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 : MODULE qs_tddfpt2_eigensolver
9 : USE cp_blacs_env, ONLY: cp_blacs_env_type
10 : USE cp_control_types, ONLY: tddfpt2_control_type
11 : USE cp_dbcsr_api, ONLY: dbcsr_get_info,&
12 : dbcsr_p_type,&
13 : dbcsr_type
14 : USE cp_dbcsr_operations, ONLY: cp_dbcsr_sm_fm_multiply
15 : USE cp_fm_basic_linalg, ONLY: cp_fm_contracted_trace,&
16 : cp_fm_scale,&
17 : cp_fm_scale_and_add,&
18 : cp_fm_trace
19 : USE cp_fm_diag, ONLY: choose_eigv_solver
20 : USE cp_fm_pool_types, ONLY: fm_pool_create_fm,&
21 : fm_pool_give_back_fm
22 : USE cp_fm_struct, ONLY: cp_fm_struct_create,&
23 : cp_fm_struct_release,&
24 : cp_fm_struct_type
25 : USE cp_fm_types, ONLY: &
26 : cp_fm_copy_general, cp_fm_create, cp_fm_get_info, cp_fm_get_submatrix, cp_fm_maxabsval, &
27 : cp_fm_release, cp_fm_set_all, cp_fm_set_submatrix, cp_fm_to_fm, cp_fm_type
28 : USE cp_log_handling, ONLY: cp_logger_type
29 : USE cp_output_handling, ONLY: cp_iterate
30 : USE input_constants, ONLY: no_sf_tddfpt,&
31 : tddfpt_kernel_full,&
32 : tddfpt_kernel_none,&
33 : tddfpt_kernel_stda
34 : USE input_section_types, ONLY: section_vals_type
35 : USE kinds, ONLY: dp,&
36 : int_8
37 : USE machine, ONLY: m_flush,&
38 : m_walltime
39 : USE memory_utilities, ONLY: reallocate
40 : USE message_passing, ONLY: mp_para_env_type
41 : USE parallel_gemm_api, ONLY: parallel_gemm
42 : USE physcon, ONLY: evolt
43 : USE qs_environment_types, ONLY: get_qs_env,&
44 : qs_environment_type
45 : USE qs_kernel_types, ONLY: full_kernel_env_type,&
46 : kernel_env_type
47 : USE qs_scf_methods, ONLY: eigensolver
48 : USE qs_tddfpt2_bse_utils, ONLY: tddfpt_apply_bse,&
49 : zeroth_order_gw
50 : USE qs_tddfpt2_fhxc, ONLY: fhxc_kernel,&
51 : stda_kernel
52 : USE qs_tddfpt2_operators, ONLY: tddfpt_apply_energy_diff,&
53 : tddfpt_apply_hfx,&
54 : tddfpt_apply_hfxlr_kernel,&
55 : tddfpt_apply_hfxsr_kernel
56 : USE qs_tddfpt2_restart, ONLY: tddfpt_write_restart
57 : USE qs_tddfpt2_smearing_methods, ONLY: add_smearing_aterm,&
58 : compute_fermib,&
59 : orthogonalize_smeared_occupation
60 : USE qs_tddfpt2_subgroups, ONLY: tddfpt_subgroup_env_type
61 : USE qs_tddfpt2_types, ONLY: tddfpt_ground_state_mos,&
62 : tddfpt_work_matrices
63 : USE qs_tddfpt2_utils, ONLY: tddfpt_total_number_of_states
64 : #include "./base/base_uses.f90"
65 :
66 : IMPLICIT NONE
67 :
68 : PRIVATE
69 :
70 : CHARACTER(len=*), PARAMETER, PRIVATE :: moduleN = 'qs_tddfpt2_eigensolver'
71 :
72 : LOGICAL, PARAMETER, PRIVATE :: debug_this_module = .FALSE.
73 : ! number of first derivative components (3: d/dx, d/dy, d/dz)
74 : INTEGER, PARAMETER, PRIVATE :: nderivs = 3
75 : INTEGER, PARAMETER, PRIVATE :: maxspins = 2
76 :
77 : PUBLIC :: tddfpt_davidson_solver, tddfpt_orthogonalize_psi1_psi0, &
78 : tddfpt_orthonormalize_psi1_psi1
79 :
80 : ! **************************************************************************************************
81 :
82 : CONTAINS
83 :
84 : ! **************************************************************************************************
85 : !> \brief Make TDDFPT trial vectors orthogonal to all occupied molecular orbitals.
86 : !> \param evects trial vectors distributed across all processors (modified on exit)
87 : !> \param S_C0_C0T matrix product S * C_0 * C_0^T, where C_0 is the ground state
88 : !> wave function for each spin expressed in atomic basis set,
89 : !> and S is the corresponding overlap matrix
90 : !> \param qs_env ...
91 : !> \param gs_mos ...
92 : !> \param evals ...
93 : !> \param tddfpt_control ...
94 : !> \param S_C0 ...
95 : !> \par History
96 : !> * 05.2016 created [Sergey Chulkov]
97 : !> * 05.2019 use a temporary work matrix [JHU]
98 : !> \note Based on the subroutine p_preortho() which was created by Thomas Chassaing on 09.2002.
99 : !> Should be useless when ground state MOs are computed with extremely high accuracy,
100 : !> as all virtual orbitals are already orthogonal to the occupied ones by design.
101 : !> However, when the norm of residual vectors is relatively small (e.g. less then SCF_EPS),
102 : !> new Krylov's vectors seem to be random and should be orthogonalised even with respect to
103 : !> the occupied MOs.
104 : ! **************************************************************************************************
105 6564 : SUBROUTINE tddfpt_orthogonalize_psi1_psi0(evects, S_C0_C0T, qs_env, gs_mos, evals, &
106 6564 : tddfpt_control, S_C0)
107 : TYPE(cp_fm_type), DIMENSION(:, :), INTENT(in) :: evects
108 : TYPE(cp_fm_type), DIMENSION(:), INTENT(in) :: S_C0_C0T
109 : TYPE(qs_environment_type), POINTER :: qs_env
110 : TYPE(tddfpt_ground_state_mos), DIMENSION(:), &
111 : INTENT(in) :: gs_mos
112 : REAL(kind=dp), DIMENSION(:), INTENT(in) :: evals
113 : TYPE(tddfpt2_control_type), INTENT(in), POINTER :: tddfpt_control
114 : TYPE(cp_fm_type), DIMENSION(:), INTENT(in) :: S_C0
115 :
116 : CHARACTER(LEN=*), PARAMETER :: routineN = 'tddfpt_orthogonalize_psi1_psi0'
117 :
118 : INTEGER :: handle, ispin, ivect, nactive, nao, &
119 : nspins, nvects, spin
120 : TYPE(cp_fm_struct_type), POINTER :: matrix_struct
121 : TYPE(cp_fm_type) :: evortho
122 6564 : TYPE(cp_fm_type), DIMENSION(:), POINTER :: mos
123 : TYPE(mp_para_env_type), POINTER :: para_env
124 :
125 6564 : CALL timeset(routineN, handle)
126 :
127 6564 : nspins = SIZE(evects, 1)
128 6564 : nvects = SIZE(evects, 2)
129 :
130 6564 : IF (nvects > 0) THEN
131 6564 : IF (.NOT. tddfpt_control%do_smearing) THEN
132 13940 : DO ispin = 1, nspins
133 7390 : IF (tddfpt_control%spinflip == no_sf_tddfpt) THEN
134 : spin = ispin
135 : ELSE
136 122 : spin = 2
137 : END IF
138 : CALL cp_fm_get_info(matrix=evects(ispin, 1), matrix_struct=matrix_struct, &
139 7390 : nrow_global=nao, ncol_global=nactive)
140 7390 : CALL cp_fm_create(evortho, matrix_struct)
141 27160 : DO ivect = 1, nvects
142 : ! evortho: C0 * C0^T * S * C1 == (S * C0 * C0^T)^T * C1
143 : CALL parallel_gemm('T', 'N', nao, nactive, nao, 1.0_dp, S_C0_C0T(spin), &
144 19770 : evects(ispin, ivect), 0.0_dp, evortho)
145 27160 : CALL cp_fm_scale_and_add(1.0_dp, evects(ispin, ivect), -1.0_dp, evortho)
146 : END DO
147 21330 : CALL cp_fm_release(evortho)
148 : END DO
149 : ELSE
150 14 : NULLIFY (para_env)
151 14 : CALL get_qs_env(qs_env, para_env=para_env)
152 14 : NULLIFY (mos)
153 56 : ALLOCATE (mos(nspins))
154 28 : DO ispin = 1, nspins
155 : CALL cp_fm_get_info(matrix=evects(ispin, 1), matrix_struct=matrix_struct, &
156 14 : nrow_global=nao, ncol_global=nactive)
157 14 : CALL cp_fm_create(mos(ispin), matrix_struct)
158 42 : CALL cp_fm_copy_general(gs_mos(ispin)%mos_occ, mos(ispin), para_env)
159 : END DO
160 28 : DO ivect = 1, nvects
161 14 : CALL compute_fermib(qs_env, gs_mos, evals(ivect))
162 28 : CALL orthogonalize_smeared_occupation(evects(:, ivect), qs_env, mos, S_C0)
163 : END DO
164 28 : DO ispin = 1, nspins
165 28 : CALL cp_fm_release(mos(ispin))
166 : END DO
167 14 : DEALLOCATE (mos)
168 : END IF
169 : END IF
170 :
171 6564 : CALL timestop(handle)
172 :
173 6564 : END SUBROUTINE tddfpt_orthogonalize_psi1_psi0
174 :
175 : ! **************************************************************************************************
176 : !> \brief Check that orthogonalised TDDFPT trial vectors remain orthogonal to
177 : !> occupied molecular orbitals.
