_add(alpha=-1.0_dp, matrix_a=bs_env%fm_V_xc_R(img, ispin), &
1812 184 : beta=1.0_dp, matrix_b=bs_env%fm_work_mo(1))
1813 : END DO
1814 : END DO
1815 : END IF
1816 :
1817 28 : CALL timestop(handle)
1818 :
1819 28 : END SUBROUTINE compute_V_xc
1820 :
1821 : ! **************************************************************************************************
1822 : !> \brief ...
1823 : !> \param bs_env ...
1824 : ! **************************************************************************************************
1825 28 : SUBROUTINE setup_time_and_frequency_minimax_grid(bs_env)
1826 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
1827 :
1828 : CHARACTER(LEN=*), PARAMETER :: routineN = 'setup_time_and_frequency_minimax_grid'
1829 :
1830 : INTEGER :: handle, homo, i_w, ierr, ispin, j_w, &
1831 : n_mo, num_time_freq_points, u
1832 : REAL(KIND=dp) :: E_max, E_max_ispin, E_min, E_min_ispin, &
1833 : E_range, max_error_min
1834 28 : REAL(KIND=dp), ALLOCATABLE, DIMENSION(:) :: points_and_weights
1835 :
1836 28 : CALL timeset(routineN, handle)
1837 :
1838 28 : n_mo = bs_env%n_ao
1839 28 : num_time_freq_points = bs_env%num_time_freq_points
1840 :
1841 84 : ALLOCATE (bs_env%imag_freq_points(num_time_freq_points))
1842 84 : ALLOCATE (bs_env%imag_time_points(num_time_freq_points))
1843 84 : ALLOCATE (bs_env%imag_time_weights_freq_zero(num_time_freq_points))
1844 112 : ALLOCATE (bs_env%weights_cos_t_to_w(num_time_freq_points, num_time_freq_points))
1845 112 : ALLOCATE (bs_env%weights_cos_w_to_t(num_time_freq_points, num_time_freq_points))
1846 112 : ALLOCATE (bs_env%weights_sin_t_to_w(num_time_freq_points, num_time_freq_points))
1847 :
1848 : ! minimum and maximum difference between eigenvalues of unoccupied and an occupied MOs
1849 28 : E_min = 1000.0_dp
1850 28 : E_max = -1000.0_dp
1851 60 : DO ispin = 1, bs_env%n_spin
1852 32 : homo = bs_env%n_occ(ispin)
1853 58 : SELECT CASE (bs_env%small_cell_full_kp_or_large_cell_Gamma)
1854 : CASE (large_cell_Gamma)
1855 : E_min_ispin = bs_env%eigenval_scf_Gamma(homo + 1, ispin) - &
1856 26 : bs_env%eigenval_scf_Gamma(homo, ispin)
1857 : E_max_ispin = bs_env%eigenval_scf_Gamma(n_mo, ispin) - &
1858 26 : bs_env%eigenval_scf_Gamma(1, ispin)
1859 : CASE (small_cell_full_kp)
1860 : E_min_ispin = MINVAL(bs_env%eigenval_scf(homo + 1, :, ispin)) - &
1861 334 : MAXVAL(bs_env%eigenval_scf(homo, :, ispin))
1862 : E_max_ispin = MAXVAL(bs_env%eigenval_scf(n_mo, :, ispin)) - &
1863 366 : MINVAL(bs_env%eigenval_scf(1, :, ispin))
1864 : END SELECT
1865 32 : E_min = MIN(E_min, E_min_ispin)
1866 60 : E_max = MAX(E_max, E_max_ispin)
1867 : END DO
1868 :
1869 28 : E_range = E_max/E_min
1870 :
1871 84 : ALLOCATE (points_and_weights(2*num_time_freq_points))
1872 :
1873 : ! frequency points
1874 28 : IF (num_time_freq_points <= 20) THEN
1875 28 : CALL get_rpa_minimax_coeff(num_time_freq_points, E_range, points_and_weights, ierr, .FALSE.)
1876 : ELSE
1877 0 : CALL get_rpa_minimax_coeff_larger_grid(num_time_freq_points, E_range, points_and_weights)
1878 : END IF
1879 :
1880 : ! one needs to scale the minimax grids, see Azizi, Wilhelm, Golze, Panades-Barrueta,
1881 : ! Giantomassi, Rinke, Draxl, Gonze et al., 2 publications
1882 396 : bs_env%imag_freq_points(:) = points_and_weights(1:num_time_freq_points)*E_min
1883 :
1884 : ! determine number of fit points in the interval [0,ω_max] for virt, or [-ω_max,0] for occ
1885 28 : bs_env%num_freq_points_fit = 0
1886 396 : DO i_w = 1, num_time_freq_points
1887 396 : IF (bs_env%imag_freq_points(i_w) < bs_env%freq_max_fit) THEN
1888 126 : bs_env%num_freq_points_fit = bs_env%num_freq_points_fit + 1
1889 : END IF
1890 : END DO
1891 :
1892 : ! iω values for the analytic continuation Σ^c_n(iω,k) -> Σ^c_n(ϵ,k)
1893 84 : ALLOCATE (bs_env%imag_freq_points_fit(bs_env%num_freq_points_fit))
1894 28 : j_w = 0
1895 396 : DO i_w = 1, num_time_freq_points
1896 396 : IF (bs_env%imag_freq_points(i_w) < bs_env%freq_max_fit) THEN
1897 126 : j_w = j_w + 1
1898 126 : bs_env%imag_freq_points_fit(j_w) = bs_env%imag_freq_points(i_w)
1899 : END IF
1900 : END DO
1901 :
1902 : ! reset the number of Padé parameters if smaller than the number of
1903 : ! imaginary-frequency points for the fit
1904 28 : IF (bs_env%num_freq_points_fit < bs_env%nparam_pade) THEN
1905 28 : bs_env%nparam_pade = bs_env%num_freq_points_fit
1906 : END IF
1907 :
1908 : ! time points
1909 28 : IF (num_time_freq_points <= 20) THEN
1910 28 : CALL get_exp_minimax_coeff(num_time_freq_points, E_range, points_and_weights)
1911 : ELSE
1912 0 : CALL get_exp_minimax_coeff_gw(num_time_freq_points, E_range, points_and_weights)
1913 : END IF
1914 :
1915 396 : bs_env%imag_time_points(:) = points_and_weights(1:num_time_freq_points)/(2.0_dp*E_min)
1916 396 : bs_env%imag_time_weights_freq_zero(:) = points_and_weights(num_time_freq_points + 1:)/(E_min)
1917 :
1918 28 : DEALLOCATE (points_and_weights)
1919 :
1920 28 : u = bs_env%unit_nr
1921 28 : IF (u > 0) THEN
1922 14 : WRITE (u, '(T2,A)') ''
1923 14 : WRITE (u, '(T2,A,F55.2)') 'SCF direct band gap (eV)', E_min*evolt
1924 14 : WRITE (u, '(T2,A,F53.2)') 'Max. SCF eigval diff. (eV)', E_max*evolt
1925 14 : WRITE (u, '(T2,A,F55.2)') 'E-Range for minimax grid', E_range
1926 14 : WRITE (u, '(T2,A,I27)') 'Number of Padé parameters for analytic continuation:', &
1927 28 : bs_env%nparam_pade
1928 14 : WRITE (u, '(T2,A)') ''
1929 : END IF
1930 :
1931 : ! in minimax grids, Fourier transforms t -> w and w -> t are split using
1932 : ! e^(iwt) = cos(wt) + i sin(wt); we thus calculate weights for trafos with a cos and
1933 : ! sine prefactor; details in Azizi, Wilhelm, Golze, Giantomassi, Panades-Barrueta,
1934 : ! Rinke, Draxl, Gonze et al., 2 publications
1935 :
1936 : ! cosine transform weights imaginary time to imaginary frequency
1937 : CALL get_l_sq_wghts_cos_tf_t_to_w(num_time_freq_points, &
1938 : bs_env%imag_time_points, &
1939 : bs_env%weights_cos_t_to_w, &
1940 : bs_env%imag_freq_points, &
1941 : E_min, E_max, max_error_min, &
1942 : bs_env%num_points_per_magnitude, &
1943 28 : bs_env%regularization_minimax)
1944 :
1945 : ! cosine transform weights imaginary frequency to imaginary time
1946 : CALL get_l_sq_wghts_cos_tf_w_to_t(num_time_freq_points, &
1947 : bs_env%imag_time_points, &
1948 : bs_env%weights_cos_w_to_t, &
1949 : bs_env%imag_freq_points, &
1950 : E_min, E_max, max_error_min, &
1951 : bs_env%num_points_per_magnitude, &
1952 28 : bs_env%regularization_minimax)
1953 :
1954 : ! sine transform weights imaginary time to imaginary frequency
1955 : CALL get_l_sq_wghts_sin_tf_t_to_w(num_time_freq_points, &
1956 : bs_env%imag_time_points, &
1957 : bs_env%weights_sin_t_to_w, &
1958 : bs_env%imag_freq_points, &
1959 : E_min, E_max, max_error_min, &
1960 : bs_env%num_points_per_magnitude, &
1961 28 : bs_env%regularization_minimax)
1962 :
1963 28 : CALL timestop(handle)
1964 :
1965 56 : END SUBROUTINE setup_time_and_frequency_minimax_grid
1966 :
1967 : ! **************************************************************************************************
1968 : !> \brief ...
1969 : !> \param qs_env ...
1970 : !> \param bs_env ...
