Line data Source code
1 : !--------------------------------------------------------------------------------------------------!
2 : ! CP2K: A general program to perform molecular dynamics simulations !
3 : ! Copyright 2000-2026 CP2K developers group <https://cp2k.org> !
4 : ! !
5 : ! SPDX-License-Identifier: GPL-2.0-or-later !
6 : !--------------------------------------------------------------------------------------------------!
7 :
8 : ! **************************************************************************************************
9 : !> \brief GW using RI-RS Approximation for molecules
10 : !> \par History
11 : !> 04.2026 created [Ritaj Tyagi]
12 : ! **************************************************************************************************
13 : MODULE gw_large_cell_Gamma_ri_rs
14 : USE atomic_kind_types, ONLY: atomic_kind_type
15 : USE basis_set_types, ONLY: gto_basis_set_type
16 : USE cell_types, ONLY: cell_type,&
17 : get_cell,&
18 : pbc
19 : USE constants_operator, ONLY: operator_coulomb
20 : USE cp_blacs_env, ONLY: cp_blacs_env_create,&
21 : cp_blacs_env_release,&
22 : cp_blacs_env_type
23 : USE cp_dbcsr_api, ONLY: &
24 : dbcsr_add, dbcsr_binary_read, dbcsr_binary_write, dbcsr_copy, dbcsr_create, &
25 : dbcsr_deallocate_matrix, dbcsr_distribution_get, dbcsr_distribution_new, &
26 : dbcsr_distribution_release, dbcsr_distribution_type, dbcsr_finalize, dbcsr_get_block_p, &
27 : dbcsr_get_info, dbcsr_iterator_blocks_left, dbcsr_iterator_next_block, &
28 : dbcsr_iterator_start, dbcsr_iterator_stop, dbcsr_iterator_type, dbcsr_multiply, &
29 : dbcsr_p_type, dbcsr_put_block, dbcsr_release, dbcsr_scale, dbcsr_set, dbcsr_type, &
30 : dbcsr_type_no_symmetry
31 : USE cp_dbcsr_contrib, ONLY: dbcsr_reserve_all_blocks
32 : USE cp_dbcsr_operations, ONLY: copy_dbcsr_to_fm,&
33 : copy_fm_to_dbcsr,&
34 : dbcsr_deallocate_matrix_set,&
35 : max_elements_per_block
36 : USE cp_fm_basic_linalg, ONLY: cp_fm_scale_and_add,&
37 : cp_fm_uplo_to_full
38 : USE cp_fm_cholesky, ONLY: cp_fm_cholesky_decompose,&
39 : cp_fm_cholesky_invert
40 : USE cp_fm_diag, ONLY: cp_fm_power
41 : USE cp_fm_struct, ONLY: cp_fm_struct_type
42 : USE cp_fm_types, ONLY: cp_fm_create,&
43 : cp_fm_get_info,&
44 : cp_fm_release,&
45 : cp_fm_set_all,&
46 : cp_fm_to_fm,&
47 : cp_fm_type
48 : USE cp_log_handling, ONLY: cp_get_default_logger,&
49 : cp_logger_type
50 : USE cp_output_handling, ONLY: cp_p_file,&
51 : cp_print_key_should_output
52 : USE gw_integrals, ONLY: build_3c_integral_block_ctx,&
53 : gw_3c_ctx_create,&
54 : gw_3c_ctx_release,&
55 : gw_3c_ctx_type,&
56 : gw_3c_ws_create,&
57 : gw_3c_ws_release,&
58 : gw_3c_ws_type
59 : USE gw_large_cell_gamma, ONLY: &
60 : Fourier_transform_w_to_t, G_occ_vir, compute_QP_energies, compute_fm_chi_Gamma_freq, &
61 : create_fm_W_MIC_time, delete_unnecessary_files, fill_fm_Sigma_c_Gamma_time, fm_write, &
62 : get_W_MIC, multiply_fm_W_MIC_time_with_Minv_Gamma
63 : USE gw_non_periodic_ri_rs, ONLY: get_basis_offsets,&
64 : precompute_ri_rs_radii,&
65 : ri_rs_grid_assembler,&
66 : solve_D_lp_distributed
67 : USE gw_utils, ONLY: de_init_bs_env
68 : USE input_constants, ONLY: rtp_method_bse
69 : USE input_section_types, ONLY: section_vals_type
70 : USE kinds, ONLY: dp
71 : USE kpoint_coulomb_2c, ONLY: build_2c_coulomb_matrix_kp
72 : USE machine, ONLY: m_walltime
73 : USE message_passing, ONLY: mp_para_env_type
74 : USE mp2_ri_2c, ONLY: RI_2c_integral_mat
75 : USE orbital_pointers, ONLY: indco,&
76 : ncoset
77 : USE parallel_gemm_api, ONLY: parallel_gemm
78 : USE particle_types, ONLY: particle_type
79 : USE post_scf_bandstructure_types, ONLY: post_scf_bandstructure_type
80 : USE qs_environment_types, ONLY: get_qs_env,&
81 : qs_environment_type
82 : USE qs_kind_types, ONLY: get_qs_kind,&
83 : qs_kind_type
84 : #include "./base/base_uses.f90"
85 :
86 : IMPLICIT NONE
87 :
88 : PRIVATE
89 :
90 : CHARACTER(len=*), PARAMETER, PRIVATE :: moduleN = 'gw_large_cell_Gamma_ri_rs'
91 :
92 : PUBLIC :: gw_calc_large_cell_Gamma_ri_rs
93 :
94 : CONTAINS
95 :
96 : ! **************************************************************************************************
97 : !> \brief GW calculation using RI-RS formalism for molecules
98 : !> \param qs_env ...
99 : !> \param bs_env ...
100 : ! **************************************************************************************************
101 :
102 0 : SUBROUTINE gw_calc_large_cell_Gamma_ri_rs(qs_env, bs_env)
103 :
104 : TYPE(qs_environment_type), POINTER :: qs_env
105 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
106 :
107 : CHARACTER(LEN=*), PARAMETER :: routineN = 'gw_calc_large_cell_Gamma_ri_rs'
108 :
109 : INTEGER :: handle
110 0 : TYPE(cp_fm_type), ALLOCATABLE, DIMENSION(:) :: fm_Sigma_x_Gamma, fm_W_time
111 0 : TYPE(cp_fm_type), ALLOCATABLE, DIMENSION(:, :, :) :: fm_Sigma_c_Gamma_time
112 :
113 0 : CALL timeset(routineN, handle)
114 :
115 : !!========================================================================
116 : !! 0. Precompute AO and RI Radii
117 : !! Per-atom cutoff radii from the most diffuse Gaussian primitives in
118 : !! the AO and RI auxiliary basis sets. Stored in bs_env%ri_rs%
119 : !! radius_ao_per_atom and radius_ri_per_atom, used for sphere-cutoff
120 : !! and phi_local screening.
121 : !!========================================================================
122 0 : CALL precompute_ri_rs_radii(qs_env, bs_env)
123 :
124 : !!========================================================================
125 : !! 1. Grid Generation for RI-RS
126 : !! (Modified Lebedev grids from Ivan Duchemin and Xavier Blase)
127 : !! Generate flattened 1D array of grid points for RI-RS.
128 : !! Equation: r_g(k) = R_A + r_g(A)
129 : !!========================================================================
130 0 : CALL ri_rs_grid_assembler(qs_env, bs_env, bs_env%ri_rs%grid_points)
131 :
132 : !!========================================================================
133 : !! 2. Atomic Basis Evaluation
134 : !! Compute values of spherical atomic basis functions at grid points.
135 : !! Expression: Φ_μl = Φ_μ(r_l) (mat_phi_mu_l)
136 : !!========================================================================
137 : CALL atomic_basis_at_grid_point(qs_env, bs_env, bs_env%ri_rs%grid_points, &
138 0 : bs_env%ri_rs%mat_phi_mu_l)
139 :
140 : !!========================================================================
141 : !! 3. Compute RI-RS Coefficients (Z_lp)
142 : !! Solve the regularized system for each atom P, where the grid domain
143 : !! is restricted to r_l within a cutoff distance of atom P:
144 : !! a. D_ll' = [ Σ_μ Φ_μ(r_l) Φ_μ(r_l') ]^2 (Equation 13)
145 : !! b. D_lP = Σ_{μν} Φ_μ(r_l) Φ_ν(r_l) (μν|P) (Equation 15)
146 : !! c. Conditioning:
147 : !! Dvec_l = 1 / sqrt(D_ll) (Diagonal scaling vector)
148 : !! D'_ll' = Dvec_l * D_ll' * Dvec_l' + λδ_ll'
149 : !! D'_lP = Dvec_l * D_lP
150 : !! d. Solve: Σ_l' D'_ll' * Z'_l'P = D'_lP (Equation 14)
151 : !! e. Rescale: Z_lP = Z'_lP * Dvec_l (Z_lP stored in mat_Z_lP)
152 : !!========================================================================
153 : CALL compute_coeff_Z_lP(qs_env, bs_env, bs_env%ri_rs%grid_points, &
154 0 : bs_env%ri_rs%mat_phi_mu_l, bs_env%ri_rs%mat_Z_lP)
155 :
156 : !!========================================================================
157 : !! 4. Compute Independent-Particle Polarizability (χ)
158 : !! G^occ_µλ(i|τ|) = sum_n^occ C_µn e^(-|(ϵ_n-ϵ_F)τ|) C_λn
159 : !! G^vir_µλ(i|τ|) = sum_n^vir C_µn e^(-|(ϵ_n-ϵ_F)τ|) C_λn
160 : !! G^occ_ll'(i|τ|) = sum_µν Φ_µ(r_l) G^occ_µν Φ_ν(r_l')
161 : !! G^vir_ll'(i|τ|) = sum_µν Φ_µ(r_l) G^vir_µν Φ_ν(r_l')
162 : !! χ_ll'(iτ) = G^occ_ll'(i|τ|) * G^vir_ll'(i|τ|)
163 : !! χ_PQ(iτ) = sum_ll' Z_lP χ_ll'(iτ) Z_l'Q
164 : !!========================================================================
165 : CALL get_mat_chi_Gamma_tau(bs_env, bs_env%mat_chi_Gamma_tau, &
166 0 : bs_env%ri_rs%mat_phi_mu_l, bs_env%ri_rs%mat_Z_lP)
167 :
168 : !!========================================================================
169 : !! 5. Compute Screened Interaction (W^MIC)
170 : !! χ_PQ(iτ) -> χ_PQ(iω) -> ε_PQ(iω) -> W_PQ(iω) -> W^MIC_PQ(iτ)
171 : !!========================================================================
172 0 : CALL get_W_MIC(bs_env, qs_env, bs_env%mat_chi_Gamma_tau, fm_W_time)
173 :
174 : !!========================================================================
175 : !! 6. Compute Exact Exchange Self-Energy (Σ^x)
176 : !! D_µν = sum_n^occ C_µn C_νn
177 : !! D_ll' = sum_µν Φ_µ(r_l) D_µν Φ_ν(r_l')
178 : !! V^trunc_ll' = sum_PQ Z_lP V^trunc_PQ Z_l'Q
179 : !! Σ^x_ll' = D_ll' * V^trunc_ll'
180 : !! Σ^x_λσ(k=0) = -sum_ll' Φ_λ(r_l) Σ^x_ll' Φ_σ(r_l')
181 : !!========================================================================
182 : CALL compute_Sigma_x(bs_env, qs_env, bs_env%ri_rs%mat_phi_mu_l, &
183 0 : bs_env%ri_rs%mat_Z_lP, fm_Sigma_x_Gamma)
184 :
185 : !!========================================================================
186 : !! 7. Compute Correlation Self-Energy (Σ^c)
187 : !! W^MIC_ll'(iτ) = sum_PQ Z_lP W^MIC_PQ(iτ) Z_l'Q
188 : !! Σ^c_ll'(iτ) = -G^occ_ll'(i|τ|) * W^MIC_ll'(iτ), for τ < 0
189 : !! Σ^c_ll'(iτ) = G^vir_ll'(i|τ|) * W^MIC_ll'(iτ), for τ > 0
190 : !! Σ^c_λσ(iτ) = sum_ll' Φ_λ(r_l) Σ^c_ll'(iτ) Φ_σ(r_l')
191 : !!========================================================================
192 : CALL compute_Sigma_c(bs_env, fm_W_time, bs_env%ri_rs%mat_phi_mu_l, &
193 0 : bs_env%ri_rs%mat_Z_lP, fm_Sigma_c_Gamma_time)
194 :
195 : !!========================================================================
196 : !! 8. Compute Quasiparticle Energies
197 : !! Σ^c_λσ(iτ) -> Σ^c_nn(ϵ)
198 : !! ϵ_nk^GW = ϵ_nk^DFT + Σ^c_nn(ϵ) + Σ^x_nn - v^xc_nn
199 : !!========================================================================
200 0 : CALL compute_QP_energies(bs_env, qs_env, fm_Sigma_x_Gamma, fm_Sigma_c_Gamma_time)
201 :
202 0 : CALL de_init_bs_env(bs_env)
203 :
204 0 : CALL timestop(handle)
205 :
206 0 : END SUBROUTINE gw_calc_large_cell_Gamma_ri_rs
207 :
208 : ! **************************************************************************************************
209 : !> \brief Evaluates atomic basis functions on a real-space grid and builds a sparse DBCSR matrix.
210 : !> \param qs_env ...
211 : !> \param bs_env ...
212 : !> \param ri_rs_grid_points ...
213 : !> \param mat_phi_mu_l ...