178 : !> \param evects trial vectors
179 : !> \param S_C0 matrix product S * C_0, where C_0 is the ground state wave function
180 : !> for each spin in atomic basis set, and S is the corresponding overlap matrix
181 : !> \param max_norm the largest possible overlap between the ground state and
182 : !> excited state wave functions
183 : !> \param spinflip ...
184 : !> \return true if trial vectors are non-orthogonal to occupied molecular orbitals
185 : !> \par History
186 : !> * 07.2016 created [Sergey Chulkov]
187 : !> * 05.2019 use temporary work matrices [JHU]
188 : ! **************************************************************************************************
189 4230 : FUNCTION tddfpt_is_nonorthogonal_psi1_psi0(evects, S_C0, max_norm, spinflip) &
190 : RESULT(is_nonortho)
191 : TYPE(cp_fm_type), DIMENSION(:, :), INTENT(in) :: evects
192 : TYPE(cp_fm_type), DIMENSION(:), INTENT(in) :: S_C0
193 : REAL(kind=dp), INTENT(in) :: max_norm
194 : INTEGER, INTENT(in) :: spinflip
195 : LOGICAL :: is_nonortho
196 :
197 : CHARACTER(LEN=*), PARAMETER :: routineN = 'tddfpt_is_nonorthogonal_psi1_psi0'
198 :
199 : INTEGER :: handle, ispin, ivect, nactive, nao, &
200 : nocc, nspins, nvects, spin
201 : REAL(kind=dp) :: maxabs_val
202 : TYPE(cp_fm_struct_type), POINTER :: matrix_struct, matrix_struct_tmp
203 : TYPE(cp_fm_type) :: aortho
204 :
205 4230 : CALL timeset(routineN, handle)
206 :
207 4230 : nspins = SIZE(evects, 1)
208 4230 : nvects = SIZE(evects, 2)
209 :
210 4230 : is_nonortho = .FALSE.
211 :
212 9032 : loop: DO ispin = 1, nspins
213 :
214 4810 : IF (spinflip == no_sf_tddfpt) THEN
215 : spin = ispin
216 : ELSE
217 74 : spin = 2
218 : END IF
219 :
220 4810 : CALL cp_fm_get_info(matrix=S_C0(spin), ncol_global=nocc)
221 : CALL cp_fm_get_info(matrix=evects(ispin, 1), matrix_struct=matrix_struct, &
222 4810 : nrow_global=nao, ncol_global=nactive)
223 : CALL cp_fm_struct_create(matrix_struct_tmp, nrow_global=nocc, &
224 4810 : ncol_global=nactive, template_fmstruct=matrix_struct)
225 4810 : CALL cp_fm_create(aortho, matrix_struct_tmp)
226 4810 : CALL cp_fm_struct_release(matrix_struct_tmp)
227 17180 : DO ivect = 1, nvects
228 : ! aortho = S_C0^T * S * C_1
229 : CALL parallel_gemm('T', 'N', nocc, nactive, nao, 1.0_dp, S_C0(spin), &
230 12378 : evects(ispin, ivect), 0.0_dp, aortho)
231 12378 : CALL cp_fm_maxabsval(aortho, maxabs_val)
232 12378 : is_nonortho = maxabs_val > max_norm
233 17180 : IF (is_nonortho) THEN
234 8 : CALL cp_fm_release(aortho)
235 8 : EXIT loop
236 : END IF
237 : END DO
238 18644 : CALL cp_fm_release(aortho)
239 : END DO loop
240 :
241 4230 : CALL timestop(handle)
242 :
243 4230 : END FUNCTION tddfpt_is_nonorthogonal_psi1_psi0
244 :
245 : ! **************************************************************************************************
246 : !> \brief Make new TDDFPT trial vectors orthonormal to all previous trial vectors.
247 : !> \param evects trial vectors (modified on exit)
248 : !> \param nvects_new number of new trial vectors to orthogonalise
249 : !> \param S_evects set of matrices to store matrix product S * evects (modified on exit)
250 : !> \param matrix_s overlap matrix
251 : !> \par History
252 : !> * 05.2016 created [Sergey Chulkov]
253 : !> * 02.2017 caching the matrix product S * evects [Sergey Chulkov]
254 : !> \note \parblock
255 : !> Based on the subroutines reorthogonalize() and normalize() which were originally created
256 : !> by Thomas Chassaing on 03.2003.
257 : !>
258 : !> In order to orthogonalise a trial vector C3 = evects(:,3) with respect to previously
259 : !> orthogonalised vectors C1 = evects(:,1) and C2 = evects(:,2) we need to compute the
260 : !> quantity C3'' using the following formulae:
261 : !> C3' = C3 - Tr(C3^T * S * C1) * C1,
262 : !> C3'' = C3' - Tr(C3'^T * S * C2) * C2,
263 : !> which can be expanded as:
264 : !> C3'' = C3 - Tr(C3^T * S * C1) * C1 - Tr(C3^T * S * C2) * C2 +
265 : !> Tr(C3^T * S * C1) * Tr(C2^T * S * C1) * C2 .
266 : !> In case of unlimited float-point precision, the last term in above expression is exactly 0,
267 : !> due to orthogonality condition between C1 and C2. In this case the expression could be
268 : !> simplified as (taking into account the identity: Tr(A * S * B) = Tr(B * S * A)):
269 : !> C3'' = C3 - Tr(C1^T * S * C3) * C1 - Tr(C2^T * S * C3) * C2 ,
270 : !> which means we do not need the variable S_evects to keep the matrix products S * Ci .
