Line data Source code
1 : !--------------------------------------------------------------------------------------------------!
2 : ! CP2K: A general program to perform molecular dynamics simulations !
3 : ! Copyright 2000-2025 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 : USE cp_blacs_env, ONLY: cp_blacs_env_create,&
23 : cp_blacs_env_release,&
24 : cp_blacs_env_type
25 : USE cp_cfm_types, ONLY: cp_cfm_create,&
26 : cp_cfm_release,&
27 : cp_cfm_to_cfm,&
28 : cp_cfm_to_fm,&
29 : cp_cfm_type
30 : USE cp_control_types, ONLY: dft_control_type
31 : USE cp_dbcsr_api, ONLY: &
32 : dbcsr_create, dbcsr_distribution_release, dbcsr_distribution_type, dbcsr_p_type, &
33 : dbcsr_release, dbcsr_set, dbcsr_type, dbcsr_type_no_symmetry, dbcsr_type_symmetric
34 : USE cp_dbcsr_operations, ONLY: copy_dbcsr_to_fm,&
35 : copy_fm_to_dbcsr,&
36 : cp_dbcsr_dist2d_to_dist,&
37 : dbcsr_allocate_matrix_set,&
38 : dbcsr_deallocate_matrix_set
39 : USE cp_files, ONLY: close_file,&
40 : open_file
41 : USE cp_fm_basic_linalg, ONLY: cp_fm_scale_and_add
42 : USE cp_fm_struct, ONLY: cp_fm_struct_create,&
43 : cp_fm_struct_release,&
44 : cp_fm_struct_type
45 : USE cp_fm_types, ONLY: cp_fm_create,&
46 : cp_fm_get_diag,&
47 : cp_fm_release,&
48 : cp_fm_set_all,&
49 : cp_fm_type
50 : USE cp_log_handling, ONLY: cp_get_default_logger,&
51 : cp_logger_type
52 : USE cp_output_handling, ONLY: cp_print_key_generate_filename
53 : USE dbt_api, ONLY: &
54 : dbt_clear, dbt_create, dbt_destroy, dbt_filter, dbt_iterator_blocks_left, &
55 : dbt_iterator_next_block, dbt_iterator_start, dbt_iterator_stop, dbt_iterator_type, &
56 : dbt_mp_environ_pgrid, dbt_pgrid_create, dbt_pgrid_destroy, dbt_pgrid_type, dbt_type
57 : USE distribution_2d_types, ONLY: distribution_2d_type
58 : USE gw_communication, ONLY: fm_to_local_array
59 : USE gw_integrals, ONLY: build_3c_integral_block
60 : USE gw_kp_to_real_space_and_back, ONLY: trafo_rs_to_ikp
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_string_length,&
73 : dp,&
74 : int_8
75 : USE kpoint_methods, ONLY: kpoint_init_cell_index
76 : USE kpoint_types, ONLY: get_kpoint_info,&
77 : kpoint_create,&
78 : kpoint_type
79 : USE libint_2c_3c, ONLY: libint_potential_type
80 : USE libint_wrapper, ONLY: cp_libint_static_cleanup,&
81 : cp_libint_static_init
82 : USE machine, ONLY: m_memory,&
83 : m_walltime
84 : USE mathconstants, ONLY: gaussi,&
85 : z_one,&
86 : z_zero
87 : USE mathlib, ONLY: diag_complex,&
88 : gcd
89 : USE message_passing, ONLY: mp_cart_type,&
90 : mp_para_env_type
91 : USE minimax_exp, ONLY: get_exp_minimax_coeff
92 : USE minimax_exp_gw, ONLY: get_exp_minimax_coeff_gw
93 : USE minimax_rpa, ONLY: get_rpa_minimax_coeff,&
94 : get_rpa_minimax_coeff_larger_grid
95 : USE mp2_gpw, ONLY: create_mat_munu
96 : USE mp2_grids, ONLY: get_l_sq_wghts_cos_tf_t_to_w,&
97 : get_l_sq_wghts_cos_tf_w_to_t,&
98 : get_l_sq_wghts_sin_tf_t_to_w
99 : USE mp2_ri_2c, ONLY: trunc_coulomb_for_exchange
100 : USE parallel_gemm_api, ONLY: parallel_gemm
101 : USE particle_methods, ONLY: get_particle_set
102 : USE particle_types, ONLY: particle_type
103 : USE physcon, ONLY: angstrom,&
104 : evolt
105 : USE post_scf_bandstructure_types, ONLY: post_scf_bandstructure_type
106 : USE post_scf_bandstructure_utils, ONLY: rsmat_to_kp
107 : USE qs_energy_types, ONLY: qs_energy_type
108 : USE qs_environment_types, ONLY: get_qs_env,&
109 : qs_env_part_release,&
110 : qs_environment_type
111 : USE qs_integral_utils, ONLY: basis_set_list_setup
112 : USE qs_interactions, ONLY: init_interaction_radii_orb_basis
113 : USE qs_kind_types, ONLY: get_qs_kind,&
114 : qs_kind_type
115 : USE qs_ks_methods, ONLY: qs_ks_build_kohn_sham_matrix
116 : USE qs_neighbor_list_types, ONLY: neighbor_list_set_p_type,&
117 : release_neighbor_list_sets
118 : USE qs_tensors, ONLY: build_2c_integrals,&
119 : build_2c_neighbor_lists,&
120 : build_3c_integrals,&
121 : build_3c_neighbor_lists,&
122 : get_tensor_occupancy,&
123 : neighbor_list_3c_destroy
124 : USE qs_tensors_types, ONLY: create_2c_tensor,&
125 : create_3c_tensor,&
126 : distribution_3d_create,&
127 : distribution_3d_type,&
128 : neighbor_list_3c_type
129 : USE rpa_gw, ONLY: continuation_pade
130 : #include "base/base_uses.f90"
131 :
132 : IMPLICIT NONE
133 :
134 : PRIVATE
135 :
136 : PUBLIC :: create_and_init_bs_env_for_gw, de_init_bs_env, get_i_j_atoms, &
137 : kpoint_init_cell_index_simple, compute_xkp, time_to_freq, analyt_conti_and_print, &
138 : add_R, is_cell_in_index_to_cell, get_V_tr_R, power
139 :
140 : CHARACTER(len=*), PARAMETER, PRIVATE :: moduleN = 'gw_utils'
141 :
142 : CONTAINS
143 :
144 : ! **************************************************************************************************
145 : !> \brief ...
146 : !> \param qs_env ...
147 : !> \param bs_env ...
148 : !> \param bs_sec ...
149 : ! **************************************************************************************************
150 34 : SUBROUTINE create_and_init_bs_env_for_gw(qs_env, bs_env, bs_sec)
151 : TYPE(qs_environment_type), POINTER :: qs_env
152 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
153 : TYPE(section_vals_type), POINTER :: bs_sec
154 :
155 : CHARACTER(LEN=*), PARAMETER :: routineN = 'create_and_init_bs_env_for_gw'
156 :
157 : INTEGER :: handle
158 :
159 34 : CALL timeset(routineN, handle)
160 :
161 34 : CALL cite_reference(Graml2024)
162 :
163 34 : CALL read_gw_input_parameters(bs_env, bs_sec)
164 :
165 34 : CALL print_header_and_input_parameters(bs_env)
166 :
167 34 : CALL setup_AO_and_RI_basis_set(qs_env, bs_env)
168 :
169 34 : CALL get_RI_basis_and_basis_function_indices(qs_env, bs_env)
170 :
171 34 : CALL set_heuristic_parameters(bs_env, qs_env)
172 :
173 34 : CALL cp_libint_static_init()
174 :
175 34 : CALL setup_kpoints_chi_eps_W(bs_env, bs_env%kpoints_chi_eps_W)
176 :
177 34 : IF (bs_env%small_cell_full_kp_or_large_cell_Gamma == small_cell_full_kp) THEN
178 6 : CALL setup_cells_3c(qs_env, bs_env)
179 : END IF
180 :
181 34 : CALL set_parallelization_parameters(qs_env, bs_env)
182 :
183 34 : CALL allocate_matrices(qs_env, bs_env)
184 :
185 34 : CALL compute_V_xc(qs_env, bs_env)
186 :
187 34 : CALL create_tensors(qs_env, bs_env)
188 :
189 62 : SELECT CASE (bs_env%small_cell_full_kp_or_large_cell_Gamma)
190 : CASE (large_cell_Gamma)
191 :
192 28 : CALL allocate_GW_eigenvalues(bs_env)
193 :
194 28 : CALL check_sparsity_3c(qs_env, bs_env)
195 :
196 28 : CALL set_sparsity_parallelization_parameters(bs_env)
197 :
198 28 : CALL check_for_restart_files(qs_env, bs_env)
199 :
200 : CASE (small_cell_full_kp)
201 :
202 6 : CALL compute_3c_integrals(qs_env, bs_env)
203 :
204 6 : CALL setup_cells_Delta_R(bs_env)
205 :
206 6 : CALL setup_parallelization_Delta_R(bs_env)
207 :
208 6 : CALL allocate_matrices_small_cell_full_kp(qs_env, bs_env)
209 :
210 6 : CALL trafo_V_xc_R_to_kp(qs_env, bs_env)
211 :
212 40 : CALL heuristic_RI_regularization(qs_env, bs_env)
213 :
214 : END SELECT
215 :
216 34 : CALL setup_time_and_frequency_minimax_grid(bs_env)
217 :
218 : ! free memory in qs_env; only if one is not calculating the LDOS because
219 : ! we need real-space grid operations in pw_env, task_list for the LDOS
220 : ! Recommendation in case of memory issues: first perform GW calculation without calculating
221 : ! LDOS (to safe memor). Then, use GW restart files
222 : ! in a subsequent calculation to calculate the LDOS
223 : ! Marek : TODO - boolean that does not interfere with RTP init but sets this to correct value
224 : IF (.NOT. bs_env%do_ldos .AND. .FALSE.) THEN
225 : CALL qs_env_part_release(qs_env)
226 : END IF
227 :
228 34 : CALL timestop(handle)
229 :
230 34 : END SUBROUTINE create_and_init_bs_env_for_gw
231 :
232 : ! **************************************************************************************************
233 : !> \brief ...
234 : !> \param bs_env ...
235 : ! **************************************************************************************************
236 34 : SUBROUTINE de_init_bs_env(bs_env)
237 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
238 :
239 : CHARACTER(LEN=*), PARAMETER :: routineN = 'de_init_bs_env'
240 :
241 : INTEGER :: handle
242 :
243 34 : CALL timeset(routineN, handle)
244 : ! deallocate quantities here which:
245 : ! 1. cannot be deallocated in bs_env_release due to circular dependencies
246 : ! 2. consume a lot of memory and should not be kept until the quantity is
247 : ! deallocated in bs_env_release
248 :
249 34 : IF (ASSOCIATED(bs_env%nl_3c%ij_list) .AND. (bs_env%rtp_method == rtp_method_bse)) THEN
250 12 : IF (bs_env%unit_nr > 0) WRITE (bs_env%unit_nr, *) "Retaining nl_3c for RTBSE"
251 : ELSE
252 22 : CALL neighbor_list_3c_destroy(bs_env%nl_3c)
253 : END IF
254 :
255 34 : CALL cp_libint_static_cleanup()
256 :
257 34 : CALL timestop(handle)
258 :
259 34 : END SUBROUTINE de_init_bs_env
260 :
261 : ! **************************************************************************************************
262 : !> \brief ...
263 : !> \param bs_env ...
264 : !> \param bs_sec ...
265 : ! **************************************************************************************************
266 34 : SUBROUTINE read_gw_input_parameters(bs_env, bs_sec)
267 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
268 : TYPE(section_vals_type), POINTER :: bs_sec
269 :
270 : CHARACTER(LEN=*), PARAMETER :: routineN = 'read_gw_input_parameters'
271 :
272 : INTEGER :: handle
273 : TYPE(section_vals_type), POINTER :: gw_sec
274 :
275 34 : CALL timeset(routineN, handle)
276 :
277 34 : NULLIFY (gw_sec)
278 34 : gw_sec => section_vals_get_subs_vals(bs_sec, "GW")
279 :
280 34 : CALL section_vals_val_get(gw_sec, "NUM_TIME_FREQ_POINTS", i_val=bs_env%num_time_freq_points)
281 34 : CALL section_vals_val_get(gw_sec, "EPS_FILTER", r_val=bs_env%eps_filter)
282 34 : CALL section_vals_val_get(gw_sec, "REGULARIZATION_RI", r_val=bs_env%input_regularization_RI)
283 34 : CALL section_vals_val_get(gw_sec, "CUTOFF_RADIUS_RI", r_val=bs_env%ri_metric%cutoff_radius)
284 34 : CALL section_vals_val_get(gw_sec, "MEMORY_PER_PROC", r_val=bs_env%input_memory_per_proc_GB)
285 34 : CALL section_vals_val_get(gw_sec, "APPROX_KP_EXTRAPOL", l_val=bs_env%approx_kp_extrapol)
286 34 : CALL section_vals_val_get(gw_sec, "SIZE_LATTICE_SUM", i_val=bs_env%size_lattice_sum_V)
287 34 : CALL section_vals_val_get(gw_sec, "KPOINTS_W", i_vals=bs_env%nkp_grid_chi_eps_W_input)
288 34 : CALL section_vals_val_get(gw_sec, "HEDIN_SHIFT", l_val=bs_env%do_hedin_shift)
289 :
290 34 : CALL timestop(handle)
291 :
292 34 : END SUBROUTINE read_gw_input_parameters
293 :
294 : ! **************************************************************************************************
295 : !> \brief ...
296 : !> \param qs_env ...
297 : !> \param bs_env ...
298 : ! **************************************************************************************************
299 34 : SUBROUTINE setup_AO_and_RI_basis_set(qs_env, bs_env)
300 : TYPE(qs_environment_type), POINTER :: qs_env
301 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
302 :
303 : CHARACTER(LEN=*), PARAMETER :: routineN = 'setup_AO_and_RI_basis_set'
304 :
305 : INTEGER :: handle, natom, nkind
306 34 : TYPE(particle_type), DIMENSION(:), POINTER :: particle_set
307 34 : TYPE(qs_kind_type), DIMENSION(:), POINTER :: qs_kind_set
308 :
309 34 : CALL timeset(routineN, handle)
310 :
311 : CALL get_qs_env(qs_env, &
312 : qs_kind_set=qs_kind_set, &
313 : particle_set=particle_set, &
314 34 : natom=natom, nkind=nkind)
315 :
316 : ! set up basis
317 170 : ALLOCATE (bs_env%sizes_RI(natom), bs_env%sizes_AO(natom))
318 282 : ALLOCATE (bs_env%basis_set_RI(nkind), bs_env%basis_set_AO(nkind))
319 :
320 34 : CALL basis_set_list_setup(bs_env%basis_set_RI, "RI_AUX", qs_kind_set)
321 34 : CALL basis_set_list_setup(bs_env%basis_set_AO, "ORB", qs_kind_set)
322 :
323 : CALL get_particle_set(particle_set, qs_kind_set, nsgf=bs_env%sizes_RI, &
324 34 : basis=bs_env%basis_set_RI)
325 : CALL get_particle_set(particle_set, qs_kind_set, nsgf=bs_env%sizes_AO, &
326 34 : basis=bs_env%basis_set_AO)
327 :
328 34 : CALL timestop(handle)
329 :
330 34 : END SUBROUTINE setup_AO_and_RI_basis_set
331 :
332 : ! **************************************************************************************************
333 : !> \brief ...
334 : !> \param qs_env ...
335 : !> \param bs_env ...
336 : ! **************************************************************************************************
337 34 : SUBROUTINE get_RI_basis_and_basis_function_indices(qs_env, bs_env)
338 : TYPE(qs_environment_type), POINTER :: qs_env
339 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
340 :
341 : CHARACTER(LEN=*), PARAMETER :: routineN = 'get_RI_basis_and_basis_function_indices'
342 :
343 : INTEGER :: handle, i_RI, iatom, ikind, iset, &
344 : max_AO_bf_per_atom, n_ao_test, n_atom, &
345 : n_kind, n_RI, nset, nsgf, u
346 34 : INTEGER, ALLOCATABLE, DIMENSION(:) :: kind_of
347 34 : INTEGER, DIMENSION(:), POINTER :: l_max, l_min, nsgf_set
348 34 : TYPE(atomic_kind_type), DIMENSION(:), POINTER :: atomic_kind_set
349 : TYPE(gto_basis_set_type), POINTER :: basis_set_a
350 34 : TYPE(qs_kind_type), DIMENSION(:), POINTER :: qs_kind_set
351 :
352 34 : CALL timeset(routineN, handle)
353 :
354 : ! determine RI basis set size
355 34 : CALL get_qs_env(qs_env, atomic_kind_set=atomic_kind_set, qs_kind_set=qs_kind_set)
356 :
357 34 : n_kind = SIZE(qs_kind_set)
358 34 : n_atom = bs_env%n_atom
359 :
360 34 : CALL get_atomic_kind_set(atomic_kind_set, kind_of=kind_of)
361 :
362 90 : DO ikind = 1, n_kind
363 : CALL get_qs_kind(qs_kind=qs_kind_set(ikind), basis_set=basis_set_a, &
364 56 : basis_type="RI_AUX")
365 90 : CPASSERT(ASSOCIATED(basis_set_a))
366 : END DO
367 :
368 102 : ALLOCATE (bs_env%i_RI_start_from_atom(n_atom))
369 68 : ALLOCATE (bs_env%i_RI_end_from_atom(n_atom))
370 68 : ALLOCATE (bs_env%i_ao_start_from_atom(n_atom))
371 68 : ALLOCATE (bs_env%i_ao_end_from_atom(n_atom))
372 :
373 34 : n_RI = 0
374 108 : DO iatom = 1, n_atom
375 74 : bs_env%i_RI_start_from_atom(iatom) = n_RI + 1
376 74 : ikind = kind_of(iatom)
377 74 : CALL get_qs_kind(qs_kind=qs_kind_set(ikind), nsgf=nsgf, basis_type="RI_AUX")
378 74 : n_RI = n_RI + nsgf
379 108 : bs_env%i_RI_end_from_atom(iatom) = n_RI
380 : END DO
381 34 : bs_env%n_RI = n_RI
382 :
383 34 : max_AO_bf_per_atom = 0
384 34 : n_ao_test = 0
385 108 : DO iatom = 1, n_atom
386 74 : bs_env%i_ao_start_from_atom(iatom) = n_ao_test + 1
387 74 : ikind = kind_of(iatom)
388 74 : CALL get_qs_kind(qs_kind=qs_kind_set(ikind), nsgf=nsgf, basis_type="ORB")
389 74 : n_ao_test = n_ao_test + nsgf
390 74 : bs_env%i_ao_end_from_atom(iatom) = n_ao_test
391 108 : max_AO_bf_per_atom = MAX(max_AO_bf_per_atom, nsgf)
392 : END DO
393 34 : CPASSERT(n_ao_test == bs_env%n_ao)
394 34 : bs_env%max_AO_bf_per_atom = max_AO_bf_per_atom
395 :
396 102 : ALLOCATE (bs_env%l_RI(n_RI))
397 34 : i_RI = 0
398 108 : DO iatom = 1, n_atom
399 74 : ikind = kind_of(iatom)
400 :
401 74 : nset = bs_env%basis_set_RI(ikind)%gto_basis_set%nset
402 74 : l_max => bs_env%basis_set_RI(ikind)%gto_basis_set%lmax
403 74 : l_min => bs_env%basis_set_RI(ikind)%gto_basis_set%lmin
404 74 : nsgf_set => bs_env%basis_set_RI(ikind)%gto_basis_set%nsgf_set
405 :
406 300 : DO iset = 1, nset
407 192 : CPASSERT(l_max(iset) == l_min(iset))
408 580 : bs_env%l_RI(i_RI + 1:i_RI + nsgf_set(iset)) = l_max(iset)
409 266 : i_RI = i_RI + nsgf_set(iset)
410 : END DO
411 :
412 : END DO
413 34 : CPASSERT(i_RI == n_RI)
414 :
415 34 : u = bs_env%unit_nr
416 :
417 34 : IF (u > 0) THEN
418 17 : WRITE (u, FMT="(T2,A)") " "
419 17 : WRITE (u, FMT="(T2,2A,T75,I8)") "Number of auxiliary Gaussian basis functions ", &
420 34 : "for χ, ε, W", n_RI
421 : END IF
422 :
423 34 : CALL timestop(handle)
424 :
425 68 : END SUBROUTINE get_RI_basis_and_basis_function_indices
426 :
427 : ! **************************************************************************************************
428 : !> \brief ...
429 : !> \param bs_env ...
430 : !> \param kpoints ...
431 : ! **************************************************************************************************
432 34 : SUBROUTINE setup_kpoints_chi_eps_W(bs_env, kpoints)
433 :
434 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
435 : TYPE(kpoint_type), POINTER :: kpoints
436 :
437 : CHARACTER(LEN=*), PARAMETER :: routineN = 'setup_kpoints_chi_eps_W'
438 :
439 : INTEGER :: handle, i_dim, n_dim, nkp, nkp_extra, &
440 : nkp_orig, u
441 : INTEGER, DIMENSION(3) :: nkp_grid, nkp_grid_extra, periodic
442 : REAL(KIND=dp) :: exp_s_p, n_dim_inv
443 :
444 34 : CALL timeset(routineN, handle)
445 :
446 : ! routine adapted from mp2_integrals.F
447 34 : NULLIFY (kpoints)
448 34 : CALL kpoint_create(kpoints)
449 :
450 34 : kpoints%kp_scheme = "GENERAL"
451 :
452 136 : periodic(1:3) = bs_env%periodic(1:3)
453 :
454 34 : CPASSERT(SIZE(bs_env%nkp_grid_chi_eps_W_input) == 3)
455 :
456 : IF (bs_env%nkp_grid_chi_eps_W_input(1) > 0 .AND. &
457 34 : bs_env%nkp_grid_chi_eps_W_input(2) > 0 .AND. &
458 : bs_env%nkp_grid_chi_eps_W_input(3) > 0) THEN
459 : ! 1. k-point mesh for χ, ε, W from input
460 0 : DO i_dim = 1, 3
461 0 : SELECT CASE (periodic(i_dim))
462 : CASE (0)
463 0 : nkp_grid(i_dim) = 1
464 0 : nkp_grid_extra(i_dim) = 1
465 : CASE (1)
466 0 : nkp_grid(i_dim) = bs_env%nkp_grid_chi_eps_W_input(i_dim)
467 0 : nkp_grid_extra(i_dim) = nkp_grid(i_dim)*2
468 : CASE DEFAULT
469 0 : CPABORT("Error in periodicity.")