1971 : ! **************************************************************************************************
1972 6 : SUBROUTINE setup_cells_3c(qs_env, bs_env)
1973 :
1974 : TYPE(qs_environment_type), POINTER :: qs_env
1975 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
1976 :
1977 : CHARACTER(LEN=*), PARAMETER :: routineN = 'setup_cells_3c'
1978 :
1979 : INTEGER :: atom_i, atom_j, atom_k, block_count, handle, i, i_cell_x, i_cell_x_max, &
1980 : i_cell_x_min, i_size, ikind, img, j, j_cell, j_cell_max, j_cell_y, j_cell_y_max, &
1981 : j_cell_y_min, j_size, k_cell, k_cell_max, k_cell_z, k_cell_z_max, k_cell_z_min, k_size, &
1982 : nimage_pairs_3c, nimages_3c, nimages_3c_max, nkind, u
1983 : INTEGER(KIND=int_8) :: mem_occ_per_proc
1984 6 : INTEGER, ALLOCATABLE, DIMENSION(:) :: kind_of, n_other_3c_images_max
1985 6 : INTEGER, ALLOCATABLE, DIMENSION(:, :) :: index_to_cell_3c_max, nblocks_3c_max
1986 : INTEGER, DIMENSION(3) :: cell_index, n_max
1987 : REAL(KIND=dp) :: avail_mem_per_proc_GB, cell_dist, cell_radius_3c, dij, dik, djk, eps, &
1988 : exp_min_ao, exp_min_RI, frobenius_norm, mem_3c_GB, mem_occ_per_proc_GB, radius_ao, &
1989 : radius_ao_product, radius_RI
1990 6 : REAL(KIND=dp), ALLOCATABLE, DIMENSION(:) :: exp_ao_kind, exp_RI_kind, &
1991 6 : radius_ao_kind, &
1992 6 : radius_ao_product_kind, radius_RI_kind
1993 6 : REAL(KIND=dp), ALLOCATABLE, DIMENSION(:, :, :) :: int_3c
1994 : REAL(KIND=dp), DIMENSION(3) :: rij, rik, rjk, vec_cell_j, vec_cell_k
1995 6 : REAL(KIND=dp), DIMENSION(:, :), POINTER :: exp_ao, exp_RI
1996 6 : TYPE(atomic_kind_type), DIMENSION(:), POINTER :: atomic_kind_set
1997 : TYPE(cell_type), POINTER :: cell
1998 6 : TYPE(particle_type), DIMENSION(:), POINTER :: particle_set
1999 :
2000 6 : CALL timeset(routineN, handle)
2001 :
2002 6 : CALL get_qs_env(qs_env, nkind=nkind, atomic_kind_set=atomic_kind_set, particle_set=particle_set, cell=cell)
2003 :
2004 : ALLOCATE (exp_ao_kind(nkind), exp_RI_kind(nkind), radius_ao_kind(nkind), &
2005 42 : radius_ao_product_kind(nkind), radius_RI_kind(nkind))
2006 :
2007 18 : exp_min_RI = 10.0_dp
2008 18 : exp_min_ao = 10.0_dp
2009 18 : exp_RI_kind = 10.0_dp
2010 18 : exp_AO_kind = 10.0_dp
2011 :
2012 6 : eps = bs_env%eps_filter*bs_env%heuristic_filter_factor
2013 :
2014 18 : DO ikind = 1, nkind
2015 :
2016 12 : CALL get_gto_basis_set(bs_env%basis_set_RI(ikind)%gto_basis_set, zet=exp_RI)
2017 12 : CALL get_gto_basis_set(bs_env%basis_set_ao(ikind)%gto_basis_set, zet=exp_ao)
2018 :
2019 : ! we need to remove all exponents lower than a lower bound, e.g. 1E-3, because
2020 : ! for contracted basis sets, there might be exponents = 0 in zet
2021 24 : DO i = 1, SIZE(exp_RI, 1)
2022 42 : DO j = 1, SIZE(exp_RI, 2)
2023 18 : IF (exp_RI(i, j) < exp_min_RI .AND. exp_RI(i, j) > 1E-3_dp) exp_min_RI = exp_RI(i, j)
2024 18 : IF (exp_RI(i, j) < exp_RI_kind(ikind) .AND. exp_RI(i, j) > 1E-3_dp) &
2025 24 : exp_RI_kind(ikind) = exp_RI(i, j)
2026 : END DO
2027 : END DO
2028 60 : DO i = 1, SIZE(exp_ao, 1)
2029 144 : DO j = 1, SIZE(exp_ao, 2)
2030 84 : IF (exp_ao(i, j) < exp_min_ao .AND. exp_ao(i, j) > 1E-3_dp) exp_min_ao = exp_ao(i, j)
2031 84 : IF (exp_ao(i, j) < exp_ao_kind(ikind) .AND. exp_ao(i, j) > 1E-3_dp) &
2032 84 : exp_ao_kind(ikind) = exp_ao(i, j)
2033 : END DO
2034 : END DO
2035 12 : radius_ao_kind(ikind) = SQRT(-LOG(eps)/exp_ao_kind(ikind))
2036 12 : radius_ao_product_kind(ikind) = SQRT(-LOG(eps)/(2.0_dp*exp_ao_kind(ikind)))
2037 18 : radius_RI_kind(ikind) = SQRT(-LOG(eps)/exp_RI_kind(ikind))
2038 : END DO
2039 :
2040 6 : radius_ao = SQRT(-LOG(eps)/exp_min_ao)
2041 6 : radius_ao_product = SQRT(-LOG(eps)/(2.0_dp*exp_min_ao))
2042 6 : radius_RI = SQRT(-LOG(eps)/exp_min_RI)
2043 :
2044 6 : CALL get_atomic_kind_set(atomic_kind_set=atomic_kind_set, kind_of=kind_of)
2045 :
2046 : ! For a 3c integral (μR υS | P0) we have that cell R and cell S need to be within radius_3c
2047 6 : cell_radius_3c = radius_ao_product + radius_RI + bs_env%ri_metric%cutoff_radius
2048 :
2049 24 : n_max(1:3) = bs_env%periodic(1:3)*30
2050 :
2051 6 : nimages_3c_max = 0
2052 :
2053 6 : i_cell_x_min = 0
2054 6 : i_cell_x_max = 0
2055 6 : j_cell_y_min = 0
2056 6 : j_cell_y_max = 0
2057 6 : k_cell_z_min = 0
2058 6 : k_cell_z_max = 0
2059 :
2060 132 : DO i_cell_x = -n_max(1), n_max(1)
2061 7818 : DO j_cell_y = -n_max(2), n_max(2)
2062 30138 : DO k_cell_z = -n_max(3), n_max(3)
2063 :
2064 89304 : cell_index(1:3) = [i_cell_x, j_cell_y, k_cell_z]
2065 :
2066 22326 : CALL get_cell_dist(cell_index, bs_env%hmat, cell_dist)
2067 :
2068 30012 : IF (cell_dist < cell_radius_3c) THEN
2069 142 : nimages_3c_max = nimages_3c_max + 1
2070 142 : i_cell_x_min = MIN(i_cell_x_min, i_cell_x)
2071 142 : i_cell_x_max = MAX(i_cell_x_max, i_cell_x)
2072 142 : j_cell_y_min = MIN(j_cell_y_min, j_cell_y)
2073 142 : j_cell_y_max = MAX(j_cell_y_max, j_cell_y)
2074 142 : k_cell_z_min = MIN(k_cell_z_min, k_cell_z)
2075 142 : k_cell_z_max = MAX(k_cell_z_max, k_cell_z)
2076 : END IF
2077 :
2078 : END DO
2079 : END DO
2080 : END DO
2081 :
2082 : ! get index_to_cell_3c_max for the maximum possible cell range;
2083 : ! compute 3c integrals later in this routine and check really which cell is needed
2084 18 : ALLOCATE (index_to_cell_3c_max(3, nimages_3c_max))
2085 :
2086 6 : img = 0
2087 132 : DO i_cell_x = -n_max(1), n_max(1)
2088 7818 : DO j_cell_y = -n_max(2), n_max(2)
2089 30138 : DO k_cell_z = -n_max(3), n_max(3)
2090 :
2091 89304 : cell_index(1:3) = [i_cell_x, j_cell_y, k_cell_z]
2092 :
2093 22326 : CALL get_cell_dist(cell_index, bs_env%hmat, cell_dist)
2094 :
2095 30012 : IF (cell_dist < cell_radius_3c) THEN
2096 142 : img = img + 1
2097 568 : index_to_cell_3c_max(1:3, img) = cell_index(1:3)
2098 : END IF
2099 :
2100 : END DO
2101 : END DO
2102 : END DO
2103 :
2104 : ! get pairs of R and S which have non-zero 3c integral (μR υS | P0)
2105 24 : ALLOCATE (nblocks_3c_max(nimages_3c_max, nimages_3c_max))
2106 3530 : nblocks_3c_max(:, :) = 0
2107 :
2108 : block_count = 0
2109 148 : DO j_cell = 1, nimages_3c_max
2110 3530 : DO k_cell = 1, nimages_3c_max
2111 :
2112 12788 : DO atom_j = 1, bs_env%n_atom
2113 38674 : DO atom_k = 1, bs_env%n_atom
2114 109848 : DO atom_i = 1, bs_env%n_atom
2115 :
2116 74556 : block_count = block_count + 1
2117 74556 : IF (MODULO(block_count, bs_env%para_env%num_pe) /= bs_env%para_env%mepos) CYCLE
2118 :
2119 149112 : CALL scaled_to_real(vec_cell_j, REAL(index_to_cell_3c_max(1:3, j_cell), kind=dp), cell)
2120 149112 : CALL scaled_to_real(vec_cell_k, REAL(index_to_cell_3c_max(1:3, k_cell), kind=dp), cell)
2121 :
2122 149112 : rij = pbc(particle_set(atom_j)%r(:), cell) - pbc(particle_set(atom_i)%r(:), cell) + vec_cell_j(:)
2123 : rjk = pbc(particle_set(atom_k)%r(:), cell) - pbc(particle_set(atom_j)%r(:), cell) &
2124 149112 : + vec_cell_k(:) - vec_cell_j(:)
2125 149112 : rik(:) = rij(:) + rjk(:)
2126 149112 : dij = NORM2(rij)
2127 149112 : dik = NORM2(rik)
2128 149112 : djk = NORM2(rjk)
2129 37278 : IF (djk > radius_ao_kind(kind_of(atom_j)) + radius_ao_kind(kind_of(atom_k))) CYCLE
2130 11682 : IF (dij > radius_ao_kind(kind_of(atom_j)) + radius_RI_kind(kind_of(atom_i)) &
2131 : + bs_env%ri_metric%cutoff_radius) CYCLE
2132 5932 : IF (dik > radius_RI_kind(kind_of(atom_i)) + radius_ao_kind(kind_of(atom_k)) &
2133 : + bs_env%ri_metric%cutoff_radius) CYCLE
2134 :
2135 3867 : j_size = bs_env%i_ao_end_from_atom(atom_j) - bs_env%i_ao_start_from_atom(atom_j) + 1
2136 3867 : k_size = bs_env%i_ao_end_from_atom(atom_k) - bs_env%i_ao_start_from_atom(atom_k) + 1
2137 3867 : i_size = bs_env%i_RI_end_from_atom(atom_i) - bs_env%i_RI_start_from_atom(atom_i) + 1
2138 :
2139 19335 : ALLOCATE (int_3c(j_size, k_size, i_size))
2140 :
2141 : ! compute 3-c int. ( μ(atom j) R , ν (atom k) S | P (atom i) 0 )
2142 : ! ("|": truncated Coulomb operator), inside build_3c_integrals: (j k | i)
2143 : CALL build_3c_integral_block(int_3c, qs_env, bs_env%ri_metric, &
2144 : basis_j=bs_env%basis_set_AO, &
2145 : basis_k=bs_env%basis_set_AO, &
2146 : basis_i=bs_env%basis_set_RI, &
2147 : cell_j=index_to_cell_3c_max(1:3, j_cell), &
2148 : cell_k=index_to_cell_3c_max(1:3, k_cell), &
2149 3867 : atom_k=atom_k, atom_j=atom_j, atom_i=atom_i)
2150 :
2151 206126 : frobenius_norm = SQRT(SUM(int_3c(:, :, :)**2))
2152 :
2153 3867 : DEALLOCATE (int_3c)
2154 :
2155 : ! we use a higher threshold here to safe memory when storing the 3c integrals
2156 : ! in every tensor group
2157 29895 : IF (frobenius_norm > eps) THEN
2158 825 : nblocks_3c_max(j_cell, k_cell) = nblocks_3c_max(j_cell, k_cell) + 1
2159 : END IF
2160 :
2161 : END DO
2162 : END DO
2163 : END DO
2164 :
2165 : END DO
2166 : END DO
2167 :
2168 6 : CALL bs_env%para_env%sum(nblocks_3c_max)
2169 :
2170 18 : ALLOCATE (n_other_3c_images_max(nimages_3c_max))
2171 148 : n_other_3c_images_max(:) = 0
2172 :
2173 6 : nimages_3c = 0
2174 6 : nimage_pairs_3c = 0
2175 :
2176 148 : DO j_cell = 1, nimages_3c_max
