Branch data Line data Source code
1 : : !!****m* ABINIT/m_gwpt
2 : : !! NAME
3 : : !! m_gwpt
4 : : !!
5 : : !! FUNCTION
6 : : !! Compute electron-phonon matrix elements within the GWPT formalism.
7 : : !!
8 : : !! COPYRIGHT
9 : : !! Copyright (C) 2008-2026 ABINIT group (MG)
10 : : !! This file is distributed under the terms of the
11 : : !! GNU General Public License, see ~abinit/COPYING
12 : : !! or http://www.gnu.org/copyleft/gpl.txt.
13 : : !!
14 : : !! SOURCE
15 : :
16 : : #if defined HAVE_CONFIG_H
17 : : #include "config.h"
18 : : #endif
19 : :
20 : : #include "abi_common.h"
21 : :
22 : : module m_gwpt
23 : :
24 : : use, intrinsic :: iso_c_binding
25 : : use defs_basis
26 : : use m_abicore
27 : : USE_MPI
28 : : use m_xmpi
29 : : use m_mpinfo
30 : : use m_errors
31 : : use m_clib
32 : : use m_hide_blas
33 : : use m_copy
34 : : use m_fft
35 : : use m_hamiltonian
36 : : use m_pawcprj
37 : : use m_sort
38 : : use m_sigtk
39 : : use m_ephtk
40 : : use netcdf
41 : : use m_nctk
42 : : use m_mkffnl
43 : : use m_xcdata
44 : :
45 : : use m_gwdefs, only : czero_gw
46 : : use defs_abitypes, only : mpi_type
47 : : use defs_datatypes, only : pseudopotential_type
48 : : use m_array, only : array2_gwpc_t
49 : : use m_dtfil, only : datafiles_type
50 : : use m_dtset, only : dataset_type
51 : : use m_gwdefs, only : GW_Q0_DEFAULT
52 : : use m_time, only : cwtime, cwtime_report, timab, sec2str, timab
53 : : use m_fstrings, only : itoa, ftoa, sjoin, ktoa, ltoa, strcat
54 : : use m_numeric_tools, only : arth, c2r, r2c, get_diag, linfit, iseven, simpson_cplx, print_arr, inrange !, bins_t
55 : : use m_io_tools, only : iomode_from_fname
56 : : use m_fftcore, only : ngfft_seq, sphereboundary, print_ngfft
57 : : use m_fft_mesh, only : setmesh
58 : : use m_cgtk, only : cgtk_rotate, cgtk_change_gsphere
59 : : use m_crystal, only : crystal_t
60 : : use m_kpts, only : kpts_ibz_from_kptrlatt, kpts_timrev_from_kptopt, kpts_map
61 : : use m_kg, only : getph, mkkin
62 : : use m_bz_mesh, only : isamek, kmesh_t
63 : : use m_gsphere, only : gsphere_t
64 : : use m_getgh1c, only : getgh1c, rf_transgrid_and_pack
65 : : use m_ioarr, only : read_rhor
66 : : use m_hdr, only : hdr_type
67 : : use m_dvdb, only : dvdb_t
68 : : use m_ifc, only : ifc_type
69 : : use m_wfd, only : wfd_t
70 : : use m_pawang, only : pawang_type
71 : : use m_pawrad, only : pawrad_type
72 : : use m_pawtab, only : pawtab_type
73 : : use m_pawrhoij, only : pawrhoij_type
74 : : use m_pawfgr, only : pawfgr_type
75 : : use m_dfpt_cgwf, only : stern_t
76 : : use m_io_screening, only : hscr_t, get_hscr_qmesh_gsph, read_screening
77 : : use m_vcoul, only : vcoul_t
78 : : use m_gstore, only : gstore_t, gqk_t, gstore_check_restart, gstore_prepare_mpiio, &
79 : : GSTORE_KQ_MISSING, GSTORE_KQ_COMPUTED, GSTORE_GTYPE_GWPT, gstore_symmetrize
80 : : use m_rhotoxc, only : rhotoxc
81 : : use m_drivexc, only : check_kxc
82 : : use m_occ, only : get_fact_spin_tol_empty
83 : : use m_ebands, only : ebands_t
84 : : use m_pstat, only : pstat_proc
85 : : use m_screen, only : em1_symmetrize_op
86 : : use m_ppmodel, only : ppmodel_t
87 : : use m_lgroup, only : lgroup_t
88 : : use m_initylmg, only : initylmg_k
89 : :
90 : : implicit none
91 : :
92 : : private
93 : : !!***
94 : :
95 : : #ifdef HAVE_MPI1
96 : : include 'mpif.h'
97 : : #endif
98 : :
99 : : public :: gwpt_run ! Main entry point to compute GWPT e-ph matrix elements
100 : :
101 : : !----------------------------------------------------------------------
102 : :
103 : : contains !=====================================================
104 : : !!***
105 : :
106 : : !----------------------------------------------------------------------
107 : :
108 : : !!****f* m_gwpt/gwpt_run
109 : : !! NAME
110 : : !! gwpt_run
111 : : !!