470 : END SELECT
471 : END DO
472 :
473 : ELSE IF (bs_env%nkp_grid_chi_eps_W_input(1) == -1 .AND. &
474 34 : bs_env%nkp_grid_chi_eps_W_input(2) == -1 .AND. &
475 : bs_env%nkp_grid_chi_eps_W_input(3) == -1) THEN
476 : ! 2. automatic k-point mesh for χ, ε, W
477 :
478 136 : DO i_dim = 1, 3
479 :
480 102 : CPASSERT(periodic(i_dim) == 0 .OR. periodic(i_dim) == 1)
481 :
482 34 : SELECT CASE (periodic(i_dim))
483 : CASE (0)
484 58 : nkp_grid(i_dim) = 1
485 58 : nkp_grid_extra(i_dim) = 1
486 : CASE (1)
487 44 : SELECT CASE (bs_env%small_cell_full_kp_or_large_cell_Gamma)
488 : CASE (large_cell_Gamma)
489 32 : nkp_grid(i_dim) = 4
490 32 : nkp_grid_extra(i_dim) = 6
491 : CASE (small_cell_full_kp)
492 12 : nkp_grid(i_dim) = bs_env%kpoints_scf_desymm%nkp_grid(i_dim)*4
493 12 : nkp_grid_extra(i_dim) = bs_env%kpoints_scf_desymm%nkp_grid(i_dim)*8
494 : END SELECT
495 : CASE DEFAULT
496 102 : CPABORT("Error in periodicity.")
497 : END SELECT
498 :
499 : END DO
500 :
501 : ELSE
502 :
503 0 : CPABORT("An error occured when setting up the k-mesh for W.")
504 :
505 : END IF
506 :
507 34 : nkp_orig = MAX(nkp_grid(1)*nkp_grid(2)*nkp_grid(3)/2, 1)
508 :
509 34 : nkp_extra = nkp_grid_extra(1)*nkp_grid_extra(2)*nkp_grid_extra(3)/2
510 :
511 34 : nkp = nkp_orig + nkp_extra
512 :
513 136 : kpoints%nkp_grid(1:3) = nkp_grid(1:3)
514 34 : kpoints%nkp = nkp
515 :
516 136 : bs_env%nkp_grid_chi_eps_W_orig(1:3) = nkp_grid(1:3)
517 136 : bs_env%nkp_grid_chi_eps_W_extra(1:3) = nkp_grid_extra(1:3)
518 34 : bs_env%nkp_chi_eps_W_orig = nkp_orig
519 34 : bs_env%nkp_chi_eps_W_extra = nkp_extra
520 34 : bs_env%nkp_chi_eps_W_orig_plus_extra = nkp
521 :
522 170 : ALLOCATE (kpoints%xkp(3, nkp), kpoints%wkp(nkp))
523 170 : ALLOCATE (bs_env%wkp_no_extra(nkp), bs_env%wkp_s_p(nkp))
524 :
525 34 : CALL compute_xkp(kpoints%xkp, 1, nkp_orig, nkp_grid)
526 34 : CALL compute_xkp(kpoints%xkp, nkp_orig + 1, nkp, nkp_grid_extra)
527 :
528 136 : n_dim = SUM(periodic)
529 34 : IF (n_dim == 0) THEN
530 : ! molecules
531 12 : kpoints%wkp(1) = 1.0_dp
532 12 : bs_env%wkp_s_p(1) = 1.0_dp
533 12 : bs_env%wkp_no_extra(1) = 1.0_dp
534 : ELSE
535 :
536 22 : n_dim_inv = 1.0_dp/REAL(n_dim, KIND=dp)
537 :
538 : ! k-point weights are chosen to automatically extrapolate the k-point mesh
539 22 : CALL compute_wkp(kpoints%wkp(1:nkp_orig), nkp_orig, nkp_extra, n_dim_inv)
540 22 : CALL compute_wkp(kpoints%wkp(nkp_orig + 1:nkp), nkp_extra, nkp_orig, n_dim_inv)
541 :
542 918 : bs_env%wkp_no_extra(1:nkp_orig) = 0.0_dp
543 3382 : bs_env%wkp_no_extra(nkp_orig + 1:nkp) = 1.0_dp/REAL(nkp_extra, KIND=dp)
544 :
545 22 : IF (n_dim == 3) THEN
546 : ! W_PQ(k) for an s-function P and a p-function Q diverges as 1/k at k=0
547 : ! (instead of 1/k^2 for P and Q both being s-functions).
548 0 : exp_s_p = 2.0_dp*n_dim_inv
549 0 : CALL compute_wkp(bs_env%wkp_s_p(1:nkp_orig), nkp_orig, nkp_extra, exp_s_p)
550 0 : CALL compute_wkp(bs_env%wkp_s_p(nkp_orig + 1:nkp), nkp_extra, nkp_orig, exp_s_p)
551 : ELSE
552 4278 : bs_env%wkp_s_p(1:nkp) = bs_env%wkp_no_extra(1:nkp)
553 : END IF
554 :
555 : END IF
556 :
557 34 : IF (bs_env%approx_kp_extrapol) THEN
558 2 : bs_env%wkp_orig = 1.0_dp/REAL(nkp_orig, KIND=dp)
559 : END IF
560 :
561 : ! heuristic parameter: how many k-points for χ, ε, and W are used simultaneously
562 : ! (less simultaneous k-points: less memory, but more computational effort because of
563 : ! recomputation of V(k))
564 34 : bs_env%nkp_chi_eps_W_batch = 4
565 :
566 : bs_env%num_chi_eps_W_batches = (bs_env%nkp_chi_eps_W_orig_plus_extra - 1)/ &
567 34 : bs_env%nkp_chi_eps_W_batch + 1
568 :
569 34 : u = bs_env%unit_nr
570 :
571 34 : IF (u > 0) THEN
572 17 : WRITE (u, FMT="(T2,A)") " "
573 17 : WRITE (u, FMT="(T2,1A,T71,3I4)") "K-point mesh 1 for χ, ε, W", nkp_grid(1:3)
574 17 : WRITE (u, FMT="(T2,2A,T71,3I4)") "K-point mesh 2 for χ, ε, W ", &
575 34 : "(for k-point extrapolation of W)", nkp_grid_extra(1:3)
576 17 : WRITE (u, FMT="(T2,A,T80,L)") "Approximate the k-point extrapolation", &
577 34 : bs_env%approx_kp_extrapol
578 : END IF
579 :
580 34 : CALL timestop(handle)
581 :
582 34 : END SUBROUTINE setup_kpoints_chi_eps_W
583 :
584 : ! **************************************************************************************************
585 : !> \brief ...
586 : !> \param kpoints ...
587 : !> \param qs_env ...
588 : ! **************************************************************************************************
589 0 : SUBROUTINE kpoint_init_cell_index_simple(kpoints, qs_env)
590 :
591 : TYPE(kpoint_type), POINTER :: kpoints
592 : TYPE(qs_environment_type), POINTER :: qs_env
593 :
594 : CHARACTER(LEN=*), PARAMETER :: routineN = 'kpoint_init_cell_index_simple'
595 :
596 : INTEGER :: handle
597 : TYPE(dft_control_type), POINTER :: dft_control
598 : TYPE(mp_para_env_type), POINTER :: para_env
599 : TYPE(neighbor_list_set_p_type), DIMENSION(:), &
600 0 : POINTER :: sab_orb
601 :
602 0 : CALL timeset(routineN, handle)
603 :
604 0 : NULLIFY (dft_control, para_env, sab_orb)
605 0 : CALL get_qs_env(qs_env=qs_env, para_env=para_env, dft_control=dft_control, sab_orb=sab_orb)
606 0 : CALL kpoint_init_cell_index(kpoints, sab_orb, para_env, dft_control)
607 :
608 0 : CALL timestop(handle)
609 :
610 0 : END SUBROUTINE kpoint_init_cell_index_simple
611 :
612 : ! **************************************************************************************************
613 : !> \brief ...
614 : !> \param xkp ...
615 : !> \param ikp_start ...
616 : !> \param ikp_end ...
617 : !> \param grid ...
618 : ! **************************************************************************************************
619 68 : SUBROUTINE compute_xkp(xkp, ikp_start, ikp_end, grid)
620 :
621 : REAL(KIND=dp), DIMENSION(:, :), POINTER :: xkp
622 : INTEGER :: ikp_start, ikp_end
623 : INTEGER, DIMENSION(3) :: grid
624 :
625 : CHARACTER(LEN=*), PARAMETER :: routineN = 'compute_xkp'
626 :
627 : INTEGER :: handle, i, ix, iy, iz
628 :
629 68 : CALL timeset(routineN, handle)
630 :
631 68 : i = ikp_start
632 384 : DO ix = 1, grid(1)
633 6120 : DO iy = 1, grid(2)
634 14588 : DO iz = 1, grid(3)
635 :
636 8536 : IF (i > ikp_end) CYCLE
637 :
638 4268 : xkp(1, i) = REAL(2*ix - grid(1) - 1, KIND=dp)/(2._dp*REAL(grid(1), KIND=dp))
639 4268 : xkp(2, i) = REAL(2*iy - grid(2) - 1, KIND=dp)/(2._dp*REAL(grid(2), KIND=dp))
640 4268 : xkp(3, i) = REAL(2*iz - grid(3) - 1, KIND=dp)/(2._dp*REAL(grid(3), KIND=dp))
641 14272 : i = i + 1
642 :
643 : END DO
644 : END DO
645 : END DO
646 :
647 68 : CALL timestop(handle)
648 :
649 68 : END SUBROUTINE compute_xkp
650 :
651 : ! **************************************************************************************************
652 : !> \brief ...
653 : !> \param wkp ...
654 : !> \param nkp_1 ...
655 : !> \param nkp_2 ...
656 : !> \param exponent ...
657 : ! **************************************************************************************************
658 44 : SUBROUTINE compute_wkp(wkp, nkp_1, nkp_2, exponent)
659 : REAL(KIND=dp), DIMENSION(:) :: wkp
660 : INTEGER :: nkp_1, nkp_2
661 : REAL(KIND=dp) :: exponent
662 :
663 : CHARACTER(LEN=*), PARAMETER :: routineN = 'compute_wkp'
664 :
665 : INTEGER :: handle
666 : REAL(KIND=dp) :: nkp_ratio
667 :
668 44 : CALL timeset(routineN, handle)
669 :
670 44 : nkp_ratio = REAL(nkp_2, KIND=dp)/REAL(nkp_1, KIND=dp)
671 :
672 4300 : wkp(:) = 1.0_dp/REAL(nkp_1, KIND=dp)/(1.0_dp - nkp_ratio**exponent)
673 :
674 44 : CALL timestop(handle)
675 :
676 44 : END SUBROUTINE compute_wkp
677 :
678 : ! **************************************************************************************************
679 : !> \brief ...
680 : !> \param qs_env ...
681 : !> \param bs_env ...
682 : ! **************************************************************************************************
683 34 : SUBROUTINE allocate_matrices(qs_env, bs_env)
684 : TYPE(qs_environment_type), POINTER :: qs_env
685 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
686 :
687 : CHARACTER(LEN=*), PARAMETER :: routineN = 'allocate_matrices'
688 :
689 : INTEGER :: handle, i_t
690 : TYPE(cp_blacs_env_type), POINTER :: blacs_env, blacs_env_tensor
691 : TYPE(cp_fm_struct_type), POINTER :: fm_struct, fm_struct_RI_global
692 : TYPE(mp_para_env_type), POINTER :: para_env
693 :
694 34 : CALL timeset(routineN, handle)
695 :
696 34 : CALL get_qs_env(qs_env, para_env=para_env, blacs_env=blacs_env)
697 :
698 34 : fm_struct => bs_env%fm_ks_Gamma(1)%matrix_struct
699 :
700 34 : CALL cp_fm_create(bs_env%fm_Gocc, fm_struct)
701 34 : CALL cp_fm_create(bs_env%fm_Gvir, fm_struct)
702 :
703 34 : NULLIFY (fm_struct_RI_global)
704 : CALL cp_fm_struct_create(fm_struct_RI_global, context=blacs_env, nrow_global=bs_env%n_RI, &
705 34 : ncol_global=bs_env%n_RI, para_env=para_env)
706 34 : CALL cp_fm_create(bs_env%fm_RI_RI, fm_struct_RI_global)
707 34 : CALL cp_fm_create(bs_env%fm_chi_Gamma_freq, fm_struct_RI_global)
708 34 : CALL cp_fm_create(bs_env%fm_W_MIC_freq, fm_struct_RI_global)
709 34 : IF (bs_env%approx_kp_extrapol) THEN
710 2 : CALL cp_fm_create(bs_env%fm_W_MIC_freq_1_extra, fm_struct_RI_global)
711 2 : CALL cp_fm_create(bs_env%fm_W_MIC_freq_1_no_extra, fm_struct_RI_global)
712 2 : CALL cp_fm_set_all(bs_env%fm_W_MIC_freq_1_extra, 0.0_dp)
713 2 : CALL cp_fm_set_all(bs_env%fm_W_MIC_freq_1_no_extra, 0.0_dp)
714 : END IF
715 34 : CALL cp_fm_struct_release(fm_struct_RI_global)
716 :
717 : ! create blacs_env for subgroups of tensor operations
718 34 : NULLIFY (blacs_env_tensor)
719 34 : CALL cp_blacs_env_create(blacs_env=blacs_env_tensor, para_env=bs_env%para_env_tensor)
720 :
721 : ! allocate dbcsr matrices in the tensor subgroup; actually, one only needs a small
722 : ! subset of blocks in the tensor subgroup, however, all atomic blocks are allocated.
723 : ! One might think of creating a dbcsr matrix with only the blocks that are needed
724 : ! in the tensor subgroup
725 : CALL create_mat_munu(bs_env%mat_ao_ao_tensor, qs_env, bs_env%eps_atom_grid_2d_mat, &
726 34 : blacs_env_tensor, do_ri_aux_basis=.FALSE.)
727 :
728 : CALL create_mat_munu(bs_env%mat_RI_RI_tensor, qs_env, bs_env%eps_atom_grid_2d_mat, &
729 34 : blacs_env_tensor, do_ri_aux_basis=.TRUE.)
730 :
731 : CALL create_mat_munu(bs_env%mat_RI_RI, qs_env, bs_env%eps_atom_grid_2d_mat, &
732 34 : blacs_env, do_ri_aux_basis=.TRUE.)
733 :
734 34 : CALL cp_blacs_env_release(blacs_env_tensor)
735 :
736 34 : NULLIFY (bs_env%mat_chi_Gamma_tau)
737 34 : CALL dbcsr_allocate_matrix_set(bs_env%mat_chi_Gamma_tau, bs_env%num_time_freq_points)
738 :
739 470 : DO i_t = 1, bs_env%num_time_freq_points
740 436 : ALLOCATE (bs_env%mat_chi_Gamma_tau(i_t)%matrix)
741 470 : CALL dbcsr_create(bs_env%mat_chi_Gamma_tau(i_t)%matrix, template=bs_env%mat_RI_RI%matrix)
742 : END DO
743 :
744 34 : CALL timestop(handle)
745 :
746 34 : END SUBROUTINE allocate_matrices
747 :
748 : ! **************************************************************************************************
749 : !> \brief ...
750 : !> \param bs_env ...
751 : ! **************************************************************************************************
752 28 : SUBROUTINE allocate_GW_eigenvalues(bs_env)
753 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
754 :
755 : CHARACTER(LEN=*), PARAMETER :: routineN = 'allocate_GW_eigenvalues'
756 :
757 : INTEGER :: handle
758 :
759 28 : CALL timeset(routineN, handle)
760 :
761 140 : ALLOCATE (bs_env%eigenval_G0W0(bs_env%n_ao, bs_env%nkp_bs_and_DOS, bs_env%n_spin))
762 140 : ALLOCATE (bs_env%eigenval_HF(bs_env%n_ao, bs_env%nkp_bs_and_DOS, bs_env%n_spin))
763 :
764 28 : CALL timestop(handle)
765 :
766 28 : END SUBROUTINE allocate_GW_eigenvalues
767 :
768 : ! **************************************************************************************************
769 : !> \brief ...
770 : !> \param qs_env ...
771 : !> \param bs_env ...
772 : ! **************************************************************************************************
773 34 : SUBROUTINE create_tensors(qs_env, bs_env)
774 : TYPE(qs_environment_type), POINTER :: qs_env
775 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
776 :
777 : CHARACTER(LEN=*), PARAMETER :: routineN = 'create_tensors'
778 :
779 : INTEGER :: handle
780 :
781 34 : CALL timeset(routineN, handle)
782 :
783 34 : CALL init_interaction_radii(bs_env)
784 :
785 : ! split blocks does not improve load balancing/efficienfy for tensor contraction, so we go
786 : ! with the standard atomic blocks
787 : CALL create_3c_t(bs_env%t_RI_AO__AO, bs_env%para_env_tensor, "(RI AO | AO)", [1, 2], [3], &
788 : bs_env%sizes_RI, bs_env%sizes_AO, &
789 34 : create_nl_3c=.TRUE., nl_3c=bs_env%nl_3c, qs_env=qs_env)
790 : CALL create_3c_t(bs_env%t_RI__AO_AO, bs_env%para_env_tensor, "(RI | AO AO)", [1], [2, 3], &
791 34 : bs_env%sizes_RI, bs_env%sizes_AO)
792 :
793 34 : CALL create_2c_t(bs_env, bs_env%sizes_RI, bs_env%sizes_AO)
794 :
795 34 : CALL timestop(handle)
796 :
797 34 : END SUBROUTINE create_tensors
798 :
799 : ! **************************************************************************************************
800 : !> \brief ...
801 : !> \param qs_env ...
802 : !> \param bs_env ...
803 : ! **************************************************************************************************
804 28 : SUBROUTINE check_sparsity_3c(qs_env, bs_env)
805 : TYPE(qs_environment_type), POINTER :: qs_env
806 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
807 :
808 : CHARACTER(LEN=*), PARAMETER :: routineN = 'check_sparsity_3c'
809 :
810 : INTEGER :: handle, n_atom_step, RI_atom
811 : INTEGER(int_8) :: mem, non_zero_elements_sum, nze
812 : REAL(dp) :: max_dist_AO_atoms, occ, occupation_sum
813 : REAL(KIND=dp) :: t1, t2
814 196 : TYPE(dbt_type) :: t_3c_global
815 28 : TYPE(dbt_type), ALLOCATABLE, DIMENSION(:, :) :: t_3c_global_array
816 : TYPE(neighbor_list_3c_type) :: nl_3c_global
817 :
818 28 : CALL timeset(routineN, handle)
819 :
820 : ! check the sparsity of 3c integral tensor (µν|P); calculate maximum distance between
821 : ! AO atoms µ, ν where at least a single integral (µν|P) is larger than the filter threshold
822 : CALL create_3c_t(t_3c_global, bs_env%para_env, "(RI AO | AO)", [1, 2], [3], &
823 : bs_env%sizes_RI, bs_env%sizes_AO, &
824 28 : create_nl_3c=.TRUE., nl_3c=nl_3c_global, qs_env=qs_env)
825 :
826 28 : CALL m_memory(mem)
827 28 : CALL bs_env%para_env%max(mem)
828 :
829 252 : ALLOCATE (t_3c_global_array(1, 1))
830 28 : CALL dbt_create(t_3c_global, t_3c_global_array(1, 1))
831 :
832 28 : CALL bs_env%para_env%sync()
833 28 : t1 = m_walltime()
834 :
835 28 : occupation_sum = 0.0_dp
836 28 : non_zero_elements_sum = 0
837 28 : max_dist_AO_atoms = 0.0_dp
838 28 : n_atom_step = INT(SQRT(REAL(bs_env%n_atom, KIND=dp)))
839 : ! do not compute full 3c integrals at once because it may cause out of memory
840 88 : DO RI_atom = 1, bs_env%n_atom, n_atom_step
841 :
842 : CALL build_3c_integrals(t_3c_global_array, &
843 : bs_env%eps_filter, &
844 : qs_env, &
845 : nl_3c_global, &
846 : int_eps=bs_env%eps_filter, &
847 : basis_i=bs_env%basis_set_RI, &
848 : basis_j=bs_env%basis_set_AO, &
849 : basis_k=bs_env%basis_set_AO, &
850 : bounds_i=[RI_atom, MIN(RI_atom + n_atom_step - 1, bs_env%n_atom)], &
851 : potential_parameter=bs_env%ri_metric, &
852 180 : desymmetrize=.FALSE.)
853 :
854 60 : CALL dbt_filter(t_3c_global_array(1, 1), bs_env%eps_filter)
855 :
856 60 : CALL bs_env%para_env%sync()
857 :
858 60 : CALL get_tensor_occupancy(t_3c_global_array(1, 1), nze, occ)
859 60 : non_zero_elements_sum = non_zero_elements_sum + nze
860 60 : occupation_sum = occupation_sum + occ
861 :
862 60 : CALL get_max_dist_AO_atoms(t_3c_global_array(1, 1), max_dist_AO_atoms, qs_env)
863 :
864 148 : CALL dbt_clear(t_3c_global_array(1, 1))
865 :
866 : END DO
867 :
868 28 : t2 = m_walltime()
869 :
870 28 : bs_env%occupation_3c_int = occupation_sum
871 28 : bs_env%max_dist_AO_atoms = max_dist_AO_atoms
872 :
873 28 : CALL dbt_destroy(t_3c_global)
874 28 : CALL dbt_destroy(t_3c_global_array(1, 1))
875 56 : DEALLOCATE (t_3c_global_array)
876 :
877 28 : CALL neighbor_list_3c_destroy(nl_3c_global)
878 :
879 28 : IF (bs_env%unit_nr > 0) THEN
880 14 : WRITE (bs_env%unit_nr, '(T2,A)') ''
881 : WRITE (bs_env%unit_nr, '(T2,A,F27.1,A)') &
882 14 : 'Computed 3-center integrals (µν|P), execution time', t2 - t1, ' s'
883 14 : WRITE (bs_env%unit_nr, '(T2,A,F48.3,A)') 'Percentage of non-zero (µν|P)', &
884 28 : occupation_sum*100, ' %'
885 14 : WRITE (bs_env%unit_nr, '(T2,A,F33.1,A)') 'Max. distance between µ,ν in non-zero (µν|P)', &
886 28 : max_dist_AO_atoms*angstrom, ' A'
887 14 : WRITE (bs_env%unit_nr, '(T2,2A,I20,A)') 'Required memory if storing all 3-center ', &
888 28 : 'integrals (µν|P)', INT(REAL(non_zero_elements_sum, KIND=dp)*8.0E-9_dp), ' GB'
889 : END IF
890 :
891 28 : CALL timestop(handle)
892 :
893 112 : END SUBROUTINE check_sparsity_3c
894 :
895 : ! **************************************************************************************************
896 : !> \brief ...