271 : !>
272 : !> In reality, however, we deal with limited float-point precision arithmetic meaning that
273 : !> the trace Tr(C2^T * S * C1) is close to 0 but does not equal to 0 exactly. The term
274 : !> Tr(C3^T * S * C1) * Tr(C2^T * S * C1) * C2
275 : !> can not be ignored anymore. Ignorance of this term will lead to numerical instability
276 : !> when the trace Tr(C3^T * S * C1) is large enough.
277 : !> \endparblock
278 : ! **************************************************************************************************
279 6564 : SUBROUTINE tddfpt_orthonormalize_psi1_psi1(evects, nvects_new, S_evects, matrix_s)
280 : TYPE(cp_fm_type), DIMENSION(:, :), INTENT(in) :: evects
281 : INTEGER, INTENT(in) :: nvects_new
282 : TYPE(cp_fm_type), DIMENSION(:, :), INTENT(INOUT) :: S_evects
283 : TYPE(dbcsr_type), POINTER :: matrix_s
284 :
285 : CHARACTER(LEN=*), PARAMETER :: routineN = 'tddfpt_orthonormalize_psi1_psi1'
286 :
287 : INTEGER :: handle, ispin, ivect, jvect, nspins, &
288 : nvects_old, nvects_total
289 : INTEGER, DIMENSION(maxspins) :: nactive
290 : REAL(kind=dp) :: norm
291 : REAL(kind=dp), DIMENSION(maxspins) :: weights
292 :
293 6564 : CALL timeset(routineN, handle)
294 :
295 6564 : nspins = SIZE(evects, 1)
296 6564 : nvects_total = SIZE(evects, 2)
297 6564 : nvects_old = nvects_total - nvects_new
298 :
299 : IF (debug_this_module) THEN
300 : CPASSERT(SIZE(S_evects, 1) == nspins)
301 : CPASSERT(SIZE(S_evects, 2) == nvects_total)
302 : CPASSERT(nvects_old >= 0)
303 : END IF
304 :
305 13968 : DO ispin = 1, nspins
306 13968 : CALL cp_fm_get_info(matrix=evects(ispin, 1), ncol_global=nactive(ispin))
307 : END DO
308 :
309 23772 : DO jvect = nvects_old + 1, nvects_total
310 : ! Orthogonalization <psi1_i | psi1_j>
311 149172 : DO ivect = 1, jvect - 1
312 131964 : CALL cp_fm_trace(evects(:, jvect), S_evects(:, ivect), weights(1:nspins), accurate=.FALSE.)
313 286942 : norm = SUM(weights(1:nspins))
314 :
315 304150 : DO ispin = 1, nspins
316 286942 : CALL cp_fm_scale_and_add(1.0_dp, evects(ispin, jvect), -norm, evects(ispin, ivect))
317 : END DO
318 : END DO
319 :
320 : ! Normalization <psi1_j | psi1_j> = 1
321 36992 : DO ispin = 1, nspins
322 : CALL cp_dbcsr_sm_fm_multiply(matrix_s, evects(ispin, jvect), S_evects(ispin, jvect), &
323 36992 : ncol=nactive(ispin), alpha=1.0_dp, beta=0.0_dp)
324 : END DO
325 :
326 17208 : CALL cp_fm_trace(evects(:, jvect), S_evects(:, jvect), weights(1:nspins), accurate=.FALSE.)
327 :
328 36992 : norm = SUM(weights(1:nspins))
329 17208 : norm = 1.0_dp/SQRT(norm)
330 :
331 43556 : DO ispin = 1, nspins
332 19784 : CALL cp_fm_scale(norm, evects(ispin, jvect))
333 36992 : CALL cp_fm_scale(norm, S_evects(ispin, jvect))
334 : END DO
335 : END DO
336 :
337 6564 : CALL timestop(handle)
338 :
339 6564 : END SUBROUTINE tddfpt_orthonormalize_psi1_psi1
340 :
341 : ! **************************************************************************************************
342 : !> \brief Compute action matrix-vector products.
343 : !> \param Aop_evects action of TDDFPT operator on trial vectors (modified on exit)
344 : !> \param evects TDDFPT trial vectors
345 : !> \param S_evects cached matrix product S * evects where S is the overlap matrix
346 : !> in primary basis set
347 : !> \param gs_mos molecular orbitals optimised for the ground state
348 : !> \param tddfpt_control control section for tddfpt
349 : !> \param matrix_ks Kohn-Sham matrix
350 : !> \param qs_env Quickstep environment
351 : !> \param kernel_env kernel environment
352 : !> \param sub_env parallel (sub)group environment
353 : !> \param work_matrices collection of work matrices (modified on exit)
354 : !> \param matrix_s ...
355 : !> \par History
356 : !> * 06.2016 created [Sergey Chulkov]
357 : !> * 03.2017 refactored [Sergey Chulkov]
358 : ! **************************************************************************************************
359 5432 : SUBROUTINE tddfpt_compute_Aop_evects(Aop_evects, evects, S_evects, gs_mos, tddfpt_control, &
360 5432 : matrix_ks, qs_env, kernel_env, &
361 : sub_env, work_matrices, matrix_s)
362 : TYPE(cp_fm_type), DIMENSION(:, :), INTENT(INOUT) :: Aop_evects
363 : TYPE(cp_fm_type), DIMENSION(:, :), INTENT(IN) :: evects, S_evects
364 : TYPE(tddfpt_ground_state_mos), DIMENSION(:), &
365 : INTENT(in) :: gs_mos
366 : TYPE(tddfpt2_control_type), POINTER :: tddfpt_control
367 : TYPE(dbcsr_p_type), DIMENSION(:), INTENT(in) :: matrix_ks
368 : TYPE(qs_environment_type), POINTER :: qs_env
369 : TYPE(kernel_env_type), INTENT(in) :: kernel_env
370 : TYPE(tddfpt_subgroup_env_type), INTENT(in) :: sub_env
371 : TYPE(tddfpt_work_matrices), INTENT(inout) :: work_matrices
372 : TYPE(dbcsr_type), POINTER :: matrix_s
373 :
374 : CHARACTER(LEN=*), PARAMETER :: routineN = 'tddfpt_compute_Aop_evects'
375 :
376 : INTEGER :: handle, ispin, ivect, nspins, nvects
377 : INTEGER, DIMENSION(maxspins) :: nmo_occ
378 : LOGICAL :: do_admm, do_bse, do_hfx, &
379 : do_lri_response, is_rks_triplets, &
380 : re_int
381 : REAL(KIND=dp) :: rcut, scale
382 : TYPE(cp_fm_type) :: fm_dummy
383 : TYPE(full_kernel_env_type), POINTER :: kernel_env_admm_aux
384 : TYPE(mp_para_env_type), POINTER :: para_env
385 :
386 5432 : CALL timeset(routineN, handle)
387 :
388 5432 : nspins = SIZE(gs_mos, 1)
389 5432 : nvects = SIZE(evects, 2)
390 5432 : do_hfx = tddfpt_control%do_hfx
391 5432 : do_admm = tddfpt_control%do_admm
392 5432 : IF (do_admm) THEN
393 642 : kernel_env_admm_aux => kernel_env%admm_kernel
394 : ELSE
395 4790 : NULLIFY (kernel_env_admm_aux)
396 : END IF
397 5432 : is_rks_triplets = tddfpt_control%rks_triplets
398 5432 : do_lri_response = tddfpt_control%do_lrigpw
399 5432 : do_bse = tddfpt_control%do_bse
400 5432 : IF (do_bse) do_hfx = .FALSE.