214 : ! **************************************************************************************************
215 :
216 0 : SUBROUTINE atomic_basis_at_grid_point(qs_env, bs_env, ri_rs_grid_points, mat_phi_mu_l)
217 :
218 : TYPE(qs_environment_type), POINTER :: qs_env
219 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
220 : REAL(KIND=dp), ALLOCATABLE, INTENT(INOUT) :: ri_rs_grid_points(:, :)
221 : TYPE(dbcsr_type), INTENT(OUT) :: mat_phi_mu_l
222 :
223 : CHARACTER(LEN=*), PARAMETER :: routineN = 'atomic_basis_at_grid_point'
224 :
225 : INTEGER :: c_size, chunk_size, dimen_ORB, handle, i, i_blk, iatom, natom, npcol, nprow, &
226 : num_grid_chunks, r_end, r_start, total_grid_npts
227 : INTEGER, ALLOCATABLE, DIMENSION(:) :: first_sgf
228 0 : INTEGER, DIMENSION(:), POINTER :: c_blk_sizes, col_dist, r_blk_sizes, &
229 0 : row_dist
230 0 : REAL(KIND=dp), ALLOCATABLE, DIMENSION(:, :) :: atom_col_buffer
231 0 : TYPE(atomic_kind_type), DIMENSION(:), POINTER :: atomic_kind_set
232 : TYPE(cell_type), POINTER :: cell
233 : TYPE(dbcsr_distribution_type) :: dist
234 : TYPE(dbcsr_distribution_type), POINTER :: dbcsr_dist_ks
235 : TYPE(mp_para_env_type), POINTER :: para_env
236 0 : TYPE(particle_type), DIMENSION(:), POINTER :: particle_set
237 0 : TYPE(qs_kind_type), DIMENSION(:), POINTER :: qs_kind_set
238 :
239 0 : CALL timeset(routineN, handle)
240 :
241 : ! Setup Grid Blocking
242 0 : chunk_size = max_elements_per_block
243 :
244 : ! Extract environment variables
245 : CALL get_qs_env(qs_env, cell=cell, atomic_kind_set=atomic_kind_set, &
246 : qs_kind_set=qs_kind_set, particle_set=particle_set, &
247 0 : para_env=para_env)
248 :
249 0 : natom = SIZE(particle_set)
250 0 : total_grid_npts = SIZE(ri_rs_grid_points, 2)
251 :
252 : ! Map the starting indices of spherical gaussian functions (SGF) for each atom
253 0 : ALLOCATE (first_sgf(natom + 1))
254 0 : CALL get_basis_offsets(particle_set, qs_kind_set, first_sgf, dimen_ORB)
255 :
256 : ! =========================================================================
257 : ! 1. SETUP DBCSR MATRIX TOPOLOGY
258 : ! =========================================================================
259 :
260 : ! A. Define Column Block Sizes (1 Block = 1 Atom's full basis set)
261 0 : ALLOCATE (c_blk_sizes(natom))
262 0 : DO iatom = 1, natom
263 0 : c_blk_sizes(iatom) = first_sgf(iatom + 1) - first_sgf(iatom)
264 : END DO
265 :
266 : ! B. Define Row Block Sizes (Grid chunks of max size 256)
267 0 : num_grid_chunks = CEILING(REAL(total_grid_npts, KIND=dp)/REAL(chunk_size, KIND=dp))
268 0 : ALLOCATE (r_blk_sizes(num_grid_chunks))
269 0 : r_blk_sizes = chunk_size
270 0 : IF (MOD(total_grid_npts, chunk_size) /= 0) THEN
271 0 : r_blk_sizes(num_grid_chunks) = MOD(total_grid_npts, chunk_size)
272 : END IF
273 :
274 : ! C. Fetch CP2K's Default Process Grid Configuration
275 0 : CALL get_qs_env(qs_env, dbcsr_dist=dbcsr_dist_ks)
276 0 : CALL dbcsr_distribution_get(dbcsr_dist_ks, nprows=nprow, npcols=npcol)
277 :
278 : ! D. Build Custom Mappings using Round-Robin across the 2D process grid
279 :
280 0 : ALLOCATE (row_dist(num_grid_chunks))
281 0 : DO i = 1, num_grid_chunks
282 0 : row_dist(i) = MOD(i - 1, nprow)
283 : END DO
284 :
285 0 : ALLOCATE (col_dist(natom))
286 0 : DO i = 1, natom
287 0 : col_dist(i) = MOD(i - 1, npcol)
288 : END DO
289 :
290 : ! E. Create the DBCSR Distribution and Initialize the Matrix
291 : CALL dbcsr_distribution_new(dist, template=dbcsr_dist_ks, &
292 0 : row_dist=row_dist, col_dist=col_dist)
293 :
294 : CALL dbcsr_create(mat_phi_mu_l, name="phi_val_sparse", dist=dist, &
295 : matrix_type=dbcsr_type_no_symmetry, &
296 0 : row_blk_size=r_blk_sizes, col_blk_size=c_blk_sizes)
297 :
298 : ! =========================================================================
299 : ! 2. STREAM DATA DIRECTLY INTO SPARSE MATRIX
300 : ! =========================================================================
301 : ! Iterate over the atoms assigned to this specific MPI rank
302 0 : DO iatom = para_env%mepos + 1, natom, para_env%num_pe
303 :
304 0 : c_size = c_blk_sizes(iatom)
305 :
306 : ! Allocate a temporary dense buffer just for this specific atom
307 0 : ALLOCATE (atom_col_buffer(total_grid_npts, c_size))
308 0 : atom_col_buffer = 0.0_dp
309 :
310 : ! Evaluate the basis functions on the grid. Skip grid points outside the spatial
311 : ! extent of the most diffuse AO Gaussian on iatom; beyond that radius the contribution
312 : ! is guaranteed below eps_filter.
313 : CALL fill_phi_for_atom(atom_col_buffer, ri_rs_grid_points, total_grid_npts, &
314 : iatom, particle_set, qs_kind_set, cell, &
315 0 : r2_threshold=bs_env%ri_rs%radius_ao_per_atom(iatom)**2)
316 :
317 : ! Slice the dense column into chunks and insert into DBCSR
318 0 : DO i_blk = 1, num_grid_chunks
319 0 : r_start = (i_blk - 1)*chunk_size + 1
320 0 : r_end = MIN(i_blk*chunk_size, total_grid_npts)
321 :
322 : ! Apply dynamic sparsity filtering: Only store blocks with physical significance
323 0 : IF (MAXVAL(ABS(atom_col_buffer(r_start:r_end, 1:c_size))) > bs_env%eps_filter) THEN
324 : CALL dbcsr_put_block(mat_phi_mu_l, row=i_blk, col=iatom, &
325 0 : block=atom_col_buffer(r_start:r_end, 1:c_size))
326 : END IF
327 : END DO
328 :
329 0 : DEALLOCATE (atom_col_buffer)
330 :
331 : END DO
332 :
333 : ! Finalize triggers internal MPI communication to route blocks to their correct 2D process owners
334 0 : CALL dbcsr_finalize(mat_phi_mu_l)
335 :
336 0 : IF (bs_env%unit_nr > 0) THEN
337 0 : WRITE (bs_env%unit_nr, *) "Done with evaluation of phi"
338 : END IF
339 :
340 : ! -------------------------------------------------------------------------
341 : ! CLEANUP
342 : ! -------------------------------------------------------------------------
343 0 : DEALLOCATE (first_sgf, r_blk_sizes, c_blk_sizes, row_dist, col_dist)
344 0 : CALL dbcsr_distribution_release(dist)
345 :
346 0 : CALL timestop(handle)
347 :
348 0 : END SUBROUTINE atomic_basis_at_grid_point
349 :
350 : ! **************************************************************************************************
351 : !> \brief Compute value of all basis functions for a single atom across all grid points.
352 : !> Sums contributions from periodic images of `iatom` (loop over (ix, iy, iz) cells gated
353 : !> by `cell%perd`). Each per-image squared distance is compared against `r2_threshold`
354 : !> (per-atom AO Gaussian extent²); images beyond that radius contribute below eps_filter
355 : !> and are skipped.
356 : !> \param phi_val ...
357 : !> \param ri_rs_grid ...
358 : !> \param npts ...
359 : !> \param iatom ...
360 : !> \param particle_set ...
361 : !> \param qs_kind_set ...
362 : !> \param cell ...
363 : !> \param r2_threshold per-image squared-distance threshold; CYCLE if r² > r2_threshold. Pass
364 : !> HUGE(1.0_dp) to disable.
365 : ! **************************************************************************************************
366 :
367 0 : SUBROUTINE fill_phi_for_atom(phi_val, ri_rs_grid, npts, iatom, &
368 : particle_set, qs_kind_set, cell, r2_threshold)
369 :
370 : REAL(KIND=dp), INTENT(INOUT) :: phi_val(:, :)
371 : INTEGER, INTENT(IN) :: npts
372 : REAL(KIND=dp), INTENT(IN) :: ri_rs_grid(3, npts)
373 : INTEGER, INTENT(IN) :: iatom
374 : TYPE(particle_type), DIMENSION(:), POINTER :: particle_set
375 : TYPE(qs_kind_type), DIMENSION(:), POINTER :: qs_kind_set
376 : TYPE(cell_type), POINTER :: cell
377 : REAL(KIND=dp), INTENT(IN) :: r2_threshold
378 :
379 : CHARACTER(LEN=*), PARAMETER :: routineN = 'fill_phi_for_atom'
380 :
381 : INTEGER :: first_sgf, handle, i_pt, ico, iend_co, ikind, ipgf, iset, isgf, ishell, &
382 : istart_co, ix, ix_max, ix_min, iy, iy_max, iy_min, iz, iz_max, iz_min, l, last_sgf, lx, &
383 : ly, lz, n_cart_total, row_idx
384 : REAL(KIND=dp) :: alpha, cell_vector(3), dist_vec(3), &
385 : dist_vec_raw(3), exp_val, poly, r2, &
386 : r_atom(3), weight
387 : REAL(KIND=dp), DIMENSION(3, 3) :: hmat
388 : TYPE(gto_basis_set_type), POINTER :: orb_basis_set
389 :
390 0 : CALL timeset(routineN, handle)
391 :
392 : ! Get Atom Info
393 0 : ikind = particle_set(iatom)%atomic_kind%kind_number
394 0 : CALL get_qs_kind(qs_kind_set(ikind), basis_set=orb_basis_set, basis_type="ORB")
395 0 : CALL get_cell(cell=cell, h=hmat)
396 :
397 0 : IF (.NOT. ASSOCIATED(orb_basis_set)) THEN
398 0 : CALL timestop(handle)
399 0 : RETURN
400 : END IF
401 :
402 0 : IF (cell%perd(1) == 1) THEN; ix_min = -1; ix_max = 1; ELSE; ix_min = 0; ix_max = 0
403 : END IF
404 0 : IF (cell%perd(2) == 1) THEN; iy_min = -1; iy_max = 1; ELSE; iy_min = 0; iy_max = 0
405 : END IF
406 0 : IF (cell%perd(3) == 1) THEN; iz_min = -1; iz_max = 1; ELSE; iz_min = 0; iz_max = 0
407 : END IF
408 :
409 0 : r_atom = particle_set(iatom)%r
410 :
411 : !$OMP PARALLEL DO DEFAULT(NONE) &
412 : !$OMP SHARED(phi_val, ri_rs_grid, npts, orb_basis_set, r_atom, hmat, &
413 : !$OMP ncoset, indco, cell, ix_min, ix_max, &
414 : !$OMP iy_min, iy_max, iz_min, iz_max, r2_threshold) &
415 : !$OMP PRIVATE(i_pt, dist_vec_raw, ix, iy, iz, cell_vector, dist_vec, r2, iset, &
416 : !$OMP n_cart_total, ishell, l, istart_co, iend_co, first_sgf, last_sgf, &
417 : !$OMP ipgf, alpha, exp_val, isgf, ico, row_idx, weight, lx, ly, lz, poly) &
418 0 : !$OMP SCHEDULE(DYNAMIC)
419 :
420 : DO i_pt = 1, npts
421 :
422 : dist_vec_raw = ri_rs_grid(:, i_pt) - r_atom
423 :
424 : DO ix = ix_min, ix_max
425 : DO iy = iy_min, iy_max
426 : DO iz = iz_min, iz_max
427 :
428 : cell_vector(1:3) = MATMUL(hmat, REAL([ix, iy, iz], dp))
429 :
430 : dist_vec = dist_vec_raw - cell_vector
431 :
432 : r2 = DOT_PRODUCT(dist_vec, dist_vec)
433 :
434 : IF (r2 > r2_threshold) CYCLE
435 :
436 : DO iset = 1, orb_basis_set%nset
437 : n_cart_total = ncoset(orb_basis_set%lmax(iset))
438 :
439 : DO ishell = 1, orb_basis_set%nshell(iset)
440 : l = orb_basis_set%l(ishell, iset)
441 : istart_co = ncoset(l - 1) + 1
442 : iend_co = ncoset(l)
443 :
444 : first_sgf = orb_basis_set%first_sgf(ishell, iset)
445 : last_sgf = orb_basis_set%last_sgf(ishell, iset)
446 :
447 : DO ipgf = 1, orb_basis_set%npgf(iset)
448 : alpha = orb_basis_set%zet(ipgf, iset)
449 : exp_val = EXP(-alpha*r2)
450 :
451 : DO isgf = first_sgf, last_sgf
452 : DO ico = istart_co, iend_co
453 : row_idx = (ipgf - 1)*n_cart_total + ico
454 : weight = orb_basis_set%sphi(row_idx, isgf)
455 : lx = indco(1, ico)
456 : ly = indco(2, ico)
457 : lz = indco(3, ico)
458 : poly = (dist_vec(1)**lx)*(dist_vec(2)**ly)*(dist_vec(3)**lz)
459 :
460 : phi_val(i_pt, isgf) = phi_val(i_pt, isgf) + (weight*poly*exp_val)
461 :
462 : END DO
463 : END DO
464 : END DO
465 : END DO
466 : END DO
467 : END DO
468 : END DO
469 : END DO
470 : END DO
471 : !$OMP END PARALLEL DO
472 :
473 0 : CALL timestop(handle)
474 :
475 : END SUBROUTINE fill_phi_for_atom
476 :
477 : ! **************************************************************************************************
478 : !> \brief Compute RI-RS Coefficients (Z_lP)
479 : !> \param qs_env ...
480 : !> \param bs_env ...
481 : !> \param ri_rs_grid_points ...
482 : !> \param mat_phi_mu_l ...
483 : !> \param mat_Z_lP ...
484 : ! **************************************************************************************************
485 :
486 0 : SUBROUTINE compute_coeff_Z_lP(qs_env, bs_env, ri_rs_grid_points, mat_phi_mu_l, mat_Z_lP)
487 :
488 : ! Arguments
489 : TYPE(qs_environment_type), POINTER :: qs_env
490 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
491 : REAL(KIND=dp), ALLOCATABLE, INTENT(INOUT) :: ri_rs_grid_points(:, :)
492 : TYPE(dbcsr_type), INTENT(INOUT) :: mat_phi_mu_l
493 : TYPE(dbcsr_type), INTENT(OUT) :: mat_Z_lP
494 :
495 : CHARACTER(LEN=*), PARAMETER :: key = 'PROPERTIES%BANDSTRUCTURE%GW%PRINT%RESTART', &
496 : routineN = 'compute_coeff_Z_lP'
497 :
498 : INTEGER :: atom_j_mepos, atom_j_stride, atom_P, atom_P_start, atom_P_stride, col_end, &
499 : col_start, current_chunk_size, g, group_handle, handle, handle_dpotrf, handle_dpotrs, &
500 : handle_dsyrk, i, i_blk, ikind, info, j, j_ri, l, loc_idx, loc_ptr, max_ao_size, &
501 : max_loc_ri, my_group, n_ao_total, n_grid_total, n_groups, n_loc_ri, n_local_grid, &
502 : n_procs_per_atom, natom, nkind, npcol_phi, num_grid_chunks, P_loop_atom, r_end, r_start, &
503 : source_atom
504 0 : INTEGER, ALLOCATABLE, DIMENSION(:) :: local_grid_idx, row_offset
505 0 : INTEGER, DIMENSION(:), POINTER :: col_dist_ri, r_blk_sizes, ri_blk_sizes, &
506 0 : row_dist_grid
507 : REAL(KIND=dp) :: cutoff_ri, cutoff_ri_2, d_sP, dist2_min, &
508 : r2_threshold, r_c, t1, t2, t3
509 0 : REAL(KIND=dp), ALLOCATABLE, DIMENSION(:) :: cutoff_ri_per_atom, cutoff_ri_per_kind, &
510 0 : d_vec_local
511 0 : REAL(KIND=dp), ALLOCATABLE, DIMENSION(:, :) :: D_local, d_lp_local, phi_local, &
512 0 : sphere_grid, Z_blk
513 : REAL(KIND=dp), DIMENSION(3) :: dist_vec_raw, pos_P
514 0 : TYPE(atomic_kind_type), DIMENSION(:), POINTER :: atomic_kind_set
515 : TYPE(cell_type), POINTER :: cell
516 : TYPE(cp_blacs_env_type), POINTER :: blacs_env_sub
517 : TYPE(cp_fm_struct_type), POINTER :: fm_struct_b, fm_struct_D
518 : TYPE(cp_fm_type) :: fm_b, fm_D
519 : TYPE(cp_logger_type), POINTER :: logger
520 : TYPE(dbcsr_distribution_type) :: dist_phi, dist_Z
521 0 : TYPE(gw_3c_ctx_type) :: ctx_3c
522 : TYPE(mp_para_env_type), POINTER :: para_env, para_env_sub
523 0 : TYPE(particle_type), DIMENSION(:), POINTER :: particle_set
524 0 : TYPE(qs_kind_type), DIMENSION(:), POINTER :: qs_kind_set
525 : TYPE(section_vals_type), POINTER :: input
526 :
527 0 : CALL timeset(routineN, handle)
528 :
529 0 : t1 = m_walltime()
530 :
531 : CALL get_qs_env(qs_env, para_env=para_env, particle_set=particle_set, input=input, &
532 0 : cell=cell, qs_kind_set=qs_kind_set, atomic_kind_set=atomic_kind_set)
533 :
534 : ! ---------------------------------------------------------------------
535 : ! Subgroup setup. Default G=1 keeps the single-rank BLAS path; G>1 splits
536 : ! ranks into atom-groups so the Cholesky on D_local distributes across G
537 : ! ranks (memory ~1/G) and the compute_d_lp build also splits across the
538 : ! subgroup. G=1 leaves para_env_sub / blacs_env_sub NULL — no subgroup
539 : ! comms created, atom_P loop uses per-rank round-robin, compute_d_lp runs
540 : ! its full atom_j range on each rank, no allreduce.