112 : : !! FUNCTION
113 : : !! Compute e-ph matrix elements with the GWPT formalism.
114 : : !!
115 : : !! INPUTS
116 : : !! wfk0_path=String with the path to the GS unperturbed WFK file.
117 : : !! ngfft(18),ngfftf(18)=Coarse and Fine FFT meshes.
118 : : !! dtset<dataset_type>=All input variables for this dataset.
119 : : !! ebands<ebands_t>=The GS KS band structure (energies, occupancies, k-weights...)
120 : : !! dvdb=Database with the DFPT SCF potentials.
121 : : !! drhovdb=Database with the DFPT SCF densities.
122 : : !! ifc<ifc_type>=interatomic force constants and corresponding real space grid info.
123 : : !! wfk_hdr=Header of the WFK file.
124 : : !! pawfgr <type(pawfgr_type)>=fine grid parameters and related data
125 : : !! pawang<pawang_type)>=PAW angular mesh and related data.
126 : : !! pawrad(ntypat*usepaw)<pawrad_type>=Paw radial mesh and related data.
127 : : !! pawtab(ntypat*usepaw)<pawtab_type>=Paw tabulated starting data.
128 : : !! psps<pseudopotential_type>=Variables related to pseudopotentials.
129 : : !! comm=MPI communicator.
130 : : !!
131 : : !! NOTES
132 : : !!
133 : : !! 1) Conventions used for g-spheres and the periodic part of the KS states:
134 : : !!
135 : : !! _kq --> k + q
136 : : !! _kmp --> k - p
137 : : !! _kqmp --> k + q - p
138 : : !!
139 : : !! 2) The routines used to symmetrize wavefunctions and DFPT scattering potentials
140 : : !! expect in input symmetry tables generated using different conventions.
141 : : !! For the wavefunctions, we use the symrel convention while for the scattering potentials we use the symrec convention.
142 : : !! We encode this in the name of the variable using e.g. mapc_qq for the symrec convention (C) and mapl_k convention (L)
143 : : !!
144 : : !! 3) The DFPT routines operate on double-precision wavefunctions stored in arrays with real/imag part e.g. cg(1:2,npw_k)
145 : : !! while the GW routines operate on complex arrays of kind=gwp where gwp is defined at configure-time.
146 : : !! The default value of gwp is single-precision.
147 : : !! We use the following conventions for the buffers used to store the wavefunctions:
148 : : !!
149 : : !! cg_kq, cr_kq
150 : : !! cg1_kqmp, cr1_kqmp
151 : : !!
152 : : !! OUTPUT
153 : : !! GSTORE.nc file.
154 : : !!
155 : : !! TODO
156 : : !! Problems to be addressed:
157 : : !!
158 : : !! 1)
159 : : !! Sigma is usually split into Sigma_c(w) and Sigma_x where Sigma_x is the static Fock operator
160 : : !! evaluated with KS orbitals. The advantage of such partitioning is that Sigma_x = iGv
161 : : !! can be computed by summing over occupied states only. Sigma_x requires more G-vectors to converge
162 : : !! as the bare Coulomb interaction goes as 1/|q+G|^2 that is not integrable in 3D but this "expensive"
163 : : !! operations are needed only inside a sum over bands that is restricted to occupied states.
164 : : !! On the other hand, Sigma_c(w) is way more expensive as we have to sum a large number of empty states
165 : : !! while taking the w-dependence of the screening into account.
166 : : !! Fortunately, all the operations can be restricted to a small G-sphere of kinetic energy ecuteps that can be handled
167 : : !! with a coarser FFT mesh.
168 : : !! Another distinct advantage of such splitting is that one can handle the divergence in v(q,G) for |q+G| --> 0
169 : : !! using well know techniques from GW and the anisotropic behavior of eps-1(q) for q --> 0 in low-dimensional systems.
170 : : !! The disavantage is that one needs to compute the GWPT e-ph matrix in two steps, first Sigma_c and then Sigma_x,
171 : : !! so certain operations such as the k-point mapping, and the computation of the form factors are performed twice
172 : : !! Note, however, that MG believes that Sigma_x is a much better approximation than v_xc when one is interested
173 : : !! in the e-ph matrix elements connecting low-energy states such as band edges to high-energy states.
174 : : !!