401 :
402 : IF (debug_this_module) THEN
403 : CPASSERT(nspins > 0)
404 : CPASSERT(SIZE(Aop_evects, 1) == SIZE(evects, 1))
405 : CPASSERT(SIZE(S_evects, 1) == SIZE(evects, 1))
406 : CPASSERT(SIZE(Aop_evects, 2) == nvects)
407 : CPASSERT(SIZE(S_evects, 2) == nvects)
408 : CPASSERT(SIZE(gs_mos) == nspins)
409 : END IF
410 :
411 11670 : DO ispin = 1, nspins
412 5432 : nmo_occ(ispin) = SIZE(gs_mos(ispin)%evals_occ)
413 : END DO
414 :
415 5432 : IF (nvects > 0) THEN
416 5432 : CALL cp_fm_get_info(evects(1, 1), para_env=para_env)
417 5432 : IF (ALLOCATED(work_matrices%evects_sub)) THEN
418 24 : DO ivect = 1, nvects
419 40 : DO ispin = 1, SIZE(evects, 1)
420 16 : ASSOCIATE (evect => evects(ispin, ivect), work_matrix => work_matrices%evects_sub(ispin, ivect))
421 16 : IF (ASSOCIATED(evect%matrix_struct)) THEN
422 16 : IF (ASSOCIATED(work_matrix%matrix_struct)) THEN
423 8 : CALL cp_fm_copy_general(evect, work_matrix, para_env)
424 : ELSE
425 8 : CALL cp_fm_copy_general(evect, fm_dummy, para_env)
426 : END IF
427 0 : ELSE IF (ASSOCIATED(work_matrix%matrix_struct)) THEN
428 0 : CALL cp_fm_copy_general(fm_dummy, work_matrix, para_env)
429 : ELSE
430 0 : CALL cp_fm_copy_general(fm_dummy, fm_dummy, para_env)
431 : END IF
432 : END ASSOCIATE
433 : END DO
434 : END DO
435 : END IF
436 :
437 5432 : IF (tddfpt_control%kernel == tddfpt_kernel_full) THEN
438 : ! full TDDFPT kernel
439 : CALL fhxc_kernel(Aop_evects, evects, is_rks_triplets, do_hfx, do_admm, qs_env, &
440 : kernel_env%full_kernel, kernel_env_admm_aux, sub_env, work_matrices, &
441 : tddfpt_control%admm_symm, tddfpt_control%admm_xc_correction, &
442 3074 : do_lri_response, tddfpt_mgrid=tddfpt_control%mgrid_is_explicit)
443 2358 : ELSE IF (tddfpt_control%kernel == tddfpt_kernel_stda) THEN
444 : ! sTDA kernel
445 : CALL stda_kernel(Aop_evects, evects, is_rks_triplets, qs_env, tddfpt_control%stda_control, &
446 2264 : kernel_env%stda_kernel, sub_env, work_matrices)
447 94 : ELSE IF (tddfpt_control%kernel == tddfpt_kernel_none) THEN
448 : ! No kernel
449 340 : DO ivect = 1, nvects
450 586 : DO ispin = 1, SIZE(evects, 1)
451 492 : CALL cp_fm_set_all(Aop_evects(ispin, ivect), 0.0_dp)
452 : END DO
453 : END DO
454 : ELSE
455 0 : CPABORT("Kernel type not implemented")
456 : END IF
457 :
458 5432 : IF (ALLOCATED(work_matrices%evects_sub)) THEN
459 24 : DO ivect = 1, nvects
460 40 : DO ispin = 1, SIZE(evects, 1)
461 : ASSOCIATE (Aop_evect => Aop_evects(ispin, ivect), &
462 16 : work_matrix => work_matrices%Aop_evects_sub(ispin, ivect))
463 16 : IF (ASSOCIATED(Aop_evect%matrix_struct)) THEN
464 16 : IF (ASSOCIATED(work_matrix%matrix_struct)) THEN
465 8 : CALL cp_fm_copy_general(work_matrix, Aop_evect, para_env)
466 : ELSE
467 8 : CALL cp_fm_copy_general(fm_dummy, Aop_evect, para_env)
468 : END IF
469 0 : ELSE IF (ASSOCIATED(work_matrix%matrix_struct)) THEN
470 0 : CALL cp_fm_copy_general(work_matrix, fm_dummy, para_env)
471 : ELSE
472 0 : CALL cp_fm_copy_general(fm_dummy, fm_dummy, para_env)
473 : END IF
474 : END ASSOCIATE
475 : END DO
476 : END DO
477 : END IF
478 :
479 : ! orbital energy difference term
480 5432 : IF (.NOT. do_bse) THEN
481 : CALL tddfpt_apply_energy_diff(Aop_evects=Aop_evects, evects=evects, S_evects=S_evects, &
482 5432 : gs_mos=gs_mos, matrix_ks=matrix_ks, tddfpt_control=tddfpt_control)
483 : ELSE
484 : CALL zeroth_order_gw(qs_env=qs_env, Aop_evects=Aop_evects, evects=evects, S_evects=S_evects, &
485 0 : gs_mos=gs_mos, matrix_s=matrix_s, matrix_ks=matrix_ks)
486 : END IF
487 :
488 : ! if smeared occupation, then add aCCSX here
489 5432 : IF (tddfpt_control%do_smearing) THEN
490 24 : DO ivect = 1, nvects
491 36 : DO ispin = 1, nspins
492 : CALL add_smearing_aterm(qs_env, Aop_evects(ispin, ivect), evects(ispin, ivect), &
493 : S_evects(ispin, ivect), gs_mos(ispin)%mos_occ, &
494 24 : tddfpt_control%smeared_occup(ispin)%fermia, matrix_s)
495 : END DO
496 : END DO
497 : END IF
498 :
499 5432 : IF (do_hfx) THEN
500 1194 : IF (tddfpt_control%kernel == tddfpt_kernel_full) THEN
501 : ! full TDDFPT kernel
502 : CALL tddfpt_apply_hfx(Aop_evects=Aop_evects, evects=evects, gs_mos=gs_mos, do_admm=do_admm, &
503 : qs_env=qs_env, wfm_rho_orb=work_matrices%hfx_fm_ao_ao, &
504 : work_hmat_symm=work_matrices%hfx_hmat_symm, &
505 : work_hmat_asymm=work_matrices%hfx_hmat_asymm, &
506 : work_rho_ia_ao_symm=work_matrices%hfx_rho_ao_symm, &
507 1194 : work_rho_ia_ao_asymm=work_matrices%hfx_rho_ao_asymm)
508 0 : ELSE IF (tddfpt_control%kernel == tddfpt_kernel_stda) THEN
509 : ! sTDA kernel
510 : ! special treatment of HFX term
511 0 : ELSE IF (tddfpt_control%kernel == tddfpt_kernel_none) THEN
512 : ! No kernel
513 : ! drop kernel contribution of HFX term
514 : ELSE
515 0 : CPABORT("Kernel type not implemented")
516 : END IF
517 : END IF
518 : ! short/long range HFX
519 5432 : IF (tddfpt_control%kernel == tddfpt_kernel_full) THEN
520 3074 : IF (tddfpt_control%do_hfxsr) THEN
521 22 : re_int = tddfpt_control%hfxsr_re_int
522 : ! symmetric dmat
523 : CALL tddfpt_apply_hfxsr_kernel(Aop_evects, evects, gs_mos, qs_env, &
524 : kernel_env%full_kernel%admm_env, &
525 : kernel_env%full_kernel%hfxsr_section, &
526 : kernel_env%full_kernel%x_data, 1, re_int, &
527 : work_rho_ia_ao=work_matrices%hfxsr_rho_ao_symm, &
528 : work_hmat=work_matrices%hfxsr_hmat_symm, &
529 22 : wfm_rho_orb=work_matrices%hfxsr_fm_ao_ao)
530 : ! antisymmetric dmat
531 : CALL tddfpt_apply_hfxsr_kernel(Aop_evects, evects, gs_mos, qs_env, &
532 : kernel_env%full_kernel%admm_env, &
533 : kernel_env%full_kernel%hfxsr_section, &
534 : kernel_env%full_kernel%x_data, -1, .FALSE., &
535 : work_rho_ia_ao=work_matrices%hfxsr_rho_ao_asymm, &
536 : work_hmat=work_matrices%hfxsr_hmat_asymm, &
537 22 : wfm_rho_orb=work_matrices%hfxsr_fm_ao_ao)
538 22 : tddfpt_control%hfxsr_re_int = .FALSE.