541 : ! ---------------------------------------------------------------------
542 0 : n_procs_per_atom = MIN(bs_env%ri_rs%n_procs_per_atom_z_lp, para_env%num_pe)
543 0 : IF (n_procs_per_atom < 1) n_procs_per_atom = 1
544 :
545 0 : NULLIFY (para_env_sub, blacs_env_sub)
546 0 : IF (n_procs_per_atom > 1) THEN
547 0 : n_groups = para_env%num_pe/n_procs_per_atom
548 0 : my_group = MIN(para_env%mepos/n_procs_per_atom, n_groups - 1)
549 0 : ALLOCATE (para_env_sub)
550 0 : CALL para_env_sub%from_split(para_env, my_group)
551 0 : CALL cp_blacs_env_create(blacs_env=blacs_env_sub, para_env=para_env_sub)
552 0 : atom_P_start = my_group + 1
553 0 : atom_P_stride = n_groups
554 0 : atom_j_mepos = para_env_sub%mepos
555 0 : atom_j_stride = para_env_sub%num_pe
556 : ELSE
557 0 : atom_P_start = para_env%mepos + 1
558 0 : atom_P_stride = para_env%num_pe
559 0 : atom_j_mepos = 0
560 0 : atom_j_stride = 1
561 : END IF
562 :
563 0 : natom = SIZE(bs_env%i_RI_start_from_atom)
564 0 : n_ao_total = bs_env%i_ao_end_from_atom(natom)
565 0 : n_grid_total = SIZE(ri_rs_grid_points, 2)
566 :
567 : ! =========================================================================
568 : ! 1. SETUP DBCSR TOPOLOGY & EXACT OFFSETS
569 : ! =========================================================================
570 0 : CALL dbcsr_get_info(mat_phi_mu_l, row_blk_size=r_blk_sizes, distribution=dist_phi)
571 : CALL dbcsr_distribution_get(dist_phi, row_dist=row_dist_grid, &
572 0 : group=group_handle, npcols=npcol_phi)
573 :
574 0 : num_grid_chunks = SIZE(r_blk_sizes)
575 :
576 0 : ALLOCATE (row_offset(num_grid_chunks))
577 0 : row_offset(1) = 0
578 0 : DO i_blk = 2, num_grid_chunks
579 0 : row_offset(i_blk) = row_offset(i_blk - 1) + r_blk_sizes(i_blk - 1)
580 : END DO
581 :
582 0 : ALLOCATE (ri_blk_sizes(natom), col_dist_ri(natom))
583 0 : DO atom_P = 1, natom
584 0 : ri_blk_sizes(atom_P) = bs_env%i_RI_end_from_atom(atom_P) - bs_env%i_RI_start_from_atom(atom_P) + 1
585 0 : col_dist_ri(atom_P) = MOD(atom_P - 1, npcol_phi)
586 : END DO
587 :
588 0 : CALL dbcsr_distribution_new(dist_Z, template=dist_phi, row_dist=row_dist_grid, col_dist=col_dist_ri)
589 :
590 0 : IF (bs_env%ri_rs%Z_lP_exists) THEN
591 : CALL dbcsr_binary_read(filepath=TRIM(bs_env%prefix)//"Z_lP.matrix", &
592 : distribution=dist_Z, &
593 0 : matrix_new=mat_Z_lP)
594 0 : IF (bs_env%unit_nr > 0) THEN
595 : WRITE (bs_env%unit_nr, '(T2,A,T57,A,F7.1,A)') &
596 0 : 'Read Z_lP from file ', ' Execution time', m_walltime() - t1, ' s'
597 0 : WRITE (bs_env%unit_nr, '(A)') ' '
598 : END IF
599 : ELSE
600 :
601 : CALL dbcsr_create(mat_Z_lP, name="mat_Z_lP", dist=dist_Z, &
602 : matrix_type=dbcsr_type_no_symmetry, &
603 0 : row_blk_size=r_blk_sizes, col_blk_size=ri_blk_sizes)
604 :
605 0 : max_ao_size = 0
606 0 : DO j = 1, SIZE(bs_env%i_ao_start_from_atom)
607 0 : max_ao_size = MAX(max_ao_size, bs_env%i_ao_end_from_atom(j) - bs_env%i_ao_start_from_atom(j) + 1)
608 : END DO
609 0 : max_loc_ri = MAXVAL(ri_blk_sizes)
610 :
611 : ! Per-atom RI-RS integration sphere:
612 : ! cutoff_ri(P) = r_c + r_AO(P)
613 : ! where r_c is the truncated-Coulomb cutoff of the RI metric. The
614 : ! CUTOFF_RADIUS_RI_RS keyword (when > 0) overrides the entire cutoff calculation.
615 0 : nkind = SIZE(atomic_kind_set)
616 0 : ALLOCATE (cutoff_ri_per_atom(natom))
617 :
618 0 : IF (bs_env%ri_rs%cutoff_radius_ri_rs > 0.0_dp) THEN
619 0 : cutoff_ri_per_atom(:) = bs_env%ri_rs%cutoff_radius_ri_rs
620 : ELSE
621 0 : r_c = bs_env%ri_metric%cutoff_radius
622 0 : DO P_loop_atom = 1, natom
623 : cutoff_ri_per_atom(P_loop_atom) = &
624 0 : r_c + bs_env%ri_rs%radius_ao_per_atom(P_loop_atom)
625 : END DO
626 : END IF
627 :
628 0 : ALLOCATE (cutoff_ri_per_kind(nkind))
629 0 : cutoff_ri_per_kind(:) = 0.0_dp
630 0 : IF (bs_env%unit_nr > 0) THEN
631 0 : DO P_loop_atom = 1, natom
632 0 : ikind = particle_set(P_loop_atom)%atomic_kind%kind_number
633 : cutoff_ri_per_kind(ikind) = MAX(cutoff_ri_per_kind(ikind), &
634 0 : cutoff_ri_per_atom(P_loop_atom))
635 : END DO
636 0 : WRITE (bs_env%unit_nr, '(T2,A)') 'Per-kind maximum RI-RS sphere cutoff (Bohr):'
637 0 : WRITE (bs_env%unit_nr, '(T4,A4,A14)') 'Kind', 'max cutoff_ri'
638 0 : DO ikind = 1, nkind
639 : WRITE (bs_env%unit_nr, '(T4,A4,F14.4)') &
640 0 : atomic_kind_set(ikind)%element_symbol, &
641 0 : cutoff_ri_per_kind(ikind)
642 : END DO
643 0 : WRITE (bs_env%unit_nr, '(A)') ' '
644 0 : DEALLOCATE (cutoff_ri_per_kind)
645 : END IF
646 :
647 : ! Shared 3c-integral context: hoists libint / t_c_g0 / md_ftable / contracted
648 : ! sphi tables out of the per-triple call so compute_d_lp threads only allocate
649 : ! a lightweight per-thread workspace. MPI-collective; must be outside any
650 : ! OMP region.
651 : CALL gw_3c_ctx_create(ctx_3c, qs_env, bs_env%ri_metric, &
652 : basis_j=bs_env%basis_set_AO, basis_k=bs_env%basis_set_AO, &
653 0 : basis_i=bs_env%basis_set_RI)
654 :
655 : ! =========================================================================
656 : ! 2. MPI LOOP OVER ATOMS (Fully independent, no MPI barriers inside)
657 : ! phi_local for each atom_P's cutoff sphere is built on the fly via
658 : ! fill_phi_for_atom — no dense replicated phi_global, no allreduce.
659 : ! =========================================================================
660 0 : DO atom_P = atom_P_start, natom, atom_P_stride
661 :
662 0 : n_loc_ri = ri_blk_sizes(atom_P)
663 0 : pos_P(:) = particle_set(atom_P)%r(:)
664 :
665 0 : cutoff_ri = cutoff_ri_per_atom(atom_P)
666 0 : cutoff_ri_2 = cutoff_ri**2
667 :
668 : ! ---------------------------------------------------------------------
669 : ! A. Determine Local Grid Domain based on cutoff_ri (PBC distance)
670 : ! ---------------------------------------------------------------------
671 0 : n_local_grid = 0
672 0 : DO l = 1, n_grid_total
673 0 : dist_vec_raw = pbc(ri_rs_grid_points(1:3, l), pos_P(1:3), cell)
674 0 : dist2_min = DOT_PRODUCT(dist_vec_raw, dist_vec_raw)
675 0 : IF (dist2_min <= cutoff_ri_2) n_local_grid = n_local_grid + 1
676 : END DO
677 :
678 0 : ALLOCATE (local_grid_idx(n_local_grid))
679 :
680 0 : n_local_grid = 0
681 0 : DO l = 1, n_grid_total
682 0 : dist_vec_raw = pbc(ri_rs_grid_points(1:3, l), pos_P(1:3), cell)
683 0 : dist2_min = DOT_PRODUCT(dist_vec_raw, dist_vec_raw)
684 0 : IF (dist2_min <= cutoff_ri_2) THEN
685 0 : n_local_grid = n_local_grid + 1
686 0 : local_grid_idx(n_local_grid) = l
687 : END IF
688 : END DO
689 :
690 : ! ---------------------------------------------------------------------
691 : ! B. Build phi_local on the fly via fill_phi_for_atom.
692 : ! Only source atoms whose AO basis can reach the cutoff sphere of
693 : ! atom_P (MIC distance) contribute; the rest are pruned. The
694 : ! periodic fill_phi_for_atom sums over (ix, iy, iz) images of
695 : ! source_atom internally.
696 : ! ---------------------------------------------------------------------
697 0 : ALLOCATE (sphere_grid(3, n_local_grid))
698 0 : DO loc_idx = 1, n_local_grid
699 0 : sphere_grid(:, loc_idx) = ri_rs_grid_points(:, local_grid_idx(loc_idx))
700 : END DO
701 :
702 0 : ALLOCATE (phi_local(n_local_grid, n_ao_total))
703 0 : phi_local = 0.0_dp
704 :
705 0 : DO source_atom = 1, natom
706 0 : dist_vec_raw = pbc(particle_set(source_atom)%r(:), pos_P(:), cell)
707 0 : d_sP = NORM2(dist_vec_raw)
708 0 : IF (d_sP > bs_env%ri_rs%radius_ao_per_atom(source_atom) + cutoff_ri) CYCLE
709 :
710 0 : col_start = bs_env%i_ao_start_from_atom(source_atom)
711 0 : col_end = bs_env%i_ao_end_from_atom(source_atom)
712 0 : r2_threshold = bs_env%ri_rs%radius_ao_per_atom(source_atom)**2
713 :
714 : CALL fill_phi_for_atom(phi_local(:, col_start:col_end), sphere_grid, &
715 : n_local_grid, source_atom, particle_set, qs_kind_set, &
716 0 : cell, r2_threshold)
717 : END DO
718 :
719 0 : DEALLOCATE (sphere_grid)
720 :
721 : ! ---------------------------------------------------------------------
722 : ! C. Build Local RHS Matrix (d_lp_local) first so the subgroup-
723 : ! distributed compute_d_lp + allreduce is not entangled with the LHS
724 : ! build. compute_d_lp does not depend on D_local or d_vec_local.
725 : ! ---------------------------------------------------------------------
726 0 : ALLOCATE (d_lp_local(n_local_grid, n_loc_ri))
727 0 : d_lp_local = 0.0_dp
728 :
729 0 : t2 = m_walltime()
730 :
731 : CALL compute_d_lp(bs_env, ctx_3c, cell, phi_local, d_lp_local, n_local_grid, &
732 0 : n_loc_ri, atom_P, max_ao_size, atom_j_mepos, atom_j_stride)
733 :
734 : ! Reduce per-subgroup-rank partials into the replicated d_lp_local.
735 : ! Skipped for G=1 (BLAS path): each rank has the full sum locally.
736 0 : IF (n_procs_per_atom > 1) THEN
737 0 : CALL para_env_sub%sum(d_lp_local)
738 : END IF
739 :
740 0 : t3 = m_walltime()
741 :
742 : ! ---------------------------------------------------------------------
743 : ! D. Build d_vec_local (Jacobi diagonal) + LHS — BLAS or ScaLAPACK
744 : ! ---------------------------------------------------------------------
745 0 : ALLOCATE (d_vec_local(n_local_grid))
746 :
747 0 : IF (n_procs_per_atom == 1) THEN
748 : ! BLAS path: build D_local densely, compute d_vec as side-effect
749 : ! of the Jacobi step (preserves bit-identical arithmetic with the
750 : ! previous branch).
751 0 : ALLOCATE (D_local(n_local_grid, n_local_grid))
752 0 : D_local = 0.0_dp
753 :
754 0 : CALL timeset(routineN//"_dsyrk", handle_dsyrk)
755 : CALL dsyrk("L", "N", n_local_grid, n_ao_total, 1.0_dp, phi_local, &
756 0 : n_local_grid, 0.0_dp, D_local, n_local_grid)
757 0 : CALL timestop(handle_dsyrk)
758 :
759 : !$OMP PARALLEL DO DEFAULT(NONE) &
760 : !$OMP SHARED(n_local_grid, D_local, d_vec_local, bs_env) &
761 : !$OMP PRIVATE(i) &
762 0 : !$OMP SCHEDULE(STATIC)
763 : DO i = 1, n_local_grid
764 : D_local(i, i) = D_local(i, i)**2
765 : d_vec_local(i) = 1.0_dp/SQRT(MAX(D_local(i, i), 1.0E-16_dp))
766 : D_local(i, i) = (D_local(i, i)*d_vec_local(i)**2) + bs_env%ri_rs%tikhonov
767 : END DO
768 : !$OMP END PARALLEL DO
769 :
770 : !$OMP PARALLEL DO DEFAULT(NONE) &
771 : !$OMP SHARED(n_local_grid, D_local, d_vec_local) &
772 : !$OMP PRIVATE(j, i) &
773 0 : !$OMP SCHEDULE(DYNAMIC)
774 : DO j = 1, n_local_grid
775 : DO i = j + 1, n_local_grid
776 : D_local(i, j) = D_local(i, j)**2
777 : D_local(i, j) = D_local(i, j)*d_vec_local(i)*d_vec_local(j)
778 : D_local(j, i) = D_local(i, j)
779 : END DO
780 : END DO
781 : !$OMP END PARALLEL DO
782 : ELSE
783 : ! ScaLAPACK path: d_vec computed directly from phi (= 1/||phi_i||^2);
784 : ! solve_D_lp_distributed builds D block-cyclic internally with
785 : ! the squared+scaled values, so no dense D_local on this rank.