897 : !> \param bs_env ...
898 : !> \param sizes_RI ...
899 : !> \param sizes_AO ...
900 : ! **************************************************************************************************
901 34 : SUBROUTINE create_2c_t(bs_env, sizes_RI, sizes_AO)
902 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
903 : INTEGER, ALLOCATABLE, DIMENSION(:) :: sizes_RI, sizes_AO
904 :
905 : CHARACTER(LEN=*), PARAMETER :: routineN = 'create_2c_t'
906 :
907 : INTEGER :: handle
908 34 : INTEGER, ALLOCATABLE, DIMENSION(:) :: dist_1, dist_2
909 : INTEGER, DIMENSION(2) :: pdims_2d
910 102 : TYPE(dbt_pgrid_type) :: pgrid_2d
911 :
912 34 : CALL timeset(routineN, handle)
913 :
914 : ! inspired from rpa_im_time.F / hfx_types.F
915 :
916 34 : pdims_2d = 0
917 34 : CALL dbt_pgrid_create(bs_env%para_env_tensor, pdims_2d, pgrid_2d)
918 :
919 : CALL create_2c_tensor(bs_env%t_G, dist_1, dist_2, pgrid_2d, sizes_AO, sizes_AO, &
920 34 : name="(AO | AO)")
921 34 : DEALLOCATE (dist_1, dist_2)
922 : CALL create_2c_tensor(bs_env%t_chi, dist_1, dist_2, pgrid_2d, sizes_RI, sizes_RI, &
923 34 : name="(RI | RI)")
924 34 : DEALLOCATE (dist_1, dist_2)
925 : CALL create_2c_tensor(bs_env%t_W, dist_1, dist_2, pgrid_2d, sizes_RI, sizes_RI, &
926 34 : name="(RI | RI)")
927 34 : DEALLOCATE (dist_1, dist_2)
928 34 : CALL dbt_pgrid_destroy(pgrid_2d)
929 :
930 34 : CALL timestop(handle)
931 :
932 34 : END SUBROUTINE create_2c_t
933 :
934 : ! **************************************************************************************************
935 : !> \brief ...
936 : !> \param tensor ...
937 : !> \param para_env ...
938 : !> \param tensor_name ...
939 : !> \param map1 ...
940 : !> \param map2 ...
941 : !> \param sizes_RI ...
942 : !> \param sizes_AO ...
943 : !> \param create_nl_3c ...
944 : !> \param nl_3c ...
945 : !> \param qs_env ...
946 : ! **************************************************************************************************
947 96 : SUBROUTINE create_3c_t(tensor, para_env, tensor_name, map1, map2, sizes_RI, sizes_AO, &
948 : create_nl_3c, nl_3c, qs_env)
949 : TYPE(dbt_type) :: tensor
950 : TYPE(mp_para_env_type), POINTER :: para_env
951 : CHARACTER(LEN=12) :: tensor_name
952 : INTEGER, DIMENSION(:) :: map1, map2
953 : INTEGER, ALLOCATABLE, DIMENSION(:) :: sizes_RI, sizes_AO
954 : LOGICAL, OPTIONAL :: create_nl_3c
955 : TYPE(neighbor_list_3c_type), OPTIONAL :: nl_3c
956 : TYPE(qs_environment_type), OPTIONAL, POINTER :: qs_env
957 :
958 : CHARACTER(LEN=*), PARAMETER :: routineN = 'create_3c_t'
959 :
960 : INTEGER :: handle, nkind
961 96 : INTEGER, ALLOCATABLE, DIMENSION(:) :: dist_AO_1, dist_AO_2, dist_RI
962 : INTEGER, DIMENSION(3) :: pcoord, pdims, pdims_3d
963 : LOGICAL :: my_create_nl_3c
964 288 : TYPE(dbt_pgrid_type) :: pgrid_3d
965 : TYPE(distribution_3d_type) :: dist_3d
966 96 : TYPE(mp_cart_type) :: mp_comm_t3c_2
967 96 : TYPE(particle_type), DIMENSION(:), POINTER :: particle_set
968 :
969 96 : CALL timeset(routineN, handle)
970 :
971 96 : pdims_3d = 0
972 96 : CALL dbt_pgrid_create(para_env, pdims_3d, pgrid_3d)
973 : CALL create_3c_tensor(tensor, dist_RI, dist_AO_1, dist_AO_2, &
974 : pgrid_3d, sizes_RI, sizes_AO, sizes_AO, &
975 96 : map1=map1, map2=map2, name=tensor_name)
976 :
977 96 : IF (PRESENT(create_nl_3c)) THEN
978 62 : my_create_nl_3c = create_nl_3c
979 : ELSE
980 : my_create_nl_3c = .FALSE.
981 : END IF
982 :
983 62 : IF (my_create_nl_3c) THEN
984 62 : CALL get_qs_env(qs_env, nkind=nkind, particle_set=particle_set)
985 62 : CALL dbt_mp_environ_pgrid(pgrid_3d, pdims, pcoord)
986 62 : CALL mp_comm_t3c_2%create(pgrid_3d%mp_comm_2d, 3, pdims)
987 : CALL distribution_3d_create(dist_3d, dist_RI, dist_AO_1, dist_AO_2, &
988 62 : nkind, particle_set, mp_comm_t3c_2, own_comm=.TRUE.)
989 :
990 : CALL build_3c_neighbor_lists(nl_3c, &
991 : qs_env%bs_env%basis_set_RI, &
992 : qs_env%bs_env%basis_set_AO, &
993 : qs_env%bs_env%basis_set_AO, &
994 : dist_3d, qs_env%bs_env%ri_metric, &
995 62 : "GW_3c_nl", qs_env, own_dist=.TRUE.)
996 : END IF
997 :
998 96 : DEALLOCATE (dist_RI, dist_AO_1, dist_AO_2)
999 96 : CALL dbt_pgrid_destroy(pgrid_3d)
1000 :
1001 96 : CALL timestop(handle)
1002 :
1003 192 : END SUBROUTINE create_3c_t
1004 :
1005 : ! **************************************************************************************************
1006 : !> \brief ...
1007 : !> \param bs_env ...
1008 : ! **************************************************************************************************
1009 34 : SUBROUTINE init_interaction_radii(bs_env)
1010 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
1011 :
1012 : CHARACTER(LEN=*), PARAMETER :: routineN = 'init_interaction_radii'
1013 :
1014 : INTEGER :: handle, ibasis
1015 : TYPE(gto_basis_set_type), POINTER :: orb_basis, ri_basis
1016 :
1017 34 : CALL timeset(routineN, handle)
1018 :
1019 90 : DO ibasis = 1, SIZE(bs_env%basis_set_AO)
1020 :
1021 56 : orb_basis => bs_env%basis_set_AO(ibasis)%gto_basis_set
1022 56 : CALL init_interaction_radii_orb_basis(orb_basis, bs_env%eps_filter)
1023 :
1024 56 : ri_basis => bs_env%basis_set_RI(ibasis)%gto_basis_set
1025 90 : CALL init_interaction_radii_orb_basis(ri_basis, bs_env%eps_filter)
1026 :
1027 : END DO
1028 :
1029 34 : CALL timestop(handle)
1030 :
1031 34 : END SUBROUTINE init_interaction_radii
1032 :
1033 : ! **************************************************************************************************
1034 : !> \brief ...
1035 : !> \param t_3c_int ...
1036 : !> \param max_dist_AO_atoms ...
1037 : !> \param qs_env ...
1038 : ! **************************************************************************************************
1039 60 : SUBROUTINE get_max_dist_AO_atoms(t_3c_int, max_dist_AO_atoms, qs_env)
1040 : TYPE(dbt_type) :: t_3c_int
1041 : REAL(KIND=dp) :: max_dist_AO_atoms
1042 : TYPE(qs_environment_type), POINTER :: qs_env
1043 :
1044 : CHARACTER(LEN=*), PARAMETER :: routineN = 'get_max_dist_AO_atoms'
1045 :
1046 : INTEGER :: atom_1, atom_2, handle, num_cells
1047 : INTEGER, DIMENSION(3) :: atom_ind
1048 60 : INTEGER, DIMENSION(:, :), POINTER :: index_to_cell
1049 : REAL(KIND=dp) :: abs_rab
1050 : REAL(KIND=dp), DIMENSION(3) :: rab
1051 : TYPE(cell_type), POINTER :: cell
1052 : TYPE(dbt_iterator_type) :: iter
1053 : TYPE(mp_para_env_type), POINTER :: para_env
1054 60 : TYPE(particle_type), DIMENSION(:), POINTER :: particle_set
1055 :
1056 60 : CALL timeset(routineN, handle)
1057 :
1058 60 : NULLIFY (cell, particle_set, para_env)
1059 60 : CALL get_qs_env(qs_env, cell=cell, particle_set=particle_set, para_env=para_env)
1060 :
1061 : !$OMP PARALLEL DEFAULT(NONE) &
1062 : !$OMP SHARED(t_3c_int, max_dist_AO_atoms, num_cells, index_to_cell, particle_set, cell) &
1063 60 : !$OMP PRIVATE(iter,atom_ind,rab, abs_rab, atom_1, atom_2)
1064 : CALL dbt_iterator_start(iter, t_3c_int)
1065 : DO WHILE (dbt_iterator_blocks_left(iter))
1066 : CALL dbt_iterator_next_block(iter, atom_ind)
1067 :
1068 : atom_1 = atom_ind(2)
1069 : atom_2 = atom_ind(3)
1070 :
1071 : rab = pbc(particle_set(atom_1)%r(1:3), particle_set(atom_2)%r(1:3), cell)
1072 :
1073 : abs_rab = SQRT(rab(1)**2 + rab(2)**2 + rab(3)**2)
1074 :
1075 : max_dist_AO_atoms = MAX(max_dist_AO_atoms, abs_rab)
1076 :
1077 : END DO
1078 : CALL dbt_iterator_stop(iter)
1079 : !$OMP END PARALLEL
1080 :
1081 60 : CALL para_env%max(max_dist_AO_atoms)
1082 :
1083 60 : CALL timestop(handle)
1084 :
1085 60 : END SUBROUTINE get_max_dist_AO_atoms
1086 :
1087 : ! **************************************************************************************************
1088 : !> \brief ...
1089 : !> \param bs_env ...
1090 : ! **************************************************************************************************
1091 28 : SUBROUTINE set_sparsity_parallelization_parameters(bs_env)
1092 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
1093 :
1094 : CHARACTER(LEN=*), PARAMETER :: routineN = 'set_sparsity_parallelization_parameters'
1095 :
1096 : INTEGER :: handle, i_ivl, IL_ivl, j_ivl, n_atom_per_IL_ivl, n_atom_per_ivl, n_intervals_i, &
1097 : n_intervals_inner_loop_atoms, n_intervals_j, u
1098 : INTEGER(KIND=int_8) :: input_memory_per_proc
1099 :
1100 28 : CALL timeset(routineN, handle)
1101 :
1102 : ! heuristic parameter to prevent out of memory
1103 28 : bs_env%safety_factor_memory = 0.10_dp
1104 :
1105 28 : input_memory_per_proc = INT(bs_env%input_memory_per_proc_GB*1.0E9_dp, KIND=int_8)
1106 :
1107 : ! choose atomic range for λ ("i_atom"), ν ("j_atom") in
1108 : ! M_λνP(iτ) = sum_µ (µν|P) G^occ_µλ(i|τ|,k=0)
1109 : ! N_νλQ(iτ) = sum_σ (σλ|Q) G^vir_σν(i|τ|,k=0)
1110 : ! such that M and N fit into the memory
1111 : n_atom_per_ivl = INT(SQRT(bs_env%safety_factor_memory*input_memory_per_proc &
1112 : *bs_env%group_size_tensor/24/bs_env%n_RI &
1113 28 : /SQRT(bs_env%occupation_3c_int)))/bs_env%max_AO_bf_per_atom
1114 :
1115 28 : n_intervals_i = (bs_env%n_atom_i - 1)/n_atom_per_ivl + 1
1116 28 : n_intervals_j = (bs_env%n_atom_j - 1)/n_atom_per_ivl + 1
1117 :
1118 28 : bs_env%n_atom_per_interval_ij = n_atom_per_ivl
1119 28 : bs_env%n_intervals_i = n_intervals_i
1120 28 : bs_env%n_intervals_j = n_intervals_j
1121 :
1122 84 : ALLOCATE (bs_env%i_atom_intervals(2, n_intervals_i))
1123 84 : ALLOCATE (bs_env%j_atom_intervals(2, n_intervals_j))
1124 :
1125 56 : DO i_ivl = 1, n_intervals_i
1126 28 : bs_env%i_atom_intervals(1, i_ivl) = (i_ivl - 1)*n_atom_per_ivl + bs_env%atoms_i(1)
1127 : bs_env%i_atom_intervals(2, i_ivl) = MIN(i_ivl*n_atom_per_ivl + bs_env%atoms_i(1) - 1, &
1128 56 : bs_env%atoms_i(2))
1129 : END DO
1130 :
1131 56 : DO j_ivl = 1, n_intervals_j
1132 28 : bs_env%j_atom_intervals(1, j_ivl) = (j_ivl - 1)*n_atom_per_ivl + bs_env%atoms_j(1)
1133 : bs_env%j_atom_intervals(2, j_ivl) = MIN(j_ivl*n_atom_per_ivl + bs_env%atoms_j(1) - 1, &
1134 56 : bs_env%atoms_j(2))
1135 : END DO
1136 :
1137 112 : ALLOCATE (bs_env%skip_Sigma_occ(n_intervals_i, n_intervals_j))
1138 84 : ALLOCATE (bs_env%skip_Sigma_vir(n_intervals_i, n_intervals_j))
1139 84 : bs_env%skip_Sigma_occ(:, :) = .FALSE.
1140 84 : bs_env%skip_Sigma_vir(:, :) = .FALSE.
1141 :
1142 : ! choose atomic range for µ and σ ("inner loop (IL) atom") in
1143 : ! M_λνP(iτ) = sum_µ (µν|P) G^occ_µλ(i|τ|,k=0)
1144 : ! N_νλQ(iτ) = sum_σ (σλ|Q) G^vir_σν(i|τ|,k=0)
1145 : n_atom_per_IL_ivl = MIN(INT(bs_env%safety_factor_memory*input_memory_per_proc &
1146 : *bs_env%group_size_tensor/n_atom_per_ivl &
1147 : /bs_env%max_AO_bf_per_atom &
1148 : /bs_env%n_RI/8/SQRT(bs_env%occupation_3c_int) &
1149 28 : /bs_env%max_AO_bf_per_atom), bs_env%n_atom)
1150 :
1151 28 : n_intervals_inner_loop_atoms = (bs_env%n_atom - 1)/n_atom_per_IL_ivl + 1
1152 :
1153 28 : bs_env%n_atom_per_IL_interval = n_atom_per_IL_ivl
1154 28 : bs_env%n_intervals_inner_loop_atoms = n_intervals_inner_loop_atoms
1155 :
1156 84 : ALLOCATE (bs_env%inner_loop_atom_intervals(2, n_intervals_inner_loop_atoms))
1157 56 : DO IL_ivl = 1, n_intervals_inner_loop_atoms
1158 28 : bs_env%inner_loop_atom_intervals(1, IL_ivl) = (IL_ivl - 1)*n_atom_per_IL_ivl + 1
1159 56 : bs_env%inner_loop_atom_intervals(2, IL_ivl) = MIN(IL_ivl*n_atom_per_IL_ivl, bs_env%n_atom)
1160 : END DO
1161 :
1162 28 : u = bs_env%unit_nr
1163 28 : IF (u > 0) THEN
1164 14 : WRITE (u, '(T2,A)') ''
1165 14 : WRITE (u, '(T2,A,I33)') 'Number of i and j atoms in M_λνP(τ), N_νλQ(τ):', n_atom_per_ivl
1166 14 : WRITE (u, '(T2,A,I18)') 'Number of inner loop atoms for µ in M_λνP = sum_µ (µν|P) G_µλ', &
1167 28 : n_atom_per_IL_ivl
1168 : END IF
1169 :
1170 28 : CALL timestop(handle)
1171 :
1172 28 : END SUBROUTINE set_sparsity_parallelization_parameters
1173 :
1174 : ! **************************************************************************************************
1175 : !> \brief ...
1176 : !> \param qs_env ...
1177 : !> \param bs_env ...
1178 : ! **************************************************************************************************
1179 28 : SUBROUTINE check_for_restart_files(qs_env, bs_env)
1180 : TYPE(qs_environment_type), POINTER :: qs_env
1181 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
1182 :
1183 : CHARACTER(LEN=*), PARAMETER :: routineN = 'check_for_restart_files'
1184 :
1185 : CHARACTER(LEN=9) :: frmt
1186 : CHARACTER(LEN=default_string_length) :: f_chi, f_S_n, f_S_p, f_S_x, f_W_t, &
1187 : prefix, project_name
1188 : INTEGER :: handle, i_spin, i_t_or_w, ind, n_spin, &
1189 : num_time_freq_points
1190 : LOGICAL :: chi_exists, Sigma_neg_time_exists, &
1191 : Sigma_pos_time_exists, &
1192 : Sigma_x_spin_exists, W_time_exists
1193 : TYPE(cp_logger_type), POINTER :: logger
1194 : TYPE(section_vals_type), POINTER :: input, print_key
1195 :
1196 28 : CALL timeset(routineN, handle)
1197 :
1198 28 : num_time_freq_points = bs_env%num_time_freq_points
1199 28 : n_spin = bs_env%n_spin
1200 :
1201 84 : ALLOCATE (bs_env%read_chi(num_time_freq_points))
1202 56 : ALLOCATE (bs_env%calc_chi(num_time_freq_points))
1203 112 : ALLOCATE (bs_env%Sigma_c_exists(num_time_freq_points, n_spin))
1204 :
1205 28 : CALL get_qs_env(qs_env, input=input)
1206 :
1207 28 : logger => cp_get_default_logger()
1208 28 : print_key => section_vals_get_subs_vals(input, 'PROPERTIES%BANDSTRUCTURE%GW%PRINT%RESTART')
1209 : project_name = cp_print_key_generate_filename(logger, print_key, extension="", &
1210 28 : my_local=.FALSE.)
1211 28 : WRITE (prefix, '(2A)') TRIM(project_name), "-RESTART_"
1212 28 : bs_env%prefix = prefix
1213 :
1214 28 : bs_env%all_W_exist = .TRUE.
1215 :
1216 412 : DO i_t_or_w = 1, num_time_freq_points
1217 :
1218 384 : IF (i_t_or_w < 10) THEN
1219 240 : WRITE (frmt, '(A)') '(3A,I1,A)'
1220 240 : WRITE (f_chi, frmt) TRIM(prefix), bs_env%chi_name, "_0", i_t_or_w, ".matrix"
1221 240 : WRITE (f_W_t, frmt) TRIM(prefix), bs_env%W_time_name, "_0", i_t_or_w, ".matrix"
1222 144 : ELSE IF (i_t_or_w < 100) THEN
1223 144 : WRITE (frmt, '(A)') '(3A,I2,A)'
1224 144 : WRITE (f_chi, frmt) TRIM(prefix), bs_env%chi_name, "_", i_t_or_w, ".matrix"
1225 144 : WRITE (f_W_t, frmt) TRIM(prefix), bs_env%W_time_name, "_", i_t_or_w, ".matrix"
1226 : ELSE
1227 0 : CPABORT('Please implement more than 99 time/frequency points.')
1228 : END IF
1229 :
1230 384 : INQUIRE (file=TRIM(f_chi), exist=chi_exists)
1231 384 : INQUIRE (file=TRIM(f_W_t), exist=W_time_exists)
1232 :
1233 384 : bs_env%read_chi(i_t_or_w) = chi_exists
1234 384 : bs_env%calc_chi(i_t_or_w) = .NOT. chi_exists
1235 :
1236 384 : bs_env%all_W_exist = bs_env%all_W_exist .AND. W_time_exists
1237 :
1238 : ! the self-energy is spin-dependent
1239 876 : DO i_spin = 1, n_spin
1240 :
1241 464 : ind = i_t_or_w + (i_spin - 1)*num_time_freq_points
1242 :
1243 464 : IF (ind < 10) THEN
1244 240 : WRITE (frmt, '(A)') '(3A,I1,A)'
1245 240 : WRITE (f_S_p, frmt) TRIM(prefix), bs_env%Sigma_p_name, "_0", ind, ".matrix"
1246 240 : WRITE (f_S_n, frmt) TRIM(prefix), bs_env%Sigma_n_name, "_0", ind, ".matrix"
1247 224 : ELSE IF (i_t_or_w < 100) THEN
1248 224 : WRITE (frmt, '(A)') '(3A,I2,A)'
1249 224 : WRITE (f_S_p, frmt) TRIM(prefix), bs_env%Sigma_p_name, "_", ind, ".matrix"
1250 224 : WRITE (f_S_n, frmt) TRIM(prefix), bs_env%Sigma_n_name, "_", ind, ".matrix"
1251 : END IF
1252 :
1253 464 : INQUIRE (file=TRIM(f_S_p), exist=Sigma_pos_time_exists)
1254 464 : INQUIRE (file=TRIM(f_S_n), exist=Sigma_neg_time_exists)
1255 :
1256 : bs_env%Sigma_c_exists(i_t_or_w, i_spin) = Sigma_pos_time_exists .AND. &
1257 1072 : Sigma_neg_time_exists
1258 :
1259 : END DO
1260 :
1261 : END DO
1262 :
1263 28 : IF (bs_env%all_W_exist) THEN
1264 192 : bs_env%read_chi(:) = .FALSE.
1265 192 : bs_env%calc_chi(:) = .FALSE.