539 : END IF
540 3074 : IF (tddfpt_control%do_hfxlr) THEN
541 36 : rcut = tddfpt_control%hfxlr_rcut
542 36 : scale = tddfpt_control%hfxlr_scale
543 108 : DO ivect = 1, nvects
544 108 : IF (ALLOCATED(work_matrices%evects_sub)) THEN
545 0 : IF (ASSOCIATED(work_matrices%evects_sub(1, ivect)%matrix_struct)) THEN
546 : CALL tddfpt_apply_hfxlr_kernel(qs_env, sub_env, rcut, scale, work_matrices, &
547 : work_matrices%evects_sub(:, ivect), &
548 0 : work_matrices%Aop_evects_sub(:, ivect))
549 : ELSE
550 : ! skip trial vectors which are assigned to different parallel groups
551 : CYCLE
552 : END IF
553 : ELSE
554 : CALL tddfpt_apply_hfxlr_kernel(qs_env, sub_env, rcut, scale, work_matrices, &
555 72 : evects(:, ivect), Aop_evects(:, ivect))
556 : END IF
557 : END DO
558 : END IF
559 : END IF
560 5432 : IF (do_bse) THEN
561 : ! add dynamical screening
562 0 : CALL tddfpt_apply_bse(Aop_evects=Aop_evects, evects=evects, gs_mos=gs_mos, qs_env=qs_env)
563 : END IF
564 :
565 : END IF
566 :
567 5432 : CALL timestop(handle)
568 :
569 5432 : END SUBROUTINE tddfpt_compute_Aop_evects
570 :
571 : ! **************************************************************************************************
572 : !> \brief Solve eigenproblem for the reduced action matrix and find new Ritz eigenvectors and
573 : !> eigenvalues.
574 : !> \param ritz_vects Ritz eigenvectors (initialised on exit)
575 : !> \param Aop_ritz approximate action of TDDFPT operator on Ritz vectors (initialised on exit)
576 : !> \param evals Ritz eigenvalues (initialised on exit)
577 : !> \param krylov_vects Krylov's vectors
578 : !> \param Aop_krylov action of TDDFPT operator on Krylov's vectors
579 : !> \param Atilde TDDFPT matrix projected into the Krylov's vectors subspace
580 : !> \param nkvo ...
581 : !> \param nkvn ...
582 : !> \par History
583 : !> * 06.2016 created [Sergey Chulkov]
584 : !> * 03.2017 altered prototype, OpenMP parallelisation [Sergey Chulkov]
585 : ! **************************************************************************************************
586 5432 : SUBROUTINE tddfpt_compute_ritz_vects(ritz_vects, Aop_ritz, evals, krylov_vects, Aop_krylov, &
587 : Atilde, nkvo, nkvn)
588 : TYPE(cp_fm_type), DIMENSION(:, :), INTENT(IN) :: ritz_vects, Aop_ritz
589 : REAL(kind=dp), DIMENSION(:), INTENT(out) :: evals
590 : TYPE(cp_fm_type), DIMENSION(:, :), INTENT(IN) :: krylov_vects, Aop_krylov
591 : REAL(kind=dp), DIMENSION(:, :), POINTER :: Atilde
592 : INTEGER, INTENT(IN) :: nkvo, nkvn
593 :
594 : CHARACTER(LEN=*), PARAMETER :: routineN = 'tddfpt_compute_ritz_vects'
595 :
596 : INTEGER :: handle, ikv, irv, ispin, nkvs, nrvs, &
597 : nspins
598 : REAL(kind=dp) :: act
599 5432 : REAL(kind=dp), ALLOCATABLE, DIMENSION(:, :) :: at12, at21, at22, evects_Atilde
600 : TYPE(cp_blacs_env_type), POINTER :: blacs_env_global
601 : TYPE(cp_fm_struct_type), POINTER :: fm_struct
602 : TYPE(cp_fm_type) :: Atilde_fm, evects_Atilde_fm
603 :
604 5432 : CALL timeset(routineN, handle)
605 :
606 5432 : nspins = SIZE(krylov_vects, 1)
607 5432 : nkvs = SIZE(krylov_vects, 2)
608 5432 : nrvs = SIZE(ritz_vects, 2)
609 5432 : CPASSERT(nkvs == nkvo + nkvn)
610 :
611 5432 : CALL cp_fm_get_info(krylov_vects(1, 1), context=blacs_env_global)
612 :
613 5432 : CALL cp_fm_struct_create(fm_struct, nrow_global=nkvs, ncol_global=nkvs, context=blacs_env_global)
614 5432 : CALL cp_fm_create(Atilde_fm, fm_struct, set_zero=.TRUE.)
615 5432 : CALL cp_fm_create(evects_Atilde_fm, fm_struct, set_zero=.TRUE.)
616 5432 : CALL cp_fm_struct_release(fm_struct)
617 :
618 : ! *** compute upper-diagonal reduced action matrix ***
619 5432 : CALL reallocate(Atilde, 1, nkvs, 1, nkvs)
620 : ! TO DO: the subroutine 'cp_fm_contracted_trace' will compute all elements of
621 : ! the matrix 'Atilde', however only upper-triangular elements are actually needed
622 : !
623 5432 : IF (nkvo == 0) THEN
624 : CALL cp_fm_contracted_trace(Aop_krylov(:, 1:nkvs), krylov_vects(:, 1:nkvs), &
625 1210 : Atilde(1:nkvs, 1:nkvs), accurate=.FALSE.)
626 : ELSE
627 37998 : ALLOCATE (at12(nkvn, nkvo), at21(nkvo, nkvn), at22(nkvn, nkvn))
628 : CALL cp_fm_contracted_trace(Aop_krylov(:, nkvo + 1:nkvs), krylov_vects(:, 1:nkvo), &
629 4222 : at12, accurate=.FALSE.)
630 144402 : Atilde(nkvo + 1:nkvs, 1:nkvo) = at12(1:nkvn, 1:nkvo)
631 : CALL cp_fm_contracted_trace(Aop_krylov(:, 1:nkvo), krylov_vects(:, nkvo + 1:nkvs), &
632 4222 : at21, accurate=.FALSE.)
633 120676 : Atilde(1:nkvo, nkvo + 1:nkvs) = at21(1:nkvo, 1:nkvn)
634 : CALL cp_fm_contracted_trace(Aop_krylov(:, nkvo + 1:nkvs), krylov_vects(:, nkvo + 1:nkvs), &
635 4222 : at22, accurate=.FALSE.)