786 : !$OMP PARALLEL DO DEFAULT(NONE) &
787 : !$OMP SHARED(n_local_grid, n_ao_total, phi_local, d_vec_local) &
788 : !$OMP PRIVATE(i, j) &
789 0 : !$OMP SCHEDULE(STATIC)
790 : DO i = 1, n_local_grid
791 : d_vec_local(i) = 0.0_dp
792 : DO j = 1, n_ao_total
793 : d_vec_local(i) = d_vec_local(i) + phi_local(i, j)*phi_local(i, j)
794 : END DO
795 : d_vec_local(i) = 1.0_dp/MAX(d_vec_local(i), 1.0E-16_dp)
796 : END DO
797 : !$OMP END PARALLEL DO
798 : END IF
799 :
800 : ! ---------------------------------------------------------------------
801 : ! E. Pre-scale d_lp by d_vec
802 : ! ---------------------------------------------------------------------
803 : !$OMP PARALLEL DO DEFAULT(NONE) &
804 : !$OMP SHARED(n_loc_ri, n_local_grid, d_lp_local, d_vec_local) &
805 : !$OMP PRIVATE(j_ri, i) &
806 0 : !$OMP SCHEDULE(STATIC)
807 : DO j_ri = 1, n_loc_ri
808 : DO i = 1, n_local_grid
809 : d_lp_local(i, j_ri) = d_lp_local(i, j_ri)*d_vec_local(i)
810 : END DO
811 : END DO
812 : !$OMP END PARALLEL DO
813 :
814 : ! ---------------------------------------------------------------------
815 : ! F. Solve — BLAS dpotrf/dpotrs or ScaLAPACK pdpotrf/pdpotrs
816 : ! ---------------------------------------------------------------------
817 0 : IF (n_procs_per_atom == 1) THEN
818 0 : CALL timeset(routineN//"_dpotrf", handle_dpotrf)
819 0 : CALL dpotrf('L', n_local_grid, D_local, n_local_grid, info)
820 0 : CALL timestop(handle_dpotrf)
821 0 : CALL timeset(routineN//"_dpotrs", handle_dpotrs)
822 : CALL dpotrs('L', n_local_grid, n_loc_ri, D_local, n_local_grid, &
823 0 : d_lp_local, n_local_grid, info)
824 0 : CALL timestop(handle_dpotrs)
825 0 : DEALLOCATE (D_local)
826 : ELSE
827 : CALL solve_D_lp_distributed(phi_local, d_vec_local, d_lp_local, &
828 : n_local_grid, n_ao_total, n_loc_ri, &
829 : bs_env%ri_rs%tikhonov, &
830 : para_env_sub, blacs_env_sub, &
831 0 : fm_struct_D, fm_struct_b, fm_D, fm_b, info)
832 : END IF
833 :
834 : ! ---------------------------------------------------------------------
835 : ! G. Post-scale solution by d_vec (common to both paths)
836 : ! ---------------------------------------------------------------------
837 : !$OMP PARALLEL DO DEFAULT(NONE) &
838 : !$OMP SHARED(n_loc_ri, n_local_grid, d_lp_local, d_vec_local) &
839 : !$OMP PRIVATE(j_ri, i) &
840 0 : !$OMP SCHEDULE(STATIC)
841 : DO j_ri = 1, n_loc_ri
842 : DO i = 1, n_local_grid
843 : d_lp_local(i, j_ri) = d_lp_local(i, j_ri)*d_vec_local(i)
844 : END DO
845 : END DO
846 : !$OMP END PARALLEL DO
847 :
848 : ! ---------------------------------------------------------------------
849 : ! H. Scatter Local Solution Back to Global DBCSR Matrix.
850 : ! Under ScaLAPACK (G>1) the d_lp_local solution is identical on all
851 : ! G subgroup ranks (gathered via cp_fm_get_submatrix); only the
852 : ! subgroup root writes to mat_Z_lP so each atom column is emitted
853 : ! exactly once. DBCSR routes blocks to their global owner on finalize.
854 : ! local_grid_idx is ascending (built by the ordered scan above), so
855 : ! a single walking pointer over chunks works.
856 : ! ---------------------------------------------------------------------
857 0 : IF (n_procs_per_atom == 1 .OR. para_env_sub%mepos == 0) THEN
858 0 : ALLOCATE (Z_blk(MAXVAL(r_blk_sizes), n_loc_ri))
859 0 : loc_ptr = 1
860 :
861 0 : DO i_blk = 1, num_grid_chunks
862 0 : r_start = row_offset(i_blk) + 1
863 0 : r_end = row_offset(i_blk) + r_blk_sizes(i_blk)
864 0 : current_chunk_size = r_blk_sizes(i_blk)
865 :
866 0 : Z_blk = 0.0_dp
867 :
868 0 : DO WHILE (loc_ptr <= n_local_grid)
869 0 : g = local_grid_idx(loc_ptr)
870 0 : IF (g > r_end) EXIT
871 0 : Z_blk(g - r_start + 1, 1:n_loc_ri) = d_lp_local(loc_ptr, 1:n_loc_ri)
872 0 : loc_ptr = loc_ptr + 1
873 : END DO
874 :
875 0 : IF (MAXVAL(ABS(Z_blk(1:current_chunk_size, 1:n_loc_ri))) > bs_env%eps_filter) THEN
876 : CALL dbcsr_put_block(mat_Z_lP, row=i_blk, col=atom_P, &
877 0 : block=Z_blk(1:current_chunk_size, 1:n_loc_ri))
878 : END IF
879 : END DO
880 :
881 0 : DEALLOCATE (Z_blk)
882 : END IF
883 :
884 0 : DEALLOCATE (d_vec_local, d_lp_local)
885 0 : DEALLOCATE (local_grid_idx, phi_local)
886 :
887 : END DO
888 :
889 0 : DEALLOCATE (cutoff_ri_per_atom)
890 0 : CALL gw_3c_ctx_release(ctx_3c)
891 :
892 0 : CALL dbcsr_finalize(mat_Z_lP)
893 :
894 0 : IF (bs_env%unit_nr > 0) THEN
895 : WRITE (bs_env%unit_nr, '(T2,A,T57,A,F7.1,A)') &
896 0 : 'Computed Z_lP ', ' Execution time', m_walltime() - t1, ' s'
897 0 : WRITE (bs_env%unit_nr, '(A)') ' '
898 : END IF
899 :
900 0 : logger => cp_get_default_logger()
901 :
902 0 : IF (BTEST(cp_print_key_should_output(logger%iter_info, input, key), cp_p_file)) THEN
903 0 : CALL dbcsr_binary_write(matrix=mat_Z_lP, filepath=TRIM(bs_env%prefix)//"Z_lP.matrix")
904 : END IF
905 :
906 : END IF
907 :
908 0 : DEALLOCATE (row_offset, ri_blk_sizes, col_dist_ri)
909 0 : CALL dbcsr_distribution_release(dist_Z)
910 :
911 0 : IF (n_procs_per_atom > 1) THEN
912 0 : CALL cp_blacs_env_release(blacs_env_sub)
913 0 : CALL para_env_sub%free()
914 0 : DEALLOCATE (para_env_sub)
915 : END IF
916 :
917 0 : DEALLOCATE (ri_rs_grid_points)
918 :
919 0 : CALL timestop(handle)
920 :
921 0 : END SUBROUTINE compute_coeff_Z_lP
922 :
923 : ! **************************************************************************************************
924 : !> \brief Computes the dense localized RHS d_lp(l,P) = Σ_{μν,R,S} Φ_μ(r_l)·Φ_ν(r_l)·(μν|P) for one
925 : !> RI atom P. OMP-threaded over (atom_j, atom_k) AO-pair blocks: per thread, sweep all
926 : !> (cell_R, cell_S) periodic images of (atom_j, atom_k) about atom_P at cell (0,0,0); each
927 : !> 3c block is built by build_3c_integral_block_ctx (cached libint / sphi tables in ctx,
928 : !> kind-radius triangle screen → `screened` short-circuits negligible image triples), and
929 : !> grid-chunked pair densities are contracted into a private d_lp partial that is reduced
930 : !> into d_lp at the end of the parallel region.
931 : !> \param bs_env ...
932 : !> \param ctx shared 3c-integral context (gw_3c_ctx_create)
933 : !> \param cell ...
934 : !> \param phi_val Φ_μ(r_l) on the local-sphere grid (n_grid_total × n_ao)
935 : !> \param d_lp output (n_grid_total × n_loc_ri), zeroed by the caller, accumulated here
936 : !> \param n_grid_total number of local-sphere grid rows
937 : !> \param n_loc_ri number of RI functions of atom_P
938 : !> \param atom_P RI atom (pinned to cell (0,0,0))
939 : !> \param max_ao_size ...
940 : !> \param atom_j_mepos ...
941 : !> \param atom_j_stride ...
942 : ! **************************************************************************************************
943 :
944 0 : SUBROUTINE compute_d_lp(bs_env, ctx, cell, phi_val, d_lp, n_grid_total, n_loc_ri, atom_P, &
945 : max_ao_size, atom_j_mepos, atom_j_stride)
946 :
947 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
948 : TYPE(gw_3c_ctx_type), INTENT(IN) :: ctx
949 : TYPE(cell_type), POINTER :: cell
950 : REAL(KIND=dp), DIMENSION(:, :), INTENT(IN) :: phi_val
951 : INTEGER, INTENT(IN) :: n_grid_total, n_loc_ri
952 : REAL(KIND=dp), INTENT(INOUT) :: d_lp(n_grid_total, n_loc_ri)
953 : INTEGER, INTENT(IN) :: atom_P, max_ao_size, atom_j_mepos, &
954 : atom_j_stride
955 :
956 : CHARACTER(LEN=*), PARAMETER :: routineN = 'compute_d_lp'
957 : INTEGER, PARAMETER :: grid_chunk = 1024
958 :
959 : INTEGER :: atom_j, atom_k, c, handle, handle_dgemm, ix_max, ix_min, ix_R, ix_S, iy_max, &
960 : iy_min, iy_R, iy_S, iz_max, iz_min, iz_R, iz_S, j, jk_idx, jsize, jstart, k, ksize, &
961 : kstart, l, l0, natom, ri
962 : INTEGER, DIMENSION(3) :: cell_R_vec, cell_S_vec
963 : LOGICAL :: any_kept, screened
964 0 : REAL(KIND=dp), ALLOCATABLE, DIMENSION(:, :) :: d_lp_prv, int_2d_prv, rho_chunk
965 0 : REAL(KIND=dp), ALLOCATABLE, DIMENSION(:, :, :) :: int_3c_prv, int_3c_sum
966 0 : TYPE(gw_3c_ws_type) :: ws
967 :
968 0 : CALL timeset(routineN, handle)
969 :
970 0 : natom = SIZE(bs_env%i_ao_start_from_atom)
971 :
972 0 : IF (cell%perd(1) == 1) THEN; ix_min = -1; ix_max = 1; ELSE; ix_min = 0; ix_max = 0
973 : END IF
974 0 : IF (cell%perd(2) == 1) THEN; iy_min = -1; iy_max = 1; ELSE; iy_min = 0; iy_max = 0
975 : END IF
976 0 : IF (cell%perd(3) == 1) THEN; iz_min = -1; iz_max = 1; ELSE; iz_min = 0; iz_max = 0
977 : END IF
978 :
979 : !$OMP PARALLEL DEFAULT(NONE) &
980 : !$OMP SHARED(bs_env, ctx, phi_val, d_lp, n_grid_total, n_loc_ri, atom_P, max_ao_size, &
981 : !$OMP natom, ix_min, ix_max, iy_min, iy_max, iz_min, iz_max, &
982 : !$OMP atom_j_mepos, atom_j_stride) &
983 : !$OMP PRIVATE(any_kept, atom_j, atom_k, c, handle_dgemm, j, jk_idx, jsize, jstart, k, &
984 : !$OMP ksize, kstart, l, l0, ri, ix_R, iy_R, iz_R, ix_S, iy_S, iz_S, cell_R_vec, &
985 0 : !$OMP cell_S_vec, screened, d_lp_prv, int_2d_prv, rho_chunk, int_3c_prv, int_3c_sum, ws)
986 :
987 : CALL gw_3c_ws_create(ws, ctx)
988 : ALLOCATE (int_3c_prv(max_ao_size, max_ao_size, n_loc_ri))
989 : ALLOCATE (int_3c_sum(max_ao_size, max_ao_size, n_loc_ri))
990 : ALLOCATE (int_2d_prv(max_ao_size*max_ao_size, n_loc_ri))
991 : ALLOCATE (rho_chunk(grid_chunk, max_ao_size*max_ao_size))
992 : ALLOCATE (d_lp_prv(n_grid_total, n_loc_ri))
993 : d_lp_prv(:, :) = 0.0_dp
994 :
995 : ! atom_P pinned at cell (0,0,0); enumerate (atom_j, cell_R) × (atom_k, cell_S). The ctx
996 : ! integral builder's kind_radius triangle screen sets screened=.TRUE. for the bulk of
997 : ! image triples (one or both AO atoms beyond the truncated-Coulomb reach of atom_P),
998 : ! so the 27 × 27 = 729 candidate cells collapse to "adjacent cells" in practice.
999 : ! MPI-stride atom_j over the subgroup (atom_j_stride = 1 for the BLAS path, > 1 for the
1000 : ! ScaLAPACK path). COLLAPSE(2) dropped because the outer stride is non-unit under
1001 : ! ScaLAPACK; the inner atom_k loop carries enough work for DYNAMIC.
1002 : !$OMP DO SCHEDULE(DYNAMIC)
1003 : DO atom_j = atom_j_mepos + 1, natom, atom_j_stride
1004 : DO atom_k = 1, natom
1005 : jstart = bs_env%i_ao_start_from_atom(atom_j)
1006 : jsize = bs_env%i_ao_end_from_atom(atom_j) - jstart + 1
1007 : kstart = bs_env%i_ao_start_from_atom(atom_k)
1008 : ksize = bs_env%i_ao_end_from_atom(atom_k) - kstart + 1
1009 :
1010 : int_3c_sum(1:jsize, 1:ksize, 1:n_loc_ri) = 0.0_dp
1011 : any_kept = .FALSE.
1012 :
1013 : DO ix_R = ix_min, ix_max
1014 : DO iy_R = iy_min, iy_max
1015 : DO iz_R = iz_min, iz_max
1016 : cell_R_vec = [ix_R, iy_R, iz_R]
1017 : DO ix_S = ix_min, ix_max
1018 : DO iy_S = iy_min, iy_max
1019 : DO iz_S = iz_min, iz_max
1020 : cell_S_vec = [ix_S, iy_S, iz_S]
1021 :
1022 : int_3c_prv(1:jsize, 1:ksize, 1:n_loc_ri) = 0.0_dp
1023 :
1024 : CALL build_3c_integral_block_ctx( &
1025 : int_3c_prv(1:jsize, 1:ksize, 1:n_loc_ri), ctx, ws, &
1026 : atom_j=atom_j, atom_k=atom_k, atom_i=atom_P, &
1027 : cell_j=cell_R_vec, cell_k=cell_S_vec, cell_i=[0, 0, 0], &
1028 : screened=screened)
1029 : IF (screened) CYCLE
1030 :
1031 : any_kept = .TRUE.