2177 3524 : DO k_cell = 1, nimages_3c_max
2178 3524 : IF (nblocks_3c_max(j_cell, k_cell) > 0) THEN
2179 290 : n_other_3c_images_max(j_cell) = n_other_3c_images_max(j_cell) + 1
2180 290 : nimage_pairs_3c = nimage_pairs_3c + 1
2181 : END IF
2182 : END DO
2183 :
2184 148 : IF (n_other_3c_images_max(j_cell) > 0) nimages_3c = nimages_3c + 1
2185 :
2186 : END DO
2187 :
2188 6 : bs_env%nimages_3c = nimages_3c
2189 18 : ALLOCATE (bs_env%index_to_cell_3c(3, nimages_3c))
2190 : ALLOCATE (bs_env%cell_to_index_3c(i_cell_x_min:i_cell_x_max, &
2191 : j_cell_y_min:j_cell_y_max, &
2192 30 : k_cell_z_min:k_cell_z_max))
2193 288 : bs_env%cell_to_index_3c(:, :, :) = -1
2194 :
2195 24 : ALLOCATE (bs_env%nblocks_3c(nimages_3c, nimages_3c))
2196 6 : bs_env%nblocks_3c(nimages_3c, nimages_3c) = 0
2197 :
2198 6 : j_cell = 0
2199 148 : DO j_cell_max = 1, nimages_3c_max
2200 142 : IF (n_other_3c_images_max(j_cell_max) == 0) CYCLE
2201 58 : j_cell = j_cell + 1
2202 232 : cell_index(1:3) = index_to_cell_3c_max(1:3, j_cell_max)
2203 232 : bs_env%index_to_cell_3c(1:3, j_cell) = cell_index(1:3)
2204 58 : bs_env%cell_to_index_3c(cell_index(1), cell_index(2), cell_index(3)) = j_cell
2205 :
2206 58 : k_cell = 0
2207 1474 : DO k_cell_max = 1, nimages_3c_max
2208 1410 : IF (n_other_3c_images_max(k_cell_max) == 0) CYCLE
2209 626 : k_cell = k_cell + 1
2210 :
2211 1552 : bs_env%nblocks_3c(j_cell, k_cell) = nblocks_3c_max(j_cell_max, k_cell_max)
2212 : END DO
2213 :
2214 : END DO
2215 :
2216 : ! we use: 8*10^-9 GB / double precision number
2217 : mem_3c_GB = REAL(bs_env%n_RI, KIND=dp)*REAL(bs_env%n_ao, KIND=dp)**2 &
2218 6 : *REAL(nimage_pairs_3c, KIND=dp)*8E-9_dp
2219 :
2220 6 : CALL m_memory(mem_occ_per_proc)
2221 6 : CALL bs_env%para_env%max(mem_occ_per_proc)
2222 :
2223 6 : mem_occ_per_proc_GB = REAL(mem_occ_per_proc, KIND=dp)/1.0E9_dp
2224 :
2225 : ! number of processors per group that entirely stores the 3c integrals and does tensor ops
2226 6 : avail_mem_per_proc_GB = bs_env%input_memory_per_proc_GB - mem_occ_per_proc_GB
2227 :
2228 : ! careful: downconvering real to integer, 1.9 -> 1; thus add 1.0 for upconversion, 1.9 -> 2
2229 6 : bs_env%group_size_tensor = MAX(INT(mem_3c_GB/avail_mem_per_proc_GB + 1.0_dp), 1)
2230 :
2231 6 : u = bs_env%unit_nr
2232 :
2233 6 : IF (u > 0) THEN
2234 3 : WRITE (u, FMT="(T2,A,F52.1,A)") "Radius of atomic orbitals", radius_ao*angstrom, " Å"
2235 3 : WRITE (u, FMT="(T2,A,F55.1,A)") "Radius of RI functions", radius_RI*angstrom, " Å"
2236 3 : WRITE (u, FMT="(T2,A,I47)") "Number of cells for 3c integrals", nimages_3c
2237 3 : WRITE (u, FMT="(T2,A,I42)") "Number of cell pairs for 3c integrals", nimage_pairs_3c
2238 3 : WRITE (u, '(T2,A)') ''
2239 3 : WRITE (u, '(T2,A,F37.1,A)') 'Input: Available memory per MPI process', &
2240 6 : bs_env%input_memory_per_proc_GB, ' GB'
2241 3 : WRITE (u, '(T2,A,F35.1,A)') 'Used memory per MPI process before GW run', &
2242 6 : mem_occ_per_proc_GB, ' GB'
2243 3 : WRITE (u, '(T2,A,F44.1,A)') 'Memory of three-center integrals', mem_3c_GB, ' GB'
2244 : END IF
2245 :
2246 6 : CALL timestop(handle)
2247 :
2248 18 : END SUBROUTINE setup_cells_3c
2249 :
2250 : ! **************************************************************************************************
2251 : !> \brief ...
2252 : !> \param index_to_cell_1 ...
2253 : !> \param index_to_cell_2 ...
2254 : !> \param nimages_1 ...
2255 : !> \param nimages_2 ...
2256 : !> \param index_to_cell ...
2257 : !> \param cell_to_index ...
2258 : !> \param nimages ...
2259 : ! **************************************************************************************************
2260 6 : SUBROUTINE sum_two_R_grids(index_to_cell_1, index_to_cell_2, nimages_1, nimages_2, &
2261 : index_to_cell, cell_to_index, nimages)
2262 :
2263 : INTEGER, DIMENSION(:, :) :: index_to_cell_1, index_to_cell_2
2264 : INTEGER :: nimages_1, nimages_2
2265 : INTEGER, ALLOCATABLE, DIMENSION(:, :) :: index_to_cell
2266 : INTEGER, DIMENSION(:, :, :), POINTER :: cell_to_index
2267 : INTEGER :: nimages
2268 :
2269 : CHARACTER(LEN=*), PARAMETER :: routineN = 'sum_two_R_grids'
2270 :
2271 : INTEGER :: handle, i_dim, img_1, img_2, nimages_max
2272 6 : INTEGER, ALLOCATABLE, DIMENSION(:, :) :: index_to_cell_tmp
2273 : INTEGER, DIMENSION(3) :: cell_1, cell_2, R, R_max, R_min
2274 :
2275 6 : CALL timeset(routineN, handle)
2276 :
2277 24 : DO i_dim = 1, 3
2278 366 : R_min(i_dim) = MINVAL(index_to_cell_1(i_dim, :)) + MINVAL(index_to_cell_2(i_dim, :))
2279 390 : R_max(i_dim) = MAXVAL(index_to_cell_1(i_dim, :)) + MAXVAL(index_to_cell_2(i_dim, :))
2280 : END DO
2281 :
2282 6 : nimages_max = (R_max(1) - R_min(1) + 1)*(R_max(2) - R_min(2) + 1)*(R_max(3) - R_min(3) + 1)
2283 :
2284 18 : ALLOCATE (index_to_cell_tmp(3, nimages_max))
2285 766 : index_to_cell_tmp(:, :) = -1
2286 :
2287 30 : ALLOCATE (cell_to_index(R_min(1):R_max(1), R_min(2):R_max(2), R_min(3):R_max(3)))
2288 376 : cell_to_index(:, :, :) = -1
2289 :
2290 6 : nimages = 0
2291 :
2292 64 : DO img_1 = 1, nimages_1
2293 :
2294 690 : DO img_2 = 1, nimages_2
2295 :
2296 2504 : cell_1(1:3) = index_to_cell_1(1:3, img_1)
2297 2504 : cell_2(1:3) = index_to_cell_2(1:3, img_2)
2298 :
2299 2504 : R(1:3) = cell_1(1:3) + cell_2(1:3)
2300 :
2301 : ! check whether we have found a new cell
2302 684 : IF (cell_to_index(R(1), R(2), R(3)) == -1) THEN
2303 :
2304 166 : nimages = nimages + 1
2305 166 : cell_to_index(R(1), R(2), R(3)) = nimages
2306 664 : index_to_cell_tmp(1:3, nimages) = R(1:3)
2307 :
2308 : END IF
2309 :
2310 : END DO
2311 :
2312 : END DO
2313 :
2314 18 : ALLOCATE (index_to_cell(3, nimages))
2315 670 : index_to_cell(:, :) = index_to_cell_tmp(1:3, 1:nimages)
2316 :
2317 6 : CALL timestop(handle)
2318 :
2319 12 : END SUBROUTINE sum_two_R_grids
2320 :
2321 : ! **************************************************************************************************
2322 : !> \brief ...
2323 : !> \param qs_env ...
2324 : !> \param bs_env ...
2325 : ! **************************************************************************************************
2326 6 : SUBROUTINE compute_3c_integrals(qs_env, bs_env)
2327 :
2328 : TYPE(qs_environment_type), POINTER :: qs_env
2329 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
2330 :
2331 : CHARACTER(LEN=*), PARAMETER :: routineN = 'compute_3c_integrals'
2332 :
2333 : INTEGER :: handle, j_cell, k_cell, nimages_3c
2334 :
2335 6 : CALL timeset(routineN, handle)
2336 :
2337 6 : nimages_3c = bs_env%nimages_3c
2338 756 : ALLOCATE (bs_env%t_3c_int(nimages_3c, nimages_3c))
2339 64 : DO j_cell = 1, nimages_3c
2340 690 : DO k_cell = 1, nimages_3c
2341 684 : CALL dbt_create(bs_env%t_RI_AO__AO, bs_env%t_3c_int(j_cell, k_cell))
2342 : END DO
2343 : END DO
2344 :
2345 : CALL build_3c_integrals(bs_env%t_3c_int, &
2346 : bs_env%eps_filter, &
2347 : qs_env, &
2348 : bs_env%nl_3c, &
2349 : int_eps=bs_env%eps_filter*0.05_dp, &
2350 : basis_i=bs_env%basis_set_RI, &
2351 : basis_j=bs_env%basis_set_AO, &
2352 : basis_k=bs_env%basis_set_AO, &
2353 : potential_parameter=bs_env%ri_metric, &
2354 : desymmetrize=.FALSE., do_kpoints=.TRUE., cell_sym=.TRUE., &
2355 6 : cell_to_index_ext=bs_env%cell_to_index_3c)
2356 :
2357 6 : CALL bs_env%para_env%sync()
2358 :
2359 6 : CALL timestop(handle)
2360 :
2361 6 : END SUBROUTINE compute_3c_integrals
2362 :
2363 : ! **************************************************************************************************
2364 : !> \brief ...
2365 : !> \param cell_index ...
2366 : !> \param hmat ...
2367 : !> \param cell_dist ...
2368 : ! **************************************************************************************************
2369 44652 : SUBROUTINE get_cell_dist(cell_index, hmat, cell_dist)
2370 :
2371 : INTEGER, DIMENSION(3) :: cell_index
2372 : REAL(KIND=dp) :: hmat(3, 3), cell_dist
2373 :
2374 : CHARACTER(LEN=*), PARAMETER :: routineN = 'get_cell_dist'
2375 :
2376 : INTEGER :: handle, i_dim
2377 : INTEGER, DIMENSION(3) :: cell_index_adj
2378 : REAL(KIND=dp) :: cell_dist_3(3)
2379 :
2380 44652 : CALL timeset(routineN, handle)
2381 :
2382 : ! the distance of cells needs to be taken to adjacent neighbors, not
2383 : ! between the center of the cells. We thus need to rescale the cell index
2384 178608 : DO i_dim = 1, 3
2385 133956 : IF (cell_index(i_dim) > 0) cell_index_adj(i_dim) = cell_index(i_dim) - 1
2386 133956 : IF (cell_index(i_dim) < 0) cell_index_adj(i_dim) = cell_index(i_dim) + 1
2387 178608 : IF (cell_index(i_dim) == 0) cell_index_adj(i_dim) = cell_index(i_dim)
2388 : END DO
2389 :
2390 714432 : cell_dist_3(1:3) = MATMUL(hmat, REAL(cell_index_adj, KIND=dp))
2391 :
2392 178608 : cell_dist = SQRT(ABS(SUM(cell_dist_3(1:3)**2)))
2393 :
2394 44652 : CALL timestop(handle)
2395 :
2396 44652 : END SUBROUTINE get_cell_dist
2397 :
2398 : ! **************************************************************************************************
2399 : !> \brief ...