636 54586 : Atilde(nkvo + 1:nkvs, nkvo + 1:nkvs) = at22(1:nkvn, 1:nkvn)
637 4222 : DEALLOCATE (at12, at21, at22)
638 : END IF
639 1679352 : Atilde = 0.5_dp*(Atilde + TRANSPOSE(Atilde))
640 5432 : CALL cp_fm_set_submatrix(Atilde_fm, Atilde)
641 :
642 : ! *** solve an eigenproblem for the reduced matrix ***
643 5432 : CALL choose_eigv_solver(Atilde_fm, evects_Atilde_fm, evals(1:nkvs))
644 :
645 21728 : ALLOCATE (evects_Atilde(nkvs, nrvs))
646 5432 : CALL cp_fm_get_submatrix(evects_Atilde_fm, evects_Atilde, start_row=1, start_col=1, n_rows=nkvs, n_cols=nrvs)
647 5432 : CALL cp_fm_release(evects_Atilde_fm)
648 5432 : CALL cp_fm_release(Atilde_fm)
649 :
650 : !$OMP PARALLEL DO DEFAULT(NONE), &
651 : !$OMP PRIVATE(act, ikv, irv, ispin), &
652 5432 : !$OMP SHARED(Aop_krylov, Aop_ritz, krylov_vects, evects_Atilde, nkvs, nrvs, nspins, ritz_vects)
653 : DO irv = 1, nrvs
654 : DO ispin = 1, nspins
655 : CALL cp_fm_set_all(ritz_vects(ispin, irv), 0.0_dp)
656 : CALL cp_fm_set_all(Aop_ritz(ispin, irv), 0.0_dp)
657 : END DO
658 :
659 : DO ikv = 1, nkvs
660 : act = evects_Atilde(ikv, irv)
661 : DO ispin = 1, nspins
662 : CALL cp_fm_scale_and_add(1.0_dp, ritz_vects(ispin, irv), &
663 : act, krylov_vects(ispin, ikv))
664 : CALL cp_fm_scale_and_add(1.0_dp, Aop_ritz(ispin, irv), &
665 : act, Aop_krylov(ispin, ikv))
666 : END DO
667 : END DO
668 : END DO
669 : !$OMP END PARALLEL DO
670 :
671 5432 : DEALLOCATE (evects_Atilde)
672 :
673 5432 : CALL timestop(handle)
674 :
675 10864 : END SUBROUTINE tddfpt_compute_ritz_vects
676 :
677 : ! **************************************************************************************************
678 : !> \brief Expand Krylov space by computing residual vectors.
679 : !> \param residual_vects residual vectors (modified on exit)
680 : !> \param evals Ritz eigenvalues (modified on exit)
681 : !> \param ritz_vects Ritz eigenvectors
682 : !> \param Aop_ritz approximate action of TDDFPT operator on Ritz vectors
683 : !> \param gs_mos molecular orbitals optimised for the ground state
684 : !> \param matrix_s overlap matrix
685 : !> \param tddfpt_control ...
686 : !> \par History
687 : !> * 06.2016 created [Sergey Chulkov]
688 : !> * 03.2017 refactored to achieve significant performance gain [Sergey Chulkov]
689 : ! **************************************************************************************************
690 4230 : SUBROUTINE tddfpt_compute_residual_vects(residual_vects, evals, ritz_vects, Aop_ritz, gs_mos, &
691 : matrix_s, tddfpt_control)
692 : TYPE(cp_fm_type), DIMENSION(:, :), INTENT(in) :: residual_vects
693 : REAL(kind=dp), DIMENSION(:), INTENT(in) :: evals
694 : TYPE(cp_fm_type), DIMENSION(:, :), INTENT(in) :: ritz_vects, Aop_ritz
695 : TYPE(tddfpt_ground_state_mos), DIMENSION(:), &
696 : INTENT(in) :: gs_mos
697 : TYPE(dbcsr_type), POINTER :: matrix_s
698 : TYPE(tddfpt2_control_type), POINTER :: tddfpt_control
699 :
700 : CHARACTER(LEN=*), PARAMETER :: routineN = 'tddfpt_compute_residual_vects'
701 : REAL(kind=dp), PARAMETER :: eref_scale = 0.99_dp, threshold = 16.0_dp*EPSILON(1.0_dp)
702 :
703 : INTEGER :: handle, ica, icb, icol_local, &
704 : irow_local, irv, ispin, nao, &
705 : ncols_local, nrows_local, nrvs, &
706 : nspins, spin2, spinflip
707 4230 : INTEGER, DIMENSION(:), POINTER :: col_indices_local, row_indices_local
708 : INTEGER, DIMENSION(maxspins) :: nactive, nmo_virt
709 : REAL(kind=dp) :: e_occ_plus_lambda, eref, lambda
710 : REAL(kind=dp), CONTIGUOUS, DIMENSION(:, :), &
711 4230 : POINTER :: weights_ldata
712 : TYPE(cp_fm_struct_type), POINTER :: ao_mo_struct, virt_mo_struct
713 4230 : TYPE(cp_fm_type), ALLOCATABLE, DIMENSION(:) :: awork, vomat
714 :
715 4230 : CALL timeset(routineN, handle)
716 :
717 4230 : nspins = SIZE(residual_vects, 1)
718 4230 : nrvs = SIZE(residual_vects, 2)
719 4230 : spinflip = tddfpt_control%spinflip
720 :
721 4230 : IF (nrvs > 0) THEN
722 4230 : CALL dbcsr_get_info(matrix_s, nfullrows_total=nao)
723 30770 : ALLOCATE (awork(nspins), vomat(nspins))
724 9040 : DO ispin = 1, nspins
725 4810 : IF (spinflip == no_sf_tddfpt) THEN
726 : spin2 = ispin
727 : ELSE
728 74 : spin2 = 2
729 : END IF
730 4810 : nmo_virt(spin2) = SIZE(gs_mos(spin2)%evals_virt)
731 : !
732 : CALL cp_fm_get_info(matrix=ritz_vects(ispin, 1), matrix_struct=ao_mo_struct, &
733 4810 : ncol_global=nactive(ispin))
734 4810 : CALL cp_fm_create(awork(ispin), ao_mo_struct)
735 : !
736 : CALL cp_fm_struct_create(virt_mo_struct, nrow_global=nmo_virt(spin2), &
737 4810 : ncol_global=nactive(ispin), template_fmstruct=ao_mo_struct)
738 4810 : CALL cp_fm_create(vomat(ispin), virt_mo_struct)
739 9040 : CALL cp_fm_struct_release(virt_mo_struct)
740 : END DO
741 :
742 : ! *** actually compute residual vectors ***
743 14866 : DO irv = 1, nrvs
744 10636 : lambda = evals(irv)
745 :
746 27270 : DO ispin = 1, nspins
747 12404 : IF (spinflip == no_sf_tddfpt) THEN
748 : spin2 = ispin
749 : ELSE
750 338 : spin2 = 2
751 : END IF
752 : CALL cp_fm_get_info(vomat(ispin), nrow_local=nrows_local, &
753 : ncol_local=ncols_local, row_indices=row_indices_local, &
754 12404 : col_indices=col_indices_local, local_data=weights_ldata)
755 :
756 : ! awork := Ab(ispin, irv) - evals(irv) b(ispin, irv), where 'b' is a Ritz vector
757 : CALL cp_dbcsr_sm_fm_multiply(matrix_s, ritz_vects(ispin, irv), awork(ispin), &
758 12404 : ncol=nactive(ispin), alpha=-lambda, beta=0.0_dp)
759 12404 : CALL cp_fm_scale_and_add(1.0_dp, awork(ispin), 1.0_dp, Aop_ritz(ispin, irv))
760 : !