1032 : int_3c_sum(1:jsize, 1:ksize, 1:n_loc_ri) = &
1033 : int_3c_sum(1:jsize, 1:ksize, 1:n_loc_ri) + &
1034 : int_3c_prv(1:jsize, 1:ksize, 1:n_loc_ri)
1035 : END DO
1036 : END DO
1037 : END DO
1038 : END DO
1039 : END DO
1040 : END DO
1041 :
1042 : IF (.NOT. any_kept) CYCLE
1043 :
1044 : ! Flatten 3D B_{μν,P} → 2D B_{(μν),P}
1045 : DO ri = 1, n_loc_ri
1046 : DO k = 1, ksize
1047 : DO j = 1, jsize
1048 : jk_idx = (k - 1)*jsize + j
1049 : int_2d_prv(jk_idx, ri) = int_3c_sum(j, k, ri)
1050 : END DO
1051 : END DO
1052 : END DO
1053 :
1054 : ! Pair density ρ(l,μν) = Φ_μ(r_l)Φ_ν(r_l) in grid chunks, contracted on the fly:
1055 : ! d_{l,P} += ρ(l,μν) B_{(μν),P} (dgemm runs serially inside the parallel region)
1056 : DO l0 = 1, n_grid_total, grid_chunk
1057 : c = MIN(grid_chunk, n_grid_total - l0 + 1)
1058 : DO k = 1, ksize
1059 : DO j = 1, jsize
1060 : jk_idx = (k - 1)*jsize + j
1061 : DO l = 1, c
1062 : rho_chunk(l, jk_idx) = phi_val(l0 + l - 1, jstart + j - 1)* &
1063 : phi_val(l0 + l - 1, kstart + k - 1)
1064 : END DO
1065 : END DO
1066 : END DO
1067 : CALL timeset(routineN//"_dgemm", handle_dgemm)
1068 : CALL dgemm("N", "N", c, n_loc_ri, jsize*ksize, &
1069 : 1.0_dp, rho_chunk, grid_chunk, &
1070 : int_2d_prv, max_ao_size*max_ao_size, &
1071 : 1.0_dp, d_lp_prv(l0, 1), n_grid_total)
1072 : CALL timestop(handle_dgemm)
1073 : END DO
1074 : END DO
1075 : END DO
1076 : !$OMP END DO
1077 :
1078 : !$OMP CRITICAL (compute_d_lp_reduce)
1079 : d_lp(1:n_grid_total, 1:n_loc_ri) = d_lp(1:n_grid_total, 1:n_loc_ri) + &
1080 : d_lp_prv(1:n_grid_total, 1:n_loc_ri)
1081 : !$OMP END CRITICAL (compute_d_lp_reduce)
1082 :
1083 : DEALLOCATE (int_3c_prv, int_3c_sum, int_2d_prv, rho_chunk, d_lp_prv)
1084 : CALL gw_3c_ws_release(ws)
1085 :
1086 : !$OMP END PARALLEL
1087 :
1088 0 : CALL timestop(handle)
1089 :
1090 0 : END SUBROUTINE compute_d_lp
1091 :
1092 : ! **************************************************************************************************
1093 : !> \brief Computes the χ(iτ, k=0) matrix
1094 : !> \param bs_env ...
1095 : !> \param mat_chi_Gamma_tau ...
1096 : !> \param mat_phi_mu_l ...
1097 : !> \param mat_Z_lP ...
1098 : ! **************************************************************************************************
1099 :
1100 0 : SUBROUTINE get_mat_chi_Gamma_tau(bs_env, mat_chi_Gamma_tau, mat_phi_mu_l, mat_Z_lP)
1101 :
1102 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
1103 : TYPE(dbcsr_p_type), DIMENSION(:), POINTER :: mat_chi_Gamma_tau
1104 : TYPE(dbcsr_type), INTENT(INOUT) :: mat_phi_mu_l, mat_Z_lP
1105 :
1106 : CHARACTER(LEN=*), PARAMETER :: routineN = 'get_mat_chi_Gamma_tau'
1107 :
1108 : INTEGER :: handle, i, i_t, ispin, npcol
1109 0 : INTEGER, DIMENSION(:), POINTER :: blk_ao, blk_grid, dist_col_grid, &
1110 0 : dist_row_grid
1111 : REAL(KIND=dp) :: t1, tau
1112 : TYPE(dbcsr_distribution_type) :: dist_grid_grid, dist_phi
1113 : TYPE(dbcsr_type) :: matrix_chi_grid, matrix_chi_grid_spin, &
1114 : matrix_G_occ_grid, matrix_G_vir_grid
1115 :
1116 0 : CALL timeset(routineN, handle)
1117 :
1118 : ! =========================================================================
1119 : ! 1. SETUP CORE TOPOLOGIES
1120 : ! =========================================================================
1121 0 : CALL dbcsr_get_info(mat_phi_mu_l, distribution=dist_phi, row_blk_size=blk_grid, col_blk_size=blk_ao)
1122 0 : CALL dbcsr_distribution_get(dist_phi, row_dist=dist_row_grid, npcols=npcol)
1123 :
1124 : ! Build a perfectly safe column distribution for the Grid dimension
1125 0 : ALLOCATE (dist_col_grid(SIZE(blk_grid)))
1126 0 : DO i = 1, SIZE(blk_grid)
1127 0 : dist_col_grid(i) = MOD(i - 1, npcol)
1128 : END DO
1129 :
1130 : CALL dbcsr_distribution_new(dist_grid_grid, template=dist_phi, &
1131 0 : row_dist=dist_row_grid, col_dist=dist_col_grid)
1132 :
1133 0 : CALL dbcsr_create(matrix_G_occ_grid, "G_occ_grid", dist_grid_grid, dbcsr_type_no_symmetry, blk_grid, blk_grid)
1134 0 : CALL dbcsr_create(matrix_G_vir_grid, "G_vir_grid", dist_grid_grid, dbcsr_type_no_symmetry, blk_grid, blk_grid)
1135 0 : CALL dbcsr_create(matrix_chi_grid, "chi_grid", dist_grid_grid, dbcsr_type_no_symmetry, blk_grid, blk_grid)
1136 0 : CALL dbcsr_create(matrix_chi_grid_spin, "chi_grid_spin", dist_grid_grid, dbcsr_type_no_symmetry, blk_grid, blk_grid)
1137 :
1138 : ! =========================================================================
1139 : ! 2. MAIN IMAGINARY TIME LOOP
1140 : ! =========================================================================
1141 0 : DO i_t = 1, bs_env%num_time_freq_points
1142 0 : t1 = m_walltime()
1143 :
1144 0 : tau = bs_env%imag_time_points(i_t)
1145 0 : CALL dbcsr_set(matrix_chi_grid, 0.0_dp)
1146 :
1147 : ! ----------------------------------------------------------------------
1148 : ! A. SPIN LOOP (Allocations safely encapsulated in wrappers)
1149 : ! ----------------------------------------------------------------------
1150 0 : DO ispin = 1, bs_env%n_spin
1151 :
1152 : ! G^occ_µλ(i|τ|,k=0) = sum_n^occ C_µn(k=0) e^(-|(ϵ_nk=0-ϵ_F)τ|) C_λn(k=0)
1153 : ! G^occ_ll'(i|τ|,k=0) = sum_µν Φ_µ(r_l) G^occ_µν Φ_ν(r_l')
1154 : CALL build_G_grid(bs_env, tau, ispin, .TRUE., .FALSE., mat_phi_mu_l, &
1155 0 : matrix_G_occ_grid, bs_env%eps_filter)
1156 :
1157 : ! G^vir_µλ(i|τ|,k=0) = sum_n^vir C_µn(k=0) e^(-|(ϵ_nk=0-ϵ_F)τ|) C_λn(k=0)
1158 : ! G^vir_ll'(i|τ|,k=0) = sum_µν Φ_µ(r_l) G^vir_µν Φ_ν(r_l')
1159 : CALL build_G_grid(bs_env, tau, ispin, .FALSE., .TRUE., mat_phi_mu_l, &
1160 0 : matrix_G_vir_grid, bs_env%eps_filter)
1161 :
1162 : ! -------------------------------------------------------------------
1163 : ! B. ELEMENT-WISE HADAMARD PRODUCT
1164 : ! -------------------------------------------------------------------
1165 : ! χ_ll'(iτ,k=0) = G^occ_ll'(i|τ|,k=0) * G^vir_ll'(i|τ|,k=0)
1166 0 : CALL hadamard_product(matrix_G_occ_grid, matrix_G_vir_grid, matrix_chi_grid_spin, bs_env%spin_degeneracy)
1167 :
1168 : ! Accumulate spin contributions
1169 0 : CALL dbcsr_add(matrix_chi_grid, matrix_chi_grid_spin, 1.0_dp, 1.0_dp)
1170 :
1171 : END DO ! ispin
1172 :
1173 : ! ----------------------------------------------------------------------
1174 : ! C. TRANSFORM TO AUXILIARY BASIS & EXPORT DIRECTLY
1175 : ! χ_aux = Z^T * χ_grid * Z
1176 : ! χ_PQ(iτ,k=0) = sum_ll' Z_lP χ_ll'(iτ,k=0) Z_l'Q
1177 : ! Result is dumped directly into the final array mat_chi_Gamma_tau!
1178 : ! ----------------------------------------------------------------------
1179 : CALL contract_A_B_A("T", "N", mat_Z_lP, matrix_chi_grid, &
1180 0 : mat_chi_Gamma_tau(i_t)%matrix, bs_env%eps_filter)
1181 :
1182 0 : IF (bs_env%unit_nr > 0) THEN
1183 : WRITE (bs_env%unit_nr, '(T2,A,I13,A,I3,A,F7.1,A)') &
1184 0 : 'Computed χ(iτ,k=0) for time point', i_t, ' /', bs_env%num_time_freq_points, &
1185 0 : ', Execution time', m_walltime() - t1, ' s'
1186 : END IF
1187 :
1188 : END DO ! i_t
1189 :
1190 : ! =========================================================================
1191 : ! 3. FINAL CLEANUP
1192 : ! =========================================================================
1193 0 : CALL dbcsr_release(matrix_G_occ_grid)
1194 0 : CALL dbcsr_release(matrix_G_vir_grid)
1195 0 : CALL dbcsr_release(matrix_chi_grid)
1196 0 : CALL dbcsr_release(matrix_chi_grid_spin)
1197 0 : CALL dbcsr_distribution_release(dist_grid_grid)
1198 0 : DEALLOCATE (dist_col_grid)
1199 :
1200 0 : IF (bs_env%unit_nr > 0) WRITE (bs_env%unit_nr, '(A)') ' '
1201 :
1202 0 : CALL timestop(handle)
1203 :
1204 0 : END SUBROUTINE get_mat_chi_Gamma_tau
1205 :
1206 : ! **************************************************************************************************
1207 : !> \brief Computes Green's Function in grid basis
1208 : !> \param bs_env ...
1209 : !> \param tau ...
1210 : !> \param ispin ...
1211 : !> \param occ ...
1212 : !> \param vir ...
1213 : !> \param mat_phi_mu_l ...
1214 : !> \param matrix_G_grid ...
1215 : !> \param eps_filter ...
1216 : ! **************************************************************************************************
1217 :
1218 0 : SUBROUTINE build_G_grid(bs_env, tau, ispin, occ, vir, mat_phi_mu_l, matrix_G_grid, eps_filter)
1219 :
1220 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
1221 : REAL(KIND=dp), INTENT(IN) :: tau
1222 : INTEGER, INTENT(IN) :: ispin
1223 : LOGICAL, INTENT(IN) :: occ, vir
1224 : TYPE(dbcsr_type), INTENT(INOUT) :: mat_phi_mu_l, matrix_G_grid
1225 : REAL(KIND=dp), INTENT(IN) :: eps_filter
1226 :
1227 : CHARACTER(LEN=*), PARAMETER :: routineN = 'build_G_grid'
1228 :
1229 : INTEGER :: handle
1230 0 : INTEGER, DIMENSION(:), POINTER :: blk_ao, dist_row_ao
1231 : TYPE(cp_fm_type), POINTER :: fm_G
1232 : TYPE(dbcsr_distribution_type) :: dist_ao_ao
1233 : TYPE(dbcsr_type) :: matrix_G_ao
1234 :
1235 0 : CALL timeset(routineN, handle)
1236 :
1237 : ! 1. Select the correct FM matrix based on occ/vir flags
1238 0 : IF (occ) THEN
1239 0 : fm_G => bs_env%fm_Gocc
1240 : ELSE
1241 0 : fm_G => bs_env%fm_Gvir
1242 : END IF
1243 :
1244 : ! 2. Compute Dense FM Green's Function
1245 : ! G^occ/vir_µλ(i|τ|,k=0) = sum_G^occ/vir_µλn^occ/vir C_µn(k=0) e^(-|(ϵ_nk=0-ϵ_F)τ|) C_λn(k=0)
1246 0 : CALL G_occ_vir(bs_env, tau, fm_G, ispin, occ=occ, vir=vir)
1247 :
1248 : ! 3. Setup AO DBCSR Topology and Create Matrix dynamically
1249 0 : CALL setup_square_topology(mat_phi_mu_l, 'COL', dist_ao_ao, blk_ao, dist_row_ao)
1250 :
1251 : CALL dbcsr_create(matrix_G_ao, name="G_ao", dist=dist_ao_ao, &
1252 : matrix_type=dbcsr_type_no_symmetry, &
1253 0 : row_blk_size=blk_ao, col_blk_size=blk_ao)
1254 :
1255 : ! 4. Convert FM to Sparse DBCSR
1256 0 : CALL copy_fm_to_dbcsr(fm_G, matrix_G_ao, keep_sparsity=.FALSE.)
1257 :
1258 : ! 5. Transform to Grid Basis: G_grid = phi * G_ao * phi^T
1259 : ! G^occ/vir_ll'(i|τ|,k=0) = sum_µν Φ_µ(r_l) G^occ/vir_µν Φ_ν(r_l')
1260 0 : CALL contract_A_B_A("N", "T", mat_phi_mu_l, matrix_G_ao, matrix_G_grid, eps_filter)
1261 :
1262 : ! 6. Release AO matrix and topology
1263 0 : CALL release_dbcsr_topology_and_matrices(dist=dist_ao_ao, mapped_dist=dist_row_ao, m1=matrix_G_ao)
1264 :
1265 0 : CALL timestop(handle)
1266 :
1267 0 : END SUBROUTINE build_G_grid
1268 :
1269 : ! **************************************************************************************************
1270 : !> \brief Generalized routine to compute OUT = A * B * A^T OR OUT = A^T * B * A using DBCSR
1271 : !> \param transA_left ...
1272 : !> \param transA_right ...
1273 : !> \param matrix_A ...
1274 : !> \param matrix_B ...
1275 : !> \param matrix_out ...
1276 : !> \param eps_filter ...
1277 : ! **************************************************************************************************
1278 :
1279 0 : SUBROUTINE contract_A_B_A(transA_left, transA_right, matrix_A, matrix_B, matrix_out, eps_filter)
1280 :
1281 : CHARACTER(LEN=1), INTENT(IN) :: transA_left, transA_right
1282 : TYPE(dbcsr_type), INTENT(INOUT) :: matrix_A, matrix_B, matrix_out
1283 : REAL(KIND=dp), INTENT(IN) :: eps_filter
1284 :
1285 : CHARACTER(LEN=*), PARAMETER :: routineN = 'contract_A_B_A'
1286 :
1287 : INTEGER :: handle
1288 : TYPE(dbcsr_type) :: matrix_tmp
1289 :
1290 0 : CALL timeset(routineN, handle)
1291 :
1292 0 : CALL dbcsr_create(matrix_tmp, template=matrix_A)
1293 :
1294 0 : IF (transA_left == "N" .AND. transA_right == "T") THEN
1295 : ! Path 1: Out = A * B * A^T
1296 : CALL dbcsr_multiply("N", "N", 1.0_dp, matrix_A, matrix_B, &
1297 0 : 0.0_dp, matrix_tmp, filter_eps=eps_filter)
1298 : CALL dbcsr_multiply("N", "T", 1.0_dp, matrix_tmp, matrix_A, &
1299 0 : 0.0_dp, matrix_out, filter_eps=eps_filter)
1300 :
1301 0 : ELSE IF (transA_left == "T" .AND. transA_right == "N") THEN
1302 : ! Path 2: Out = A^T * B * A
1303 : CALL dbcsr_multiply("N", "N", 1.0_dp, matrix_B, matrix_A, &
1304 0 : 0.0_dp, matrix_tmp, filter_eps=eps_filter)
1305 : CALL dbcsr_multiply("T", "N", 1.0_dp, matrix_A, matrix_tmp, &
1306 0 : 0.0_dp, matrix_out, filter_eps=eps_filter)
1307 : ELSE
1308 0 : CPABORT("Unsupported transposition pair in contract_A_B_A")
1309 : END IF
1310 :
1311 0 : CALL dbcsr_release(matrix_tmp)
1312 :
1313 0 : CALL timestop(handle)
1314 :
1315 0 : END SUBROUTINE contract_A_B_A
1316 :
1317 : ! **************************************************************************************************
1318 : !> \brief Computes C = A ◦ B (Element-wise Hadamard product) for sparse DBCSR matrices.