2400 : !> \param qs_env ...
2401 : !> \param bs_env ...
2402 : !> \param kpoints ...
2403 : !> \param do_print ...
2404 : ! **************************************************************************************************
2405 0 : SUBROUTINE setup_kpoints_scf_desymm(qs_env, bs_env, kpoints, do_print)
2406 : TYPE(qs_environment_type), POINTER :: qs_env
2407 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
2408 : TYPE(kpoint_type), POINTER :: kpoints
2409 :
2410 : CHARACTER(LEN=*), PARAMETER :: routineN = 'setup_kpoints_scf_desymm'
2411 :
2412 : INTEGER :: handle, i_cell_x, i_dim, img, j_cell_y, &
2413 : k_cell_z, nimages, nkp, u
2414 : INTEGER, DIMENSION(3) :: cell_grid, cixd, nkp_grid
2415 : TYPE(kpoint_type), POINTER :: kpoints_scf
2416 :
2417 : LOGICAL:: do_print
2418 :
2419 0 : CALL timeset(routineN, handle)
2420 :
2421 0 : NULLIFY (kpoints)
2422 0 : CALL kpoint_create(kpoints)
2423 :
2424 0 : CALL get_qs_env(qs_env=qs_env, kpoints=kpoints_scf)
2425 :
2426 0 : nkp_grid(1:3) = kpoints_scf%nkp_grid(1:3)
2427 0 : nkp = nkp_grid(1)*nkp_grid(2)*nkp_grid(3)
2428 :
2429 : ! we need in periodic directions at least 2 k-points in the SCF
2430 0 : DO i_dim = 1, 3
2431 0 : IF (bs_env%periodic(i_dim) == 1) THEN
2432 0 : CPASSERT(nkp_grid(i_dim) > 1)
2433 : END IF
2434 : END DO
2435 :
2436 0 : kpoints%kp_scheme = "GENERAL"
2437 0 : kpoints%nkp_grid(1:3) = nkp_grid(1:3)
2438 0 : kpoints%nkp = nkp
2439 0 : bs_env%nkp_scf_desymm = nkp
2440 :
2441 0 : ALLOCATE (kpoints%xkp(1:3, nkp))
2442 0 : CALL compute_xkp(kpoints%xkp, 1, nkp, nkp_grid)
2443 :
2444 0 : ALLOCATE (kpoints%wkp(nkp))
2445 0 : kpoints%wkp(:) = 1.0_dp/REAL(nkp, KIND=dp)
2446 :
2447 : ! for example 4x3x6 kpoint grid -> 3x3x5 cell grid because we need the same number of
2448 : ! neighbor cells on both sides of the unit cell
2449 0 : cell_grid(1:3) = nkp_grid(1:3) - MODULO(nkp_grid(1:3) + 1, 2)
2450 : ! cell index: for example for x: from -n_x/2 to +n_x/2, n_x: number of cells in x direction
2451 0 : cixd(1:3) = cell_grid(1:3)/2
2452 :
2453 0 : nimages = cell_grid(1)*cell_grid(2)*cell_grid(3)
2454 :
2455 0 : bs_env%nimages_scf_desymm = nimages
2456 :
2457 0 : ALLOCATE (kpoints%cell_to_index(-cixd(1):cixd(1), -cixd(2):cixd(2), -cixd(3):cixd(3)))
2458 0 : ALLOCATE (kpoints%index_to_cell(3, nimages))
2459 :
2460 0 : img = 0
2461 0 : DO i_cell_x = -cixd(1), cixd(1)
2462 0 : DO j_cell_y = -cixd(2), cixd(2)
2463 0 : DO k_cell_z = -cixd(3), cixd(3)
2464 0 : img = img + 1
2465 0 : kpoints%cell_to_index(i_cell_x, j_cell_y, k_cell_z) = img
2466 0 : kpoints%index_to_cell(1:3, img) = [i_cell_x, j_cell_y, k_cell_z]
2467 : END DO
2468 : END DO
2469 : END DO
2470 :
2471 0 : u = bs_env%unit_nr
2472 0 : IF (u > 0 .AND. do_print) THEN
2473 0 : WRITE (u, FMT="(T2,A,I49)") "Number of cells for G, χ, W, Σ", nimages
2474 : END IF
2475 :
2476 0 : CALL timestop(handle)
2477 :
2478 0 : END SUBROUTINE setup_kpoints_scf_desymm
2479 :
2480 : ! **************************************************************************************************
2481 : !> \brief ...
2482 : !> \param bs_env ...
2483 : ! **************************************************************************************************
2484 6 : SUBROUTINE setup_cells_Delta_R(bs_env)
2485 :
2486 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
2487 :
2488 : CHARACTER(LEN=*), PARAMETER :: routineN = 'setup_cells_Delta_R'
2489 :
2490 : INTEGER :: handle
2491 :
2492 6 : CALL timeset(routineN, handle)
2493 :
2494 : ! cell sums batch wise for fixed ΔR = S_1 - R_1; for example:
2495 : ! Σ_λσ^R = sum_PR1νS1 M^G_λ0,νS1,PR1 M^W_σR,νS1,PR1
2496 :
2497 : CALL sum_two_R_grids(bs_env%index_to_cell_3c, &
2498 : bs_env%index_to_cell_3c, &
2499 : bs_env%nimages_3c, bs_env%nimages_3c, &
2500 : bs_env%index_to_cell_Delta_R, &
2501 : bs_env%cell_to_index_Delta_R, &
2502 6 : bs_env%nimages_Delta_R)
2503 :
2504 6 : IF (bs_env%unit_nr > 0) THEN
2505 3 : WRITE (bs_env%unit_nr, FMT="(T2,A,I61)") "Number of cells ΔR", bs_env%nimages_Delta_R
2506 : END IF
2507 :
2508 6 : CALL timestop(handle)
2509 :
2510 6 : END SUBROUTINE setup_cells_Delta_R
2511 :
2512 : ! **************************************************************************************************
2513 : !> \brief ...
2514 : !> \param bs_env ...
2515 : ! **************************************************************************************************
2516 6 : SUBROUTINE setup_parallelization_Delta_R(bs_env)
2517 :
2518 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
2519 :
2520 : CHARACTER(LEN=*), PARAMETER :: routineN = 'setup_parallelization_Delta_R'
2521 :
2522 : INTEGER :: handle, i_cell_Delta_R, i_task_local, &
2523 : n_tasks_local
2524 6 : INTEGER, ALLOCATABLE, DIMENSION(:) :: i_cell_Delta_R_group, &
2525 6 : n_tensor_ops_Delta_R
2526 :
2527 6 : CALL timeset(routineN, handle)
2528 :
2529 6 : CALL compute_n_tensor_ops_Delta_R(bs_env, n_tensor_ops_Delta_R)
2530 :
2531 6 : CALL compute_Delta_R_dist(bs_env, n_tensor_ops_Delta_R, i_cell_Delta_R_group, n_tasks_local)
2532 :
2533 6 : bs_env%n_tasks_Delta_R_local = n_tasks_local
2534 :
2535 18 : ALLOCATE (bs_env%task_Delta_R(n_tasks_local))
2536 :
2537 6 : i_task_local = 0
2538 172 : DO i_cell_Delta_R = 1, bs_env%nimages_Delta_R
2539 :
2540 166 : IF (i_cell_Delta_R_group(i_cell_Delta_R) /= bs_env%tensor_group_color) CYCLE
2541 :
2542 73 : i_task_local = i_task_local + 1
2543 :
2544 172 : bs_env%task_Delta_R(i_task_local) = i_cell_Delta_R
2545 :
2546 : END DO
2547 :
2548 18 : ALLOCATE (bs_env%skip_DR_chi(n_tasks_local))
2549 79 : bs_env%skip_DR_chi(:) = .FALSE.
2550 18 : ALLOCATE (bs_env%skip_DR_Sigma(n_tasks_local))
2551 79 : bs_env%skip_DR_Sigma(:) = .FALSE.
2552 :
2553 6 : CALL allocate_skip_3xR(bs_env%skip_DR_R12_S_Goccx3c_chi, bs_env)
2554 6 : CALL allocate_skip_3xR(bs_env%skip_DR_R12_S_Gvirx3c_chi, bs_env)
2555 6 : CALL allocate_skip_3xR(bs_env%skip_DR_R_R2_MxM_chi, bs_env)
2556 :
2557 6 : CALL allocate_skip_3xR(bs_env%skip_DR_R1_S2_Gx3c_Sigma, bs_env)
2558 6 : CALL allocate_skip_3xR(bs_env%skip_DR_R1_R_MxM_Sigma, bs_env)
2559 :
2560 6 : CALL timestop(handle)
2561 :
2562 12 : END SUBROUTINE setup_parallelization_Delta_R
2563 :
2564 : ! **************************************************************************************************
2565 : !> \brief ...
2566 : !> \param skip ...
2567 : !> \param bs_env ...
2568 : ! **************************************************************************************************
2569 30 : SUBROUTINE allocate_skip_3xR(skip, bs_env)
2570 : LOGICAL, ALLOCATABLE, DIMENSION(:, :, :) :: skip
2571 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
2572 :
2573 : CHARACTER(LEN=*), PARAMETER :: routineN = 'allocate_skip_3xR'
2574 :
2575 : INTEGER :: handle
2576 :
2577 30 : CALL timeset(routineN, handle)
2578 :
2579 150 : ALLOCATE (skip(bs_env%n_tasks_Delta_R_local, bs_env%nimages_3c, bs_env%nimages_scf_desymm))
2580 38235 : skip(:, :, :) = .FALSE.
2581 :
2582 30 : CALL timestop(handle)
2583 :
2584 30 : END SUBROUTINE allocate_skip_3xR
2585 :
2586 : ! **************************************************************************************************
2587 : !> \brief ...
2588 : !> \param bs_env ...
2589 : !> \param n_tensor_ops_Delta_R ...
2590 : !> \param i_cell_Delta_R_group ...
2591 : !> \param n_tasks_local ...
2592 : ! **************************************************************************************************
2593 6 : SUBROUTINE compute_Delta_R_dist(bs_env, n_tensor_ops_Delta_R, i_cell_Delta_R_group, n_tasks_local)
2594 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
2595 : INTEGER, ALLOCATABLE, DIMENSION(:) :: n_tensor_ops_Delta_R, &
2596 : i_cell_Delta_R_group
2597 : INTEGER :: n_tasks_local
2598 :
2599 : CHARACTER(LEN=*), PARAMETER :: routineN = 'compute_Delta_R_dist'
2600 :
2601 : INTEGER :: handle, i_Delta_R_max_op, i_group_min, &
2602 : nimages_Delta_R, u
2603 6 : INTEGER, ALLOCATABLE, DIMENSION(:) :: n_tensor_ops_Delta_R_in_group
2604 :
2605 6 : CALL timeset(routineN, handle)
2606 :
2607 6 : nimages_Delta_R = bs_env%nimages_Delta_R
2608 :
2609 6 : u = bs_env%unit_nr
2610 :
2611 6 : IF (u > 0 .AND. nimages_Delta_R < bs_env%num_tensor_groups) THEN
2612 0 : WRITE (u, FMT="(T2,A,I5,A,I5,A)") "There are only ", nimages_Delta_R, &
2613 0 : " tasks to work on but there are ", bs_env%num_tensor_groups, " groups."
2614 0 : WRITE (u, FMT="(T2,A)") "Please reduce the number of MPI processes."