175 : : !! 2)
176 : : !! We need to solve the NSCF Sternheimer for q and -q. In principle one can solve the equation only at q
177 : : !! and then use spatial inversion or TR to get the solution at -q but this requires solving the Sternheimer
178 : : !! for all the pp wavevectors in the BZ (or better in the IBZ_{q,k,alpha}. The use of symmetries is rendered complicated
179 : : !! by the parallelism over pp but perhaps one can precompute \Delta psi with all MPI procs and write the results to temporary files.
180 : : !!
181 : : !! 3) Check gstore_brange less than nband
182 : : !!
183 : : !! Debugging options:
184 : : !! useria = 1 # write gvals=gxc to GSTORE.nc
185 : : !! useria = 0 (default) # write gvals=gks-gxc+gsigx+gsigc to GSTORE.nc
186 : : !! useria = -1 # write gvals=gsigx (when userid=0) + gsigc (when useric=0) to GSTORE.nc
187 : : !!
188 : : !! userib = 0 (default) # calculate all k and q
189 : : !! userib = 1 # filter k and q, only calculate k=Lambda, and q=L (for comparesion with finite difference GW)
190 : : !!
191 : : !! userra = 0 (default) # neglect the dG0/de additional term at q=Gamma
192 : : !! userra = 1 # include the additional term at q=Gamma arising from the first-order change of the KS eigenvalues
193 : : !!
194 : : !! userrb = 0 (default) # sum over all n'' terms in chi, i.e. the basis-invariant degenerate-block form -chi_{n'} dWtilde/de
195 : : !! userrb = 1 # debug: sum over only the n'' = n' term in chi, i.e. the non-degenerate form -(d eps_{n'}) dWtilde/de
196 : : !!
197 : : !! useric = 0 (default) # include the correlation part of Sigma
198 : : !! useric = 1 # exclude the correlation part of Sigma
199 : : !!
200 : : !! userid = 0 (default) # include the exchange part of Sigma
201 : : !! userid = 1 # exclude the exchange part of Sigma
202 : : !!
203 : : !! userie = 1 : Use only pp = Gamma
204 : : !! userie = 0 (default): Use all pp of the given p-grid
205 : : !! userie = -1 : Exclude pp = Gamma
206 : : !!
207 : : !! SOURCE
208 : :
209 : 3 : subroutine gwpt_run(wfk0_path, dtfil, ngfft, ngfftf, dtset, cryst, ebands, dvdb, drhodb, ifc, wfk_hdr, &
210 : 3 : pawfgr, pawang, pawrad, pawtab, psps, mpi_enreg, comm)
211 : :
212 : : !Arguments ------------------------------------
213 : : !scalars
214 : : character(len=*),intent(in) :: wfk0_path
215 : : integer,intent(in) :: comm
216 : : type(datafiles_type),intent(in) :: dtfil
217 : : type(dataset_type),intent(inout) :: dtset
218 : : type(crystal_t),intent(in) :: cryst
219 : : type(ebands_t),target,intent(in) :: ebands
220 : : type(dvdb_t),intent(inout) :: dvdb, drhodb
221 : : type(pawang_type),intent(in) :: pawang
222 : : type(pseudopotential_type),intent(in) :: psps
223 : : type(pawfgr_type),intent(in) :: pawfgr
224 : : type(ifc_type),intent(in) :: ifc
225 : : type(hdr_type),intent(in) :: wfk_hdr
226 : : type(mpi_type),intent(inout) :: mpi_enreg
227 : : !arrays
228 : : integer,intent(in) :: ngfft(18), ngfftf(18)
229 : : type(pawrad_type),intent(in) :: pawrad(psps%ntypat*psps%usepaw)
230 : : type(pawtab_type),intent(in) :: pawtab(psps%ntypat*psps%usepaw)
231 : :
232 : : !Local variables ------------------------------
233 : : !scalars
234 : : integer,parameter :: LOG_MODQ = 1, LOG_MODK = 4, LOG_MODP = 4, ENOUGH_STERN = 5