1319 : !> \param matrix_A ...
1320 : !> \param matrix_B ...
1321 : !> \param matrix_C ...
1322 : !> \param fac (Scaling factor applied to the product)
1323 : ! **************************************************************************************************
1324 :
1325 0 : SUBROUTINE hadamard_product(matrix_A, matrix_B, matrix_C, fac)
1326 :
1327 : TYPE(dbcsr_type), INTENT(INOUT) :: matrix_A, matrix_B, matrix_C
1328 : REAL(KIND=dp), INTENT(IN) :: fac
1329 :
1330 : CHARACTER(LEN=*), PARAMETER :: routineN = 'hadamard_product'
1331 :
1332 : INTEGER :: col, handle, row
1333 : LOGICAL :: found
1334 0 : REAL(KIND=dp), DIMENSION(:, :), POINTER :: blk_B, blk_C
1335 : TYPE(dbcsr_iterator_type) :: iter
1336 :
1337 0 : CALL timeset(routineN, handle)
1338 :
1339 0 : CALL dbcsr_copy(matrix_C, matrix_A)
1340 :
1341 0 : CALL dbcsr_iterator_start(iter, matrix_C)
1342 0 : DO WHILE (dbcsr_iterator_blocks_left(iter))
1343 0 : CALL dbcsr_iterator_next_block(iter, row, col, blk_C)
1344 :
1345 0 : CALL dbcsr_get_block_p(matrix_B, row, col, blk_B, found)
1346 :
1347 0 : IF (found) THEN
1348 0 : blk_C(:, :) = fac*blk_C(:, :)*blk_B(:, :)
1349 : ELSE
1350 : ! If B is sparse here, the product is zero
1351 0 : blk_C(:, :) = 0.0_dp
1352 : END IF
1353 : END DO
1354 0 : CALL dbcsr_iterator_stop(iter)
1355 :
1356 0 : CALL timestop(handle)
1357 :
1358 0 : END SUBROUTINE hadamard_product
1359 :
1360 : ! **************************************************************************************************
1361 : !> \brief Compute screened Coulomb interaction matrix
1362 : !> \param bs_env ...
1363 : !> \param qs_env ...
1364 : !> \param mat_chi_Gamma_tau ...
1365 : !> \param fm_W_time ...
1366 : ! **************************************************************************************************
1367 :
1368 0 : SUBROUTINE compute_W(bs_env, qs_env, mat_chi_Gamma_tau, fm_W_time)
1369 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
1370 : TYPE(qs_environment_type), POINTER :: qs_env
1371 : TYPE(dbcsr_p_type), DIMENSION(:), POINTER :: mat_chi_Gamma_tau
1372 : TYPE(cp_fm_type), ALLOCATABLE, DIMENSION(:) :: fm_W_time
1373 :
1374 : CHARACTER(LEN=*), PARAMETER :: routineN = 'compute_W'
1375 :
1376 : INTEGER :: handle, i_t, j_w
1377 : REAL(KIND=dp) :: t1
1378 : TYPE(cp_fm_type) :: fm_M_inv_V_sqrt, fm_V, fm_V_sqrt
1379 :
1380 0 : CALL timeset(routineN, handle)
1381 :
1382 0 : t1 = m_walltime()
1383 :
1384 0 : CALL create_fm_W_MIC_time(bs_env, fm_W_time)
1385 :
1386 : ! 1. Allocate V and M matrices
1387 0 : CALL cp_fm_create(fm_V, bs_env%fm_RI_RI%matrix_struct)
1388 0 : CALL cp_fm_create(fm_V_sqrt, bs_env%fm_RI_RI%matrix_struct)
1389 0 : CALL cp_fm_create(fm_M_inv_V_sqrt, bs_env%fm_RI_RI%matrix_struct)
1390 :
1391 : ! Compute V and M^-1 * V^0.5
1392 0 : CALL compute_V_MinvVsqrt(bs_env, qs_env, fm_V, fm_V_sqrt, fm_M_inv_V_sqrt)
1393 :
1394 : ! 2. Loop over frequencies
1395 0 : DO j_w = 1, bs_env%num_time_freq_points
1396 : ! Fourier transformation of χ_PQ(iτ) to χ_PQ(iω_j)
1397 0 : CALL compute_fm_chi_Gamma_freq(bs_env, bs_env%fm_chi_Gamma_freq, j_w, mat_chi_Gamma_tau)
1398 :
1399 : ! ε(iω_j) = Id - V^0.5*M^-1*χ(iω_j)*M^-1*V^0.5
1400 : ! W(iω_j) = V^0.5*(ε^-1(iω_j)-Id)*V^0.5
1401 : CALL compute_fm_W_freq(bs_env, bs_env%fm_chi_Gamma_freq, fm_V_sqrt, &
1402 0 : fm_M_inv_V_sqrt, bs_env%fm_W_MIC_freq)
1403 :
1404 : ! Fourier transform from W_PQ^MIC(iω_j) to W_PQ^MIC(iτ)
1405 0 : CALL Fourier_transform_w_to_t(bs_env, fm_W_time, bs_env%fm_W_MIC_freq, j_w)
1406 : END DO
1407 :
1408 : ! M^-1*W^MIC(iτ)*M^-1
1409 0 : CALL multiply_fm_W_MIC_time_with_Minv_Gamma(bs_env, qs_env, fm_W_time)
1410 :
1411 0 : IF (bs_env%unit_nr > 0) THEN
1412 : WRITE (bs_env%unit_nr, '(T2,A,T55,A,F10.1,A)') &
1413 0 : 'Computed W(iτ),', ' Execution time', m_walltime() - t1, ' s'
1414 : END IF
1415 :
1416 0 : CALL dbcsr_deallocate_matrix_set(mat_chi_Gamma_tau)
1417 :
1418 : ! Cleanup
1419 0 : CALL cp_fm_release(fm_V)
1420 0 : CALL cp_fm_release(fm_V_sqrt)
1421 0 : CALL cp_fm_release(fm_M_inv_V_sqrt)
1422 :
1423 : ! Marek : Fourier transform W^MIC(itau) back to get it at a specific im.frequency point - iomega = 0
1424 0 : IF (bs_env%rtp_method == rtp_method_bse) THEN
1425 0 : t1 = m_walltime()
1426 0 : CALL cp_fm_create(bs_env%fm_W_MIC_freq_zero, bs_env%fm_W_MIC_freq%matrix_struct)
1427 : ! Set to zero
1428 0 : CALL cp_fm_set_all(bs_env%fm_W_MIC_freq_zero, 0.0_dp)
1429 : ! Sum over all times
1430 0 : DO i_t = 1, bs_env%num_time_freq_points
1431 : ! Add the relevant structure with correct weight
1432 : CALL cp_fm_scale_and_add(1.0_dp, bs_env%fm_W_MIC_freq_zero, &
1433 0 : bs_env%imag_time_weights_freq_zero(i_t), fm_W_time(i_t))
1434 : END DO
1435 : ! Done, save to file
1436 0 : CALL fm_write(bs_env%fm_W_MIC_freq_zero, 0, "W_freq_rtp", qs_env)
1437 : ! Report calculation
1438 0 : IF (bs_env%unit_nr > 0) THEN
1439 : WRITE (bs_env%unit_nr, '(T2,A,T55,A,F10.1,A)') &
1440 0 : 'Computed W(0),', ' Execution time', m_walltime() - t1, ' s'
1441 : END IF
1442 : END IF
1443 :
1444 0 : IF (bs_env%unit_nr > 0) WRITE (bs_env%unit_nr, '(A)') ' '
1445 :
1446 0 : CALL timestop(handle)
1447 :
1448 0 : END SUBROUTINE compute_W
1449 :
1450 : ! **************************************************************************************************
1451 : !> \brief Computes V, V^0.5, and M^-1 * V^0.5
1452 : !> \param bs_env ...
1453 : !> \param qs_env ...
1454 : !> \param fm_V ...
1455 : !> \param fm_V_sqrt ...
1456 : !> \param fm_Minv_Vsqrt ...
1457 : ! **************************************************************************************************
1458 :
1459 0 : SUBROUTINE compute_V_MinvVsqrt(bs_env, qs_env, fm_V, fm_V_sqrt, fm_Minv_Vsqrt)
1460 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
1461 : TYPE(qs_environment_type), POINTER :: qs_env
1462 : TYPE(cp_fm_type), INTENT(INOUT) :: fm_V, fm_V_sqrt, fm_Minv_Vsqrt
1463 :
1464 : CHARACTER(LEN=*), PARAMETER :: routineN = 'compute_V_MinvVsqrt'
1465 :
1466 : INTEGER :: handle, info, n_RI, ndep
1467 0 : TYPE(atomic_kind_type), DIMENSION(:), POINTER :: atomic_kind_set
1468 : TYPE(cell_type), POINTER :: cell
1469 : TYPE(cp_fm_type) :: fm_work
1470 0 : TYPE(cp_fm_type), ALLOCATABLE, DIMENSION(:, :) :: fm_M
1471 0 : TYPE(dbcsr_p_type), DIMENSION(:, :), POINTER :: mat_V_kp
1472 0 : TYPE(particle_type), DIMENSION(:), POINTER :: particle_set
1473 0 : TYPE(qs_kind_type), DIMENSION(:), POINTER :: qs_kind_set
1474 :
1475 0 : CALL timeset(routineN, handle)
1476 :
1477 0 : n_RI = bs_env%n_RI
1478 0 : CALL cp_fm_create(fm_work, fm_V%matrix_struct)
1479 :
1480 : ! -----------------------------------------------------------------------
1481 : ! 1. Build Coulomb Matrix V(k=0) using the kp-routine but only for ikp=1
1482 : ! -----------------------------------------------------------------------
1483 : CALL get_qs_env(qs_env=qs_env, particle_set=particle_set, cell=cell, &
1484 0 : qs_kind_set=qs_kind_set, atomic_kind_set=atomic_kind_set)
1485 :
1486 0 : ALLOCATE (mat_V_kp(1:1, 1:2))
1487 0 : NULLIFY (mat_V_kp(1, 1)%matrix, mat_V_kp(1, 2)%matrix)
1488 0 : ALLOCATE (mat_V_kp(1, 1)%matrix, mat_V_kp(1, 2)%matrix)
1489 :
1490 0 : CALL dbcsr_create(mat_V_kp(1, 1)%matrix, template=bs_env%mat_RI_RI%matrix)
1491 0 : CALL dbcsr_reserve_all_blocks(mat_V_kp(1, 1)%matrix)
1492 0 : CALL dbcsr_set(mat_V_kp(1, 1)%matrix, 0.0_dp)
1493 :
1494 : ! Dummy imaginary part just to satisfy the routine
1495 0 : CALL dbcsr_create(mat_V_kp(1, 2)%matrix, template=bs_env%mat_RI_RI%matrix)
1496 0 : CALL dbcsr_reserve_all_blocks(mat_V_kp(1, 2)%matrix)
1497 0 : CALL dbcsr_set(mat_V_kp(1, 2)%matrix, 0.0_dp)
1498 :
1499 0 : bs_env%kpoints_chi_eps_W%nkp_grid = bs_env%nkp_grid_chi_eps_W_orig
1500 :
1501 : CALL build_2c_coulomb_matrix_kp(mat_V_kp, bs_env%kpoints_chi_eps_W, "RI_AUX", cell, &
1502 : particle_set, qs_kind_set, atomic_kind_set, &
1503 0 : bs_env%size_lattice_sum_V, operator_coulomb, 1, 1)
1504 :
1505 : ! Copy real part to fm_V
1506 0 : CALL copy_dbcsr_to_fm(mat_V_kp(1, 1)%matrix, fm_V)
1507 :
1508 0 : CALL dbcsr_deallocate_matrix(mat_V_kp(1, 1)%matrix)
1509 0 : CALL dbcsr_deallocate_matrix(mat_V_kp(1, 2)%matrix)
1510 0 : DEALLOCATE (mat_V_kp)
1511 :
1512 : ! -----------------------------------------------------------------------
1513 : ! 2. Get RI-Metric Matrix M(k=0)
1514 : ! -----------------------------------------------------------------------
1515 : CALL RI_2c_integral_mat(qs_env, fm_M, fm_V, n_RI, bs_env%ri_metric, &
1516 0 : do_kpoints=.FALSE., regularization_RI=bs_env%regularization_RI)
1517 :
1518 : ! -----------------------------------------------------------------------
1519 : ! 3. M -> M^-1
1520 : ! -----------------------------------------------------------------------
1521 0 : CALL cp_fm_cholesky_decompose(fm_M(1, 1), info_out=info)
1522 0 : IF (info == 0) THEN
1523 0 : CALL cp_fm_cholesky_invert(fm_M(1, 1))
1524 0 : CALL cp_fm_uplo_to_full(fm_M(1, 1), fm_work)
1525 : ELSE
1526 : ! Fallback if Cholesky fails due to conditioning
1527 0 : CALL cp_fm_power(fm_M(1, 1), fm_work, -1.0_dp, bs_env%eps_eigval_mat_RI, ndep)
1528 0 : CALL cp_fm_to_fm(fm_work, fm_M(1, 1))
1529 : END IF
1530 :
1531 : ! -----------------------------------------------------------------------
1532 : ! 4. V -> V^0.5
1533 : ! -----------------------------------------------------------------------
1534 0 : CALL cp_fm_to_fm(fm_V, fm_V_sqrt)
1535 0 : CALL cp_fm_cholesky_decompose(fm_V_sqrt, info_out=info)
1536 0 : IF (info == 0) THEN
1537 0 : CALL clean_lower_part(fm_V_sqrt)
1538 : ELSE
1539 0 : CALL cp_fm_power(fm_V, fm_V_sqrt, 0.5_dp, bs_env%eps_eigval_mat_RI, ndep)
1540 : END IF
1541 :
1542 : ! -----------------------------------------------------------------------
1543 : ! 5. M^-1 * V^0.5
1544 : ! -----------------------------------------------------------------------
1545 : CALL parallel_gemm("N", "T", n_RI, n_RI, n_RI, 1.0_dp, fm_M(1, 1), fm_V_sqrt, &
1546 0 : 0.0_dp, fm_Minv_Vsqrt)
1547 :
1548 0 : CALL cp_fm_release(fm_M)
1549 0 : CALL cp_fm_release(fm_work)
1550 :
1551 0 : CALL timestop(handle)
1552 :
1553 0 : END SUBROUTINE compute_V_MinvVsqrt
1554 :
1555 : ! **************************************************************************************************
1556 : !> \brief Computes W(iω) from χ_PQ(iω_j)
1557 : !> \param bs_env ...
1558 : !> \param fm_chi_freq_j ...
1559 : !> \param fm_V_sqrt ...
1560 : !> \param fm_Minv_Vsqrt ...
1561 : !> \param fm_W_freq_j ...