2615 0 : WRITE (u, '(T2,A)') ''
2616 : END IF
2617 :
2618 18 : ALLOCATE (n_tensor_ops_Delta_R_in_group(bs_env%num_tensor_groups))
2619 18 : n_tensor_ops_Delta_R_in_group(:) = 0
2620 18 : ALLOCATE (i_cell_Delta_R_group(nimages_Delta_R))
2621 172 : i_cell_Delta_R_group(:) = -1
2622 :
2623 6 : n_tasks_local = 0
2624 :
2625 624 : DO WHILE (ANY(n_tensor_ops_Delta_R(:) /= 0))
2626 :
2627 : ! get largest element of n_tensor_ops_Delta_R
2628 4684 : i_Delta_R_max_op = MAXLOC(n_tensor_ops_Delta_R, 1)
2629 :
2630 : ! distribute i_Delta_R_max_op to tensor group which has currently the smallest load
2631 584 : i_group_min = MINLOC(n_tensor_ops_Delta_R_in_group, 1)
2632 :
2633 : ! the tensor groups are 0-index based; but i_group_min is 1-index based
2634 146 : i_cell_Delta_R_group(i_Delta_R_max_op) = i_group_min - 1
2635 : n_tensor_ops_Delta_R_in_group(i_group_min) = n_tensor_ops_Delta_R_in_group(i_group_min) + &
2636 146 : n_tensor_ops_Delta_R(i_Delta_R_max_op)
2637 :
2638 : ! remove i_Delta_R_max_op from n_tensor_ops_Delta_R
2639 146 : n_tensor_ops_Delta_R(i_Delta_R_max_op) = 0
2640 :
2641 152 : IF (bs_env%tensor_group_color == i_group_min - 1) n_tasks_local = n_tasks_local + 1
2642 :
2643 : END DO
2644 :
2645 6 : CALL timestop(handle)
2646 :
2647 12 : END SUBROUTINE compute_Delta_R_dist
2648 :
2649 : ! **************************************************************************************************
2650 : !> \brief ...
2651 : !> \param bs_env ...
2652 : !> \param n_tensor_ops_Delta_R ...
2653 : ! **************************************************************************************************
2654 6 : SUBROUTINE compute_n_tensor_ops_Delta_R(bs_env, n_tensor_ops_Delta_R)
2655 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
2656 : INTEGER, ALLOCATABLE, DIMENSION(:) :: n_tensor_ops_Delta_R
2657 :
2658 : CHARACTER(LEN=*), PARAMETER :: routineN = 'compute_n_tensor_ops_Delta_R'
2659 :
2660 : INTEGER :: handle, i_cell_Delta_R, i_cell_R, i_cell_R1, i_cell_R1_minus_R, i_cell_R2, &
2661 : i_cell_R2_m_R1, i_cell_S1, i_cell_S1_m_R1_p_R2, i_cell_S1_minus_R, i_cell_S2, &
2662 : nimages_Delta_R
2663 : INTEGER, DIMENSION(3) :: cell_DR, cell_m_R1, cell_R, cell_R1, cell_R1_minus_R, cell_R2, &
2664 : cell_R2_m_R1, cell_S1, cell_S1_m_R2_p_R1, cell_S1_minus_R, cell_S1_p_S2_m_R1, cell_S2
2665 : LOGICAL :: cell_found
2666 :
2667 6 : CALL timeset(routineN, handle)
2668 :
2669 6 : nimages_Delta_R = bs_env%nimages_Delta_R
2670 :
2671 18 : ALLOCATE (n_tensor_ops_Delta_R(nimages_Delta_R))
2672 172 : n_tensor_ops_Delta_R(:) = 0
2673 :
2674 : ! compute number of tensor operations for specific Delta_R
2675 172 : DO i_cell_Delta_R = 1, nimages_Delta_R
2676 :
2677 166 : IF (MODULO(i_cell_Delta_R, bs_env%num_tensor_groups) /= bs_env%tensor_group_color) CYCLE
2678 :
2679 994 : DO i_cell_R1 = 1, bs_env%nimages_3c
2680 :
2681 3620 : cell_R1(1:3) = bs_env%index_to_cell_3c(1:3, i_cell_R1)
2682 3620 : cell_DR(1:3) = bs_env%index_to_cell_Delta_R(1:3, i_cell_Delta_R)
2683 :
2684 : ! S_1 = R_1 + ΔR (from ΔR = S_1 - R_1)
2685 : CALL add_R(cell_R1, cell_DR, bs_env%index_to_cell_3c, cell_S1, &
2686 905 : cell_found, bs_env%cell_to_index_3c, i_cell_S1)
2687 905 : IF (.NOT. cell_found) CYCLE
2688 :
2689 2950 : DO i_cell_R2 = 1, bs_env%nimages_scf_desymm
2690 :
2691 10620 : cell_R2(1:3) = bs_env%kpoints_scf_desymm%index_to_cell(1:3, i_cell_R2)
2692 :
2693 : ! R_2 - R_1
2694 : CALL add_R(cell_R2, -cell_R1, bs_env%index_to_cell_3c, cell_R2_m_R1, &
2695 10620 : cell_found, bs_env%cell_to_index_3c, i_cell_R2_m_R1)
2696 2655 : IF (.NOT. cell_found) CYCLE
2697 :
2698 : ! S_1 - R_1 + R_2
2699 : CALL add_R(cell_S1, cell_R2_m_R1, bs_env%index_to_cell_3c, cell_S1_m_R2_p_R1, &
2700 1575 : cell_found, bs_env%cell_to_index_3c, i_cell_S1_m_R1_p_R2)
2701 1575 : IF (.NOT. cell_found) CYCLE
2702 :
2703 3993 : n_tensor_ops_Delta_R(i_cell_Delta_R) = n_tensor_ops_Delta_R(i_cell_Delta_R) + 1
2704 :
2705 : END DO ! i_cell_R2
2706 :
2707 2950 : DO i_cell_S2 = 1, bs_env%nimages_scf_desymm
2708 :
2709 10620 : cell_S2(1:3) = bs_env%kpoints_scf_desymm%index_to_cell(1:3, i_cell_S2)
2710 10620 : cell_m_R1(1:3) = -cell_R1(1:3)
2711 10620 : cell_S1_p_S2_m_R1(1:3) = cell_S1(1:3) + cell_S2(1:3) - cell_R1(1:3)
2712 :
2713 2655 : CALL is_cell_in_index_to_cell(cell_m_R1, bs_env%index_to_cell_3c, cell_found)
2714 2655 : IF (.NOT. cell_found) CYCLE
2715 :
2716 2169 : CALL is_cell_in_index_to_cell(cell_S1_p_S2_m_R1, bs_env%index_to_cell_3c, cell_found)
2717 295 : IF (.NOT. cell_found) CYCLE
2718 :
2719 : END DO ! i_cell_S2
2720 :
2721 4021 : DO i_cell_R = 1, bs_env%nimages_scf_desymm
2722 :
2723 10620 : cell_R = bs_env%kpoints_scf_desymm%index_to_cell(1:3, i_cell_R)
2724 :
2725 : ! R_1 - R
2726 : CALL add_R(cell_R1, -cell_R, bs_env%index_to_cell_3c, cell_R1_minus_R, &
2727 10620 : cell_found, bs_env%cell_to_index_3c, i_cell_R1_minus_R)
2728 2655 : IF (.NOT. cell_found) CYCLE
2729 :
2730 : ! S_1 - R
2731 : CALL add_R(cell_S1, -cell_R, bs_env%index_to_cell_3c, cell_S1_minus_R, &
2732 6804 : cell_found, bs_env%cell_to_index_3c, i_cell_S1_minus_R)
2733 905 : IF (.NOT. cell_found) CYCLE
2734 :
2735 : END DO ! i_cell_R
2736 :
2737 : END DO ! i_cell_R1
2738 :
2739 : END DO ! i_cell_Delta_R
2740 :
2741 6 : CALL bs_env%para_env%sum(n_tensor_ops_Delta_R)
2742 :
2743 6 : CALL timestop(handle)
2744 :
2745 6 : END SUBROUTINE compute_n_tensor_ops_Delta_R
2746 :
2747 : ! **************************************************************************************************
2748 : !> \brief ...
2749 : !> \param cell_1 ...
2750 : !> \param cell_2 ...
2751 : !> \param index_to_cell ...
2752 : !> \param cell_1_plus_2 ...
2753 : !> \param cell_found ...
2754 : !> \param cell_to_index ...
2755 : !> \param i_cell_1_plus_2 ...
2756 : ! **************************************************************************************************
2757 85114 : SUBROUTINE add_R(cell_1, cell_2, index_to_cell, cell_1_plus_2, cell_found, &
2758 : cell_to_index, i_cell_1_plus_2)
2759 :
2760 : INTEGER, DIMENSION(3) :: cell_1, cell_2
2761 : INTEGER, DIMENSION(:, :) :: index_to_cell
2762 : INTEGER, DIMENSION(3) :: cell_1_plus_2
2763 : LOGICAL :: cell_found
2764 : INTEGER, DIMENSION(:, :, :), INTENT(IN), &
2765 : OPTIONAL, POINTER :: cell_to_index
2766 : INTEGER, INTENT(OUT), OPTIONAL :: i_cell_1_plus_2
2767 :
2768 : CHARACTER(LEN=*), PARAMETER :: routineN = 'add_R'
2769 :
2770 : INTEGER :: handle
2771 :
2772 85114 : CALL timeset(routineN, handle)
2773 :
2774 340456 : cell_1_plus_2(1:3) = cell_1(1:3) + cell_2(1:3)
2775 :
2776 85114 : CALL is_cell_in_index_to_cell(cell_1_plus_2, index_to_cell, cell_found)
2777 :
2778 85114 : IF (PRESENT(i_cell_1_plus_2)) THEN
2779 85114 : IF (cell_found) THEN
2780 48214 : CPASSERT(PRESENT(cell_to_index))
2781 48214 : i_cell_1_plus_2 = cell_to_index(cell_1_plus_2(1), cell_1_plus_2(2), cell_1_plus_2(3))
2782 : ELSE
2783 36900 : i_cell_1_plus_2 = -1000
2784 : END IF
2785 : END IF
2786 :
2787 85114 : CALL timestop(handle)
2788 :
2789 85114 : END SUBROUTINE add_R
2790 :
2791 : ! **************************************************************************************************
2792 : !> \brief ...
2793 : !> \param cell ...
2794 : !> \param index_to_cell ...
2795 : !> \param cell_found ...
2796 : ! **************************************************************************************************
2797 133779 : SUBROUTINE is_cell_in_index_to_cell(cell, index_to_cell, cell_found)
2798 : INTEGER, DIMENSION(3) :: cell
2799 : INTEGER, DIMENSION(:, :) :: index_to_cell
2800 : LOGICAL :: cell_found
2801 :
2802 : CHARACTER(LEN=*), PARAMETER :: routineN = 'is_cell_in_index_to_cell'
2803 :
2804 : INTEGER :: handle, i_cell, nimg
2805 : INTEGER, DIMENSION(3) :: cell_i
2806 :
2807 133779 : CALL timeset(routineN, handle)
2808 :
2809 133779 : nimg = SIZE(index_to_cell, 2)
2810 :
2811 133779 : cell_found = .FALSE.
2812 :
2813 1653594 : DO i_cell = 1, nimg
2814 :
2815 6079260 : cell_i(1:3) = index_to_cell(1:3, i_cell)
2816 :
2817 1653594 : IF (cell_i(1) == cell(1) .AND. cell_i(2) == cell(2) .AND. cell_i(3) == cell(3)) THEN
2818 79661 : cell_found = .TRUE.
2819 : END IF
2820 :
2821 : END DO
2822 :
2823 133779 : CALL timestop(handle)
2824 :
2825 133779 : END SUBROUTINE is_cell_in_index_to_cell
2826 :
2827 : ! **************************************************************************************************
2828 : !> \brief ...