235 : : integer,parameter :: tim_getgh1c1 = 1, berryopt0 = 0, ider0 = 0, idir0 = 0, istwfk1 = 1, cplex1 = 1, pawread0 = 0
236 : : integer,parameter :: master = 0, ndat1 = 1, with_cplex0 = 0, n3xccc0 = 0, optder0 = 0
237 : : integer :: band, band_me, nband_me, stern_comm, nkpt, my_rank, nsppol, iq_ibz, iq_bz, my_npert
238 : : integer :: nb_k, nb_kq, bstart_k, bstop_k, bstart_kq, bstop_kq, matblk, method, enforce_sym
239 : : integer :: cplex,drho_cplex,nkxc,nk3xc,option,usexcnhat,db_iqpt,natom,natom3,ipc,nspinor,nproc
240 : : integer :: ib_sum, ii, u1_band !,u1c_ib_k, jj, iw !ib_kq, band_ks, ib_k, ibsum_kq, u1_master, ip
241 : : integer :: my_is, spin, idir,ipert, ig, max_npw_xc, min_npw_xc, npw_x, npw_c, nw_nk, nw_mkq
242 : : integer :: isym_q, trev_q, ip_ibz
243 : : integer :: ik_ibz, isym_k, trev_k, npw_k, istwf_k, npw_k_ibz, istwf_k_ibz, ik_glob, ik_bz
244 : : integer :: ikq_ibz, isym_kq, trev_kq, npw_kq, istwf_kq, npw_kq_ibz, istwf_kq_ibz
245 : : integer :: ikmp_ibz, isym_kmp, trev_kmp, npw_kmp, istwf_kmp
246 : : integer :: ikqmp_ibz, isym_kqmp, trev_kqmp, npw_kqmp, istwf_kqmp, npw_kqmp_ibz, istwf_kqmp_ibz, mpw,ierr,nqbz,ncerr !,spad
247 : : integer :: n1,n2,n3,n4,n5,n6,nspden, mqmem, im_kq, m_kq, in_k, n_k, restart, root_ncid, spin_ncid, usecprj
248 : : integer :: nfft,nfftf,mgfft,mgfftf,nkpg_k,nkpg_kq,nkpg_kqmp,nkpg_kmp,imyp, cnt, nvloc, iw_nk, iw_mkq, ndone, nmiss, jb_sum
249 : : integer :: my_ipp, ipp_bz, ipp_ibz, isym_pp, itim_pp, comm_rpt, nqlwl, scr_iomode, stern_qq_ierr, stern_mq_ierr
250 : : integer :: qptopt, my_iq, my_ik, qbuf_size, iqbuf_cnt, timrev_k, timrev_q
251 : : real(dp) :: cpu_all, wall_all, gflops_all, cpu_qq, wall_qq, gflops_qq, cpu_kk, wall_kk, gflops_kk, cpu_pp, wall_pp, gflops_pp
252 : : real(dp) :: drude_plsmf, my_plsmf !, rtmp
253 : : real(dp) :: en_high, dyn_shift, e_high
254 : : real(dp) :: fact_spin, theta_mu_minus_e0i, tol_empty, tol_empty_in, e_nk, e_mkq, ebsum_kmp, ebsum_kqmp ! e0i
255 : : real(dp) :: eta_sum, delta_E, inv_delta_E, wgt_dg0
256 : 3 : real(dp),contiguous, pointer :: qp_ene(:,:,:), qp_occ(:,:,:)
257 : : real(dp) :: weight_q,bigexc,bigsxc,vxcavg ! ediff, eshift, q0rad, bz_vol
258 : : logical :: isirr_k, isirr_kq, isirr_kmp, isirr_kqmp, qq_is_gamma, pp_is_gamma, isirr_q, use_dg0
259 : : logical :: stern_use_cache, need_ftinterp, symmetrize, use_lgk
260 : : logical :: print_time_qq, print_time_kk, print_time_pp, non_magnetic_xc, need_x_kmp, need_x_kqmp, test_sigma
261 : : complex(dp) :: ieta !, idelta_sum
262 : 3 : type(wfd_t) :: wfd
263 : 3 : type(gs_hamiltonian_type) :: gs_ham_kqmp, gs_ham_kmp
264 : 3 : type(rf_hamiltonian_type) :: rf_ham_kqmp, rf_ham_kmp
265 [ + + + + : 378 : type(crystal_t) :: pot_cryst, den_cryst
+ + + + +
+ + + + +
+ + + + +
+ + + + +
+ + + + +
+ + + + +
+ + + + +
+ ]
266 : 6 : type(hdr_type) :: pot_hdr, den_hdr
267 : 3 : type(stern_t) :: stern_kmp, stern_kqmp
268 [ + + + + : 78 : type(kmesh_t) :: pp_mesh, kmesh
+ + + + ]
269 : 3 : type(gsphere_t),target :: gsph_x, gsph_c
270 [ + + ]: 12 : type(hscr_t),target :: hscr
271 [ + + + + : 66 : type(vcoul_t) :: vcp
+ + + + +
+ ]
272 [ + + ]: 12 : type(gstore_t),target :: gstore
273 : : type(gqk_t),pointer :: gqk