1266 : END IF
1267 :
1268 28 : bs_env%Sigma_x_exists = .TRUE.
1269 64 : DO i_spin = 1, n_spin
1270 36 : WRITE (f_S_x, '(3A,I1,A)') TRIM(prefix), bs_env%Sigma_x_name, "_0", i_spin, ".matrix"
1271 36 : INQUIRE (file=TRIM(f_S_x), exist=Sigma_x_spin_exists)
1272 82 : bs_env%Sigma_x_exists = bs_env%Sigma_x_exists .AND. Sigma_x_spin_exists
1273 : END DO
1274 :
1275 28 : CALL timestop(handle)
1276 :
1277 28 : END SUBROUTINE check_for_restart_files
1278 :
1279 : ! **************************************************************************************************
1280 : !> \brief ...
1281 : !> \param qs_env ...
1282 : !> \param bs_env ...
1283 : ! **************************************************************************************************
1284 34 : SUBROUTINE set_parallelization_parameters(qs_env, bs_env)
1285 : TYPE(qs_environment_type), POINTER :: qs_env
1286 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
1287 :
1288 : CHARACTER(LEN=*), PARAMETER :: routineN = 'set_parallelization_parameters'
1289 :
1290 : INTEGER :: color_sub, dummy_1, dummy_2, handle, &
1291 : num_pe, num_t_groups, u
1292 : INTEGER(KIND=int_8) :: mem
1293 : TYPE(mp_para_env_type), POINTER :: para_env
1294 :
1295 34 : CALL timeset(routineN, handle)
1296 :
1297 34 : CALL get_qs_env(qs_env, para_env=para_env)
1298 :
1299 34 : num_pe = para_env%num_pe
1300 : ! if not already set, use all processors for the group (for large-cell GW, performance
1301 : ! seems to be best for a single group with all MPI processes per group)
1302 34 : IF (bs_env%group_size_tensor < 0 .OR. bs_env%group_size_tensor > num_pe) &
1303 28 : bs_env%group_size_tensor = num_pe
1304 :
1305 : ! group_size_tensor must divide num_pe without rest; otherwise everything will be complicated
1306 34 : IF (MODULO(num_pe, bs_env%group_size_tensor) .NE. 0) THEN
1307 0 : CALL find_good_group_size(num_pe, bs_env%group_size_tensor)
1308 : END IF
1309 :
1310 : ! para_env_tensor for tensor subgroups
1311 34 : color_sub = para_env%mepos/bs_env%group_size_tensor
1312 34 : bs_env%tensor_group_color = color_sub
1313 :
1314 34 : ALLOCATE (bs_env%para_env_tensor)
1315 34 : CALL bs_env%para_env_tensor%from_split(para_env, color_sub)
1316 :
1317 34 : num_t_groups = para_env%num_pe/bs_env%group_size_tensor
1318 34 : bs_env%num_tensor_groups = num_t_groups
1319 :
1320 : CALL get_i_j_atoms(bs_env%atoms_i, bs_env%atoms_j, bs_env%n_atom_i, bs_env%n_atom_j, &
1321 34 : color_sub, bs_env)
1322 :
1323 102 : ALLOCATE (bs_env%atoms_i_t_group(2, num_t_groups))
1324 102 : ALLOCATE (bs_env%atoms_j_t_group(2, num_t_groups))
1325 74 : DO color_sub = 0, num_t_groups - 1
1326 : CALL get_i_j_atoms(bs_env%atoms_i_t_group(1:2, color_sub + 1), &
1327 : bs_env%atoms_j_t_group(1:2, color_sub + 1), &
1328 74 : dummy_1, dummy_2, color_sub, bs_env)
1329 : END DO
1330 :
1331 34 : CALL m_memory(mem)
1332 34 : CALL bs_env%para_env%max(mem)
1333 :
1334 34 : u = bs_env%unit_nr
1335 34 : IF (u > 0) THEN
1336 17 : WRITE (u, '(T2,A,I47)') 'Group size for tensor operations', bs_env%group_size_tensor
1337 17 : IF (bs_env%group_size_tensor > 1 .AND. bs_env%n_atom < 5) THEN
1338 14 : WRITE (u, '(T2,A)') 'The requested group size is > 1 which can lead to bad performance.'
1339 14 : WRITE (u, '(T2,A)') 'Using more memory per MPI process might improve performance.'
1340 14 : WRITE (u, '(T2,A)') '(Also increase MEMORY_PER_PROC when using more memory per process.)'
1341 : END IF
1342 : END IF
1343 :
1344 34 : CALL timestop(handle)
1345 :
1346 68 : END SUBROUTINE set_parallelization_parameters
1347 :
1348 : ! **************************************************************************************************
1349 : !> \brief ...
1350 : !> \param num_pe ...
1351 : !> \param group_size ...
1352 : ! **************************************************************************************************
1353 0 : SUBROUTINE find_good_group_size(num_pe, group_size)
1354 :
1355 : INTEGER :: num_pe, group_size
1356 :
1357 : CHARACTER(LEN=*), PARAMETER :: routineN = 'find_good_group_size'
1358 :
1359 : INTEGER :: group_size_minus, group_size_orig, &
1360 : group_size_plus, handle, i_diff
1361 :
1362 0 : CALL timeset(routineN, handle)
1363 :
1364 0 : group_size_orig = group_size
1365 :
1366 0 : DO i_diff = 1, num_pe
1367 :
1368 0 : group_size_minus = group_size - i_diff
1369 :
1370 0 : IF (MODULO(num_pe, group_size_minus) == 0 .AND. group_size_minus > 0) THEN
1371 0 : group_size = group_size_minus
1372 0 : EXIT
1373 : END IF
1374 :
1375 0 : group_size_plus = group_size + i_diff
1376 :
1377 0 : IF (MODULO(num_pe, group_size_plus) == 0 .AND. group_size_plus <= num_pe) THEN
1378 0 : group_size = group_size_plus
1379 0 : EXIT
1380 : END IF
1381 :
1382 : END DO
1383 :
1384 0 : IF (group_size_orig == group_size) CPABORT("Group size error")
1385 :
1386 0 : CALL timestop(handle)
1387 :
1388 0 : END SUBROUTINE find_good_group_size
1389 :
1390 : ! **************************************************************************************************
1391 : !> \brief ...
1392 : !> \param atoms_i ...
1393 : !> \param atoms_j ...
1394 : !> \param n_atom_i ...
1395 : !> \param n_atom_j ...
1396 : !> \param color_sub ...
1397 : !> \param bs_env ...
1398 : ! **************************************************************************************************
1399 74 : SUBROUTINE get_i_j_atoms(atoms_i, atoms_j, n_atom_i, n_atom_j, color_sub, bs_env)
1400 :
1401 : INTEGER, DIMENSION(2) :: atoms_i, atoms_j
1402 : INTEGER :: n_atom_i, n_atom_j, color_sub
1403 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
1404 :
1405 : CHARACTER(LEN=*), PARAMETER :: routineN = 'get_i_j_atoms'
1406 :
1407 : INTEGER :: handle, i_atoms_per_group, i_group, &
1408 : ipcol, ipcol_loop, iprow, iprow_loop, &
1409 : j_atoms_per_group, npcol, nprow
1410 :
1411 74 : CALL timeset(routineN, handle)
1412 :
1413 : ! create a square mesh of tensor groups for iatom and jatom; code from blacs_env_create
1414 74 : CALL square_mesh(nprow, npcol, bs_env%num_tensor_groups)
1415 :
1416 74 : i_group = 0
1417 148 : DO ipcol_loop = 0, npcol - 1
1418 240 : DO iprow_loop = 0, nprow - 1
1419 92 : IF (i_group == color_sub) THEN
1420 74 : iprow = iprow_loop
1421 74 : ipcol = ipcol_loop
1422 : END IF
1423 166 : i_group = i_group + 1
1424 : END DO
1425 : END DO
1426 :
1427 74 : IF (MODULO(bs_env%n_atom, nprow) == 0) THEN
1428 68 : i_atoms_per_group = bs_env%n_atom/nprow
1429 : ELSE
1430 6 : i_atoms_per_group = bs_env%n_atom/nprow + 1
1431 : END IF
1432 :
1433 74 : IF (MODULO(bs_env%n_atom, npcol) == 0) THEN
1434 74 : j_atoms_per_group = bs_env%n_atom/npcol
1435 : ELSE
1436 0 : j_atoms_per_group = bs_env%n_atom/npcol + 1
1437 : END IF
1438 :
1439 74 : atoms_i(1) = iprow*i_atoms_per_group + 1
1440 74 : atoms_i(2) = MIN((iprow + 1)*i_atoms_per_group, bs_env%n_atom)
1441 74 : n_atom_i = atoms_i(2) - atoms_i(1) + 1
1442 :
1443 74 : atoms_j(1) = ipcol*j_atoms_per_group + 1
1444 74 : atoms_j(2) = MIN((ipcol + 1)*j_atoms_per_group, bs_env%n_atom)
1445 74 : n_atom_j = atoms_j(2) - atoms_j(1) + 1
1446 :
1447 74 : CALL timestop(handle)
1448 :
1449 74 : END SUBROUTINE get_i_j_atoms
1450 :
1451 : ! **************************************************************************************************
1452 : !> \brief ...
1453 : !> \param nprow ...
1454 : !> \param npcol ...
1455 : !> \param nproc ...
1456 : ! **************************************************************************************************
1457 74 : SUBROUTINE square_mesh(nprow, npcol, nproc)
1458 : INTEGER :: nprow, npcol, nproc
1459 :
1460 : CHARACTER(LEN=*), PARAMETER :: routineN = 'square_mesh'
1461 :
1462 : INTEGER :: gcd_max, handle, ipe, jpe
1463 :
1464 74 : CALL timeset(routineN, handle)
1465 :
1466 74 : gcd_max = -1
1467 166 : DO ipe = 1, CEILING(SQRT(REAL(nproc, dp)))
1468 92 : jpe = nproc/ipe
1469 92 : IF (ipe*jpe .NE. nproc) CYCLE
1470 166 : IF (gcd(ipe, jpe) >= gcd_max) THEN
1471 92 : nprow = ipe
1472 92 : npcol = jpe
1473 92 : gcd_max = gcd(ipe, jpe)
1474 : END IF
1475 : END DO
1476 :
1477 74 : CALL timestop(handle)
1478 :
1479 74 : END SUBROUTINE square_mesh
1480 :
1481 : ! **************************************************************************************************
1482 : !> \brief ...
1483 : !> \param bs_env ...
1484 : !> \param qs_env ...
1485 : ! **************************************************************************************************
1486 34 : SUBROUTINE set_heuristic_parameters(bs_env, qs_env)
1487 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
1488 : TYPE(qs_environment_type), OPTIONAL, POINTER :: qs_env
1489 :
1490 : CHARACTER(LEN=*), PARAMETER :: routineN = 'set_heuristic_parameters'
1491 :
1492 : INTEGER :: handle, u
1493 : LOGICAL :: do_BvK_cell
1494 :
1495 34 : CALL timeset(routineN, handle)
1496 :
1497 : ! for generating numerically stable minimax Fourier integration weights
1498 34 : bs_env%num_points_per_magnitude = 200
1499 :
1500 : ! for periodic systems and for 20 minimax points, we use a regularized minimax mesh
1501 : ! (from experience: regularized minimax meshes converges faster for periodic systems
1502 : ! and for 20 pts)
1503 136 : IF (SUM(bs_env%periodic) .NE. 0 .OR. bs_env%num_time_freq_points == 20) THEN
1504 34 : bs_env%regularization_minimax = 1.0E-6_dp
1505 : ELSE
1506 0 : bs_env%regularization_minimax = 0.0_dp
1507 : END IF
1508 :
1509 34 : bs_env%stabilize_exp = 70.0_dp
1510 34 : bs_env%eps_atom_grid_2d_mat = 1.0E-50_dp
1511 :
1512 : ! only use interval ω in [0, 1 Ha] (1 Hartree = 27.211 eV) for virt, and ω in [-1 Ha, 0]
1513 : ! for occ for use in analytic continuation of self-energy Σ^c_n(iω,k) -> Σ^c_n(ϵ,k)
1514 34 : bs_env%freq_max_fit = 1.0_dp
1515 :
1516 : ! use a 16-parameter Padé fit
1517 34 : bs_env%nparam_pade = 16
1518 :
1519 : ! resolution of the identity with the truncated Coulomb metric, cutoff radius 3 Angström
1520 34 : bs_env%ri_metric%potential_type = do_potential_truncated
1521 34 : bs_env%ri_metric%omega = 0.0_dp
1522 : ! cutoff radius is specified in the input
1523 34 : bs_env%ri_metric%filename = "t_c_g.dat"
1524 :
1525 34 : bs_env%eps_eigval_mat_RI = 0.0_dp
1526 :
1527 34 : IF (bs_env%input_regularization_RI > -1.0E-12_dp) THEN
1528 0 : bs_env%regularization_RI = bs_env%input_regularization_RI
1529 : ELSE
1530 : ! default case:
1531 :
1532 : ! 1. for periodic systems, we use the regularized resolution of the identity per default
1533 34 : bs_env%regularization_RI = 1.0E-2_dp
1534 :
1535 : ! 2. for molecules, no regularization is necessary
1536 136 : IF (SUM(bs_env%periodic) == 0) bs_env%regularization_RI = 0.0_dp
1537 :
1538 : END IF
1539 :
1540 : ! truncated Coulomb operator for exchange self-energy
1541 : ! (see details in Guidon, VandeVondele, Hutter, JCTC 5, 3010 (2009) and references therein)
1542 34 : do_BvK_cell = bs_env%small_cell_full_kp_or_large_cell_Gamma == small_cell_full_kp
1543 : CALL trunc_coulomb_for_exchange(qs_env, bs_env%trunc_coulomb, &
1544 : rel_cutoff_trunc_coulomb_ri_x=0.5_dp, &
1545 : cell_grid=bs_env%cell_grid_scf_desymm, &
1546 34 : do_BvK_cell=do_BvK_cell)
1547 :
1548 : ! for small-cell GW, we need more cells than normally used by the filter bs_env%eps_filter
1549 : ! (in particular for computing the self-energy because of higher number of cells needed)
1550 34 : bs_env%heuristic_filter_factor = 1.0E-4
1551 :
1552 34 : u = bs_env%unit_nr
1553 34 : IF (u > 0) THEN
1554 17 : WRITE (u, FMT="(T2,2A,F21.1,A)") "Cutoff radius for the truncated Coulomb ", &
1555 34 : "operator in Σ^x:", bs_env%trunc_coulomb%cutoff_radius*angstrom, " Å"
1556 17 : WRITE (u, FMT="(T2,2A,F15.1,A)") "Cutoff radius for the truncated Coulomb ", &
1557 34 : "operator in RI metric:", bs_env%ri_metric%cutoff_radius*angstrom, " Å"
1558 17 : WRITE (u, FMT="(T2,A,ES48.1)") "Regularization parameter of RI ", bs_env%regularization_RI
1559 17 : WRITE (u, FMT="(T2,A,I53)") "Lattice sum size for V(k):", bs_env%size_lattice_sum_V
1560 : END IF
1561 :
1562 34 : CALL timestop(handle)
1563 :
1564 34 : END SUBROUTINE set_heuristic_parameters
1565 :
1566 : ! **************************************************************************************************
1567 : !> \brief ...
1568 : !> \param bs_env ...
1569 : ! **************************************************************************************************
1570 34 : SUBROUTINE print_header_and_input_parameters(bs_env)
1571 :
1572 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
1573 :
1574 : CHARACTER(LEN=*), PARAMETER :: routineN = 'print_header_and_input_parameters'
1575 :
1576 : INTEGER :: handle, u
1577 :
1578 34 : CALL timeset(routineN, handle)
1579 :
1580 34 : u = bs_env%unit_nr
1581 :
1582 34 : IF (u > 0) THEN
1583 17 : WRITE (u, *) ' '
1584 17 : WRITE (u, '(T2,2A)') '------------------------------------------------', &
1585 34 : '-------------------------------'
1586 17 : WRITE (u, '(T2,2A)') '- ', &
1587 34 : ' -'
1588 17 : WRITE (u, '(T2,2A)') '- GW CALCULATION ', &
1589 34 : ' -'
1590 17 : WRITE (u, '(T2,2A)') '- ', &
1591 34 : ' -'
1592 17 : WRITE (u, '(T2,2A)') '------------------------------------------------', &
1593 34 : '-------------------------------'
1594 17 : WRITE (u, '(T2,A)') ' '
1595 17 : WRITE (u, '(T2,A,I45)') 'Input: Number of time/freq. points', bs_env%num_time_freq_points
1596 17 : WRITE (u, '(T2,A,ES27.1)') 'Input: Filter threshold for sparse tensor operations', &
1597 34 : bs_env%eps_filter
1598 17 : WRITE (bs_env%unit_nr, FMT="(T2,A,L62)") "Apply Hedin shift", bs_env%do_hedin_shift
1599 : END IF
1600 :
1601 34 : CALL timestop(handle)
1602 :
1603 34 : END SUBROUTINE print_header_and_input_parameters
1604 :
1605 : ! **************************************************************************************************
1606 : !> \brief ...
1607 : !> \param qs_env ...
1608 : !> \param bs_env ...
1609 : ! **************************************************************************************************
1610 68 : SUBROUTINE compute_V_xc(qs_env, bs_env)
1611 : TYPE(qs_environment_type), POINTER :: qs_env
1612 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
1613 :
1614 : CHARACTER(LEN=*), PARAMETER :: routineN = 'compute_V_xc'
1615 :
1616 : INTEGER :: handle, img, ispin, myfun, nimages
1617 : LOGICAL :: hf_present
1618 : REAL(KIND=dp) :: energy_ex, energy_exc, energy_total, &
1619 : myfraction
1620 34 : TYPE(dbcsr_p_type), DIMENSION(:), POINTER :: mat_ks_without_v_xc
1621 34 : TYPE(dbcsr_p_type), DIMENSION(:, :), POINTER :: matrix_ks_kp
1622 : TYPE(dft_control_type), POINTER :: dft_control
1623 : TYPE(qs_energy_type), POINTER :: energy
1624 : TYPE(section_vals_type), POINTER :: hf_section, input, xc_section
1625 :
1626 34 : CALL timeset(routineN, handle)
1627 :
1628 34 : CALL get_qs_env(qs_env, input=input, energy=energy, dft_control=dft_control)
1629 :
1630 : ! previously, dft_control%nimages set to # neighbor cells, revert for Γ-only KS matrix
1631 34 : nimages = dft_control%nimages
1632 34 : dft_control%nimages = bs_env%nimages_scf
1633 :
1634 : ! we need to reset XC functional, therefore, get XC input
1635 34 : xc_section => section_vals_get_subs_vals(input, "DFT%XC")
1636 34 : CALL section_vals_val_get(xc_section, "XC_FUNCTIONAL%_SECTION_PARAMETERS_", i_val=myfun)
1637 34 : CALL section_vals_val_set(xc_section, "XC_FUNCTIONAL%_SECTION_PARAMETERS_", i_val=xc_none)
1638 : ! IF (ASSOCIATED(section_vals_get_subs_vals(xc_section, "HF", can_return_null=.TRUE.))) THEN
1639 34 : hf_section => section_vals_get_subs_vals(input, "DFT%XC%HF", can_return_null=.TRUE.)
1640 34 : hf_present = .FALSE.
1641 34 : IF (ASSOCIATED(hf_section)) THEN
1642 34 : CALL section_vals_get(hf_section, explicit=hf_present)
1643 : END IF
1644 34 : IF (hf_present) THEN
1645 : ! Special case for handling hfx
1646 0 : CALL section_vals_val_get(xc_section, "HF%FRACTION", r_val=myfraction)
1647 0 : CALL section_vals_val_set(xc_section, "HF%FRACTION", r_val=0.0_dp)
1648 : END IF
1649 :
1650 : ! save the energy before the energy gets updated
1651 34 : energy_total = energy%total
1652 34 : energy_exc = energy%exc
1653 34 : energy_ex = energy%ex
1654 :
1655 62 : SELECT CASE (bs_env%small_cell_full_kp_or_large_cell_Gamma)
1656 : CASE (large_cell_Gamma)
1657 :
1658 28 : NULLIFY (mat_ks_without_v_xc)
1659 28 : CALL dbcsr_allocate_matrix_set(mat_ks_without_v_xc, bs_env%n_spin)
1660 :
1661 64 : DO ispin = 1, bs_env%n_spin
1662 36 : ALLOCATE (mat_ks_without_v_xc(ispin)%matrix)
1663 64 : IF (hf_present) THEN
1664 : CALL dbcsr_create(mat_ks_without_v_xc(ispin)%matrix, template=bs_env%mat_ao_ao%matrix, &
1665 0 : matrix_type=dbcsr_type_symmetric)
1666 : ELSE
1667 36 : CALL dbcsr_create(mat_ks_without_v_xc(ispin)%matrix, template=bs_env%mat_ao_ao%matrix)
1668 : END IF
1669 : END DO
1670 :
1671 : ! calculate KS-matrix without XC
1672 : CALL qs_ks_build_kohn_sham_matrix(qs_env, calculate_forces=.FALSE., just_energy=.FALSE., &
1673 28 : ext_ks_matrix=mat_ks_without_v_xc)
1674 :
1675 64 : DO ispin = 1, bs_env%n_spin
1676 : ! transfer dbcsr matrix to fm
1677 36 : CALL cp_fm_create(bs_env%fm_V_xc_Gamma(ispin), bs_env%fm_s_Gamma%matrix_struct)
1678 36 : CALL copy_dbcsr_to_fm(mat_ks_without_v_xc(ispin)%matrix, bs_env%fm_V_xc_Gamma(ispin))
1679 :
1680 : ! v_xc = h_ks - h_ks(v_xc = 0)
1681 : CALL cp_fm_scale_and_add(alpha=-1.0_dp, matrix_a=bs_env%fm_V_xc_Gamma(ispin), &
1682 64 : beta=1.0_dp, matrix_b=bs_env%fm_ks_Gamma(ispin))
1683 : END DO
1684 :
1685 28 : CALL dbcsr_deallocate_matrix_set(mat_ks_without_v_xc)
1686 :
1687 : CASE (small_cell_full_kp)
1688 :
1689 : ! calculate KS-matrix without XC
1690 6 : CALL qs_ks_build_kohn_sham_matrix(qs_env, calculate_forces=.FALSE., just_energy=.FALSE.)