1562 : ! **************************************************************************************************
1563 :
1564 0 : SUBROUTINE compute_fm_W_freq(bs_env, fm_chi_freq_j, fm_V_sqrt, fm_Minv_Vsqrt, fm_W_freq_j)
1565 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
1566 : TYPE(cp_fm_type), INTENT(IN) :: fm_chi_freq_j, fm_V_sqrt, fm_Minv_Vsqrt
1567 : TYPE(cp_fm_type), INTENT(INOUT) :: fm_W_freq_j
1568 :
1569 : CHARACTER(LEN=*), PARAMETER :: routineN = 'compute_fm_W_freq'
1570 :
1571 : INTEGER :: handle, info, n_RI, ndep
1572 : TYPE(cp_fm_type) :: fm_eps_freq_j, fm_work
1573 :
1574 0 : CALL timeset(routineN, handle)
1575 :
1576 0 : n_RI = bs_env%n_RI
1577 :
1578 0 : CALL cp_fm_create(fm_eps_freq_j, fm_chi_freq_j%matrix_struct)
1579 0 : CALL cp_fm_create(fm_work, fm_chi_freq_j%matrix_struct)
1580 :
1581 : ! -----------------------------------------------------------------------
1582 : ! 1. ε(iω_j) = Id - (M^-1 * V^0.5)^T * χ(iω_j) * (M^-1 * V^0.5)
1583 : ! -----------------------------------------------------------------------
1584 : ! work = χ(iω_j) * (M^-1 * V^0.5)
1585 : CALL parallel_gemm('N', 'N', n_RI, n_RI, n_RI, 1.0_dp, &
1586 0 : fm_chi_freq_j, fm_Minv_Vsqrt, 0.0_dp, fm_work)
1587 :
1588 : ! eps_work = (M^-1 * V^0.5)^T * work
1589 : CALL parallel_gemm('T', 'N', n_RI, n_RI, n_RI, 1.0_dp, &
1590 0 : fm_Minv_Vsqrt, fm_work, 0.0_dp, fm_eps_freq_j)
1591 :
1592 : ! ε(iω_j) = Id - eps_work --> -eps_work + Id
1593 0 : CALL fm_add_on_diag(fm_eps_freq_j, 1.0_dp)
1594 :
1595 : ! Force perfect symmetry before Cholesky to avoid info != 0 due to GEMM noise
1596 0 : CALL cp_fm_uplo_to_full(fm_eps_freq_j, fm_work)
1597 :
1598 : ! -----------------------------------------------------------------------
1599 : ! 2. W(iω_j) = V^0.5^T * (ε^-1(iω_j) - Id) * V^0.5
1600 : ! -----------------------------------------------------------------------
1601 :
1602 : ! a) Cholesky decomposition of ε(iω_j)
1603 0 : CALL cp_fm_cholesky_decompose(fm_eps_freq_j, info_out=info)
1604 :
1605 : ! b) Inversion
1606 0 : IF (info == 0) THEN
1607 0 : CALL cp_fm_cholesky_invert(fm_eps_freq_j)
1608 0 : CALL cp_fm_uplo_to_full(fm_eps_freq_j, fm_work)
1609 : ELSE
1610 : ! Fallback to expensive diagonalization if Cholesky fails
1611 0 : CALL cp_fm_power(fm_eps_freq_j, fm_work, -1.0_dp, bs_env%eps_eigval_mat_RI, ndep)
1612 0 : CALL cp_fm_to_fm(fm_work, fm_eps_freq_j)
1613 : END IF
1614 :
1615 : ! c) ε^-1(iω_j) - Id
1616 0 : CALL fm_add_on_diag(fm_eps_freq_j, -1.0_dp)
1617 :
1618 : ! d) work = (ε^-1(iω_j) - Id) * V^0.5
1619 : CALL parallel_gemm('N', 'N', n_RI, n_RI, n_RI, 1.0_dp, fm_eps_freq_j, fm_V_sqrt, &
1620 0 : 0.0_dp, fm_work)
1621 :
1622 : ! e) W(iw) = V^0.5^T * work
1623 : CALL parallel_gemm('T', 'N', n_RI, n_RI, n_RI, 1.0_dp, fm_V_sqrt, fm_work, &
1624 0 : 0.0_dp, fm_W_freq_j)
1625 :
1626 : ! Cleanup
1627 0 : CALL cp_fm_release(fm_work)
1628 0 : CALL cp_fm_release(fm_eps_freq_j)
1629 :
1630 0 : CALL timestop(handle)
1631 :
1632 0 : END SUBROUTINE compute_fm_W_freq
1633 :
1634 : ! **************************************************************************************************
1635 : !> \brief Adds a real scalar value to the diagonal of a real full matrix (fm)
1636 : !> \param fm ...
1637 : !> \param alpha ...
1638 : ! **************************************************************************************************
1639 :
1640 0 : SUBROUTINE fm_add_on_diag(fm, alpha)
1641 : TYPE(cp_fm_type), INTENT(INOUT) :: fm
1642 : REAL(KIND=dp), INTENT(IN) :: alpha
1643 :
1644 : CHARACTER(LEN=*), PARAMETER :: routineN = 'fm_add_on_diag'
1645 :
1646 : INTEGER :: handle, i_global, i_row, j_col, &
1647 : j_global, ncol_local, nrow_local
1648 0 : INTEGER, DIMENSION(:), POINTER :: col_indices, row_indices
1649 :
1650 0 : CALL timeset(routineN, handle)
1651 :
1652 : CALL cp_fm_get_info(matrix=fm, &
1653 : nrow_local=nrow_local, &
1654 : ncol_local=ncol_local, &
1655 : row_indices=row_indices, &
1656 0 : col_indices=col_indices)
1657 :
1658 0 : DO j_col = 1, ncol_local
1659 0 : j_global = col_indices(j_col)
1660 0 : DO i_row = 1, nrow_local
1661 0 : i_global = row_indices(i_row)
1662 0 : IF (j_global == i_global) THEN
1663 0 : fm%local_data(i_row, j_col) = fm%local_data(i_row, j_col) + alpha
1664 : END IF
1665 : END DO
1666 : END DO
1667 :
1668 0 : CALL timestop(handle)
1669 :
1670 0 : END SUBROUTINE fm_add_on_diag
1671 :
1672 : ! **************************************************************************************************
1673 : !> \brief Zeroes out the strictly lower triangular part of a real matrix
1674 : !> \param fm_mat ...
1675 : ! **************************************************************************************************
1676 0 : SUBROUTINE clean_lower_part(fm_mat)
1677 : TYPE(cp_fm_type) :: fm_mat
1678 :
1679 : CHARACTER(LEN=*), PARAMETER :: routineN = 'clean_lower_part'
1680 :
1681 : INTEGER :: handle, i_row, j_col, j_global, &
1682 : ncol_local, nrow_local
1683 0 : INTEGER, DIMENSION(:), POINTER :: col_indices, row_indices
1684 :
1685 0 : CALL timeset(routineN, handle)
1686 :
1687 : CALL cp_fm_get_info(matrix=fm_mat, &
1688 : nrow_local=nrow_local, ncol_local=ncol_local, &
1689 0 : row_indices=row_indices, col_indices=col_indices)
1690 :
1691 0 : DO j_col = 1, ncol_local
1692 0 : j_global = col_indices(j_col)
1693 0 : DO i_row = 1, nrow_local
1694 0 : IF (j_global < row_indices(i_row)) fm_mat%local_data(i_row, j_col) = 0.0_dp
1695 : END DO
1696 : END DO
1697 :
1698 0 : CALL timestop(handle)
1699 :
1700 0 : END SUBROUTINE clean_lower_part
1701 :
1702 : ! **************************************************************************************************
1703 : !> \brief Computes the exact exchange part of the GW self-energy
1704 : !> \param bs_env ...
1705 : !> \param qs_env ...
1706 : !> \param mat_phi_mu_l ...
1707 : !> \param mat_Z_lP ...
1708 : !> \param fm_Sigma_x_Gamma ...
1709 : ! **************************************************************************************************
1710 :
1711 0 : SUBROUTINE compute_Sigma_x(bs_env, qs_env, mat_phi_mu_l, mat_Z_lP, fm_Sigma_x_Gamma)
1712 :
1713 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
1714 : TYPE(qs_environment_type), POINTER :: qs_env
1715 : TYPE(dbcsr_type), INTENT(INOUT) :: mat_phi_mu_l, mat_Z_lP
1716 : TYPE(cp_fm_type), ALLOCATABLE, DIMENSION(:) :: fm_Sigma_x_Gamma
1717 :
1718 : CHARACTER(LEN=*), PARAMETER :: routineN = 'compute_Sigma_x'
1719 :
1720 : INTEGER :: handle, ispin
1721 0 : INTEGER, DIMENSION(:), POINTER :: blk_aux, blk_grid, dist_col_grid, &
1722 0 : dist_row_aux
1723 : REAL(KIND=dp) :: t1
1724 0 : TYPE(cp_fm_type), ALLOCATABLE, DIMENSION(:, :) :: fm_Vtr_Gamma
1725 : TYPE(dbcsr_distribution_type) :: dist_aux_aux, dist_grid_grid
1726 : TYPE(dbcsr_type) :: mat_Sigma_x_Gamma, matrix_D_grid, &
1727 : matrix_Sigma_x_grid, matrix_V_aux, &
1728 : matrix_V_grid
1729 :
1730 0 : CALL timeset(routineN, handle)
1731 :
1732 0 : t1 = m_walltime()
1733 :
1734 0 : ALLOCATE (fm_Sigma_x_Gamma(bs_env%n_spin))
1735 0 : DO ispin = 1, bs_env%n_spin
1736 0 : CALL cp_fm_create(fm_Sigma_x_Gamma(ispin), bs_env%fm_s_Gamma%matrix_struct)
1737 : END DO
1738 :
1739 0 : CALL dbcsr_create(mat_Sigma_x_Gamma, template=bs_env%mat_ao_ao%matrix)
1740 :
1741 : ! =========================================================================
1742 : ! 1. SETUP CORE TOPOLOGIES
1743 : ! =========================================================================
1744 0 : CALL setup_square_topology(mat_phi_mu_l, 'ROW', dist_grid_grid, blk_grid, dist_col_grid)
1745 0 : CALL setup_square_topology(mat_Z_lP, 'COL', dist_aux_aux, blk_aux, dist_row_aux)
1746 :
1747 : ! =========================================================================
1748 : ! 2. COMPUTE V^tr_ll'
1749 : ! =========================================================================
1750 : CALL RI_2c_integral_mat(qs_env, fm_Vtr_Gamma, bs_env%fm_RI_RI, bs_env%n_RI, &
1751 0 : bs_env%trunc_coulomb, do_kpoints=.FALSE.)
1752 :
1753 : ! M^-1 * V^tr * M^-1 directly modifies fm_Vtr_Gamma(:, 1)
1754 0 : CALL multiply_fm_W_MIC_time_with_Minv_Gamma(bs_env, qs_env, fm_Vtr_Gamma(:, 1))
1755 :
1756 0 : CALL dbcsr_create(matrix_V_aux, "V_aux", dist_aux_aux, dbcsr_type_no_symmetry, blk_aux, blk_aux)
1757 0 : CALL dbcsr_create(matrix_V_grid, "V_grid", dist_grid_grid, dbcsr_type_no_symmetry, blk_grid, blk_grid)
1758 :
1759 0 : CALL copy_fm_to_dbcsr(fm_Vtr_Gamma(1, 1), matrix_V_aux, keep_sparsity=.FALSE.)
1760 :
1761 : ! V^tr_ll' = sum_PQ Z_lP V^trunc_PQ Z_l'Q
1762 0 : CALL contract_A_B_A("N", "T", mat_Z_lP, matrix_V_aux, matrix_V_grid, bs_env%eps_filter)
1763 0 : CALL dbcsr_release(matrix_V_aux)
1764 :
1765 : ! =========================================================================
1766 : ! 3. SPIN LOOP FOR EXACT EXCHANGE
1767 : ! =========================================================================
1768 0 : DO ispin = 1, bs_env%n_spin
1769 :
1770 : ! Density matrix on grid is essentially G_occ at tau = 0.0
1771 : ! D_µν = sum_n^occ C_µn(k=0) C_νn(k=0)
1772 : ! D_ll' = sum_µν Φ_µ(r_l) D_µν Φ_ν(r_l')
1773 0 : CALL dbcsr_create(matrix_D_grid, "D_grid", dist_grid_grid, dbcsr_type_no_symmetry, blk_grid, blk_grid)
1774 0 : CALL build_G_grid(bs_env, 0.0_dp, ispin, .TRUE., .FALSE., mat_phi_mu_l, matrix_D_grid, bs_env%eps_filter)
1775 :
1776 : ! Element-wise Hadamard product: Σ^x_grid = D_grid ◦ V_grid
1777 : ! Σ^x_ll' = D_ll' * V^tr_ll'
1778 0 : CALL dbcsr_create(matrix_Sigma_x_grid, template=matrix_V_grid)
1779 0 : CALL hadamard_product(matrix_D_grid, matrix_V_grid, matrix_Sigma_x_grid, 1.0_dp)
1780 :
1781 0 : CALL dbcsr_release(matrix_D_grid)
1782 :
1783 : ! Transform back to AO basis: Σ^x_ao = -1.0 * phi^T * Σ^x_grid * phi
1784 : ! Σ^x_λσ(k=0) = -sum_ll' Φ_λ(r_l) Σ^x_ll' Φ_σ(r_l')
1785 0 : CALL contract_A_B_A("T", "N", mat_phi_mu_l, matrix_Sigma_x_grid, mat_Sigma_x_Gamma, bs_env%eps_filter)
1786 0 : CALL dbcsr_scale(mat_Sigma_x_Gamma, -1.0_dp)
1787 :
1788 0 : CALL dbcsr_release(matrix_Sigma_x_grid)
1789 :
1790 : ! Data I/O and Export to CP2K Full Matrices
1791 0 : CALL copy_dbcsr_to_fm(mat_Sigma_x_Gamma, fm_Sigma_x_Gamma(ispin))
1792 :
1793 : END DO ! ispin
1794 :
1795 0 : IF (bs_env%unit_nr > 0) THEN
1796 : WRITE (bs_env%unit_nr, '(T2,A,T58,A,F7.1,A)') &
1797 0 : 'Computed Σ^x(k=0),', ' Execution time', m_walltime() - t1, ' s'
1798 0 : WRITE (bs_env%unit_nr, '(A)') ' '
1799 : END IF
1800 :
1801 : ! =========================================================================
1802 : ! 4. CLEANUP
1803 : ! =========================================================================
1804 : CALL release_dbcsr_topology_and_matrices(dist=dist_grid_grid, mapped_dist=dist_col_grid, &
1805 0 : m1=mat_Sigma_x_Gamma, m2=matrix_V_grid)
1806 0 : CALL release_dbcsr_topology_and_matrices(dist=dist_aux_aux, mapped_dist=dist_row_aux)
1807 :
1808 0 : CALL cp_fm_release(fm_Vtr_Gamma)
1809 :
1810 0 : CALL timestop(handle)
1811 :
1812 0 : END SUBROUTINE compute_Sigma_x
1813 :
1814 : ! **************************************************************************************************
1815 : !> \brief Computes the correlation part of the GW self-energy
1816 : !> \param bs_env ...
1817 : !> \param fm_W_time ...
1818 : !> \param mat_phi_mu_l ...
1819 : !> \param mat_Z_lP ...
1820 : !> \param fm_Sigma_c_Gamma_time ...