274 : 3 : type(lgroup_t) :: lg_myq
275 : : type(xcdata_type) :: xcdata
276 : 3 : type(ppmodel_t) :: ppm
277 : : character(len=fnlen) :: screen_filepath, gstore_filepath
278 : : character(len=5000) :: msg, qq_bz_string, kk_string, qkp_string, pp_string
279 : : character(len=500) :: init_mode
280 : : !arrays
281 : 6 : integer :: nbsum, my_bsum_start(dtset%nsppol), my_bsum_stop(dtset%nsppol), my_nbsum(dtset%nsppol)
282 : : integer :: g0_k(3), g0_q(3), g0_kq(3), g0_kmp(3), g0_kqmp(3), units(2), work_ngfft(18), gmax(3)
283 : : integer :: mapl_k(6), mapl_kq(6), mapl_kqmp(6), mapl_kmp(6), mapc_qq(6), mapc_qq2dvdb(6)
284 : 9 : integer :: my_pp_start_spin(dtset%nsppol), my_pp_stop_spin(dtset%nsppol), my_npp(dtset%nsppol)
285 : 3 : integer(i1b),allocatable :: itreatq_dvdb(:)
286 : 6 : integer,allocatable :: kg_k(:,:), kg_kq(:,:), kg_kmp(:,:), kg_kqmp(:,:), my_pp_inds(:)
287 : 3 : integer,allocatable :: gbound_k(:,:), gbound_kq(:,:), gbound_kmp(:,:), gbound_kqmp(:,:), gbound_c(:,:), gbound_x(:,:)
288 : 9 : integer,allocatable :: nband(:,:), wfd_istwfk(:), qibz2dvdb(:) ! count_bk(:,:),
289 : 6 : integer,allocatable :: iq_buf(:,:), done_qbz_spin(:,:), state_kq(:,:)
290 : 3 : integer(i1b),allocatable :: itreat_qibz(:)
291 : 3 : integer, contiguous, pointer :: kg_c(:,:), kg_x(:,:)
292 : : !real(dp) :: eig0nk !, cpu, wall, gflops !, cpu_q, wall_q, gflops_q, cpu_all, wall_all, gflops_all
293 : : complex(gwp) :: ctmp_gwpc, xdot_tmp, dg0_tmp
294 : : complex(dp) :: ctmp_dp
295 : : !type(bins_t) :: bins
296 : : !arrays
297 : 6 : real(dp) :: fermie1_idir_ipert(3,cryst%natom), ylmgr_dum(1,1,1), dum_nhat(0), dum_xccc3d(0), tsec(2)
298 : : real(dp) :: kk(3),kq(3),kk_ibz(3),kq_ibz(3), kqmp(3), kmp(3), pp(3), kmp_ibz(3), kqmp_ibz(3), qq_ibz(3), qq_bz(3)
299 : 6 : real(dp),allocatable :: qlwl(:,:), kpg_k(:,:),kpg_kq(:,:),kpg_kmp(:,:),kpg_kqmp(:,:)
300 : 9 : real(dp),allocatable :: ffnl_kmp(:,:,:,:),ffnl_kqmp(:,:,:,:), kinpw_kqmp(:), kinpw_kmp(:), ph3d_kqmp(:,:,:), ph3d_kmp(:,:,:)
301 : 3 : real(dp),allocatable, target :: vxc1_qq(:,:,:,:)
302 : 3 : real(dp),allocatable :: gxc_atm(:,:,:,:), gks_atm(:,:,:,:), gks_atm2(:,:,:,:), gsig_atm(:,:,:,:)
303 : 9 : real(dp),allocatable :: cg_work(:,:), ug_k(:,:), ug_kq(:,:), ph1d(:,:)
304 : 6 : real(dp),allocatable :: vlocal(:,:,:,:), vlocal1_qq(:,:,:,:,:), v1scf_qq(:,:,:,:), vlocal1_mqq(:,:,:,:,:), v1scf_mq(:,:,:,:)
305 : 9 : real(dp),allocatable :: ylm_kmp(:,:), ylm_kqmp(:,:), vtrial(:,:), work(:,:,:,:), rhor(:,:), vxc(:,:), kxc(:,:)
306 : 3 : real(dp),allocatable :: omegame0i_nk(:), omegame0i_mkq(:), omegas_nk(:), omegas_mkq(:)
307 : 3 : real(dp),allocatable :: my_gbuf(:,:,:,:,:,:), my_gbuf_ks(:,:,:,:,:,:)
308 : 9 : real(dp),allocatable :: cg_kmp(:,:), cg_kqmp(:,:), cg1_kqmp(:,:), cg1_kmp(:,:), full_cg1_kqmp(:,:), full_cg1_kmp(:,:), vxc_nk(:,:)
309 : 3 : complex(dp), contiguous, pointer :: cvxc1_qq_ptr(:,:,:)
310 : 6 : complex(gwp),allocatable :: ur_star_kmp(:), ur_star_kqmp(:), cwork_ur(:), rhotwg_c(:), rhotwg_x(:), vc_sqrt_gx(:)
311 : 6 : complex(gwp),allocatable :: full_ur1_kqmp(:), full_ur1_star_kmp(:), sigcme_nk(:), sigcme_mkq(:), ur_nk(:,:), ur_mkq(:,:)
312 : 3 : complex(gwp),allocatable :: vec_gwc_nk(:,:,:), vec_gwc_mkq(:,:,:), vec_gx_nk(:,:), vec_gx_mkq(:,:)