1691 6 : CALL get_qs_env(qs_env=qs_env, matrix_ks_kp=matrix_ks_kp)
1692 :
1693 176 : ALLOCATE (bs_env%fm_V_xc_R(dft_control%nimages, bs_env%n_spin))
1694 46 : DO ispin = 1, bs_env%n_spin
1695 158 : DO img = 1, dft_control%nimages
1696 : ! safe fm_V_xc_R in fm_matrix because saving in dbcsr matrix caused trouble...
1697 146 : CALL copy_dbcsr_to_fm(matrix_ks_kp(ispin, img)%matrix, bs_env%fm_work_mo(1))
1698 146 : CALL cp_fm_create(bs_env%fm_V_xc_R(img, ispin), bs_env%fm_work_mo(1)%matrix_struct)
1699 : ! store h_ks(v_xc = 0) in fm_V_xc_R
1700 : CALL cp_fm_scale_and_add(alpha=1.0_dp, matrix_a=bs_env%fm_V_xc_R(img, ispin), &
1701 152 : beta=1.0_dp, matrix_b=bs_env%fm_work_mo(1))
1702 : END DO
1703 : END DO
1704 :
1705 : END SELECT
1706 :
1707 : ! set back the energy
1708 34 : energy%total = energy_total
1709 34 : energy%exc = energy_exc
1710 34 : energy%ex = energy_ex
1711 :
1712 : ! set back nimages
1713 34 : dft_control%nimages = nimages
1714 :
1715 : ! set the DFT functional and HF fraction back
1716 : CALL section_vals_val_set(xc_section, "XC_FUNCTIONAL%_SECTION_PARAMETERS_", &
1717 34 : i_val=myfun)
1718 34 : IF (hf_present) THEN
1719 : CALL section_vals_val_set(xc_section, "HF%FRACTION", &
1720 0 : r_val=myfraction)
1721 : END IF
1722 :
1723 34 : IF (bs_env%small_cell_full_kp_or_large_cell_Gamma == small_cell_full_kp) THEN
1724 : ! calculate KS-matrix again with XC
1725 6 : CALL qs_ks_build_kohn_sham_matrix(qs_env, calculate_forces=.FALSE., just_energy=.FALSE.)
1726 12 : DO ispin = 1, bs_env%n_spin
1727 158 : DO img = 1, dft_control%nimages
1728 : ! store h_ks in fm_work_mo
1729 146 : CALL copy_dbcsr_to_fm(matrix_ks_kp(ispin, img)%matrix, bs_env%fm_work_mo(1))
1730 : ! v_xc = h_ks - h_ks(v_xc = 0)
1731 : CALL cp_fm_scale_and_add(alpha=-1.0_dp, matrix_a=bs_env%fm_V_xc_R(img, ispin), &
1732 152 : beta=1.0_dp, matrix_b=bs_env%fm_work_mo(1))
1733 : END DO
1734 : END DO
1735 : END IF
1736 :
1737 34 : CALL timestop(handle)
1738 :
1739 34 : END SUBROUTINE compute_V_xc
1740 :
1741 : ! **************************************************************************************************
1742 : !> \brief ...
1743 : !> \param bs_env ...
1744 : ! **************************************************************************************************
1745 34 : SUBROUTINE setup_time_and_frequency_minimax_grid(bs_env)
1746 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
1747 :
1748 : CHARACTER(LEN=*), PARAMETER :: routineN = 'setup_time_and_frequency_minimax_grid'
1749 :
1750 : INTEGER :: handle, homo, i_w, ierr, ispin, j_w, &
1751 : n_mo, num_time_freq_points, u
1752 : REAL(KIND=dp) :: E_max, E_max_ispin, E_min, E_min_ispin, &
1753 : E_range, max_error_min
1754 34 : REAL(KIND=dp), ALLOCATABLE, DIMENSION(:) :: points_and_weights
1755 :
1756 34 : CALL timeset(routineN, handle)
1757 :
1758 34 : n_mo = bs_env%n_ao
1759 34 : num_time_freq_points = bs_env%num_time_freq_points
1760 :
1761 102 : ALLOCATE (bs_env%imag_freq_points(num_time_freq_points))
1762 102 : ALLOCATE (bs_env%imag_time_points(num_time_freq_points))
1763 102 : ALLOCATE (bs_env%imag_time_weights_freq_zero(num_time_freq_points))
1764 136 : ALLOCATE (bs_env%weights_cos_t_to_w(num_time_freq_points, num_time_freq_points))
1765 136 : ALLOCATE (bs_env%weights_cos_w_to_t(num_time_freq_points, num_time_freq_points))
1766 136 : ALLOCATE (bs_env%weights_sin_t_to_w(num_time_freq_points, num_time_freq_points))
1767 :
1768 : ! minimum and maximum difference between eigenvalues of unoccupied and an occupied MOs
1769 34 : E_min = 1000.0_dp
1770 34 : E_max = -1000.0_dp
1771 76 : DO ispin = 1, bs_env%n_spin
1772 42 : homo = bs_env%n_occ(ispin)
1773 78 : SELECT CASE (bs_env%small_cell_full_kp_or_large_cell_Gamma)
1774 : CASE (large_cell_Gamma)
1775 : E_min_ispin = bs_env%eigenval_scf_Gamma(homo + 1, ispin) - &
1776 36 : bs_env%eigenval_scf_Gamma(homo, ispin)
1777 : E_max_ispin = bs_env%eigenval_scf_Gamma(n_mo, ispin) - &
1778 36 : bs_env%eigenval_scf_Gamma(1, ispin)
1779 : CASE (small_cell_full_kp)
1780 : E_min_ispin = MINVAL(bs_env%eigenval_scf(homo + 1, :, ispin)) - &
1781 458 : MAXVAL(bs_env%eigenval_scf(homo, :, ispin))
1782 : E_max_ispin = MAXVAL(bs_env%eigenval_scf(n_mo, :, ispin)) - &
1783 500 : MINVAL(bs_env%eigenval_scf(1, :, ispin))
1784 : END SELECT
1785 42 : E_min = MIN(E_min, E_min_ispin)
1786 76 : E_max = MAX(E_max, E_max_ispin)
1787 : END DO
1788 :
1789 34 : E_range = E_max/E_min
1790 :
1791 102 : ALLOCATE (points_and_weights(2*num_time_freq_points))
1792 :
1793 : ! frequency points
1794 34 : IF (num_time_freq_points .LE. 20) THEN
1795 34 : CALL get_rpa_minimax_coeff(num_time_freq_points, E_range, points_and_weights, ierr, .FALSE.)
1796 : ELSE
1797 0 : CALL get_rpa_minimax_coeff_larger_grid(num_time_freq_points, E_range, points_and_weights)
1798 : END IF
1799 :
1800 : ! one needs to scale the minimax grids, see Azizi, Wilhelm, Golze, Panades-Barrueta,
1801 : ! Giantomassi, Rinke, Draxl, Gonze et al., 2 publications
1802 470 : bs_env%imag_freq_points(:) = points_and_weights(1:num_time_freq_points)*E_min
1803 :
1804 : ! determine number of fit points in the interval [0,ω_max] for virt, or [-ω_max,0] for occ
1805 34 : bs_env%num_freq_points_fit = 0
1806 470 : DO i_w = 1, num_time_freq_points
1807 470 : IF (bs_env%imag_freq_points(i_w) < bs_env%freq_max_fit) THEN
1808 180 : bs_env%num_freq_points_fit = bs_env%num_freq_points_fit + 1
1809 : END IF
1810 : END DO
1811 :
1812 : ! iω values for the analytic continuation Σ^c_n(iω,k) -> Σ^c_n(ϵ,k)
1813 102 : ALLOCATE (bs_env%imag_freq_points_fit(bs_env%num_freq_points_fit))
1814 34 : j_w = 0
1815 470 : DO i_w = 1, num_time_freq_points
1816 470 : IF (bs_env%imag_freq_points(i_w) < bs_env%freq_max_fit) THEN
1817 180 : j_w = j_w + 1
1818 180 : bs_env%imag_freq_points_fit(j_w) = bs_env%imag_freq_points(i_w)
1819 : END IF
1820 : END DO
1821 :
1822 : ! reset the number of Padé parameters if smaller than the number of
1823 : ! imaginary-frequency points for the fit
1824 34 : IF (bs_env%num_freq_points_fit < bs_env%nparam_pade) THEN
1825 34 : bs_env%nparam_pade = bs_env%num_freq_points_fit
1826 : END IF
1827 :
1828 : ! time points
1829 34 : IF (num_time_freq_points .LE. 20) THEN
1830 34 : CALL get_exp_minimax_coeff(num_time_freq_points, E_range, points_and_weights)
1831 : ELSE
1832 0 : CALL get_exp_minimax_coeff_gw(num_time_freq_points, E_range, points_and_weights)
1833 : END IF
1834 :
1835 470 : bs_env%imag_time_points(:) = points_and_weights(1:num_time_freq_points)/(2.0_dp*E_min)
1836 470 : bs_env%imag_time_weights_freq_zero(:) = points_and_weights(num_time_freq_points + 1:)/(E_min)
1837 :
1838 34 : DEALLOCATE (points_and_weights)
1839 :
1840 34 : u = bs_env%unit_nr
1841 34 : IF (u > 0) THEN
1842 17 : WRITE (u, '(T2,A)') ''
1843 17 : WRITE (u, '(T2,A,F55.2)') 'SCF direct band gap (eV)', E_min*evolt
1844 17 : WRITE (u, '(T2,A,F53.2)') 'Max. SCF eigval diff. (eV)', E_max*evolt
1845 17 : WRITE (u, '(T2,A,F55.2)') 'E-Range for minimax grid', E_range
1846 17 : WRITE (u, '(T2,A,I27)') 'Number of Padé parameters for analytic continuation:', &
1847 34 : bs_env%nparam_pade
1848 17 : WRITE (u, '(T2,A)') ''
1849 : END IF
1850 :
1851 : ! in minimax grids, Fourier transforms t -> w and w -> t are split using
1852 : ! e^(iwt) = cos(wt) + i sin(wt); we thus calculate weights for trafos with a cos and
1853 : ! sine prefactor; details in Azizi, Wilhelm, Golze, Giantomassi, Panades-Barrueta,
1854 : ! Rinke, Draxl, Gonze et al., 2 publications
1855 :
1856 : ! cosine transform weights imaginary time to imaginary frequency
1857 : CALL get_l_sq_wghts_cos_tf_t_to_w(num_time_freq_points, &
1858 : bs_env%imag_time_points, &
1859 : bs_env%weights_cos_t_to_w, &
1860 : bs_env%imag_freq_points, &
1861 : E_min, E_max, max_error_min, &
1862 : bs_env%num_points_per_magnitude, &
1863 34 : bs_env%regularization_minimax)
1864 :
1865 : ! cosine transform weights imaginary frequency to imaginary time
1866 : CALL get_l_sq_wghts_cos_tf_w_to_t(num_time_freq_points, &
1867 : bs_env%imag_time_points, &
1868 : bs_env%weights_cos_w_to_t, &
1869 : bs_env%imag_freq_points, &
1870 : E_min, E_max, max_error_min, &
1871 : bs_env%num_points_per_magnitude, &
1872 34 : bs_env%regularization_minimax)
1873 :
1874 : ! sine transform weights imaginary time to imaginary frequency
1875 : CALL get_l_sq_wghts_sin_tf_t_to_w(num_time_freq_points, &
1876 : bs_env%imag_time_points, &
1877 : bs_env%weights_sin_t_to_w, &
1878 : bs_env%imag_freq_points, &
1879 : E_min, E_max, max_error_min, &
1880 : bs_env%num_points_per_magnitude, &
1881 34 : bs_env%regularization_minimax)
1882 :
1883 34 : CALL timestop(handle)
1884 :
1885 68 : END SUBROUTINE setup_time_and_frequency_minimax_grid
1886 :
1887 : ! **************************************************************************************************
1888 : !> \brief ...
1889 : !> \param qs_env ...
1890 : !> \param bs_env ...
1891 : ! **************************************************************************************************
1892 6 : SUBROUTINE setup_cells_3c(qs_env, bs_env)
1893 :
1894 : TYPE(qs_environment_type), POINTER :: qs_env
1895 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
1896 :
1897 : CHARACTER(LEN=*), PARAMETER :: routineN = 'setup_cells_3c'
1898 :
1899 : INTEGER :: atom_i, atom_j, atom_k, cell_pair_count, handle, i, i_cell_x, i_cell_x_max, &
1900 : i_cell_x_min, i_size, ikind, img, j, j_cell, j_cell_max, j_cell_y, j_cell_y_max, &
1901 : j_cell_y_min, j_size, k_cell, k_cell_max, k_cell_z, k_cell_z_max, k_cell_z_min, k_size, &
1902 : nimage_pairs_3c, nimages_3c, nimages_3c_max, nkind, u
1903 : INTEGER(KIND=int_8) :: mem_occ_per_proc
1904 6 : INTEGER, ALLOCATABLE, DIMENSION(:) :: n_other_3c_images_max
1905 6 : INTEGER, ALLOCATABLE, DIMENSION(:, :) :: index_to_cell_3c_max, nblocks_3c_max
1906 : INTEGER, DIMENSION(3) :: cell_index, n_max
1907 : REAL(KIND=dp) :: avail_mem_per_proc_GB, cell_dist, cell_radius_3c, eps, exp_min_ao, &
1908 : exp_min_RI, frobenius_norm, mem_3c_GB, mem_occ_per_proc_GB, radius_ao, radius_ao_product, &
1909 : radius_RI
1910 6 : REAL(KIND=dp), ALLOCATABLE, DIMENSION(:, :, :) :: int_3c
1911 6 : REAL(KIND=dp), DIMENSION(:, :), POINTER :: exp_ao, exp_RI
1912 :
1913 6 : CALL timeset(routineN, handle)
1914 :
1915 6 : CALL get_qs_env(qs_env, nkind=nkind)
1916 :
1917 6 : exp_min_RI = 10.0_dp
1918 6 : exp_min_ao = 10.0_dp
1919 :
1920 18 : DO ikind = 1, nkind
1921 :
1922 12 : CALL get_gto_basis_set(bs_env%basis_set_RI(ikind)%gto_basis_set, zet=exp_RI)
1923 12 : CALL get_gto_basis_set(bs_env%basis_set_ao(ikind)%gto_basis_set, zet=exp_ao)
1924 :
1925 : ! we need to remove all exponents lower than a lower bound, e.g. 1E-3, because
1926 : ! for contracted basis sets, there might be exponents = 0 in zet
1927 24 : DO i = 1, SIZE(exp_RI, 1)
1928 42 : DO j = 1, SIZE(exp_RI, 2)
1929 30 : IF (exp_RI(i, j) < exp_min_RI .AND. exp_RI(i, j) > 1E-3_dp) exp_min_RI = exp_RI(i, j)
1930 : END DO
1931 : END DO
1932 66 : DO i = 1, SIZE(exp_ao, 1)
1933 144 : DO j = 1, SIZE(exp_ao, 2)
1934 132 : IF (exp_ao(i, j) < exp_min_ao .AND. exp_ao(i, j) > 1E-3_dp) exp_min_ao = exp_ao(i, j)
1935 : END DO
1936 : END DO
1937 :
1938 : END DO
1939 :
1940 6 : eps = bs_env%eps_filter*bs_env%heuristic_filter_factor
1941 :
1942 6 : radius_ao = SQRT(-LOG(eps)/exp_min_ao)
1943 6 : radius_ao_product = SQRT(-LOG(eps)/(2.0_dp*exp_min_ao))
1944 6 : radius_RI = SQRT(-LOG(eps)/exp_min_RI)
1945 :
1946 : ! For a 3c integral (μR υS | P0) we have that cell R and cell S need to be within radius_3c
1947 6 : cell_radius_3c = radius_ao_product + radius_RI + bs_env%ri_metric%cutoff_radius
1948 :
1949 24 : n_max(1:3) = bs_env%periodic(1:3)*30
1950 :
1951 6 : nimages_3c_max = 0
1952 :
1953 6 : i_cell_x_min = 0
1954 6 : i_cell_x_max = 0
1955 6 : j_cell_y_min = 0
1956 6 : j_cell_y_max = 0
1957 6 : k_cell_z_min = 0
1958 6 : k_cell_z_max = 0
1959 :
1960 252 : DO i_cell_x = -n_max(1), n_max(1)
1961 15258 : DO j_cell_y = -n_max(2), n_max(2)
1962 37578 : DO k_cell_z = -n_max(3), n_max(3)
1963 :
1964 89304 : cell_index(1:3) = (/i_cell_x, j_cell_y, k_cell_z/)
1965 :
1966 22326 : CALL get_cell_dist(cell_index, bs_env%hmat, cell_dist)
1967 :
1968 37332 : IF (cell_dist < cell_radius_3c) THEN
1969 134 : nimages_3c_max = nimages_3c_max + 1
1970 134 : i_cell_x_min = MIN(i_cell_x_min, i_cell_x)
1971 134 : i_cell_x_max = MAX(i_cell_x_max, i_cell_x)
1972 134 : j_cell_y_min = MIN(j_cell_y_min, j_cell_y)
1973 134 : j_cell_y_max = MAX(j_cell_y_max, j_cell_y)
1974 134 : k_cell_z_min = MIN(k_cell_z_min, k_cell_z)
1975 134 : k_cell_z_max = MAX(k_cell_z_max, k_cell_z)
1976 : END IF
1977 :
1978 : END DO
1979 : END DO
1980 : END DO
1981 :
1982 : ! get index_to_cell_3c_max for the maximum possible cell range;
1983 : ! compute 3c integrals later in this routine and check really which cell is needed
1984 18 : ALLOCATE (index_to_cell_3c_max(nimages_3c_max, 3))
1985 :
1986 6 : img = 0
1987 252 : DO i_cell_x = -n_max(1), n_max(1)
1988 15258 : DO j_cell_y = -n_max(2), n_max(2)
1989 37578 : DO k_cell_z = -n_max(3), n_max(3)
1990 :
1991 89304 : cell_index(1:3) = (/i_cell_x, j_cell_y, k_cell_z/)
1992 :
1993 22326 : CALL get_cell_dist(cell_index, bs_env%hmat, cell_dist)
1994 :
1995 37332 : IF (cell_dist < cell_radius_3c) THEN
1996 134 : img = img + 1
1997 536 : index_to_cell_3c_max(img, 1:3) = cell_index(1:3)
1998 : END IF
1999 :
2000 : END DO
2001 : END DO
2002 : END DO
2003 :
2004 : ! get pairs of R and S which have non-zero 3c integral (μR υS | P0)
2005 24 : ALLOCATE (nblocks_3c_max(nimages_3c_max, nimages_3c_max))
2006 3154 : nblocks_3c_max(:, :) = 0
2007 :
2008 : cell_pair_count = 0
2009 140 : DO j_cell = 1, nimages_3c_max
2010 3154 : DO k_cell = 1, nimages_3c_max
2011 :
2012 3014 : cell_pair_count = cell_pair_count + 1
2013 :
2014 : ! trivial parallelization over cell pairs
2015 3014 : IF (MODULO(cell_pair_count, bs_env%para_env%num_pe) .NE. bs_env%para_env%mepos) CYCLE
2016 :
2017 5280 : DO atom_j = 1, bs_env%n_atom
2018 15806 : DO atom_k = 1, bs_env%n_atom
2019 36723 : DO atom_i = 1, bs_env%n_atom
2020 :
2021 23931 : j_size = bs_env%i_ao_end_from_atom(atom_j) - bs_env%i_ao_start_from_atom(atom_j) + 1
2022 23931 : k_size = bs_env%i_ao_end_from_atom(atom_k) - bs_env%i_ao_start_from_atom(atom_k) + 1
2023 23931 : i_size = bs_env%i_RI_end_from_atom(atom_i) - bs_env%i_RI_start_from_atom(atom_i) + 1
2024 :
2025 119655 : ALLOCATE (int_3c(j_size, k_size, i_size))
2026 :
2027 : ! compute 3-c int. ( μ(atom j) R , ν (atom k) S | P (atom i) 0 )
2028 : ! ("|": truncated Coulomb operator), inside build_3c_integrals: (j k | i)
2029 : CALL build_3c_integral_block(int_3c, qs_env, bs_env%ri_metric, &
2030 : basis_j=bs_env%basis_set_AO, &
2031 : basis_k=bs_env%basis_set_AO, &
2032 : basis_i=bs_env%basis_set_RI, &
2033 : cell_j=index_to_cell_3c_max(j_cell, 1:3), &
2034 : cell_k=index_to_cell_3c_max(k_cell, 1:3), &
2035 167517 : atom_k=atom_k, atom_j=atom_j, atom_i=atom_i)
2036 :
2037 1404411 : frobenius_norm = SQRT(SUM(int_3c(:, :, :)**2))
2038 :
2039 23931 : DEALLOCATE (int_3c)
2040 :
2041 : ! we use a higher threshold here to safe memory when storing the 3c integrals
2042 : ! in every tensor group
2043 33084 : IF (frobenius_norm > eps) THEN
2044 513 : nblocks_3c_max(j_cell, k_cell) = nblocks_3c_max(j_cell, k_cell) + 1
2045 : END IF
2046 :
2047 : END DO
2048 : END DO
2049 : END DO
2050 :
2051 : END DO
2052 : END DO
2053 :
2054 6 : CALL bs_env%para_env%sum(nblocks_3c_max)
2055 :
2056 18 : ALLOCATE (n_other_3c_images_max(nimages_3c_max))
2057 140 : n_other_3c_images_max(:) = 0
2058 :
2059 6 : nimages_3c = 0
2060 6 : nimage_pairs_3c = 0
2061 :
2062 140 : DO j_cell = 1, nimages_3c_max
2063 3148 : DO k_cell = 1, nimages_3c_max
2064 3148 : IF (nblocks_3c_max(j_cell, k_cell) > 0) THEN
2065 178 : n_other_3c_images_max(j_cell) = n_other_3c_images_max(j_cell) + 1
2066 178 : nimage_pairs_3c = nimage_pairs_3c + 1
2067 : END IF
2068 : END DO
2069 :
2070 140 : IF (n_other_3c_images_max(j_cell) > 0) nimages_3c = nimages_3c + 1
2071 :
2072 : END DO
2073 :
2074 6 : bs_env%nimages_3c = nimages_3c
2075 18 : ALLOCATE (bs_env%index_to_cell_3c(nimages_3c, 3))
2076 : ALLOCATE (bs_env%cell_to_index_3c(i_cell_x_min:i_cell_x_max, &
2077 : j_cell_y_min:j_cell_y_max, &
2078 30 : k_cell_z_min:k_cell_z_max))
2079 240 : bs_env%cell_to_index_3c(:, :, :) = -1
2080 :
2081 24 : ALLOCATE (bs_env%nblocks_3c(nimages_3c, nimages_3c))
2082 6 : bs_env%nblocks_3c(nimages_3c, nimages_3c) = 0
2083 :
2084 6 : j_cell = 0
2085 140 : DO j_cell_max = 1, nimages_3c_max
2086 134 : IF (n_other_3c_images_max(j_cell_max) == 0) CYCLE
2087 44 : j_cell = j_cell + 1
2088 176 : cell_index(1:3) = index_to_cell_3c_max(j_cell_max, 1:3)
2089 176 : bs_env%index_to_cell_3c(j_cell, 1:3) = cell_index(1:3)
2090 44 : bs_env%cell_to_index_3c(cell_index(1), cell_index(2), cell_index(3)) = j_cell
2091 :
2092 44 : k_cell = 0
2093 1070 : DO k_cell_max = 1, nimages_3c_max
2094 1020 : IF (n_other_3c_images_max(k_cell_max) == 0) CYCLE
2095 388 : k_cell = k_cell + 1
2096 :
2097 1154 : bs_env%nblocks_3c(j_cell, k_cell) = nblocks_3c_max(j_cell_max, k_cell_max)
2098 : END DO
2099 :
2100 : END DO
2101 :
2102 : ! we use: 8*10^-9 GB / double precision number
2103 : mem_3c_GB = REAL(bs_env%n_RI, KIND=dp)*REAL(bs_env%n_ao, KIND=dp)**2 &
2104 6 : *REAL(nimage_pairs_3c, KIND=dp)*8E-9_dp
2105 :
2106 6 : CALL m_memory(mem_occ_per_proc)
2107 6 : CALL bs_env%para_env%max(mem_occ_per_proc)
2108 :
2109 6 : mem_occ_per_proc_GB = REAL(mem_occ_per_proc, KIND=dp)/1.0E9_dp
2110 :
2111 : ! number of processors per group that entirely stores the 3c integrals and does tensor ops
2112 6 : avail_mem_per_proc_GB = bs_env%input_memory_per_proc_GB - mem_occ_per_proc_GB
2113 :
2114 : ! careful: downconvering real to integer, 1.9 -> 1; thus add 1.0 for upconversion, 1.9 -> 2
2115 6 : bs_env%group_size_tensor = MAX(INT(mem_3c_GB/avail_mem_per_proc_GB + 1.0_dp), 1)
2116 :
2117 6 : u = bs_env%unit_nr
2118 :
2119 6 : IF (u > 0) THEN
2120 3 : WRITE (u, FMT="(T2,A,F52.1,A)") "Radius of atomic orbitals", radius_ao*angstrom, " Å"
2121 3 : WRITE (u, FMT="(T2,A,F55.1,A)") "Radius of RI functions", radius_RI*angstrom, " Å"
2122 3 : WRITE (u, FMT="(T2,A,I47)") "Number of cells for 3c integrals", nimages_3c
2123 3 : WRITE (u, FMT="(T2,A,I42)") "Number of cell pairs for 3c integrals", nimage_pairs_3c
2124 3 : WRITE (u, '(T2,A)') ''
2125 3 : WRITE (u, '(T2,A,F37.1,A)') 'Input: Available memory per MPI process', &
2126 6 : bs_env%input_memory_per_proc_GB, ' GB'
2127 3 : WRITE (u, '(T2,A,F35.1,A)') 'Used memory per MPI process before GW run', &
2128 6 : mem_occ_per_proc_GB, ' GB'
2129 3 : WRITE (u, '(T2,A,F44.1,A)') 'Memory of three-center integrals', mem_3c_GB, ' GB'
2130 : END IF
2131 :
2132 6 : CALL timestop(handle)
2133 :
2134 18 : END SUBROUTINE setup_cells_3c
2135 :
2136 : ! **************************************************************************************************
2137 : !> \brief ...