2829 : !> \param qs_env ...
2830 : !> \param bs_env ...
2831 : ! **************************************************************************************************
2832 6 : SUBROUTINE allocate_matrices_small_cell_full_kp(qs_env, bs_env)
2833 : TYPE(qs_environment_type), POINTER :: qs_env
2834 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
2835 :
2836 : CHARACTER(LEN=*), PARAMETER :: routineN = 'allocate_matrices_small_cell_full_kp'
2837 :
2838 : INTEGER :: handle, i_spin, i_t, img, n_spin, &
2839 : nimages_scf, num_time_freq_points
2840 : TYPE(cp_blacs_env_type), POINTER :: blacs_env
2841 : TYPE(mp_para_env_type), POINTER :: para_env
2842 :
2843 6 : CALL timeset(routineN, handle)
2844 :
2845 6 : nimages_scf = bs_env%nimages_scf_desymm
2846 6 : num_time_freq_points = bs_env%num_time_freq_points
2847 6 : n_spin = bs_env%n_spin
2848 :
2849 6 : CALL get_qs_env(qs_env, para_env=para_env, blacs_env=blacs_env)
2850 :
2851 72 : ALLOCATE (bs_env%fm_G_S(nimages_scf))
2852 72 : ALLOCATE (bs_env%fm_Sigma_x_R(nimages_scf))
2853 464 : ALLOCATE (bs_env%fm_chi_R_t(nimages_scf, num_time_freq_points))
2854 464 : ALLOCATE (bs_env%fm_MWM_R_t(nimages_scf, num_time_freq_points))
2855 476 : ALLOCATE (bs_env%fm_Sigma_c_R_neg_tau(nimages_scf, num_time_freq_points, n_spin))
2856 476 : ALLOCATE (bs_env%fm_Sigma_c_R_pos_tau(nimages_scf, num_time_freq_points, n_spin))
2857 60 : DO img = 1, nimages_scf
2858 54 : CALL cp_fm_create(bs_env%fm_G_S(img), bs_env%fm_work_mo(1)%matrix_struct)
2859 54 : CALL cp_fm_create(bs_env%fm_Sigma_x_R(img), bs_env%fm_work_mo(1)%matrix_struct)
2860 456 : DO i_t = 1, num_time_freq_points
2861 396 : CALL cp_fm_create(bs_env%fm_chi_R_t(img, i_t), bs_env%fm_RI_RI%matrix_struct)
2862 396 : CALL cp_fm_create(bs_env%fm_MWM_R_t(img, i_t), bs_env%fm_RI_RI%matrix_struct)
2863 396 : CALL cp_fm_set_all(bs_env%fm_MWM_R_t(img, i_t), 0.0_dp)
2864 846 : DO i_spin = 1, n_spin
2865 : CALL cp_fm_create(bs_env%fm_Sigma_c_R_neg_tau(img, i_t, i_spin), &
2866 396 : bs_env%fm_work_mo(1)%matrix_struct)
2867 : CALL cp_fm_create(bs_env%fm_Sigma_c_R_pos_tau(img, i_t, i_spin), &
2868 396 : bs_env%fm_work_mo(1)%matrix_struct)
2869 396 : CALL cp_fm_set_all(bs_env%fm_Sigma_c_R_neg_tau(img, i_t, i_spin), 0.0_dp)
2870 792 : CALL cp_fm_set_all(bs_env%fm_Sigma_c_R_pos_tau(img, i_t, i_spin), 0.0_dp)
2871 : END DO
2872 : END DO
2873 : END DO
2874 :
2875 6 : CALL timestop(handle)
2876 :
2877 6 : END SUBROUTINE allocate_matrices_small_cell_full_kp
2878 :
2879 : ! **************************************************************************************************
2880 : !> \brief ...
2881 : !> \param qs_env ...
2882 : !> \param bs_env ...
2883 : ! **************************************************************************************************
2884 6 : SUBROUTINE trafo_V_xc_R_to_kp(qs_env, bs_env)
2885 : TYPE(qs_environment_type), POINTER :: qs_env
2886 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
2887 :
2888 : CHARACTER(LEN=*), PARAMETER :: routineN = 'trafo_V_xc_R_to_kp'
2889 :
2890 : INTEGER :: handle, ikp, img, ispin, n_ao
2891 6 : INTEGER, DIMENSION(:, :, :), POINTER :: cell_to_index_scf
2892 : TYPE(cp_cfm_type) :: cfm_mo_coeff, cfm_tmp, cfm_V_xc
2893 : TYPE(cp_fm_type) :: fm_V_xc_re
2894 6 : TYPE(dbcsr_p_type), DIMENSION(:, :), POINTER :: matrix_ks
2895 : TYPE(kpoint_type), POINTER :: kpoints_scf
2896 : TYPE(neighbor_list_set_p_type), DIMENSION(:), &
2897 6 : POINTER :: sab_nl
2898 :
2899 6 : CALL timeset(routineN, handle)
2900 :
2901 6 : n_ao = bs_env%n_ao
2902 :
2903 6 : CALL get_qs_env(qs_env, matrix_ks_kp=matrix_ks, kpoints=kpoints_scf)
2904 :
2905 6 : NULLIFY (sab_nl)
2906 6 : CALL get_kpoint_info(kpoints_scf, sab_nl=sab_nl, cell_to_index=cell_to_index_scf)
2907 :
2908 6 : CALL cp_cfm_create(cfm_V_xc, bs_env%cfm_work_mo%matrix_struct)
2909 6 : CALL cp_cfm_create(cfm_mo_coeff, bs_env%cfm_work_mo%matrix_struct)
2910 6 : CALL cp_cfm_create(cfm_tmp, bs_env%cfm_work_mo%matrix_struct)
2911 6 : CALL cp_fm_create(fm_V_xc_re, bs_env%cfm_work_mo%matrix_struct)
2912 :
2913 184 : DO img = 1, bs_env%nimages_scf
2914 362 : DO ispin = 1, bs_env%n_spin
2915 : ! JW kind of hack because the format of matrix_ks remains dubious...
2916 178 : CALL dbcsr_set(matrix_ks(ispin, img)%matrix, 0.0_dp)
2917 356 : CALL copy_fm_to_dbcsr(bs_env%fm_V_xc_R(img, ispin), matrix_ks(ispin, img)%matrix)
2918 : END DO
2919 : END DO
2920 :
2921 30 : ALLOCATE (bs_env%v_xc_n(n_ao, bs_env%nkp_bs_and_DOS, bs_env%n_spin))
2922 :
2923 12 : DO ispin = 1, bs_env%n_spin
2924 170 : DO ikp = 1, bs_env%nkp_bs_and_DOS
2925 :
2926 : ! v^xc^R -> v^xc(k) (matrix_ks stores v^xc^R, see SUBROUTINE compute_V_xc)
2927 : CALL rsmat_to_kp(matrix_ks, ispin, bs_env%kpoints_DOS%xkp(1:3, ikp), &
2928 158 : cell_to_index_scf, sab_nl, bs_env, cfm_V_xc)
2929 :
2930 : ! get C_µn(k)
2931 158 : CALL cp_cfm_to_cfm(bs_env%cfm_mo_coeff_kp(ikp, ispin), cfm_mo_coeff)
2932 :
2933 : ! v^xc_nm(k_i) = sum_µν C^*_µn(k_i) v^xc_µν(k_i) C_νn(k_i)
2934 : CALL parallel_gemm('N', 'N', n_ao, n_ao, n_ao, z_one, cfm_V_xc, cfm_mo_coeff, &
2935 158 : z_zero, cfm_tmp)
2936 : CALL parallel_gemm('C', 'N', n_ao, n_ao, n_ao, z_one, cfm_mo_coeff, cfm_tmp, &
2937 158 : z_zero, cfm_V_xc)
2938 :
2939 : ! get v^xc_nn(k_i) which is a real quantity as v^xc is Hermitian
2940 158 : CALL cp_cfm_to_fm(cfm_V_xc, fm_V_xc_re)
2941 164 : CALL cp_fm_get_diag(fm_V_xc_re, bs_env%v_xc_n(:, ikp, ispin))
2942 :
2943 : END DO
2944 :
2945 : END DO
2946 :
2947 : ! just rebuild the overwritten KS matrix again
2948 6 : CALL qs_ks_build_kohn_sham_matrix(qs_env, calculate_forces=.FALSE., just_energy=.FALSE.)
2949 :
2950 6 : CALL cp_cfm_release(cfm_V_xc)
2951 6 : CALL cp_cfm_release(cfm_mo_coeff)
2952 6 : CALL cp_cfm_release(cfm_tmp)
2953 6 : CALL cp_fm_release(fm_V_xc_re)
2954 :
2955 6 : CALL timestop(handle)
2956 :
2957 12 : END SUBROUTINE trafo_V_xc_R_to_kp
2958 :
2959 : ! **************************************************************************************************
2960 : !> \brief ...
2961 : !> \param qs_env ...
2962 : !> \param bs_env ...
2963 : ! **************************************************************************************************
2964 6 : SUBROUTINE heuristic_RI_regularization(qs_env, bs_env)
2965 : TYPE(qs_environment_type), POINTER :: qs_env
2966 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
2967 :
2968 : CHARACTER(LEN=*), PARAMETER :: routineN = 'heuristic_RI_regularization'
2969 :
2970 6 : COMPLEX(KIND=dp), ALLOCATABLE, DIMENSION(:, :, :) :: M
2971 : INTEGER :: handle, ikp, ikp_local, n_RI, nkp, &
2972 : nkp_local, u
2973 : REAL(KIND=dp) :: cond_nr, cond_nr_max, max_ev, &
2974 : max_ev_ikp, min_ev, min_ev_ikp
2975 6 : REAL(KIND=dp), ALLOCATABLE, DIMENSION(:, :, :) :: M_R
2976 :
2977 6 : CALL timeset(routineN, handle)
2978 :
2979 : ! compute M^R_PQ = <phi_P,0|V^tr(rc)|phi_Q,R> for RI metric
2980 6 : CALL get_V_tr_R(M_R, bs_env%ri_metric, 0.0_dp, bs_env, qs_env)
2981 :
2982 6 : nkp = bs_env%nkp_chi_eps_W_orig_plus_extra
2983 6 : n_RI = bs_env%n_RI
2984 :
2985 6 : nkp_local = 0
2986 3846 : DO ikp = 1, nkp
2987 : ! trivial parallelization over k-points
2988 3840 : IF (MODULO(ikp, bs_env%para_env%num_pe) /= bs_env%para_env%mepos) CYCLE
2989 3846 : nkp_local = nkp_local + 1
2990 : END DO
2991 :
2992 30 : ALLOCATE (M(n_RI, n_RI, nkp_local))
2993 :
2994 6 : ikp_local = 0
2995 6 : cond_nr_max = 0.0_dp
2996 6 : min_ev = 1000.0_dp
2997 6 : max_ev = -1000.0_dp
2998 :
2999 3846 : DO ikp = 1, nkp
3000 :
3001 : ! trivial parallelization
3002 3840 : IF (MODULO(ikp, bs_env%para_env%num_pe) /= bs_env%para_env%mepos) CYCLE
3003 :
3004 1920 : ikp_local = ikp_local + 1
3005 :
3006 : ! M(k) = sum_R e^ikR M^R
3007 : CALL rs_to_kp(M_R, M(:, :, ikp_local), &
3008 : bs_env%kpoints_scf_desymm%index_to_cell, &
3009 1920 : bs_env%kpoints_chi_eps_W%xkp(1:3, ikp))
3010 :
3011 : ! compute condition number of M_PQ(k)
3012 1920 : CALL power(M(:, :, ikp_local), 1.0_dp, 0.0_dp, cond_nr, min_ev_ikp, max_ev_ikp)
3013 :
3014 1920 : IF (cond_nr > cond_nr_max) cond_nr_max = cond_nr
3015 1920 : IF (max_ev_ikp > max_ev) max_ev = max_ev_ikp
3016 1926 : IF (min_ev_ikp < min_ev) min_ev = min_ev_ikp
3017 :
3018 : END DO ! ikp
3019 :
3020 6 : CALL bs_env%para_env%max(cond_nr_max)
3021 :
3022 6 : u = bs_env%unit_nr
3023 6 : IF (u > 0) THEN
3024 3 : WRITE (u, FMT="(T2,A,ES34.1)") "Min. abs. eigenvalue of RI metric matrix M(k)", min_ev
3025 3 : WRITE (u, FMT="(T2,A,ES34.1)") "Max. abs. eigenvalue of RI metric matrix M(k)", max_ev
3026 3 : WRITE (u, FMT="(T2,A,ES50.1)") "Max. condition number of M(k)", cond_nr_max
3027 : END IF
3028 :
3029 6 : CALL timestop(handle)
3030 :
3031 12 : END SUBROUTINE heuristic_RI_regularization
3032 :
3033 : ! **************************************************************************************************
3034 : !> \brief ...