313 : 3 : complex(gwp),allocatable :: vec_comp_nk(:,:), vec_comp_mkq(:,:)
314 : 9 : complex(gwp),allocatable :: vec_dgwc_nk(:,:,:), chi_ug_kqmp(:), chi_ur_kqmp(:), rhotwg_chi(:)
315 : 3 : complex(gwp),allocatable :: vec_coh_nk(:,:), vec_coh_mkq(:,:), wc0_pbz(:,:)
316 : 9 : complex(gwp),allocatable :: botsq_pbz(:,:), otq_pbz(:,:), dmeig_pbz(:,:), epsm1_ggw(:,:,:)
317 : 3 : complex(gwp),allocatable :: trans_botsq_pbz(:,:), trans_otq_pbz(:,:), trans_dmeig_pbz(:,:), sigx_nk(:,:), sigce0_nk(:,:)
318 : 6 : logical,allocatable :: bks_mask(:,:,:), keep_ur(:,:,:)
319 : 3 : type(pawcprj_type),allocatable :: cwaveprj0(:,:), cwaveprj(:,:)
320 : 6 : type(pawrhoij_type),allocatable :: pot_pawrhoij(:), den_pawrhoij(:)
321 : 3 : type(lgroup_t),allocatable :: lg_myk(:)
322 : 3 : type(array2_gwpc_t),allocatable :: wc0_pibz(:)
323 : : !************************************************************************
324 : :
325 [ - + ]: 3 : if (psps%usepaw == 1) then
326 : 0 : ABI_ERROR("PAW not implemented")
327 : : ABI_UNUSED((/pawang%nsym, pawrad(1)%mesh_size/))
328 : : end if
329 : :
330 [ - + ]: 3 : ABI_CHECK_IEQ(dtset%nspinor, 1, "GWPT with nspinor 2 not coded")
331 [ - + ]: 3 : ABI_CHECK_IEQ(dtset%nsppol, 1, "GWPT with nsppol 2 not tested")
332 : :
333 : 3 : test_sigma = .False.
334 : : !test_sigma = my_rank == 2
335 : :
336 : : !dtset%useric = 1 ! exclude the correlation part of Sigma
337 : : !dtset%userid = 1 ! exclude the exchange part of Sigma
338 : :
339 : 3 : call cwtime(cpu_all, wall_all, gflops_all, "start")
340 : :
341 : : ! Copy important dimensions
342 [ + + ]: 9 : my_rank = xmpi_comm_rank(comm); nproc = xmpi_comm_size(comm); units = [std_out, ab_out]
343 : 3 : natom = cryst%natom; natom3 = 3 * natom; nsppol = ebands%nsppol; nspinor = ebands%nspinor; nspden = dtset%nspden
344 : 3 : ieta = +j_dpc * dtset%zcut
345 : : !idelta_sum = +j_dpc * dtset%elph2_imagden
346 : 3 : eta_sum = dtset%elph2_imagden
347 : :
348 : : ! Set tolerance used to decide if a band is empty
349 : : ! and normalization of theta_mu_minus_esum. If nsppol == 2, qp_occ $\in [0,1]$
350 : 3 : tol_empty_in = 0.01
351 : 3 : call get_fact_spin_tol_empty(nsppol, nspinor, tol_empty_in, fact_spin, tol_empty)
352 : :
353 : 3 : qp_ene => ebands%eig; qp_occ => ebands%occ
354 : :
355 : : ! Check if a previous GSTORE.nc file is present to restart the calculation if dtset%eph_restart == 1,
356 : : ! and use done_qbz_spin mask to cycle the loops below if restart /= 0.
357 [ + - ]: 3 : gstore_filepath = strcat(dtfil%filnam_ds(4), "_GSTORE.nc")
358 : :
359 : : !gstore_filepath = dtfil%filgstorein
360 [ - + ]: 3 : call gstore_check_restart(gstore_filepath, dtset, nqbz, done_qbz_spin, restart, comm)
361 : :
362 [ + - ]: 3 : if (restart == 0) then
363 : : ! Build new gstore object from dtset input variables.
364 : 3 : call gstore%init(gstore_filepath, dtset, dtfil, wfk_hdr, cryst, ebands, ifc, comm, gtype=GSTORE_GTYPE_GWPT)
365 [ - + - + : 12 : ABI_REMALLOC(done_qbz_spin, (gstore%nqbz, nsppol))
+ - - + +
- + - + -
- + - + -
+ ]
366 [ + + + + ]: 86 : done_qbz_spin = 0
367 : : else
368 : : ! Init gstore from pre-existent file. gstore_gname and read_dw are not relevant here.