2138 : !> \param index_to_cell_1 ...
2139 : !> \param index_to_cell_2 ...
2140 : !> \param nimages_1 ...
2141 : !> \param nimages_2 ...
2142 : !> \param index_to_cell ...
2143 : !> \param cell_to_index ...
2144 : !> \param nimages ...
2145 : ! **************************************************************************************************
2146 6 : SUBROUTINE sum_two_R_grids(index_to_cell_1, index_to_cell_2, nimages_1, nimages_2, &
2147 : index_to_cell, cell_to_index, nimages)
2148 :
2149 : INTEGER, DIMENSION(:, :) :: index_to_cell_1, index_to_cell_2
2150 : INTEGER :: nimages_1, nimages_2
2151 : INTEGER, ALLOCATABLE, DIMENSION(:, :) :: index_to_cell
2152 : INTEGER, DIMENSION(:, :, :), POINTER :: cell_to_index
2153 : INTEGER :: nimages
2154 :
2155 : CHARACTER(LEN=*), PARAMETER :: routineN = 'sum_two_R_grids'
2156 :
2157 : INTEGER :: handle, i_dim, img_1, img_2, nimages_max
2158 6 : INTEGER, ALLOCATABLE, DIMENSION(:, :) :: index_to_cell_tmp
2159 : INTEGER, DIMENSION(3) :: cell_1, cell_2, R, R_max, R_min
2160 :
2161 6 : CALL timeset(routineN, handle)
2162 :
2163 24 : DO i_dim = 1, 3
2164 282 : R_min(i_dim) = MINVAL(index_to_cell_1(:, i_dim)) + MINVAL(index_to_cell_2(:, i_dim))
2165 306 : R_max(i_dim) = MAXVAL(index_to_cell_1(:, i_dim)) + MAXVAL(index_to_cell_2(:, i_dim))
2166 : END DO
2167 :
2168 6 : nimages_max = (R_max(1) - R_min(1) + 1)*(R_max(2) - R_min(2) + 1)*(R_max(3) - R_min(3) + 1)
2169 :
2170 18 : ALLOCATE (index_to_cell_tmp(nimages_max, 3))
2171 534 : index_to_cell_tmp(:, :) = -1
2172 :
2173 30 : ALLOCATE (cell_to_index(R_min(1):R_max(1), R_min(2):R_max(2), R_min(3):R_max(3)))
2174 284 : cell_to_index(:, :, :) = -1
2175 :
2176 6 : nimages = 0
2177 :
2178 50 : DO img_1 = 1, nimages_1
2179 :
2180 438 : DO img_2 = 1, nimages_2
2181 :
2182 1552 : cell_1(1:3) = index_to_cell_1(img_1, 1:3)
2183 1552 : cell_2(1:3) = index_to_cell_2(img_2, 1:3)
2184 :
2185 1552 : R(1:3) = cell_1(1:3) + cell_2(1:3)
2186 :
2187 : ! check whether we have found a new cell
2188 432 : IF (cell_to_index(R(1), R(2), R(3)) == -1) THEN
2189 :
2190 122 : nimages = nimages + 1
2191 122 : cell_to_index(R(1), R(2), R(3)) = nimages
2192 488 : index_to_cell_tmp(nimages, 1:3) = R(1:3)
2193 :
2194 : END IF
2195 :
2196 : END DO
2197 :
2198 : END DO
2199 :
2200 18 : ALLOCATE (index_to_cell(nimages, 3))
2201 390 : index_to_cell(:, :) = index_to_cell_tmp(1:nimages, 1:3)
2202 :
2203 6 : CALL timestop(handle)
2204 :
2205 12 : END SUBROUTINE sum_two_R_grids
2206 :
2207 : ! **************************************************************************************************
2208 : !> \brief ...
2209 : !> \param qs_env ...
2210 : !> \param bs_env ...
2211 : ! **************************************************************************************************
2212 6 : SUBROUTINE compute_3c_integrals(qs_env, bs_env)
2213 :
2214 : TYPE(qs_environment_type), POINTER :: qs_env
2215 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
2216 :
2217 : CHARACTER(LEN=*), PARAMETER :: routineN = 'compute_3c_integrals'
2218 :
2219 : INTEGER :: handle, j_cell, k_cell, nimages_3c
2220 :
2221 6 : CALL timeset(routineN, handle)
2222 :
2223 6 : nimages_3c = bs_env%nimages_3c
2224 504 : ALLOCATE (bs_env%t_3c_int(nimages_3c, nimages_3c))
2225 50 : DO j_cell = 1, nimages_3c
2226 438 : DO k_cell = 1, nimages_3c
2227 432 : CALL dbt_create(bs_env%t_RI_AO__AO, bs_env%t_3c_int(j_cell, k_cell))
2228 : END DO
2229 : END DO
2230 :
2231 : CALL build_3c_integrals(bs_env%t_3c_int, &
2232 : bs_env%eps_filter, &
2233 : qs_env, &
2234 : bs_env%nl_3c, &
2235 : int_eps=bs_env%eps_filter*0.05_dp, &
2236 : basis_i=bs_env%basis_set_RI, &
2237 : basis_j=bs_env%basis_set_AO, &
2238 : basis_k=bs_env%basis_set_AO, &
2239 : potential_parameter=bs_env%ri_metric, &
2240 : desymmetrize=.FALSE., do_kpoints=.TRUE., cell_sym=.TRUE., &
2241 6 : cell_to_index_ext=bs_env%cell_to_index_3c)
2242 :
2243 6 : CALL bs_env%para_env%sync()
2244 :
2245 6 : CALL timestop(handle)
2246 :
2247 6 : END SUBROUTINE compute_3c_integrals
2248 :
2249 : ! **************************************************************************************************
2250 : !> \brief ...
2251 : !> \param cell_index ...
2252 : !> \param hmat ...
2253 : !> \param cell_dist ...
2254 : ! **************************************************************************************************
2255 44652 : SUBROUTINE get_cell_dist(cell_index, hmat, cell_dist)
2256 :
2257 : INTEGER, DIMENSION(3) :: cell_index
2258 : REAL(KIND=dp) :: hmat(3, 3), cell_dist
2259 :
2260 : CHARACTER(LEN=*), PARAMETER :: routineN = 'get_cell_dist'
2261 :
2262 : INTEGER :: handle, i_dim
2263 : INTEGER, DIMENSION(3) :: cell_index_adj
2264 : REAL(KIND=dp) :: cell_dist_3(3)
2265 :
2266 44652 : CALL timeset(routineN, handle)
2267 :
2268 : ! the distance of cells needs to be taken to adjacent neighbors, not
2269 : ! between the center of the cells. We thus need to rescale the cell index
2270 178608 : DO i_dim = 1, 3
2271 133956 : IF (cell_index(i_dim) > 0) cell_index_adj(i_dim) = cell_index(i_dim) - 1
2272 133956 : IF (cell_index(i_dim) < 0) cell_index_adj(i_dim) = cell_index(i_dim) + 1
2273 178608 : IF (cell_index(i_dim) == 0) cell_index_adj(i_dim) = cell_index(i_dim)
2274 : END DO
2275 :
2276 714432 : cell_dist_3(1:3) = MATMUL(hmat, REAL(cell_index_adj, KIND=dp))
2277 :
2278 178608 : cell_dist = SQRT(ABS(SUM(cell_dist_3(1:3)**2)))
2279 :
2280 44652 : CALL timestop(handle)
2281 :
2282 44652 : END SUBROUTINE get_cell_dist
2283 :
2284 : ! **************************************************************************************************
2285 : !> \brief ...
2286 : !> \param qs_env ...
2287 : !> \param bs_env ...
2288 : !> \param kpoints ...
2289 : !> \param do_print ...
2290 : ! **************************************************************************************************
2291 0 : SUBROUTINE setup_kpoints_scf_desymm(qs_env, bs_env, kpoints, do_print)
2292 : TYPE(qs_environment_type), POINTER :: qs_env
2293 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
2294 : TYPE(kpoint_type), POINTER :: kpoints
2295 :
2296 : CHARACTER(LEN=*), PARAMETER :: routineN = 'setup_kpoints_scf_desymm'
2297 :
2298 : INTEGER :: handle, i_cell_x, i_dim, img, j_cell_y, &
2299 : k_cell_z, nimages, nkp, u
2300 : INTEGER, DIMENSION(3) :: cell_grid, cixd, nkp_grid
2301 : TYPE(kpoint_type), POINTER :: kpoints_scf
2302 :
2303 : LOGICAL:: do_print
2304 :
2305 0 : CALL timeset(routineN, handle)
2306 :
2307 0 : NULLIFY (kpoints)
2308 0 : CALL kpoint_create(kpoints)
2309 :
2310 0 : CALL get_qs_env(qs_env=qs_env, kpoints=kpoints_scf)
2311 :
2312 0 : nkp_grid(1:3) = kpoints_scf%nkp_grid(1:3)
2313 0 : nkp = nkp_grid(1)*nkp_grid(2)*nkp_grid(3)
2314 :
2315 : ! we need in periodic directions at least 2 k-points in the SCF
2316 0 : DO i_dim = 1, 3
2317 0 : IF (bs_env%periodic(i_dim) == 1) THEN
2318 0 : CPASSERT(nkp_grid(i_dim) > 1)
2319 : END IF
2320 : END DO
2321 :
2322 0 : kpoints%kp_scheme = "GENERAL"
2323 0 : kpoints%nkp_grid(1:3) = nkp_grid(1:3)
2324 0 : kpoints%nkp = nkp
2325 0 : bs_env%nkp_scf_desymm = nkp
2326 :
2327 0 : ALLOCATE (kpoints%xkp(1:3, nkp))
2328 0 : CALL compute_xkp(kpoints%xkp, 1, nkp, nkp_grid)
2329 :
2330 0 : ALLOCATE (kpoints%wkp(nkp))
2331 0 : kpoints%wkp(:) = 1.0_dp/REAL(nkp, KIND=dp)
2332 :
2333 : ! for example 4x3x6 kpoint grid -> 3x3x5 cell grid because we need the same number of
2334 : ! neighbor cells on both sides of the unit cell
2335 0 : cell_grid(1:3) = nkp_grid(1:3) - MODULO(nkp_grid(1:3) + 1, 2)
2336 : ! cell index: for example for x: from -n_x/2 to +n_x/2, n_x: number of cells in x direction
2337 0 : cixd(1:3) = cell_grid(1:3)/2
2338 :
2339 0 : nimages = cell_grid(1)*cell_grid(2)*cell_grid(3)
2340 :
2341 0 : bs_env%nimages_scf_desymm = nimages
2342 :
2343 0 : ALLOCATE (kpoints%cell_to_index(-cixd(1):cixd(1), -cixd(2):cixd(2), -cixd(3):cixd(3)))
2344 0 : ALLOCATE (kpoints%index_to_cell(nimages, 3))
2345 :
2346 0 : img = 0
2347 0 : DO i_cell_x = -cixd(1), cixd(1)
2348 0 : DO j_cell_y = -cixd(2), cixd(2)
2349 0 : DO k_cell_z = -cixd(3), cixd(3)
2350 0 : img = img + 1
2351 0 : kpoints%cell_to_index(i_cell_x, j_cell_y, k_cell_z) = img
2352 0 : kpoints%index_to_cell(img, 1:3) = (/i_cell_x, j_cell_y, k_cell_z/)
2353 : END DO
2354 : END DO
2355 : END DO
2356 :
2357 0 : u = bs_env%unit_nr
2358 0 : IF (u > 0 .AND. do_print) THEN
2359 0 : WRITE (u, FMT="(T2,A,I49)") "Number of cells for G, χ, W, Σ", nimages
2360 : END IF
2361 :
2362 0 : CALL timestop(handle)
2363 :
2364 0 : END SUBROUTINE setup_kpoints_scf_desymm
2365 :
2366 : ! **************************************************************************************************
2367 : !> \brief ...
2368 : !> \param bs_env ...
2369 : ! **************************************************************************************************
2370 6 : SUBROUTINE setup_cells_Delta_R(bs_env)
2371 :
2372 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
2373 :
2374 : CHARACTER(LEN=*), PARAMETER :: routineN = 'setup_cells_Delta_R'
2375 :
2376 : INTEGER :: handle
2377 :
2378 6 : CALL timeset(routineN, handle)
2379 :
2380 : ! cell sums batch wise for fixed ΔR = S_1 - R_1; for example:
2381 : ! Σ_λσ^R = sum_PR1νS1 M^G_λ0,νS1,PR1 M^W_σR,νS1,PR1
2382 :
2383 : CALL sum_two_R_grids(bs_env%index_to_cell_3c, &
2384 : bs_env%index_to_cell_3c, &
2385 : bs_env%nimages_3c, bs_env%nimages_3c, &
2386 : bs_env%index_to_cell_Delta_R, &
2387 : bs_env%cell_to_index_Delta_R, &
2388 6 : bs_env%nimages_Delta_R)
2389 :
2390 6 : IF (bs_env%unit_nr > 0) THEN
2391 3 : WRITE (bs_env%unit_nr, FMT="(T2,A,I61)") "Number of cells ΔR", bs_env%nimages_Delta_R
2392 : END IF
2393 :
2394 6 : CALL timestop(handle)
2395 :
2396 6 : END SUBROUTINE setup_cells_Delta_R
2397 :
2398 : ! **************************************************************************************************
2399 : !> \brief ...
2400 : !> \param bs_env ...
2401 : ! **************************************************************************************************
2402 6 : SUBROUTINE setup_parallelization_Delta_R(bs_env)
2403 :
2404 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
2405 :
2406 : CHARACTER(LEN=*), PARAMETER :: routineN = 'setup_parallelization_Delta_R'
2407 :
2408 : INTEGER :: handle, i_cell_Delta_R, i_task_local, &
2409 : n_tasks_local
2410 6 : INTEGER, ALLOCATABLE, DIMENSION(:) :: i_cell_Delta_R_group, &
2411 6 : n_tensor_ops_Delta_R
2412 :
2413 6 : CALL timeset(routineN, handle)
2414 :
2415 6 : CALL compute_n_tensor_ops_Delta_R(bs_env, n_tensor_ops_Delta_R)
2416 :
2417 6 : CALL compute_Delta_R_dist(bs_env, n_tensor_ops_Delta_R, i_cell_Delta_R_group, n_tasks_local)
2418 :
2419 6 : bs_env%n_tasks_Delta_R_local = n_tasks_local
2420 :
2421 18 : ALLOCATE (bs_env%task_Delta_R(n_tasks_local))
2422 :
2423 6 : i_task_local = 0
2424 128 : DO i_cell_Delta_R = 1, bs_env%nimages_Delta_R
2425 :
2426 122 : IF (i_cell_Delta_R_group(i_cell_Delta_R) /= bs_env%tensor_group_color) CYCLE
2427 :
2428 56 : i_task_local = i_task_local + 1
2429 :
2430 128 : bs_env%task_Delta_R(i_task_local) = i_cell_Delta_R
2431 :
2432 : END DO
2433 :
2434 6 : CALL timestop(handle)
2435 :
2436 12 : END SUBROUTINE setup_parallelization_Delta_R
2437 :
2438 : ! **************************************************************************************************
2439 : !> \brief ...
2440 : !> \param bs_env ...
2441 : !> \param n_tensor_ops_Delta_R ...
2442 : !> \param i_cell_Delta_R_group ...
2443 : !> \param n_tasks_local ...
2444 : ! **************************************************************************************************
2445 6 : SUBROUTINE compute_Delta_R_dist(bs_env, n_tensor_ops_Delta_R, i_cell_Delta_R_group, n_tasks_local)
2446 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
2447 : INTEGER, ALLOCATABLE, DIMENSION(:) :: n_tensor_ops_Delta_R, &
2448 : i_cell_Delta_R_group
2449 : INTEGER :: n_tasks_local
2450 :
2451 : CHARACTER(LEN=*), PARAMETER :: routineN = 'compute_Delta_R_dist'
2452 :
2453 : INTEGER :: handle, i_Delta_R_max_op, i_group_min, &
2454 : nimages_Delta_R, u
2455 6 : INTEGER, ALLOCATABLE, DIMENSION(:) :: n_tensor_ops_Delta_R_in_group
2456 :
2457 6 : CALL timeset(routineN, handle)
2458 :
2459 6 : nimages_Delta_R = bs_env%nimages_Delta_R
2460 :
2461 6 : u = bs_env%unit_nr
2462 :
2463 6 : IF (u > 0 .AND. nimages_Delta_R < bs_env%num_tensor_groups) THEN
2464 0 : WRITE (u, FMT="(T2,A,I5,A,I5,A)") "There are only ", nimages_Delta_R, &
2465 0 : " tasks to work on but there are ", bs_env%num_tensor_groups, " groups."
2466 0 : WRITE (u, FMT="(T2,A)") "Please reduce the number of MPI processes."
2467 0 : WRITE (u, '(T2,A)') ''
2468 : END IF
2469 :
2470 18 : ALLOCATE (n_tensor_ops_Delta_R_in_group(bs_env%num_tensor_groups))
2471 18 : n_tensor_ops_Delta_R_in_group(:) = 0
2472 18 : ALLOCATE (i_cell_Delta_R_group(nimages_Delta_R))
2473 128 : i_cell_Delta_R_group(:) = -1
2474 :
2475 6 : n_tasks_local = 0
2476 :
2477 474 : DO WHILE (ANY(n_tensor_ops_Delta_R(:) .NE. 0))
2478 :
2479 : ! get largest element of n_tensor_ops_Delta_R
2480 2888 : i_Delta_R_max_op = MAXLOC(n_tensor_ops_Delta_R, 1)
2481 :
2482 : ! distribute i_Delta_R_max_op to tensor group which has currently the smallest load
2483 448 : i_group_min = MINLOC(n_tensor_ops_Delta_R_in_group, 1)
2484 :
2485 : ! the tensor groups are 0-index based; but i_group_min is 1-index based
2486 112 : i_cell_Delta_R_group(i_Delta_R_max_op) = i_group_min - 1
2487 : n_tensor_ops_Delta_R_in_group(i_group_min) = n_tensor_ops_Delta_R_in_group(i_group_min) + &
2488 112 : n_tensor_ops_Delta_R(i_Delta_R_max_op)
2489 :
2490 : ! remove i_Delta_R_max_op from n_tensor_ops_Delta_R
2491 112 : n_tensor_ops_Delta_R(i_Delta_R_max_op) = 0
2492 :
2493 118 : IF (bs_env%tensor_group_color == i_group_min - 1) n_tasks_local = n_tasks_local + 1
2494 :
2495 : END DO
2496 :
2497 6 : CALL timestop(handle)
2498 :
2499 12 : END SUBROUTINE compute_Delta_R_dist
2500 :
2501 : ! **************************************************************************************************
2502 : !> \brief ...