761 : CALL parallel_gemm('T', 'N', nmo_virt(spin2), nactive(ispin), nao, 1.0_dp, gs_mos(spin2)%mos_virt, &
762 12404 : awork(ispin), 0.0_dp, vomat(ispin))
763 :
764 : ! Apply Davidson preconditioner to the residue vectors vomat to obtain new directions
765 108898 : DO icol_local = 1, ncols_local
766 96494 : ica = col_indices_local(icol_local)
767 96494 : icb = gs_mos(ispin)%index_active(ica)
768 96494 : e_occ_plus_lambda = gs_mos(ispin)%evals_occ(icb) + lambda
769 :
770 3428112 : DO irow_local = 1, nrows_local
771 3319214 : eref = gs_mos(spin2)%evals_virt(row_indices_local(irow_local)) - e_occ_plus_lambda
772 :
773 : ! eref = e_virt - e_occ - lambda = e_virt - e_occ - (eref_scale*lambda + (1-eref_scale)*lambda);
774 : ! eref_new = e_virt - e_occ - eref_scale*lambda = eref + (1 - eref_scale)*lambda
775 3319214 : IF (ABS(eref) < threshold) &
776 74 : eref = eref + (1.0_dp - eref_scale)*lambda
777 :
778 3415708 : weights_ldata(irow_local, icol_local) = weights_ldata(irow_local, icol_local)/eref
779 : END DO
780 : END DO
781 :
782 : CALL parallel_gemm('N', 'N', nao, nactive(ispin), nmo_virt(spin2), 1.0_dp, gs_mos(spin2)%mos_virt, &
783 35444 : vomat(ispin), 0.0_dp, residual_vects(ispin, irv))
784 : END DO
785 : END DO
786 : !
787 4230 : CALL cp_fm_release(awork)
788 8460 : CALL cp_fm_release(vomat)
789 : END IF
790 :
791 4230 : CALL timestop(handle)
792 :
793 8460 : END SUBROUTINE tddfpt_compute_residual_vects
794 :
795 : ! **************************************************************************************************
796 : !> \brief Perform Davidson iterations.
797 : !> \param evects TDDFPT trial vectors (modified on exit)
798 : !> \param evals TDDFPT eigenvalues (modified on exit)
799 : !> \param S_evects cached matrix product S * evects (modified on exit)
800 : !> \param gs_mos molecular orbitals optimised for the ground state
801 : !> \param tddfpt_control TDDFPT control parameters
802 : !> \param matrix_ks Kohn-Sham matrix
803 : !> \param qs_env Quickstep environment
804 : !> \param kernel_env kernel environment
805 : !> \param sub_env parallel (sub)group environment
806 : !> \param logger CP2K logger
807 : !> \param iter_unit I/O unit to write basic iteration information
808 : !> \param energy_unit I/O unit to write detailed energy information
809 : !> \param tddfpt_print_section TDDFPT print input section (need to write TDDFPT restart files)
810 : !> \param work_matrices collection of work matrices (modified on exit)
811 : !> \return energy convergence achieved (in Hartree)
812 : !> \par History
813 : !> * 03.2017 code related to Davidson eigensolver has been moved here from the main subroutine
814 : !> tddfpt() [Sergey Chulkov]
815 : !> \note Based on the subroutines apply_op() and iterative_solver() originally created by
816 : !> Thomas Chassaing in 2002.
817 : ! **************************************************************************************************
818 1210 : FUNCTION tddfpt_davidson_solver(evects, evals, S_evects, gs_mos, tddfpt_control, &
819 : matrix_ks, qs_env, kernel_env, &
820 : sub_env, logger, iter_unit, energy_unit, &
821 : tddfpt_print_section, work_matrices) RESULT(conv)
822 : TYPE(cp_fm_type), DIMENSION(:, :), INTENT(inout) :: evects
823 : REAL(kind=dp), DIMENSION(:), INTENT(inout) :: evals
824 : TYPE(cp_fm_type), DIMENSION(:, :), INTENT(inout) :: S_evects
825 : TYPE(tddfpt_ground_state_mos), DIMENSION(:), &
826 : INTENT(in) :: gs_mos
827 : TYPE(tddfpt2_control_type), POINTER :: tddfpt_control
828 : TYPE(dbcsr_p_type), DIMENSION(:), POINTER :: matrix_ks
829 : TYPE(qs_environment_type), POINTER :: qs_env
830 : TYPE(kernel_env_type), INTENT(in) :: kernel_env
831 : TYPE(tddfpt_subgroup_env_type), INTENT(in) :: sub_env
832 : TYPE(cp_logger_type), POINTER :: logger
833 : INTEGER, INTENT(in) :: iter_unit, energy_unit
834 : TYPE(section_vals_type), POINTER :: tddfpt_print_section
835 : TYPE(tddfpt_work_matrices), INTENT(inout) :: work_matrices
836 : REAL(kind=dp) :: conv
837 :
838 : CHARACTER(LEN=*), PARAMETER :: routineN = 'tddfpt_davidson_solver'
839 :
840 : INTEGER :: handle, ispin, istate, iter, &
841 : max_krylov_vects, nspins, nstates, &
842 : nstates_conv, nvects_exist, nvects_new
843 : INTEGER(kind=int_8) :: nstates_total
844 : LOGICAL :: is_nonortho
845 : REAL(kind=dp) :: t1, t2
846 1210 : REAL(kind=dp), ALLOCATABLE, DIMENSION(:) :: evals_last
847 1210 : REAL(kind=dp), DIMENSION(:, :), POINTER :: Atilde
848 1210 : TYPE(cp_fm_type), ALLOCATABLE, DIMENSION(:, :) :: Aop_krylov, Aop_ritz, krylov_vects, &
849 1210 : S_krylov
850 1210 : TYPE(dbcsr_p_type), DIMENSION(:), POINTER :: matrix_s
851 :
852 1210 : CALL timeset(routineN, handle)
853 :
854 1210 : nspins = SIZE(evects, 1)
855 1210 : nstates = tddfpt_control%nstates
856 1210 : nstates_total = tddfpt_total_number_of_states(tddfpt_control, gs_mos)
857 :
858 : IF (debug_this_module) THEN
859 : CPASSERT(SIZE(evects, 1) == nspins)
860 : CPASSERT(SIZE(evects, 2) == nstates)
861 : CPASSERT(SIZE(evals) == nstates)
862 : END IF
863 :
864 1210 : CALL get_qs_env(qs_env, matrix_s=matrix_s)
865 :
866 : ! adjust the number of Krylov vectors
867 1210 : max_krylov_vects = MIN(MAX(tddfpt_control%nkvs, nstates), INT(nstates_total))
868 :
869 12240 : ALLOCATE (Aop_ritz(nspins, nstates))
870 4708 : DO istate = 1, nstates
871 8610 : DO ispin = 1, nspins
872 7400 : CALL fm_pool_create_fm(work_matrices%fm_pool_ao_mo_active(ispin)%pool, Aop_ritz(ispin, istate))
873 : END DO
874 : END DO
875 :
876 3630 : ALLOCATE (evals_last(max_krylov_vects))
877 : ALLOCATE (Aop_krylov(nspins, max_krylov_vects), krylov_vects(nspins, max_krylov_vects), &
878 845858 : S_krylov(nspins, max_krylov_vects))
879 :
880 4708 : DO istate = 1, nstates
881 8610 : DO ispin = 1, nspins
882 3902 : CALL fm_pool_create_fm(work_matrices%fm_pool_ao_mo_active(ispin)%pool, krylov_vects(ispin, istate))
883 3902 : CALL cp_fm_to_fm(evects(ispin, istate), krylov_vects(ispin, istate))
884 :
885 3902 : CALL fm_pool_create_fm(work_matrices%fm_pool_ao_mo_active(ispin)%pool, S_krylov(ispin, istate))