369 : : call gstore%from_ncpath(gstore_filepath, with_cplex0, dtset, dtfil, cryst, ebands, ifc, &
370 : 0 : "atom", dtset%gstore_gname, .False., comm)
371 : : end if
372 : :
373 [ - + ]: 3 : if (nqbz /= 0) then
374 [ # # ]: 0 : ABI_CHECK_IEQ(nqbz, gstore%nqbz, "nqbz from file differs from gstore%nqbz")
375 : : else
376 : 3 : nqbz = gstore%nqbz
377 : : end if
378 : :
379 : : ! Open GSTORE.nc file and go to data mode.
380 : : ! SC: Open with xmpi_comm_self (independent IO) instead of comm, otherwise there will be a deadlock on lemaitre4
381 : : !NCF_CHECK(nctk_open_modify(root_ncid, gstore%path, xmpi_comm_self))
382 : : ! Follow-up: Open with comm again. xmpi_comm_self avoided a hang in the open caused by the MPI
383 : : ! stack of the lemaitre4 compute nodes, which has changed since. The run that hung in June 2026
384 : : ! (32 procs) no longer hangs (checked on the compute nodes on 2026-10-05).
385 [ - + ]: 3 : NCF_CHECK(nctk_open_modify(root_ncid, gstore%path, comm))
386 [ - + ]: 3 : NCF_CHECK(nctk_set_datamode(root_ncid))
387 : : ! SC: Load the metadata of all the variables now, with all the procs of comm.
388 : 3 : call gstore_prepare_mpiio(root_ncid, gstore%nsppol)
389 : :
390 : 3 : call gstore%get_missing_qbz_spin(done_qbz_spin, ndone, nmiss)
391 : : !call wrtout(units, sjoin("- Number of q-points/spin completed:", itoa(count(done_qbz_spin == 1)))
392 : :
393 : : ! TODO: Check restart capabilities. It seems they don't work properly.
394 : : ! NB: Write phonon data here as we are not guaranteed to have all the IBZ q-points
395 : : ! inside the loop over my_iq if filtering has been used.
396 : : ! Make sure internal table with gstore_done_qbz_spin is properly filled.
397 [ + - ]: 3 : if (ndone == 0) then
398 : 3 : call gstore%compute_and_write_ph(root_ncid)
399 : : else
400 : 0 : call wrtout(units, sjoin("- Restarting GSTORE calculation from file:", gstore%path))
401 : 0 : call wrtout(units, sjoin(" Found: ", itoa(ndone), " (qpt, spin) entries already computed"))
402 : : end if
403 : :
404 : : ! ================
405 : : ! HANDLE SCREENING
406 : : ! ================
407 : : ! Read g-sphere and pp_mesh from SCR file and init gsph_c for the correlated part.
408 : 3 : screen_filepath = dtfil%fnameabi_scr
409 [ - + ]: 3 : ABI_CHECK(dtfil%fnameabi_scr /= ABI_NOFILE, "SCR file must be specified")
410 [ - + ]: 3 : call get_hscr_qmesh_gsph(screen_filepath , dtset, cryst, hscr, pp_mesh, gsph_c, qlwl, comm)
411 : 3 : call hscr%print(units, dtset%prtvol, header="Header of the SCR file")
412 : :
413 : 3 : nqlwl = size(qlwl, dim=2)
414 [ - + ]: 3 : if (nqlwl == 0) then
415 : 0 : nqlwl=1
416 [ # # # # ]: 0 : ABI_MALLOC(qlwl,(3,nqlwl))
417 [ # # ]: 0 : qlwl(:,nqlwl)= GW_Q0_DEFAULT
418 : : write(msg,'(3a,i0,a,3f9.6)')&
419 : 0 : "The Header of the screening file does not contain the list of q-point for the optical limit ",ch10,&
420 : 0 : "Using nqlwl= ",nqlwl," and qlwl = ",qlwl(:,1)
421 : 0 : ABI_COMMENT(msg)
422 : : end if
423 : :
424 : : ! Init g-sphere for the exchange part from ecutsigx.
425 : 3 : call gsph_c%extend(cryst, dtset%ecutsigx, gsph_x)
426 : :
427 : 3 : call gsph_c%print(units, dtset%prtvol, header="Header of the correlation g-sphere")
428 : 3 : call gsph_x%print(units, dtset%prtvol, header="Header of the exchange g-sphere")
429 : :
430 : : ! TODO:
431 : : ! Here we sort the pp_mesh by stars so that we can split the pp wavevectors in blocks and therefore
432 : : ! reduce the number of wavevectors in the IBZ that must be stored in memory.