2503 : !> \param bs_env ...
2504 : !> \param n_tensor_ops_Delta_R ...
2505 : ! **************************************************************************************************
2506 6 : SUBROUTINE compute_n_tensor_ops_Delta_R(bs_env, n_tensor_ops_Delta_R)
2507 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
2508 : INTEGER, ALLOCATABLE, DIMENSION(:) :: n_tensor_ops_Delta_R
2509 :
2510 : CHARACTER(LEN=*), PARAMETER :: routineN = 'compute_n_tensor_ops_Delta_R'
2511 :
2512 : INTEGER :: handle, i_cell_Delta_R, i_cell_R, i_cell_R1, i_cell_R1_minus_R, i_cell_R2, &
2513 : i_cell_R2_m_R1, i_cell_S1, i_cell_S1_m_R1_p_R2, i_cell_S1_minus_R, i_cell_S2, &
2514 : nimages_Delta_R
2515 : INTEGER, DIMENSION(3) :: cell_DR, cell_m_R1, cell_R, cell_R1, cell_R1_minus_R, cell_R2, &
2516 : cell_R2_m_R1, cell_S1, cell_S1_m_R2_p_R1, cell_S1_minus_R, cell_S1_p_S2_m_R1, cell_S2
2517 : LOGICAL :: cell_found
2518 :
2519 6 : CALL timeset(routineN, handle)
2520 :
2521 6 : nimages_Delta_R = bs_env%nimages_Delta_R
2522 :
2523 18 : ALLOCATE (n_tensor_ops_Delta_R(nimages_Delta_R))
2524 128 : n_tensor_ops_Delta_R(:) = 0
2525 :
2526 : ! compute number of tensor operations for specific Delta_R
2527 128 : DO i_cell_Delta_R = 1, nimages_Delta_R
2528 :
2529 122 : IF (MODULO(i_cell_Delta_R, bs_env%num_tensor_groups) /= bs_env%tensor_group_color) CYCLE
2530 :
2531 617 : DO i_cell_R1 = 1, bs_env%nimages_3c
2532 :
2533 2200 : cell_R1(1:3) = bs_env%index_to_cell_3c(i_cell_R1, 1:3)
2534 2200 : cell_DR(1:3) = bs_env%index_to_cell_Delta_R(i_cell_Delta_R, 1:3)
2535 :
2536 : ! S_1 = R_1 + ΔR (from ΔR = S_1 - R_1)
2537 : CALL add_R(cell_R1, cell_DR, bs_env%index_to_cell_3c, cell_S1, &
2538 550 : cell_found, bs_env%cell_to_index_3c, i_cell_S1)
2539 550 : IF (.NOT. cell_found) CYCLE
2540 :
2541 1850 : DO i_cell_R2 = 1, bs_env%nimages_scf_desymm
2542 :
2543 6660 : cell_R2(1:3) = bs_env%kpoints_scf_desymm%index_to_cell(i_cell_R2, 1:3)
2544 :
2545 : ! R_2 - R_1
2546 : CALL add_R(cell_R2, -cell_R1, bs_env%index_to_cell_3c, cell_R2_m_R1, &
2547 6660 : cell_found, bs_env%cell_to_index_3c, i_cell_R2_m_R1)
2548 1665 : IF (.NOT. cell_found) CYCLE
2549 :
2550 : ! S_1 - R_1 + R_2
2551 : CALL add_R(cell_S1, cell_R2_m_R1, bs_env%index_to_cell_3c, cell_S1_m_R2_p_R1, &
2552 945 : cell_found, bs_env%cell_to_index_3c, i_cell_S1_m_R1_p_R2)
2553 945 : IF (.NOT. cell_found) CYCLE
2554 :
2555 2457 : n_tensor_ops_Delta_R(i_cell_Delta_R) = n_tensor_ops_Delta_R(i_cell_Delta_R) + 1
2556 :
2557 : END DO ! i_cell_R2
2558 :
2559 1850 : DO i_cell_S2 = 1, bs_env%nimages_scf_desymm
2560 :
2561 6660 : cell_S2(1:3) = bs_env%kpoints_scf_desymm%index_to_cell(i_cell_S2, 1:3)
2562 6660 : cell_m_R1(1:3) = -cell_R1(1:3)
2563 6660 : cell_S1_p_S2_m_R1(1:3) = cell_S1(1:3) + cell_S2(1:3) - cell_R1(1:3)
2564 :
2565 1665 : CALL is_cell_in_index_to_cell(cell_m_R1, bs_env%index_to_cell_3c, cell_found)
2566 1665 : IF (.NOT. cell_found) CYCLE
2567 :
2568 1422 : CALL is_cell_in_index_to_cell(cell_S1_p_S2_m_R1, bs_env%index_to_cell_3c, cell_found)
2569 185 : IF (.NOT. cell_found) CYCLE
2570 :
2571 : END DO ! i_cell_S2
2572 :
2573 2522 : DO i_cell_R = 1, bs_env%nimages_scf_desymm
2574 :
2575 6660 : cell_R = bs_env%kpoints_scf_desymm%index_to_cell(i_cell_R, 1:3)
2576 :
2577 : ! R_1 - R
2578 : CALL add_R(cell_R1, -cell_R, bs_env%index_to_cell_3c, cell_R1_minus_R, &
2579 6660 : cell_found, bs_env%cell_to_index_3c, i_cell_R1_minus_R)
2580 1665 : IF (.NOT. cell_found) CYCLE
2581 :
2582 : ! S_1 - R
2583 : CALL add_R(cell_S1, -cell_R, bs_env%index_to_cell_3c, cell_S1_minus_R, &
2584 4032 : cell_found, bs_env%cell_to_index_3c, i_cell_S1_minus_R)
2585 550 : IF (.NOT. cell_found) CYCLE
2586 :
2587 : END DO ! i_cell_R
2588 :
2589 : END DO ! i_cell_R1
2590 :
2591 : END DO ! i_cell_Delta_R
2592 :
2593 6 : CALL bs_env%para_env%sum(n_tensor_ops_Delta_R)
2594 :
2595 6 : CALL timestop(handle)
2596 :
2597 6 : END SUBROUTINE compute_n_tensor_ops_Delta_R
2598 :
2599 : ! **************************************************************************************************
2600 : !> \brief ...
2601 : !> \param cell_1 ...
2602 : !> \param cell_2 ...
2603 : !> \param index_to_cell ...
2604 : !> \param cell_1_plus_2 ...
2605 : !> \param cell_found ...
2606 : !> \param cell_to_index ...
2607 : !> \param i_cell_1_plus_2 ...
2608 : ! **************************************************************************************************
2609 131410 : SUBROUTINE add_R(cell_1, cell_2, index_to_cell, cell_1_plus_2, cell_found, &
2610 : cell_to_index, i_cell_1_plus_2)
2611 :
2612 : INTEGER, DIMENSION(3) :: cell_1, cell_2
2613 : INTEGER, DIMENSION(:, :) :: index_to_cell
2614 : INTEGER, DIMENSION(3) :: cell_1_plus_2
2615 : LOGICAL :: cell_found
2616 : INTEGER, DIMENSION(:, :, :), INTENT(IN), &
2617 : OPTIONAL, POINTER :: cell_to_index
2618 : INTEGER, INTENT(OUT), OPTIONAL :: i_cell_1_plus_2
2619 :
2620 : CHARACTER(LEN=*), PARAMETER :: routineN = 'add_R'
2621 :
2622 : INTEGER :: handle
2623 :
2624 131410 : CALL timeset(routineN, handle)
2625 :
2626 525640 : cell_1_plus_2(1:3) = cell_1(1:3) + cell_2(1:3)
2627 :
2628 131410 : CALL is_cell_in_index_to_cell(cell_1_plus_2, index_to_cell, cell_found)
2629 :
2630 131410 : IF (PRESENT(i_cell_1_plus_2)) THEN
2631 131410 : IF (cell_found) THEN
2632 68845 : CPASSERT(PRESENT(cell_to_index))
2633 68845 : i_cell_1_plus_2 = cell_to_index(cell_1_plus_2(1), cell_1_plus_2(2), cell_1_plus_2(3))
2634 : ELSE
2635 62565 : i_cell_1_plus_2 = -1000
2636 : END IF
2637 : END IF
2638 :
2639 131410 : CALL timestop(handle)
2640 :
2641 131410 : END SUBROUTINE add_R
2642 :
2643 : ! **************************************************************************************************
2644 : !> \brief ...
2645 : !> \param cell ...
2646 : !> \param index_to_cell ...
2647 : !> \param cell_found ...
2648 : ! **************************************************************************************************
2649 205408 : SUBROUTINE is_cell_in_index_to_cell(cell, index_to_cell, cell_found)
2650 : INTEGER, DIMENSION(3) :: cell
2651 : INTEGER, DIMENSION(:, :) :: index_to_cell
2652 : LOGICAL :: cell_found
2653 :
2654 : CHARACTER(LEN=*), PARAMETER :: routineN = 'is_cell_in_index_to_cell'
2655 :
2656 : INTEGER :: handle, i_cell, nimg
2657 : INTEGER, DIMENSION(3) :: cell_i
2658 :
2659 205408 : CALL timeset(routineN, handle)
2660 :
2661 205408 : nimg = SIZE(index_to_cell, 1)
2662 :
2663 205408 : cell_found = .FALSE.
2664 :
2665 2144394 : DO i_cell = 1, nimg
2666 :
2667 7755944 : cell_i(1:3) = index_to_cell(i_cell, 1:3)
2668 :
2669 2144394 : IF (cell_i(1) == cell(1) .AND. cell_i(2) == cell(2) .AND. cell_i(3) == cell(3)) THEN
2670 117317 : cell_found = .TRUE.
2671 : END IF
2672 :
2673 : END DO
2674 :
2675 205408 : CALL timestop(handle)
2676 :
2677 205408 : END SUBROUTINE is_cell_in_index_to_cell
2678 :
2679 : ! **************************************************************************************************
2680 : !> \brief ...
2681 : !> \param qs_env ...
2682 : !> \param bs_env ...
2683 : ! **************************************************************************************************
2684 6 : SUBROUTINE allocate_matrices_small_cell_full_kp(qs_env, bs_env)
2685 : TYPE(qs_environment_type), POINTER :: qs_env
2686 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
2687 :
2688 : CHARACTER(LEN=*), PARAMETER :: routineN = 'allocate_matrices_small_cell_full_kp'
2689 :
2690 : INTEGER :: handle, i_spin, i_t, img, n_spin, &
2691 : nimages_scf, num_time_freq_points
2692 : TYPE(cp_blacs_env_type), POINTER :: blacs_env
2693 : TYPE(mp_para_env_type), POINTER :: para_env
2694 :
2695 6 : CALL timeset(routineN, handle)
2696 :
2697 6 : nimages_scf = bs_env%nimages_scf_desymm
2698 6 : num_time_freq_points = bs_env%num_time_freq_points
2699 6 : n_spin = bs_env%n_spin
2700 :
2701 6 : CALL get_qs_env(qs_env, para_env=para_env, blacs_env=blacs_env)
2702 :
2703 72 : ALLOCATE (bs_env%fm_G_S(nimages_scf))
2704 72 : ALLOCATE (bs_env%fm_Sigma_x_R(nimages_scf))
2705 544 : ALLOCATE (bs_env%fm_chi_R_t(nimages_scf, num_time_freq_points))
2706 544 : ALLOCATE (bs_env%fm_MWM_R_t(nimages_scf, num_time_freq_points))
2707 556 : ALLOCATE (bs_env%fm_Sigma_c_R_neg_tau(nimages_scf, num_time_freq_points, n_spin))
2708 556 : ALLOCATE (bs_env%fm_Sigma_c_R_pos_tau(nimages_scf, num_time_freq_points, n_spin))
2709 60 : DO img = 1, nimages_scf
2710 54 : CALL cp_fm_create(bs_env%fm_G_S(img), bs_env%fm_work_mo(1)%matrix_struct)
2711 54 : CALL cp_fm_create(bs_env%fm_Sigma_x_R(img), bs_env%fm_work_mo(1)%matrix_struct)
2712 528 : DO i_t = 1, num_time_freq_points
2713 468 : CALL cp_fm_create(bs_env%fm_chi_R_t(img, i_t), bs_env%fm_RI_RI%matrix_struct)
2714 468 : CALL cp_fm_create(bs_env%fm_MWM_R_t(img, i_t), bs_env%fm_RI_RI%matrix_struct)
2715 468 : CALL cp_fm_set_all(bs_env%fm_MWM_R_t(img, i_t), 0.0_dp)
2716 990 : DO i_spin = 1, n_spin
2717 : CALL cp_fm_create(bs_env%fm_Sigma_c_R_neg_tau(img, i_t, i_spin), &
2718 468 : bs_env%fm_work_mo(1)%matrix_struct)
2719 : CALL cp_fm_create(bs_env%fm_Sigma_c_R_pos_tau(img, i_t, i_spin), &
2720 468 : bs_env%fm_work_mo(1)%matrix_struct)
2721 468 : CALL cp_fm_set_all(bs_env%fm_Sigma_c_R_neg_tau(img, i_t, i_spin), 0.0_dp)
2722 936 : CALL cp_fm_set_all(bs_env%fm_Sigma_c_R_pos_tau(img, i_t, i_spin), 0.0_dp)
2723 : END DO
2724 : END DO
2725 : END DO
2726 :
2727 6 : CALL timestop(handle)
2728 :
2729 6 : END SUBROUTINE allocate_matrices_small_cell_full_kp
2730 :
2731 : ! **************************************************************************************************
2732 : !> \brief ...
2733 : !> \param qs_env ...
2734 : !> \param bs_env ...
2735 : ! **************************************************************************************************
2736 6 : SUBROUTINE trafo_V_xc_R_to_kp(qs_env, bs_env)
2737 : TYPE(qs_environment_type), POINTER :: qs_env
2738 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
2739 :
2740 : CHARACTER(LEN=*), PARAMETER :: routineN = 'trafo_V_xc_R_to_kp'
2741 :
2742 : INTEGER :: handle, ikp, img, ispin, n_ao
2743 6 : INTEGER, DIMENSION(:, :, :), POINTER :: cell_to_index_scf
2744 : TYPE(cp_cfm_type) :: cfm_mo_coeff, cfm_tmp, cfm_V_xc
2745 : TYPE(cp_fm_type) :: fm_V_xc_re
2746 6 : TYPE(dbcsr_p_type), DIMENSION(:, :), POINTER :: matrix_ks
2747 : TYPE(kpoint_type), POINTER :: kpoints_scf
2748 : TYPE(neighbor_list_set_p_type), DIMENSION(:), &
2749 6 : POINTER :: sab_nl
2750 :
2751 6 : CALL timeset(routineN, handle)
2752 :
2753 6 : n_ao = bs_env%n_ao
2754 :
2755 6 : CALL get_qs_env(qs_env, matrix_ks_kp=matrix_ks, kpoints=kpoints_scf)
2756 :
2757 6 : NULLIFY (sab_nl)
2758 6 : CALL get_kpoint_info(kpoints_scf, sab_nl=sab_nl, cell_to_index=cell_to_index_scf)
2759 :
2760 6 : CALL cp_cfm_create(cfm_V_xc, bs_env%cfm_work_mo%matrix_struct)
2761 6 : CALL cp_cfm_create(cfm_mo_coeff, bs_env%cfm_work_mo%matrix_struct)
2762 6 : CALL cp_cfm_create(cfm_tmp, bs_env%cfm_work_mo%matrix_struct)
2763 6 : CALL cp_fm_create(fm_V_xc_re, bs_env%cfm_work_mo%matrix_struct)
2764 :
2765 152 : DO img = 1, bs_env%nimages_scf
2766 298 : DO ispin = 1, bs_env%n_spin
2767 : ! JW kind of hack because the format of matrix_ks remains dubious...
2768 146 : CALL dbcsr_set(matrix_ks(ispin, img)%matrix, 0.0_dp)
2769 292 : CALL copy_fm_to_dbcsr(bs_env%fm_V_xc_R(img, ispin), matrix_ks(ispin, img)%matrix)
2770 : END DO
2771 : END DO
2772 :
2773 30 : ALLOCATE (bs_env%v_xc_n(n_ao, bs_env%nkp_bs_and_DOS, bs_env%n_spin))
2774 :
2775 12 : DO ispin = 1, bs_env%n_spin
2776 232 : DO ikp = 1, bs_env%nkp_bs_and_DOS
2777 :
2778 : ! v^xc^R -> v^xc(k) (matrix_ks stores v^xc^R, see SUBROUTINE compute_V_xc)
2779 : CALL rsmat_to_kp(matrix_ks, ispin, bs_env%kpoints_DOS%xkp(1:3, ikp), &
2780 220 : cell_to_index_scf, sab_nl, bs_env, cfm_V_xc)
2781 :
2782 : ! get C_µn(k)
2783 220 : CALL cp_cfm_to_cfm(bs_env%cfm_mo_coeff_kp(ikp, ispin), cfm_mo_coeff)
2784 :
2785 : ! v^xc_nm(k_i) = sum_µν C^*_µn(k_i) v^xc_µν(k_i) C_νn(k_i)
2786 : CALL parallel_gemm('N', 'N', n_ao, n_ao, n_ao, z_one, cfm_V_xc, cfm_mo_coeff, &
2787 220 : z_zero, cfm_tmp)
2788 : CALL parallel_gemm('C', 'N', n_ao, n_ao, n_ao, z_one, cfm_mo_coeff, cfm_tmp, &
2789 220 : z_zero, cfm_V_xc)
2790 :
2791 : ! get v^xc_nn(k_i) which is a real quantity as v^xc is Hermitian
2792 220 : CALL cp_cfm_to_fm(cfm_V_xc, fm_V_xc_re)
2793 226 : CALL cp_fm_get_diag(fm_V_xc_re, bs_env%v_xc_n(:, ikp, ispin))
2794 :
2795 : END DO
2796 :
2797 : END DO
2798 :
2799 : ! just rebuild the overwritten KS matrix again
2800 6 : CALL qs_ks_build_kohn_sham_matrix(qs_env, calculate_forces=.FALSE., just_energy=.FALSE.)
2801 :
2802 6 : CALL cp_cfm_release(cfm_V_xc)
2803 6 : CALL cp_cfm_release(cfm_mo_coeff)
2804 6 : CALL cp_cfm_release(cfm_tmp)
2805 6 : CALL cp_fm_release(fm_V_xc_re)
2806 :
2807 6 : CALL timestop(handle)
2808 :
2809 12 : END SUBROUTINE trafo_V_xc_R_to_kp
2810 :
2811 : ! **************************************************************************************************
2812 : !> \brief ...
2813 : !> \param qs_env ...
2814 : !> \param bs_env ...
2815 : ! **************************************************************************************************
2816 6 : SUBROUTINE heuristic_RI_regularization(qs_env, bs_env)
2817 : TYPE(qs_environment_type), POINTER :: qs_env
2818 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
2819 :
2820 : CHARACTER(LEN=*), PARAMETER :: routineN = 'heuristic_RI_regularization'
2821 :
2822 6 : COMPLEX(KIND=dp), ALLOCATABLE, DIMENSION(:, :, :) :: M
2823 : INTEGER :: handle, ikp, ikp_local, n_RI, nkp, &
2824 : nkp_local, u
2825 : REAL(KIND=dp) :: cond_nr, cond_nr_max, max_ev, &
2826 : max_ev_ikp, min_ev, min_ev_ikp
2827 6 : REAL(KIND=dp), ALLOCATABLE, DIMENSION(:, :, :) :: M_R
2828 :
2829 6 : CALL timeset(routineN, handle)
2830 :
2831 : ! compute M^R_PQ = <phi_P,0|V^tr(rc)|phi_Q,R> for RI metric
2832 6 : CALL get_V_tr_R(M_R, bs_env%ri_metric, 0.0_dp, bs_env, qs_env)
2833 :
2834 6 : nkp = bs_env%nkp_chi_eps_W_orig_plus_extra
2835 6 : n_RI = bs_env%n_RI
2836 :
2837 6 : nkp_local = 0
2838 3846 : DO ikp = 1, nkp
2839 : ! trivial parallelization over k-points
2840 3840 : IF (MODULO(ikp, bs_env%para_env%num_pe) .NE. bs_env%para_env%mepos) CYCLE
2841 3846 : nkp_local = nkp_local + 1
2842 : END DO
2843 :
2844 30 : ALLOCATE (M(n_RI, n_RI, nkp_local))
2845 :
2846 6 : ikp_local = 0
2847 6 : cond_nr_max = 0.0_dp
2848 6 : min_ev = 1000.0_dp
2849 6 : max_ev = -1000.0_dp
2850 :
2851 3846 : DO ikp = 1, nkp
2852 :
2853 : ! trivial parallelization
2854 3840 : IF (MODULO(ikp, bs_env%para_env%num_pe) .NE. bs_env%para_env%mepos) CYCLE
2855 :
2856 1920 : ikp_local = ikp_local + 1
2857 :
2858 : ! M(k) = sum_R e^ikR M^R
2859 : CALL trafo_rs_to_ikp(M_R, M(:, :, ikp_local), &
2860 : bs_env%kpoints_scf_desymm%index_to_cell, &
2861 1920 : bs_env%kpoints_chi_eps_W%xkp(1:3, ikp))
2862 :
2863 : ! compute condition number of M_PQ(k)
2864 1920 : CALL power(M(:, :, ikp_local), 1.0_dp, 0.0_dp, cond_nr, min_ev_ikp, max_ev_ikp)
2865 :
2866 1920 : IF (cond_nr > cond_nr_max) cond_nr_max = cond_nr
2867 1920 : IF (max_ev_ikp > max_ev) max_ev = max_ev_ikp
2868 1926 : IF (min_ev_ikp < min_ev) min_ev = min_ev_ikp
2869 :
2870 : END DO ! ikp
2871 :
2872 6 : CALL bs_env%para_env%max(cond_nr_max)
2873 :
2874 6 : u = bs_env%unit_nr
2875 6 : IF (u > 0) THEN
2876 3 : WRITE (u, FMT="(T2,A,ES34.1)") "Min. abs. eigenvalue of RI metric matrix M(k)", min_ev
2877 3 : WRITE (u, FMT="(T2,A,ES34.1)") "Max. abs. eigenvalue of RI metric matrix M(k)", max_ev
2878 3 : WRITE (u, FMT="(T2,A,ES50.1)") "Max. condition number of M(k)", cond_nr_max
2879 : END IF
2880 :
2881 6 : CALL timestop(handle)
2882 :
2883 12 : END SUBROUTINE heuristic_RI_regularization
2884 :
2885 : ! **************************************************************************************************
2886 : !> \brief ...