3035 : !> \param V_tr_R ...
3036 : !> \param pot_type ...
3037 : !> \param regularization_RI ...
3038 : !> \param bs_env ...
3039 : !> \param qs_env ...
3040 : ! **************************************************************************************************
3041 68 : SUBROUTINE get_V_tr_R(V_tr_R, pot_type, regularization_RI, bs_env, qs_env)
3042 : REAL(KIND=dp), ALLOCATABLE, DIMENSION(:, :, :) :: V_tr_R
3043 : TYPE(libint_potential_type) :: pot_type
3044 : REAL(KIND=dp) :: regularization_RI
3045 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
3046 : TYPE(qs_environment_type), POINTER :: qs_env
3047 :
3048 : CHARACTER(LEN=*), PARAMETER :: routineN = 'get_V_tr_R'
3049 :
3050 : INTEGER :: handle, img, nimages_scf_desymm
3051 : INTEGER, ALLOCATABLE, DIMENSION(:) :: sizes_RI
3052 68 : INTEGER, DIMENSION(:), POINTER :: col_bsize, row_bsize
3053 : TYPE(cp_blacs_env_type), POINTER :: blacs_env
3054 68 : TYPE(cp_fm_type), ALLOCATABLE, DIMENSION(:) :: fm_V_tr_R
3055 : TYPE(dbcsr_distribution_type) :: dbcsr_dist
3056 68 : TYPE(dbcsr_type), ALLOCATABLE, DIMENSION(:) :: mat_V_tr_R
3057 : TYPE(distribution_2d_type), POINTER :: dist_2d
3058 : TYPE(neighbor_list_set_p_type), DIMENSION(:), &
3059 68 : POINTER :: sab_RI
3060 68 : TYPE(particle_type), DIMENSION(:), POINTER :: particle_set
3061 68 : TYPE(qs_kind_type), DIMENSION(:), POINTER :: qs_kind_set
3062 :
3063 68 : CALL timeset(routineN, handle)
3064 :
3065 68 : NULLIFY (sab_RI, dist_2d)
3066 :
3067 : CALL get_qs_env(qs_env=qs_env, &
3068 : blacs_env=blacs_env, &
3069 : distribution_2d=dist_2d, &
3070 : qs_kind_set=qs_kind_set, &
3071 68 : particle_set=particle_set)
3072 :
3073 204 : ALLOCATE (sizes_RI(bs_env%n_atom))
3074 68 : CALL get_particle_set(particle_set, qs_kind_set, nsgf=sizes_RI, basis=bs_env%basis_set_RI)
3075 : CALL build_2c_neighbor_lists(sab_RI, bs_env%basis_set_RI, bs_env%basis_set_RI, &
3076 : pot_type, "2c_nl_RI", qs_env, sym_ij=.FALSE., &
3077 68 : dist_2d=dist_2d)
3078 68 : CALL cp_dbcsr_dist2d_to_dist(dist_2d, dbcsr_dist)
3079 204 : ALLOCATE (row_bsize(SIZE(sizes_RI)))
3080 204 : ALLOCATE (col_bsize(SIZE(sizes_RI)))
3081 244 : row_bsize(:) = sizes_RI
3082 244 : col_bsize(:) = sizes_RI
3083 :
3084 68 : nimages_scf_desymm = bs_env%nimages_scf_desymm
3085 816 : ALLOCATE (mat_V_tr_R(nimages_scf_desymm))
3086 : CALL dbcsr_create(mat_V_tr_R(1), "(RI|RI)", dbcsr_dist, dbcsr_type_no_symmetry, &
3087 68 : row_bsize, col_bsize)
3088 68 : DEALLOCATE (row_bsize, col_bsize)
3089 :
3090 612 : DO img = 2, nimages_scf_desymm
3091 612 : CALL dbcsr_create(mat_V_tr_R(img), template=mat_V_tr_R(1))
3092 : END DO
3093 :
3094 : CALL build_2c_integrals(mat_V_tr_R, 0.0_dp, qs_env, sab_RI, bs_env%basis_set_RI, &
3095 : bs_env%basis_set_RI, pot_type, do_kpoints=.TRUE., &
3096 : ext_kpoints=bs_env%kpoints_scf_desymm, &
3097 68 : regularization_RI=regularization_RI)
3098 :
3099 816 : ALLOCATE (fm_V_tr_R(nimages_scf_desymm))
3100 680 : DO img = 1, nimages_scf_desymm
3101 612 : CALL cp_fm_create(fm_V_tr_R(img), bs_env%fm_RI_RI%matrix_struct)
3102 612 : CALL copy_dbcsr_to_fm(mat_V_tr_R(img), fm_V_tr_R(img))
3103 680 : CALL dbcsr_release(mat_V_tr_R(img))
3104 : END DO
3105 :
3106 68 : IF (.NOT. ALLOCATED(V_tr_R)) THEN
3107 340 : ALLOCATE (V_tr_R(bs_env%n_RI, bs_env%n_RI, nimages_scf_desymm))
3108 : END IF
3109 :
3110 68 : CALL fm_to_local_array(fm_V_tr_R, V_tr_R)
3111 :
3112 68 : CALL cp_fm_release(fm_V_tr_R)
3113 68 : CALL dbcsr_distribution_release(dbcsr_dist)
3114 68 : CALL release_neighbor_list_sets(sab_RI)
3115 :
3116 68 : CALL timestop(handle)
3117 :
3118 204 : END SUBROUTINE get_V_tr_R
3119 :
3120 : ! **************************************************************************************************
3121 : !> \brief ...
3122 : !> \param matrix ...
3123 : !> \param exponent ...
3124 : !> \param eps ...
3125 : !> \param cond_nr ...
3126 : !> \param min_ev ...
3127 : !> \param max_ev ...
3128 : ! **************************************************************************************************
3129 34320 : SUBROUTINE power(matrix, exponent, eps, cond_nr, min_ev, max_ev)
3130 : COMPLEX(KIND=dp), DIMENSION(:, :) :: matrix
3131 : REAL(KIND=dp) :: exponent, eps
3132 : REAL(KIND=dp), OPTIONAL :: cond_nr, min_ev, max_ev
3133 :
3134 : CHARACTER(len=*), PARAMETER :: routineN = 'power'
3135 :
3136 34320 : COMPLEX(KIND=dp), ALLOCATABLE, DIMENSION(:, :) :: eigenvectors
3137 : INTEGER :: handle, i, n
3138 : REAL(KIND=dp) :: pos_eval
3139 34320 : REAL(KIND=dp), ALLOCATABLE, DIMENSION(:) :: eigenvalues
3140 :
3141 34320 : CALL timeset(routineN, handle)
3142 :
3143 : ! make matrix perfectly Hermitian
3144 2559696 : matrix(:, :) = 0.5_dp*(matrix(:, :) + CONJG(TRANSPOSE(matrix(:, :))))
3145 :
3146 34320 : n = SIZE(matrix, 1)
3147 205920 : ALLOCATE (eigenvalues(n), eigenvectors(n, n))
3148 34320 : CALL diag_complex(matrix, eigenvectors, eigenvalues)
3149 :
3150 59920 : IF (PRESENT(cond_nr)) cond_nr = MAXVAL(ABS(eigenvalues))/MINVAL(ABS(eigenvalues))
3151 47120 : IF (PRESENT(min_ev)) min_ev = MINVAL(ABS(eigenvalues))
3152 47120 : IF (PRESENT(max_ev)) max_ev = MAXVAL(ABS(eigenvalues))
3153 :
3154 225344 : DO i = 1, n
3155 191024 : IF (eps < eigenvalues(i)) THEN
3156 191024 : pos_eval = (eigenvalues(i))**(0.5_dp*exponent)
3157 : ELSE
3158 : pos_eval = 0.0_dp
3159 : END IF
3160 1297008 : eigenvectors(:, i) = eigenvectors(:, i)*pos_eval
3161 : END DO
3162 :
3163 34320 : CALL ZGEMM("N", "C", n, n, n, z_one, eigenvectors, n, eigenvectors, n, z_zero, matrix, n)
3164 :
3165 34320 : DEALLOCATE (eigenvalues, eigenvectors)
3166 :
3167 34320 : CALL timestop(handle)
3168 :
3169 34320 : END SUBROUTINE power
3170 :
3171 : ! **************************************************************************************************
3172 : !> \brief ...
3173 : !> \param bs_env ...
3174 : !> \param Sigma_c_n_time ...
3175 : !> \param Sigma_c_n_freq ...
3176 : !> \param ispin ...
3177 : ! **************************************************************************************************
3178 196 : SUBROUTINE time_to_freq(bs_env, Sigma_c_n_time, Sigma_c_n_freq, ispin)
3179 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
3180 : REAL(KIND=dp), DIMENSION(:, :, :) :: Sigma_c_n_time, Sigma_c_n_freq
3181 : INTEGER :: ispin
3182 :
3183 : CHARACTER(LEN=*), PARAMETER :: routineN = 'time_to_freq'
3184 :
3185 : INTEGER :: handle, i_t, j_w, n_occ
3186 : REAL(KIND=dp) :: freq_j, time_i, w_cos_ij, w_sin_ij
3187 196 : REAL(KIND=dp), ALLOCATABLE, DIMENSION(:, :) :: Sigma_c_n_cos_time, Sigma_c_n_sin_time
3188 :
3189 196 : CALL timeset(routineN, handle)
3190 :
3191 784 : ALLOCATE (Sigma_c_n_cos_time(bs_env%n_ao, bs_env%num_time_freq_points))
3192 588 : ALLOCATE (Sigma_c_n_sin_time(bs_env%n_ao, bs_env%num_time_freq_points))
3193 :
3194 19028 : Sigma_c_n_cos_time(:, :) = 0.5_dp*(Sigma_c_n_time(:, :, 1) + Sigma_c_n_time(:, :, 2))
3195 19028 : Sigma_c_n_sin_time(:, :) = 0.5_dp*(Sigma_c_n_time(:, :, 1) - Sigma_c_n_time(:, :, 2))
3196 :
3197 38252 : Sigma_c_n_freq(:, :, :) = 0.0_dp
3198 :
3199 1988 : DO i_t = 1, bs_env%num_time_freq_points
3200 :
3201 20580 : DO j_w = 1, bs_env%num_time_freq_points
3202 :
3203 18592 : freq_j = bs_env%imag_freq_points(j_w)
3204 18592 : time_i = bs_env%imag_time_points(i_t)
3205 : ! integration weights for cosine and sine transform
3206 18592 : w_cos_ij = bs_env%weights_cos_t_to_w(j_w, i_t)*COS(freq_j*time_i)
3207 18592 : w_sin_ij = bs_env%weights_sin_t_to_w(j_w, i_t)*SIN(freq_j*time_i)
3208 :
3209 : ! 1. Re(Σ^c_nn(k_i,iω)) from cosine transform
3210 : Sigma_c_n_freq(:, j_w, 1) = Sigma_c_n_freq(:, j_w, 1) + &
3211 167872 : w_cos_ij*Sigma_c_n_cos_time(:, i_t)
3212 :
3213 : ! 2. Im(Σ^c_nn(k_i,iω)) from sine transform
3214 : Sigma_c_n_freq(:, j_w, 2) = Sigma_c_n_freq(:, j_w, 2) + &
3215 169664 : w_sin_ij*Sigma_c_n_sin_time(:, i_t)
3216 :
3217 : END DO
3218 :
3219 : END DO
3220 :
3221 : ! for occupied levels, we need the correlation self-energy for negative omega.