1821 : ! **************************************************************************************************
1822 :
1823 0 : SUBROUTINE compute_Sigma_c(bs_env, fm_W_time, mat_phi_mu_l, mat_Z_lP, fm_Sigma_c_Gamma_time)
1824 :
1825 : TYPE(post_scf_bandstructure_type), POINTER :: bs_env
1826 : TYPE(cp_fm_type), ALLOCATABLE, DIMENSION(:) :: fm_W_time
1827 : TYPE(dbcsr_type), INTENT(INOUT) :: mat_phi_mu_l, mat_Z_lP
1828 : TYPE(cp_fm_type), ALLOCATABLE, DIMENSION(:, :, :) :: fm_Sigma_c_Gamma_time
1829 :
1830 : CHARACTER(LEN=*), PARAMETER :: routineN = 'compute_Sigma_c'
1831 :
1832 : INTEGER :: handle, i_t, ispin
1833 0 : INTEGER, DIMENSION(:), POINTER :: blk_aux, blk_grid, dist_col_grid, &
1834 0 : dist_row_aux
1835 : REAL(KIND=dp) :: t1, tau
1836 : TYPE(dbcsr_distribution_type) :: dist_aux_aux, dist_grid_grid
1837 0 : TYPE(dbcsr_p_type), DIMENSION(:, :), POINTER :: mat_Sigma_neg_tau, mat_Sigma_pos_tau
1838 : TYPE(dbcsr_type) :: matrix_G_occ_grid, matrix_G_vir_grid, matrix_Sigma_neg_grid, &
1839 : matrix_Sigma_pos_grid, matrix_W_aux, matrix_W_grid
1840 :
1841 0 : CALL timeset(routineN, handle)
1842 :
1843 : ! =========================================================================
1844 : ! 1. SETUP CORE TOPOLOGIES AND PRE-ALLOCATE OUTPUT ARRAYS
1845 : ! =========================================================================
1846 0 : CALL setup_square_topology(mat_phi_mu_l, 'ROW', dist_grid_grid, blk_grid, dist_col_grid)
1847 0 : CALL setup_square_topology(mat_Z_lP, 'COL', dist_aux_aux, blk_aux, dist_row_aux)
1848 :
1849 : ! Pre-allocate local DBCSR matrices to act as targets for final output
1850 0 : NULLIFY (mat_Sigma_neg_tau, mat_Sigma_pos_tau)
1851 0 : ALLOCATE (mat_Sigma_neg_tau(bs_env%num_time_freq_points, bs_env%n_spin))
1852 0 : ALLOCATE (mat_Sigma_pos_tau(bs_env%num_time_freq_points, bs_env%n_spin))
1853 :
1854 0 : DO i_t = 1, bs_env%num_time_freq_points
1855 0 : DO ispin = 1, bs_env%n_spin
1856 0 : ALLOCATE (mat_Sigma_neg_tau(i_t, ispin)%matrix)
1857 0 : ALLOCATE (mat_Sigma_pos_tau(i_t, ispin)%matrix)
1858 0 : CALL dbcsr_create(mat_Sigma_neg_tau(i_t, ispin)%matrix, template=bs_env%mat_ao_ao%matrix)
1859 0 : CALL dbcsr_create(mat_Sigma_pos_tau(i_t, ispin)%matrix, template=bs_env%mat_ao_ao%matrix)
1860 : END DO
1861 : END DO
1862 :
1863 : ! =========================================================================
1864 : ! 2. MAIN IMAGINARY TIME LOOP
1865 : ! =========================================================================
1866 0 : DO i_t = 1, bs_env%num_time_freq_points
1867 0 : tau = bs_env%imag_time_points(i_t)
1868 :
1869 : ! -------------------------------------------------------------------
1870 : ! Compute W_grid = Z * W_aux * Z^T
1871 : ! -------------------------------------------------------------------
1872 0 : CALL dbcsr_create(matrix_W_aux, "W_aux", dist_aux_aux, dbcsr_type_no_symmetry, blk_aux, blk_aux)
1873 0 : CALL dbcsr_create(matrix_W_grid, "W_grid", dist_grid_grid, dbcsr_type_no_symmetry, blk_grid, blk_grid)
1874 :
1875 0 : CALL copy_fm_to_dbcsr(fm_W_time(i_t), matrix_W_aux, keep_sparsity=.FALSE.)
1876 :
1877 : ! W^MIC_ll'(iτ,k=0) = sum_PQ Z_lP W^MIC_PQ(iτ) Z_l'Q
1878 0 : CALL contract_A_B_A("N", "T", mat_Z_lP, matrix_W_aux, matrix_W_grid, bs_env%eps_filter)
1879 :
1880 0 : CALL dbcsr_release(matrix_W_aux) ! Clean up aux basis immediately
1881 :
1882 0 : DO ispin = 1, bs_env%n_spin
1883 0 : t1 = m_walltime()
1884 :
1885 : ! -------------------------------------------------------------------
1886 : ! A. Transform Green's Functions to the Grid
1887 : ! -------------------------------------------------------------------
1888 0 : CALL dbcsr_create(matrix_G_occ_grid, "G_occ_grid", dist_grid_grid, dbcsr_type_no_symmetry, blk_grid, blk_grid)
1889 0 : CALL dbcsr_create(matrix_G_vir_grid, "G_vir_grid", dist_grid_grid, dbcsr_type_no_symmetry, blk_grid, blk_grid)
1890 :
1891 : ! G^occ_µλ(i|τ|,k=0) = sum_G^occ_µλn^occ C_µn(k=0) e^(-|(ϵ_nk=0-ϵ_F)τ|) C_λn(k=0)
1892 : ! G^occ_ll'(i|τ|,k=0) = sum_µν Φ_µ(r_l) G^occ_µν Φ_ν(r_l')
1893 0 : CALL build_G_grid(bs_env, tau, ispin, .TRUE., .FALSE., mat_phi_mu_l, matrix_G_occ_grid, bs_env%eps_filter)
1894 :
1895 : ! G^vir_µλ(i|τ|,k=0) = sum_n^vir C_µn(k=0) e^(-|(ϵ_nk=0-ϵ_F)τ|) C_λn(k=0)
1896 : ! G^vir_ll'(i|τ|,k=0) = sum_µν Φ_µ(r_l) G^vir_µν Φ_ν(r_l')
1897 0 : CALL build_G_grid(bs_env, tau, ispin, .FALSE., .TRUE., mat_phi_mu_l, matrix_G_vir_grid, bs_env%eps_filter)
1898 :
1899 : ! -------------------------------------------------------------------
1900 : ! B. Element-wise Hadamard Products for Sigma_c on Grid
1901 : ! Σ_neg_grid = G_occ_grid ◦ W_grid
1902 : ! Σ_pos_grid = G_vir_grid ◦ W_grid
1903 : ! -------------------------------------------------------------------
1904 0 : CALL dbcsr_create(matrix_Sigma_neg_grid, template=matrix_W_grid)
1905 0 : CALL dbcsr_create(matrix_Sigma_pos_grid, template=matrix_W_grid)
1906 :
1907 : ! Σ^c_ll'(iτ,k=0) = -G^occ_ll'(i|τ|,k=0) * W^MIC_ll'(iτ,k=0), for τ < 0
1908 0 : CALL hadamard_product(matrix_G_occ_grid, matrix_W_grid, matrix_Sigma_neg_grid, 1.0_dp)
1909 :
1910 : ! Σ^c_ll'(iτ,k=0) = G^vir_ll'(i|τ|,k=0) * W^MIC_ll'(iτ,k=0), for τ > 0
1911 0 : CALL hadamard_product(matrix_G_vir_grid, matrix_W_grid, matrix_Sigma_pos_grid, 1.0_dp)
1912 :
1913 : ! Instantly purge massive G_grid arrays to save memory
1914 0 : CALL dbcsr_release(matrix_G_occ_grid)
1915 0 : CALL dbcsr_release(matrix_G_vir_grid)
1916 :
1917 : ! -------------------------------------------------------------------
1918 : ! C. Transform Sigma back to AO Basis
1919 : ! Σ_AO = phi^T * Σ_grid * phi
1920 : ! -------------------------------------------------------------------
1921 :
1922 : ! Σ^c_λσ(iτ,k=0) = sum_ll' Φ_λ(r_l) Σ^c_ll'(iτ,k=0) Φ_σ(r_l'), for τ < 0
1923 : CALL contract_A_B_A("T", "N", mat_phi_mu_l, matrix_Sigma_neg_grid, &
1924 0 : mat_Sigma_neg_tau(i_t, ispin)%matrix, bs_env%eps_filter)
1925 0 : CALL dbcsr_scale(mat_Sigma_neg_tau(i_t, ispin)%matrix, -1.0_dp)
1926 :
1927 : ! Σ^c_λσ(iτ,k=0) = sum_ll' Φ_λ(r_l) Σ^c_ll'(iτ,k=0) Φ_σ(r_l'), for τ > 0
1928 : CALL contract_A_B_A("T", "N", mat_phi_mu_l, matrix_Sigma_pos_grid, &
1929 0 : mat_Sigma_pos_tau(i_t, ispin)%matrix, bs_env%eps_filter)
1930 :
1931 : ! Purge Grid Sigma arrays
1932 0 : CALL dbcsr_release(matrix_Sigma_neg_grid)
1933 0 : CALL dbcsr_release(matrix_Sigma_pos_grid)
1934 :
1935 0 : IF (bs_env%unit_nr > 0) THEN
1936 : WRITE (bs_env%unit_nr, '(T2,A,I10,A,I3,A,F7.1,A)') &
1937 0 : 'Computed Σ^c(iτ,k=0) for time point ', i_t, ' /', bs_env%num_time_freq_points, &
1938 0 : ', Execution time', m_walltime() - t1, ' s'
1939 : END IF
1940 :
1941 : END DO ! ispin
1942 :
1943 : ! Release the W_grid for this time point
1944 0 : CALL dbcsr_release(matrix_W_grid)
1945 :
1946 : END DO ! i_t
1947 :
1948 0 : IF (bs_env%unit_nr > 0) WRITE (bs_env%unit_nr, '(A)') ' '
1949 :
1950 : ! -------------------------------------------------------------------------
1951 : ! 3. FINALIZE AND CLEANUP
1952 : ! -------------------------------------------------------------------------
1953 : CALL fill_fm_Sigma_c_Gamma_time(fm_Sigma_c_Gamma_time, bs_env, &
1954 0 : mat_Sigma_pos_tau, mat_Sigma_neg_tau)
1955 :
1956 0 : CALL cp_fm_release(fm_W_time)
1957 :
1958 0 : CALL dbcsr_deallocate_matrix_set(mat_Sigma_neg_tau)
1959 0 : CALL dbcsr_deallocate_matrix_set(mat_Sigma_pos_tau)
1960 :
1961 0 : CALL release_dbcsr_topology_and_matrices(dist=dist_grid_grid, mapped_dist=dist_col_grid)
1962 0 : CALL release_dbcsr_topology_and_matrices(dist=dist_aux_aux, mapped_dist=dist_row_aux)
1963 :
1964 0 : CALL delete_unnecessary_files(bs_env)
1965 0 : CALL timestop(handle)
1966 :
1967 0 : END SUBROUTINE compute_Sigma_c
1968 :
1969 : ! **************************************************************************************************
1970 : !> \brief DBCSR Topology Generation
1971 : !> \param matrix_template ...
1972 : !> \param dim_type ...
1973 : !> \param square_dist ...
1974 : !> \param blk_sizes ...
1975 : !> \param mapped_dist ...
1976 : ! **************************************************************************************************
1977 :
1978 0 : SUBROUTINE setup_square_topology(matrix_template, dim_type, square_dist, blk_sizes, mapped_dist)
1979 :
1980 : TYPE(dbcsr_type), INTENT(IN) :: matrix_template
1981 : CHARACTER(LEN=*), INTENT(IN) :: dim_type
1982 : TYPE(dbcsr_distribution_type), INTENT(OUT) :: square_dist
1983 : INTEGER, DIMENSION(:), INTENT(OUT), POINTER :: blk_sizes, mapped_dist
1984 :
1985 : CHARACTER(LEN=*), PARAMETER :: routineN = 'setup_square_topology'
1986 :
1987 : INTEGER :: handle, i, np, npcols, nprows
1988 0 : INTEGER, DIMENSION(:), POINTER :: col_blk, col_dist, row_blk, row_dist
1989 : TYPE(dbcsr_distribution_type) :: dist_template
1990 :
1991 0 : CALL timeset(routineN, handle)
1992 :
1993 : CALL dbcsr_get_info(matrix_template, distribution=dist_template, &
1994 0 : row_blk_size=row_blk, col_blk_size=col_blk)
1995 : CALL dbcsr_distribution_get(dist_template, row_dist=row_dist, col_dist=col_dist, &
1996 0 : nprows=nprows, npcols=npcols)
1997 :
1998 0 : IF (TRIM(dim_type) == 'ROW') THEN
1999 : ! Creates ROW x ROW (e.g., Grid x Grid from mat_phi_mu_l)
2000 0 : blk_sizes => row_blk
2001 0 : np = npcols
2002 0 : ALLOCATE (mapped_dist(SIZE(blk_sizes)))
2003 0 : DO i = 1, SIZE(blk_sizes)
2004 0 : mapped_dist(i) = MOD(i - 1, np)
2005 : END DO
2006 : CALL dbcsr_distribution_new(square_dist, template=dist_template, &
2007 0 : row_dist=row_dist, col_dist=mapped_dist)
2008 :
2009 0 : ELSE IF (TRIM(dim_type) == 'COL') THEN
2010 : ! Creates COL x COL (e.g., Aux x Aux from mat_Z_lP)
2011 0 : blk_sizes => col_blk
2012 0 : np = nprows
2013 0 : ALLOCATE (mapped_dist(SIZE(blk_sizes)))
2014 0 : DO i = 1, SIZE(blk_sizes)
2015 0 : mapped_dist(i) = MOD(i - 1, np)
2016 : END DO
2017 : CALL dbcsr_distribution_new(square_dist, template=dist_template, &
2018 0 : row_dist=mapped_dist, col_dist=col_dist)
2019 : END IF
2020 :
2021 0 : CALL timestop(handle)
2022 :
2023 0 : END SUBROUTINE setup_square_topology
2024 :
2025 : ! **************************************************************************************************
2026 : !> \brief DBCSR matrices deallocation
2027 : !> \param dist ...
2028 : !> \param mapped_dist ...
2029 : !> \param m1 ...
2030 : !> \param m2 ...
2031 : !> \param m3 ...
2032 : !> \param m4 ...
2033 : ! **************************************************************************************************
2034 :
2035 0 : SUBROUTINE release_dbcsr_topology_and_matrices(dist, mapped_dist, m1, m2, m3, m4)
2036 :
2037 : TYPE(dbcsr_distribution_type), INTENT(INOUT), &
2038 : OPTIONAL :: dist
2039 : INTEGER, DIMENSION(:), INTENT(INOUT), OPTIONAL, &
2040 : POINTER :: mapped_dist
2041 : TYPE(dbcsr_type), INTENT(INOUT), OPTIONAL :: m1, m2, m3, m4
2042 :
2043 : CHARACTER(LEN=*), PARAMETER :: routineN = 'release_dbcsr_topology_and_matrices'
2044 :
2045 : INTEGER :: handle
2046 :
2047 0 : CALL timeset(routineN, handle)
2048 :
2049 0 : IF (PRESENT(dist)) CALL dbcsr_distribution_release(dist)
2050 0 : IF (PRESENT(mapped_dist)) THEN
2051 0 : IF (ASSOCIATED(mapped_dist)) THEN
2052 0 : DEALLOCATE (mapped_dist)
2053 : NULLIFY (mapped_dist)
2054 : END IF
2055 : END IF
2056 0 : IF (PRESENT(m1)) CALL dbcsr_release(m1)
2057 0 : IF (PRESENT(m2)) CALL dbcsr_release(m2)
2058 0 : IF (PRESENT(m3)) CALL dbcsr_release(m3)
2059 0 : IF (PRESENT(m4)) CALL dbcsr_release(m4)
2060 :
2061 0 : CALL timestop(handle)
2062 :
2063 0 : END SUBROUTINE release_dbcsr_topology_and_matrices
2064 :
2065 : END MODULE gw_large_cell_Gamma_ri_rs
|