2887 : !> \param V_tr_R ...
2888 : !> \param pot_type ...
2889 : !> \param regularization_RI ...
2890 : !> \param bs_env ...
2891 : !> \param qs_env ...
2892 : ! **************************************************************************************************
2893 76 : SUBROUTINE get_V_tr_R(V_tr_R, pot_type, regularization_RI, bs_env, qs_env)
2894 : REAL(KIND=dp), ALLOCATABLE, DIMENSION(:, :, :) :: V_tr_R
2895 : TYPE(libint_potential_type) :: pot_type
2896 : REAL(KIND=dp) :: regularization_RI
2897 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
2898 : TYPE(qs_environment_type), POINTER :: qs_env
2899 :
2900 : CHARACTER(LEN=*), PARAMETER :: routineN = 'get_V_tr_R'
2901 :
2902 : INTEGER :: handle, img, nimages_scf_desymm
2903 : INTEGER, ALLOCATABLE, DIMENSION(:) :: sizes_RI
2904 76 : INTEGER, DIMENSION(:), POINTER :: col_bsize, row_bsize
2905 : TYPE(cp_blacs_env_type), POINTER :: blacs_env
2906 76 : TYPE(cp_fm_type), ALLOCATABLE, DIMENSION(:) :: fm_V_tr_R
2907 : TYPE(dbcsr_distribution_type) :: dbcsr_dist
2908 76 : TYPE(dbcsr_type), ALLOCATABLE, DIMENSION(:) :: mat_V_tr_R
2909 : TYPE(distribution_2d_type), POINTER :: dist_2d
2910 : TYPE(neighbor_list_set_p_type), DIMENSION(:), &
2911 76 : POINTER :: sab_RI
2912 76 : TYPE(particle_type), DIMENSION(:), POINTER :: particle_set
2913 76 : TYPE(qs_kind_type), DIMENSION(:), POINTER :: qs_kind_set
2914 :
2915 76 : CALL timeset(routineN, handle)
2916 :
2917 76 : NULLIFY (sab_RI, dist_2d)
2918 :
2919 : CALL get_qs_env(qs_env=qs_env, &
2920 : blacs_env=blacs_env, &
2921 : distribution_2d=dist_2d, &
2922 : qs_kind_set=qs_kind_set, &
2923 76 : particle_set=particle_set)
2924 :
2925 228 : ALLOCATE (sizes_RI(bs_env%n_atom))
2926 76 : CALL get_particle_set(particle_set, qs_kind_set, nsgf=sizes_RI, basis=bs_env%basis_set_RI)
2927 : CALL build_2c_neighbor_lists(sab_RI, bs_env%basis_set_RI, bs_env%basis_set_RI, &
2928 : pot_type, "2c_nl_RI", qs_env, sym_ij=.FALSE., &
2929 76 : dist_2d=dist_2d)
2930 76 : CALL cp_dbcsr_dist2d_to_dist(dist_2d, dbcsr_dist)
2931 228 : ALLOCATE (row_bsize(SIZE(sizes_RI)))
2932 228 : ALLOCATE (col_bsize(SIZE(sizes_RI)))
2933 248 : row_bsize(:) = sizes_RI
2934 248 : col_bsize(:) = sizes_RI
2935 :
2936 76 : nimages_scf_desymm = bs_env%nimages_scf_desymm
2937 912 : ALLOCATE (mat_V_tr_R(nimages_scf_desymm))
2938 : CALL dbcsr_create(mat_V_tr_R(1), "(RI|RI)", dbcsr_dist, dbcsr_type_no_symmetry, &
2939 76 : row_bsize, col_bsize)
2940 76 : DEALLOCATE (row_bsize, col_bsize)
2941 :
2942 684 : DO img = 2, nimages_scf_desymm
2943 684 : CALL dbcsr_create(mat_V_tr_R(img), template=mat_V_tr_R(1))
2944 : END DO
2945 :
2946 : CALL build_2c_integrals(mat_V_tr_R, 0.0_dp, qs_env, sab_RI, bs_env%basis_set_RI, &
2947 : bs_env%basis_set_RI, pot_type, do_kpoints=.TRUE., &
2948 : ext_kpoints=bs_env%kpoints_scf_desymm, &
2949 76 : regularization_RI=regularization_RI)
2950 :
2951 912 : ALLOCATE (fm_V_tr_R(nimages_scf_desymm))
2952 760 : DO img = 1, nimages_scf_desymm
2953 684 : CALL cp_fm_create(fm_V_tr_R(img), bs_env%fm_RI_RI%matrix_struct)
2954 684 : CALL copy_dbcsr_to_fm(mat_V_tr_R(img), fm_V_tr_R(img))
2955 760 : CALL dbcsr_release(mat_V_tr_R(img))
2956 : END DO
2957 :
2958 76 : IF (.NOT. ALLOCATED(V_tr_R)) THEN
2959 380 : ALLOCATE (V_tr_R(bs_env%n_RI, bs_env%n_RI, nimages_scf_desymm))
2960 : END IF
2961 :
2962 76 : CALL fm_to_local_array(fm_V_tr_R, V_tr_R)
2963 :
2964 76 : CALL cp_fm_release(fm_V_tr_R)
2965 76 : CALL dbcsr_distribution_release(dbcsr_dist)
2966 76 : CALL release_neighbor_list_sets(sab_RI)
2967 :
2968 76 : CALL timestop(handle)
2969 :
2970 228 : END SUBROUTINE get_V_tr_R
2971 :
2972 : ! **************************************************************************************************
2973 : !> \brief ...
2974 : !> \param matrix ...
2975 : !> \param exponent ...
2976 : !> \param eps ...
2977 : !> \param cond_nr ...
2978 : !> \param min_ev ...
2979 : !> \param max_ev ...
2980 : ! **************************************************************************************************
2981 39504 : SUBROUTINE power(matrix, exponent, eps, cond_nr, min_ev, max_ev)
2982 : COMPLEX(KIND=dp), DIMENSION(:, :) :: matrix
2983 : REAL(KIND=dp) :: exponent, eps
2984 : REAL(KIND=dp), OPTIONAL :: cond_nr, min_ev, max_ev
2985 :
2986 : CHARACTER(len=*), PARAMETER :: routineN = 'power'
2987 :
2988 39504 : COMPLEX(KIND=dp), ALLOCATABLE, DIMENSION(:, :) :: eigenvectors
2989 : INTEGER :: handle, i, n
2990 : REAL(KIND=dp) :: pos_eval
2991 39504 : REAL(KIND=dp), ALLOCATABLE, DIMENSION(:) :: eigenvalues
2992 :
2993 39504 : CALL timeset(routineN, handle)
2994 :
2995 : ! make matrix perfectly Hermitian
2996 2642832 : matrix(:, :) = 0.5_dp*(matrix(:, :) + CONJG(TRANSPOSE(matrix(:, :))))
2997 :
2998 39504 : n = SIZE(matrix, 1)
2999 237024 : ALLOCATE (eigenvalues(n), eigenvectors(n, n))
3000 39504 : CALL diag_complex(matrix, eigenvectors, eigenvalues)
3001 :
3002 63824 : IF (PRESENT(cond_nr)) cond_nr = MAXVAL(ABS(eigenvalues))/MINVAL(ABS(eigenvalues))
3003 51664 : IF (PRESENT(min_ev)) min_ev = MINVAL(ABS(eigenvalues))
3004 51664 : IF (PRESENT(max_ev)) max_ev = MAXVAL(ABS(eigenvalues))
3005 :
3006 246736 : DO i = 1, n
3007 207232 : IF (eps < eigenvalues(i)) THEN
3008 207232 : pos_eval = (eigenvalues(i))**(0.5_dp*exponent)
3009 : ELSE
3010 : pos_eval = 0.0_dp
3011 : END IF
3012 1341168 : eigenvectors(:, i) = eigenvectors(:, i)*pos_eval
3013 : END DO
3014 :
3015 39504 : CALL ZGEMM("N", "C", n, n, n, z_one, eigenvectors, n, eigenvectors, n, z_zero, matrix, n)
3016 :
3017 39504 : DEALLOCATE (eigenvalues, eigenvectors)
3018 :
3019 39504 : CALL timestop(handle)
3020 :
3021 39504 : END SUBROUTINE power
3022 :
3023 : ! **************************************************************************************************
3024 : !> \brief ...
3025 : !> \param bs_env ...
3026 : !> \param Sigma_c_n_time ...
3027 : !> \param Sigma_c_n_freq ...
3028 : !> \param ispin ...
3029 : ! **************************************************************************************************
3030 276 : SUBROUTINE time_to_freq(bs_env, Sigma_c_n_time, Sigma_c_n_freq, ispin)
3031 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
3032 : REAL(KIND=dp), DIMENSION(:, :, :) :: Sigma_c_n_time, Sigma_c_n_freq
3033 : INTEGER :: ispin
3034 :
3035 : CHARACTER(LEN=*), PARAMETER :: routineN = 'time_to_freq'
3036 :
3037 : INTEGER :: handle, i_t, j_w, n_occ
3038 : REAL(KIND=dp) :: freq_j, time_i, w_cos_ij, w_sin_ij
3039 276 : REAL(KIND=dp), ALLOCATABLE, DIMENSION(:, :) :: Sigma_c_n_cos_time, Sigma_c_n_sin_time
3040 :
3041 276 : CALL timeset(routineN, handle)
3042 :
3043 1104 : ALLOCATE (Sigma_c_n_cos_time(bs_env%n_ao, bs_env%num_time_freq_points))
3044 828 : ALLOCATE (Sigma_c_n_sin_time(bs_env%n_ao, bs_env%num_time_freq_points))
3045 :
3046 27428 : Sigma_c_n_cos_time(:, :) = 0.5_dp*(Sigma_c_n_time(:, :, 1) + Sigma_c_n_time(:, :, 2))
3047 27428 : Sigma_c_n_sin_time(:, :) = 0.5_dp*(Sigma_c_n_time(:, :, 1) - Sigma_c_n_time(:, :, 2))
3048 :
3049 55132 : Sigma_c_n_freq(:, :, :) = 0.0_dp
3050 :
3051 2996 : DO i_t = 1, bs_env%num_time_freq_points
3052 :
3053 31636 : DO j_w = 1, bs_env%num_time_freq_points
3054 :
3055 28640 : freq_j = bs_env%imag_freq_points(j_w)
3056 28640 : time_i = bs_env%imag_time_points(i_t)
3057 : ! integration weights for cosine and sine transform
3058 28640 : w_cos_ij = bs_env%weights_cos_t_to_w(j_w, i_t)*COS(freq_j*time_i)
3059 28640 : w_sin_ij = bs_env%weights_sin_t_to_w(j_w, i_t)*SIN(freq_j*time_i)
3060 :
3061 : ! 1. Re(Σ^c_nn(k_i,iω)) from cosine transform
3062 : Sigma_c_n_freq(:, j_w, 1) = Sigma_c_n_freq(:, j_w, 1) + &
3063 262592 : w_cos_ij*Sigma_c_n_cos_time(:, i_t)
3064 :
3065 : ! 2. Im(Σ^c_nn(k_i,iω)) from sine transform
3066 : Sigma_c_n_freq(:, j_w, 2) = Sigma_c_n_freq(:, j_w, 2) + &
3067 265312 : w_sin_ij*Sigma_c_n_sin_time(:, i_t)
3068 :
3069 : END DO
3070 :
3071 : END DO
3072 :
3073 : ! for occupied levels, we need the correlation self-energy for negative omega.
3074 : ! Therefore, weight_sin should be computed with -omega, which results in an
3075 : ! additional minus for the imaginary part:
3076 276 : n_occ = bs_env%n_occ(ispin)
3077 12996 : Sigma_c_n_freq(1:n_occ, :, 2) = -Sigma_c_n_freq(1:n_occ, :, 2)
3078 :
3079 276 : CALL timestop(handle)
3080 :
3081 552 : END SUBROUTINE time_to_freq
3082 :
3083 : ! **************************************************************************************************
3084 : !> \brief ...
3085 : !> \param bs_env ...
3086 : !> \param Sigma_c_ikp_n_freq ...
3087 : !> \param Sigma_x_ikp_n ...
3088 : !> \param V_xc_ikp_n ...
3089 : !> \param eigenval_scf ...
3090 : !> \param ikp ...
3091 : !> \param ispin ...
3092 : ! **************************************************************************************************
3093 276 : SUBROUTINE analyt_conti_and_print(bs_env, Sigma_c_ikp_n_freq, Sigma_x_ikp_n, V_xc_ikp_n, &
3094 276 : eigenval_scf, ikp, ispin)
3095 :
3096 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
3097 : REAL(KIND=dp), DIMENSION(:, :, :) :: Sigma_c_ikp_n_freq
3098 : REAL(KIND=dp), DIMENSION(:) :: Sigma_x_ikp_n, V_xc_ikp_n, eigenval_scf
3099 : INTEGER :: ikp, ispin
3100 :
3101 : CHARACTER(LEN=*), PARAMETER :: routineN = 'analyt_conti_and_print'
3102 :
3103 : CHARACTER(len=3) :: occ_vir
3104 : CHARACTER(len=default_string_length) :: fname
3105 : INTEGER :: handle, i_mo, ikp_for_print, iunit, &
3106 : n_mo, nkp
3107 : LOGICAL :: is_bandstruc_kpoint, print_DOS_kpoints, &
3108 : print_ikp
3109 : REAL(KIND=dp), ALLOCATABLE, DIMENSION(:) :: dummy, Sigma_c_ikp_n_qp
3110 :
3111 276 : CALL timeset(routineN, handle)
3112 :
3113 276 : n_mo = bs_env%n_ao
3114 1104 : ALLOCATE (dummy(n_mo), Sigma_c_ikp_n_qp(n_mo))
3115 2948 : Sigma_c_ikp_n_qp(:) = 0.0_dp
3116 :
3117 2948 : DO i_mo = 1, n_mo
3118 :
3119 : ! parallelization
3120 2672 : IF (MODULO(i_mo, bs_env%para_env%num_pe) /= bs_env%para_env%mepos) CYCLE
3121 :
3122 : CALL continuation_pade(Sigma_c_ikp_n_qp, &
3123 : bs_env%imag_freq_points_fit, dummy, dummy, &
3124 : Sigma_c_ikp_n_freq(:, 1:bs_env%num_freq_points_fit, 1)*z_one + &
3125 : Sigma_c_ikp_n_freq(:, 1:bs_env%num_freq_points_fit, 2)*gaussi, &
3126 : Sigma_x_ikp_n(:) - V_xc_ikp_n(:), &
3127 : eigenval_scf(:), eigenval_scf(:), &
3128 : bs_env%do_hedin_shift, &
3129 : i_mo, bs_env%n_occ(ispin), bs_env%n_vir(ispin), &
3130 : bs_env%nparam_pade, bs_env%num_freq_points_fit, &
3131 : ri_rpa_g0w0_crossing_newton, bs_env%n_occ(ispin), &
3132 87412 : 0.0_dp, .TRUE., .FALSE., 1, e_fermi_ext=bs_env%e_fermi(ispin))
3133 : END DO
3134 :
3135 276 : CALL bs_env%para_env%sum(Sigma_c_ikp_n_qp)
3136 :
3137 276 : CALL correct_obvious_fitting_fails(Sigma_c_ikp_n_qp, ispin, bs_env)
3138 :
3139 : bs_env%eigenval_G0W0(:, ikp, ispin) = eigenval_scf(:) + &
3140 : Sigma_c_ikp_n_qp(:) + &
3141 : Sigma_x_ikp_n(:) - &
3142 2948 : V_xc_ikp_n(:)
3143 :
3144 2948 : bs_env%eigenval_HF(:, ikp, ispin) = eigenval_scf(:) + Sigma_x_ikp_n(:) - V_xc_ikp_n(:)
3145 :
3146 : ! only print eigenvalues of DOS k-points in case no bandstructure path has been given
3147 276 : print_DOS_kpoints = (bs_env%nkp_only_bs .LE. 0)
3148 : ! in kpoints_DOS, the last nkp_only_bs are bandstructure k-points
3149 276 : is_bandstruc_kpoint = (ikp > bs_env%nkp_only_DOS)
3150 276 : print_ikp = print_DOS_kpoints .OR. is_bandstruc_kpoint
3151 :
3152 276 : IF (bs_env%para_env%is_source() .AND. print_ikp) THEN
3153 :
3154 106 : IF (print_DOS_kpoints) THEN
3155 44 : nkp = bs_env%nkp_only_DOS
3156 44 : ikp_for_print = ikp
3157 : ELSE
3158 62 : nkp = bs_env%nkp_only_bs
3159 62 : ikp_for_print = ikp - bs_env%nkp_only_DOS
3160 : END IF
3161 :
3162 106 : fname = "bandstructure_SCF_and_G0W0"
3163 :
3164 106 : IF (ikp_for_print == 1) THEN
3165 : CALL open_file(TRIM(fname), unit_number=iunit, file_status="REPLACE", &
3166 21 : file_action="WRITE")
3167 : ELSE
3168 : CALL open_file(TRIM(fname), unit_number=iunit, file_status="OLD", &
3169 85 : file_action="WRITE", file_position="APPEND")
3170 : END IF
3171 :
3172 106 : WRITE (iunit, "(A)") " "
3173 106 : WRITE (iunit, "(A10,I7,A25,3F10.4)") "kpoint: ", ikp_for_print, "coordinate: ", &
3174 212 : bs_env%kpoints_DOS%xkp(:, ikp)
3175 106 : WRITE (iunit, "(A)") " "
3176 106 : WRITE (iunit, "(A5,A12,3A17,A16,A18)") "n", "k", "ϵ_nk^DFT (eV)", "Σ^c_nk (eV)", &
3177 212 : "Σ^x_nk (eV)", "v_nk^xc (eV)", "ϵ_nk^G0W0 (eV)"
3178 106 : WRITE (iunit, "(A)") " "
3179 :
3180 1154 : DO i_mo = 1, n_mo
3181 1048 : IF (i_mo .LE. bs_env%n_occ(ispin)) occ_vir = 'occ'
3182 1048 : IF (i_mo > bs_env%n_occ(ispin)) occ_vir = 'vir'
3183 1048 : WRITE (iunit, "(I5,3A,I5,4F16.3,F17.3)") i_mo, ' (', occ_vir, ') ', ikp_for_print, &
3184 1048 : eigenval_scf(i_mo)*evolt, &
3185 1048 : Sigma_c_ikp_n_qp(i_mo)*evolt, &
3186 1048 : Sigma_x_ikp_n(i_mo)*evolt, &
3187 1048 : V_xc_ikp_n(i_mo)*evolt, &
3188 2202 : bs_env%eigenval_G0W0(i_mo, ikp, ispin)*evolt
3189 : END DO
3190 :
3191 106 : WRITE (iunit, "(A)") " "
3192 :
3193 106 : CALL close_file(iunit)
3194 :
3195 : END IF
3196 :
3197 276 : CALL timestop(handle)
3198 :
3199 552 : END SUBROUTINE analyt_conti_and_print
3200 :
3201 : ! **************************************************************************************************
3202 : !> \brief ...
3203 : !> \param Sigma_c_ikp_n_qp ...
3204 : !> \param ispin ...
3205 : !> \param bs_env ...
3206 : ! **************************************************************************************************
3207 276 : SUBROUTINE correct_obvious_fitting_fails(Sigma_c_ikp_n_qp, ispin, bs_env)
3208 : REAL(KIND=dp), ALLOCATABLE, DIMENSION(:) :: Sigma_c_ikp_n_qp
3209 : INTEGER :: ispin
3210 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
3211 :
3212 : CHARACTER(LEN=*), PARAMETER :: routineN = 'correct_obvious_fitting_fails'
3213 :
3214 : INTEGER :: handle, homo, i_mo, j_mo, &
3215 : n_levels_scissor, n_mo
3216 : LOGICAL :: is_occ, is_vir
3217 : REAL(KIND=dp) :: sum_Sigma_c
3218 :
3219 276 : CALL timeset(routineN, handle)
3220 :
3221 276 : n_mo = bs_env%n_ao
3222 276 : homo = bs_env%n_occ(ispin)
3223 :
3224 2948 : DO i_mo = 1, n_mo
3225 :
3226 : ! if |𝚺^c| > 13 eV, we use a scissors shift
3227 2948 : IF (ABS(Sigma_c_ikp_n_qp(i_mo)) > 13.0_dp/evolt) THEN
3228 :
3229 0 : is_occ = (i_mo .LE. homo)
3230 0 : is_vir = (i_mo > homo)
3231 :
3232 0 : n_levels_scissor = 0
3233 0 : sum_Sigma_c = 0.0_dp
3234 :
3235 : ! compute scissor
3236 0 : DO j_mo = 1, n_mo
3237 :
3238 : ! only compute scissor from other GW levels close in energy
3239 0 : IF (is_occ .AND. j_mo > homo) CYCLE
3240 0 : IF (is_vir .AND. j_mo .LE. homo) CYCLE
3241 0 : IF (ABS(i_mo - j_mo) > 10) CYCLE
3242 0 : IF (i_mo == j_mo) CYCLE
3243 :
3244 0 : n_levels_scissor = n_levels_scissor + 1
3245 0 : sum_Sigma_c = sum_Sigma_c + Sigma_c_ikp_n_qp(j_mo)
3246 :
3247 : END DO
3248 :
3249 : ! overwrite the self-energy with scissor shift
3250 0 : Sigma_c_ikp_n_qp(i_mo) = sum_Sigma_c/REAL(n_levels_scissor, KIND=dp)
3251 :
3252 : END IF
3253 :
3254 : END DO ! i_mo
3255 :
3256 276 : CALL timestop(handle)
3257 :
3258 276 : END SUBROUTINE correct_obvious_fitting_fails
3259 :
3260 : END MODULE gw_utils
|