3222 : ! Therefore, weight_sin should be computed with -omega, which results in an
3223 : ! additional minus for the imaginary part:
3224 196 : n_occ = bs_env%n_occ(ispin)
3225 8356 : Sigma_c_n_freq(1:n_occ, :, 2) = -Sigma_c_n_freq(1:n_occ, :, 2)
3226 :
3227 196 : CALL timestop(handle)
3228 :
3229 392 : END SUBROUTINE time_to_freq
3230 :
3231 : ! **************************************************************************************************
3232 : !> \brief ...
3233 : !> \param bs_env ...
3234 : !> \param Sigma_c_ikp_n_freq ...
3235 : !> \param Sigma_x_ikp_n ...
3236 : !> \param V_xc_ikp_n ...
3237 : !> \param eigenval_scf ...
3238 : !> \param ikp ...
3239 : !> \param ispin ...
3240 : ! **************************************************************************************************
3241 196 : SUBROUTINE analyt_conti_and_print(bs_env, Sigma_c_ikp_n_freq, Sigma_x_ikp_n, V_xc_ikp_n, &
3242 196 : eigenval_scf, ikp, ispin)
3243 :
3244 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
3245 : REAL(KIND=dp), DIMENSION(:, :, :) :: Sigma_c_ikp_n_freq
3246 : REAL(KIND=dp), DIMENSION(:) :: Sigma_x_ikp_n, V_xc_ikp_n, eigenval_scf
3247 : INTEGER :: ikp, ispin
3248 :
3249 : CHARACTER(LEN=*), PARAMETER :: routineN = 'analyt_conti_and_print'
3250 :
3251 : CHARACTER(len=3) :: occ_vir
3252 : CHARACTER(len=default_string_length) :: fname
3253 : INTEGER :: handle, i_mo, ikp_for_print, iunit, &
3254 : n_mo, nkp
3255 : LOGICAL :: is_bandstruc_kpoint, print_DOS_kpoints, &
3256 : print_ikp
3257 : REAL(KIND=dp), ALLOCATABLE, DIMENSION(:) :: dummy, Sigma_c_ikp_n_qp
3258 :
3259 196 : CALL timeset(routineN, handle)
3260 :
3261 196 : n_mo = bs_env%n_ao
3262 784 : ALLOCATE (dummy(n_mo), Sigma_c_ikp_n_qp(n_mo))
3263 2308 : Sigma_c_ikp_n_qp(:) = 0.0_dp
3264 :
3265 2308 : DO i_mo = 1, n_mo
3266 :
3267 : ! parallelization
3268 2112 : IF (MODULO(i_mo, bs_env%para_env%num_pe) /= bs_env%para_env%mepos) CYCLE
3269 :
3270 : CALL continuation_pade(Sigma_c_ikp_n_qp, &
3271 : bs_env%imag_freq_points_fit, dummy, dummy, &
3272 : Sigma_c_ikp_n_freq(:, 1:bs_env%num_freq_points_fit, 1)*z_one + &
3273 : Sigma_c_ikp_n_freq(:, 1:bs_env%num_freq_points_fit, 2)*gaussi, &
3274 : Sigma_x_ikp_n(:) - V_xc_ikp_n(:), &
3275 : eigenval_scf(:), eigenval_scf(:), &
3276 : bs_env%do_hedin_shift, &
3277 : i_mo, bs_env%n_occ(ispin), bs_env%n_vir(ispin), &
3278 : bs_env%nparam_pade, bs_env%num_freq_points_fit, &
3279 : ri_rpa_g0w0_crossing_newton, bs_env%n_occ(ispin), &
3280 68230 : 0.0_dp, .TRUE., .FALSE., 1, e_fermi_ext=bs_env%e_fermi(ispin))
3281 : END DO
3282 :
3283 196 : CALL bs_env%para_env%sum(Sigma_c_ikp_n_qp)
3284 :
3285 196 : CALL correct_obvious_fitting_fails(Sigma_c_ikp_n_qp, ispin, bs_env)
3286 :
3287 : bs_env%eigenval_G0W0(:, ikp, ispin) = eigenval_scf(:) + &
3288 : Sigma_c_ikp_n_qp(:) + &
3289 : Sigma_x_ikp_n(:) - &
3290 2308 : V_xc_ikp_n(:)
3291 :
3292 2308 : bs_env%eigenval_HF(:, ikp, ispin) = eigenval_scf(:) + Sigma_x_ikp_n(:) - V_xc_ikp_n(:)
3293 :
3294 : ! only print eigenvalues of DOS k-points in case no bandstructure path has been given
3295 196 : print_DOS_kpoints = (bs_env%nkp_only_bs <= 0)
3296 : ! in kpoints_DOS, the last nkp_only_bs are bandstructure k-points
3297 196 : is_bandstruc_kpoint = (ikp > bs_env%nkp_only_DOS)
3298 196 : print_ikp = print_DOS_kpoints .OR. is_bandstruc_kpoint
3299 :
3300 196 : IF (bs_env%para_env%is_source() .AND. print_ikp) THEN
3301 :
3302 82 : IF (print_DOS_kpoints) THEN
3303 51 : nkp = bs_env%nkp_only_DOS
3304 51 : ikp_for_print = ikp
3305 : ELSE
3306 31 : nkp = bs_env%nkp_only_bs
3307 31 : ikp_for_print = ikp - bs_env%nkp_only_DOS
3308 : END IF
3309 :
3310 82 : fname = "bandstructure_SCF_and_G0W0"
3311 :
3312 82 : IF (ikp_for_print == 1) THEN
3313 : CALL open_file(TRIM(fname), unit_number=iunit, file_status="REPLACE", &
3314 16 : file_action="WRITE")
3315 : ELSE
3316 : CALL open_file(TRIM(fname), unit_number=iunit, file_status="OLD", &
3317 66 : file_action="WRITE", file_position="APPEND")
3318 : END IF
3319 :
3320 82 : WRITE (iunit, "(A)") " "
3321 82 : WRITE (iunit, "(A10,I7,A25,3F10.4)") "kpoint: ", ikp_for_print, "coordinate: ", &
3322 164 : bs_env%kpoints_DOS%xkp(:, ikp)
3323 82 : WRITE (iunit, "(A)") " "
3324 82 : WRITE (iunit, "(A5,A12,3A17,A16,A18)") "n", "k", "ϵ_nk^DFT (eV)", "Σ^c_nk (eV)", &
3325 164 : "Σ^x_nk (eV)", "v_nk^xc (eV)", "ϵ_nk^G0W0 (eV)"
3326 82 : WRITE (iunit, "(A)") " "
3327 :
3328 994 : DO i_mo = 1, n_mo
3329 912 : IF (i_mo <= bs_env%n_occ(ispin)) occ_vir = 'occ'
3330 912 : IF (i_mo > bs_env%n_occ(ispin)) occ_vir = 'vir'
3331 912 : WRITE (iunit, "(I5,3A,I5,4F16.3,F17.3)") i_mo, ' (', occ_vir, ') ', ikp_for_print, &
3332 912 : eigenval_scf(i_mo)*evolt, &
3333 912 : Sigma_c_ikp_n_qp(i_mo)*evolt, &
3334 912 : Sigma_x_ikp_n(i_mo)*evolt, &
3335 912 : V_xc_ikp_n(i_mo)*evolt, &
3336 1906 : bs_env%eigenval_G0W0(i_mo, ikp, ispin)*evolt
3337 : END DO
3338 :
3339 82 : WRITE (iunit, "(A)") " "
3340 :
3341 82 : CALL close_file(iunit)
3342 :
3343 : END IF
3344 :
3345 196 : CALL timestop(handle)
3346 :
3347 392 : END SUBROUTINE analyt_conti_and_print
3348 :
3349 : ! **************************************************************************************************
3350 : !> \brief ...
3351 : !> \param Sigma_c_ikp_n_qp ...
3352 : !> \param ispin ...
3353 : !> \param bs_env ...
3354 : ! **************************************************************************************************
3355 196 : SUBROUTINE correct_obvious_fitting_fails(Sigma_c_ikp_n_qp, ispin, bs_env)
3356 : REAL(KIND=dp), ALLOCATABLE, DIMENSION(:) :: Sigma_c_ikp_n_qp
3357 : INTEGER :: ispin
3358 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
3359 :
3360 : CHARACTER(LEN=*), PARAMETER :: routineN = 'correct_obvious_fitting_fails'
3361 :
3362 : INTEGER :: handle, homo, i_mo, j_mo, &
3363 : n_levels_scissor, n_mo
3364 : LOGICAL :: is_occ, is_vir
3365 : REAL(KIND=dp) :: sum_Sigma_c
3366 :
3367 196 : CALL timeset(routineN, handle)
3368 :
3369 196 : n_mo = bs_env%n_ao
3370 196 : homo = bs_env%n_occ(ispin)
3371 :
3372 2308 : DO i_mo = 1, n_mo
3373 :
3374 : ! if |𝚺^c| > 13 eV, we use a scissors shift
3375 2308 : IF (ABS(Sigma_c_ikp_n_qp(i_mo)) > 13.0_dp/evolt) THEN
3376 :
3377 0 : is_occ = (i_mo <= homo)
3378 0 : is_vir = (i_mo > homo)
3379 :
3380 0 : n_levels_scissor = 0
3381 0 : sum_Sigma_c = 0.0_dp
3382 :
3383 : ! compute scissor
3384 0 : DO j_mo = 1, n_mo
3385 :
3386 : ! only compute scissor from other GW levels close in energy
3387 0 : IF (is_occ .AND. j_mo > homo) CYCLE
3388 0 : IF (is_vir .AND. j_mo <= homo) CYCLE
3389 0 : IF (ABS(i_mo - j_mo) > 10) CYCLE
3390 0 : IF (i_mo == j_mo) CYCLE
3391 :
3392 0 : n_levels_scissor = n_levels_scissor + 1
3393 0 : sum_Sigma_c = sum_Sigma_c + Sigma_c_ikp_n_qp(j_mo)
3394 :
3395 : END DO
3396 :
3397 : ! overwrite the self-energy with scissor shift
3398 0 : Sigma_c_ikp_n_qp(i_mo) = sum_Sigma_c/REAL(n_levels_scissor, KIND=dp)
3399 :
3400 : END IF
3401 :
3402 : END DO ! i_mo
3403 :
3404 196 : CALL timestop(handle)
3405 :
3406 196 : END SUBROUTINE correct_obvious_fitting_fails
3407 :
3408 : END MODULE gw_utils
|