886 3902 : CALL cp_fm_to_fm(S_evects(ispin, istate), S_krylov(ispin, istate))
887 :
888 7400 : CALL fm_pool_create_fm(work_matrices%fm_pool_ao_mo_active(ispin)%pool, Aop_krylov(ispin, istate))
889 : END DO
890 : END DO
891 :
892 1210 : nvects_exist = 0
893 1210 : nvects_new = nstates
894 :
895 1210 : t1 = m_walltime()
896 :
897 1210 : ALLOCATE (Atilde(1, 1))
898 :
899 5432 : DO
900 : ! davidson iteration
901 5432 : CALL cp_iterate(logger%iter_info, iter_nr_out=iter)
902 :
903 : ! Matrix-vector operations
904 : CALL tddfpt_compute_Aop_evects(Aop_evects=Aop_krylov(:, nvects_exist + 1:nvects_exist + nvects_new), &
905 : evects=krylov_vects(:, nvects_exist + 1:nvects_exist + nvects_new), &
906 : S_evects=S_krylov(:, nvects_exist + 1:nvects_exist + nvects_new), &
907 : gs_mos=gs_mos, tddfpt_control=tddfpt_control, &
908 : matrix_ks=matrix_ks, &
909 : qs_env=qs_env, kernel_env=kernel_env, &
910 : sub_env=sub_env, &
911 : work_matrices=work_matrices, &
912 5432 : matrix_s=matrix_s(1)%matrix)
913 :
914 : CALL tddfpt_compute_ritz_vects(ritz_vects=evects, Aop_ritz=Aop_ritz, &
915 : evals=evals_last(1:nvects_exist + nvects_new), &
916 : krylov_vects=krylov_vects(:, 1:nvects_exist + nvects_new), &
917 : Aop_krylov=Aop_krylov(:, 1:nvects_exist + nvects_new), &
918 5432 : Atilde=Atilde, nkvo=nvects_exist, nkvn=nvects_new)
919 :
920 : CALL tddfpt_write_restart(evects=evects, evals=evals_last(1:nstates), gs_mos=gs_mos, &
921 5432 : logger=logger, tddfpt_print_section=tddfpt_print_section)
922 :
923 25662 : conv = MAXVAL(ABS(evals_last(1:nstates) - evals(1:nstates)))
924 :
925 5432 : nvects_exist = nvects_exist + nvects_new
926 5432 : IF (nvects_exist + nvects_new > max_krylov_vects) &
927 402 : nvects_new = max_krylov_vects - nvects_exist
928 5432 : IF (iter >= tddfpt_control%niters) nvects_new = 0
929 :
930 5432 : IF (conv > tddfpt_control%conv .AND. nvects_new > 0) THEN
931 : ! compute residual vectors for the next iteration
932 14866 : DO istate = 1, nvects_new
933 27270 : DO ispin = 1, nspins
934 : CALL fm_pool_create_fm(work_matrices%fm_pool_ao_mo_active(ispin)%pool, &
935 12404 : krylov_vects(ispin, nvects_exist + istate))
936 : CALL fm_pool_create_fm(work_matrices%fm_pool_ao_mo_active(ispin)%pool, &
937 12404 : S_krylov(ispin, nvects_exist + istate))
938 : CALL fm_pool_create_fm(work_matrices%fm_pool_ao_mo_active(ispin)%pool, &
939 23040 : Aop_krylov(ispin, nvects_exist + istate))
940 : END DO
941 : END DO
942 :
943 : CALL tddfpt_compute_residual_vects(residual_vects=krylov_vects(:, nvects_exist + 1:nvects_exist + nvects_new), &
944 : evals=evals_last(1:nvects_new), &
945 : ritz_vects=evects(:, 1:nvects_new), Aop_ritz=Aop_ritz(:, 1:nvects_new), &
946 4230 : gs_mos=gs_mos, matrix_s=matrix_s(1)%matrix, tddfpt_control=tddfpt_control)
947 :
948 : CALL tddfpt_orthogonalize_psi1_psi0(krylov_vects(:, nvects_exist + 1:nvects_exist + nvects_new), &
949 : work_matrices%S_C0_C0T, qs_env, &
950 4230 : gs_mos, evals(1:nstates), tddfpt_control, work_matrices%S_C0)
951 :
952 : CALL tddfpt_orthonormalize_psi1_psi1(krylov_vects(:, 1:nvects_exist + nvects_new), nvects_new, &
953 4230 : S_krylov(:, 1:nvects_exist + nvects_new), matrix_s(1)%matrix)
954 :
955 : is_nonortho = tddfpt_is_nonorthogonal_psi1_psi0(krylov_vects(:, nvects_exist + 1:nvects_exist + nvects_new), &
956 : work_matrices%S_C0, tddfpt_control%orthogonal_eps, &
957 4230 : tddfpt_control%spinflip)
958 : ELSE
959 : ! convergence or the maximum number of Krylov vectors have been achieved
960 1202 : nvects_new = 0
961 1202 : is_nonortho = .FALSE.
962 : END IF
963 :
964 5432 : t2 = m_walltime()
965 5432 : IF (energy_unit > 0) THEN
966 315 : WRITE (energy_unit, '(/,4X,A,T14,A,T36,A)') "State", "Exc. energy (eV)", "Convergence (eV)"
967 727 : DO istate = 1, nstates
968 412 : WRITE (energy_unit, '(1X,I8,T12,F14.7,T38,ES11.4)') istate, &
969 1139 : evals_last(istate)*evolt, (evals_last(istate) - evals(istate))*evolt
970 : END DO
971 315 : WRITE (energy_unit, *)
972 315 : CALL m_flush(energy_unit)
973 : END IF
974 :
975 5432 : IF (iter_unit > 0) THEN
976 2716 : nstates_conv = 0
977 10115 : DO istate = 1, nstates
978 7399 : IF (ABS(evals_last(istate) - evals(istate)) <= tddfpt_control%conv) &
979 5164 : nstates_conv = nstates_conv + 1
980 : END DO
981 :
982 2716 : WRITE (iter_unit, '(T7,I8,T24,F7.1,T40,ES11.4,T66,I8)') iter, t2 - t1, conv, nstates_conv
983 2716 : CALL m_flush(iter_unit)
984 : END IF
985 :
986 20230 : t1 = t2
987 20230 : evals(1:nstates) = evals_last(1:nstates)
988 :
989 : ! nvects_new == 0 if iter >= tddfpt_control%niters
990 5432 : IF (nvects_new == 0 .OR. is_nonortho) THEN
991 : ! restart Davidson iterations
992 : CALL tddfpt_orthogonalize_psi1_psi0(evects, work_matrices%S_C0_C0T, qs_env, &
993 : gs_mos, &
994 1210 : evals(1:nstates), tddfpt_control, work_matrices%S_C0)
995 1210 : CALL tddfpt_orthonormalize_psi1_psi1(evects, nstates, S_evects, matrix_s(1)%matrix)
996 :
997 : EXIT
998 : END IF
999 : END DO
1000 :
1001 1210 : DEALLOCATE (Atilde)
1002 :
1003 15344 : DO istate = nvects_exist + nvects_new, 1, -1
1004 31650 : DO ispin = nspins, 1, -1
1005 16306 : CALL fm_pool_give_back_fm(work_matrices%fm_pool_ao_mo_active(ispin)%pool, Aop_krylov(ispin, istate))
1006 16306 : CALL fm_pool_give_back_fm(work_matrices%fm_pool_ao_mo_active(ispin)%pool, S_krylov(ispin, istate))
1007 30440 : CALL fm_pool_give_back_fm(work_matrices%fm_pool_ao_mo_active(ispin)%pool, krylov_vects(ispin, istate))
1008 : END DO
1009 : END DO
1010 1210 : DEALLOCATE (Aop_krylov, krylov_vects, S_krylov)
1011 1210 : DEALLOCATE (evals_last)
1012 :
1013 4708 : DO istate = nstates, 1, -1
1014 8610 : DO ispin = nspins, 1, -1
1015 7400 : CALL fm_pool_give_back_fm(work_matrices%fm_pool_ao_mo_active(ispin)%pool, Aop_ritz(ispin, istate))
1016 : END DO
1017 : END DO
1018 1210 : DEALLOCATE (Aop_ritz)
1019 :
1020 1210 : CALL timestop(handle)
1021 :
1022 1210 : END FUNCTION tddfpt_davidson_solver
1023 :
1024 : END MODULE qs_tddfpt2_eigensolver
|