433 : : !call pp_mesh%pack_by_stars()
434 : :
435 : : ! Distribute the sum over pp wavevectors inside pp_sum_comm using block distribution.
436 [ + + + + ]: 9 : my_pp_start_spin = 1; my_pp_stop_spin = 0
437 [ + + ]: 6 : do my_is=1,gstore%my_nspins
438 : 3 : spin = gstore%my_spins(my_is); gqk => gstore%gqk(my_is)
439 [ - + ]: 3 : call xmpi_split_block(pp_mesh%nbz, gqk%pp_sum_comm%value, my_npp(spin), my_pp_inds)
440 [ + - ]: 3 : if (my_npp(spin) > 0) then
441 : 3 : my_pp_start_spin(spin) = my_pp_inds(1); my_pp_stop_spin(spin) = my_pp_inds(my_npp(spin))
442 : : end if
443 [ + - ]: 6 : ABI_SFREE(my_pp_inds)
444 : : end do ! my_is
445 : :
446 : : ! TODO: Use same FFT mesh as GW code but make sure that all potentials are properly interpolated.
447 : :
448 : : ! === Setup of the FFT mesh for the oscillator strengths ===
449 : : ! * Init gwc_ngfft(7:18) and gwx_ngfft(7:18) with Dtset%ngfft(7:18)
450 : : ! * Here we redefine gwc_ngfft(1:6) according to the following options:
451 : : !
452 : : ! method == 0 --> FFT grid read from fft.in (debugging purpose)
453 : : ! method == 1 --> Normal FFT mesh
454 : : ! method == 2 --> Slightly augmented FFT grid to calculate exactly rho_tw_g (see setmesh.F90)
455 : : ! method == 3 --> Doubled FFT grid, same as the the FFT for the density,
456 : : !
457 : : ! enforce_sym == 1 --> Enforce a FFT mesh compatible with all the symmetry operation and FFT library
458 : : ! enforce_sym == 0 --> Find the smallest FFT grid compatible with the library, do not care about symmetries
459 : : !
460 : : !gwc_ngfft(1:18) = dtset%ngfft(1:18)
461 : : !gwx_ngfft(1:18) = dtset%ngfft(1:18)
462 : :
463 : : method = 2
464 : 3 : if (dtset%fftgw == 00 .or. dtset%fftgw == 01) method = 0
465 : : if (dtset%fftgw == 10 .or. dtset%fftgw == 11) method = 1
466 : : if (dtset%fftgw == 20 .or. dtset%fftgw == 21) method = 2
467 : 3 : if (dtset%fftgw == 30 .or. dtset%fftgw == 31) method = 3
468 : 3 : enforce_sym = mod(dtset%fftgw, 10)
469 : :
470 : : #if 0
471 : : ! FFT mesh for sigma_x.
472 : : call setmesh(cryst%gmet, Gsph_Max%gvec, gwx_ngfft, Sigp%npwvec, Sigp%npwx, Sigp%npwwfn, &
473 : : gwx_nfftot, method, Sigp%mG0, cryst, enforce_sym)
474 : :
475 : : ! FFT mesh for sigma_c.
476 : : call setmesh(cryst%gmet, Gsph_Max%gvec, gwc_ngfft, Sigp%npwvec, epsm1%npwe, Sigp%npwwfn,&
477 : : gwc_nfftot, method, Sigp%mG0, cryst, enforce_sym, unit=dev_null)
478 : : #endif
479 : :
480 : : ! Initialize Coulomb term on the IBZ of the pp_mesh. Use largest G-sphere.
481 : 3 : call kmesh%init(cryst, wfk_hdr%nkpt, wfk_hdr%kptns, dtset%kptopt)
482 : :
483 : : ! Some required information are not filled up inside kmesh_init
484 : : ! So doing it here, even though it is not clean.
485 [ + + + + ]: 39 : Kmesh%kptrlatt(:,:) =Dtset%kptrlatt(:,:)
486 : 3 : Kmesh%nshift =Dtset%nshiftk
487 [ - + + - : 9 : ABI_MALLOC(Kmesh%shift,(3,Kmesh%nshift))
+ - + - -
+ - + -
+ ]
488 [ + + + + ]: 15 : Kmesh%shift(:,:) =Dtset%shiftk(:,1:Dtset%nshiftk)
489 : :
490 : 3 : npw_x = gsph_x%ng; npw_c = gsph_c%ng
491 : 3 : min_npw_xc = min(npw_x, npw_c)
492 : 3 : max_npw_xc = max(npw_x, npw_c)
493 [ + - ]: 3 : if (gsph_x%ng >= gsph_c%ng) then
494 : : call vcp%init(gsph_x, cryst, pp_mesh, kmesh, dtset%gw_rcut, dtset%gw_icutcoul, dtset%vcutgeo, dtset%ecuteps, gsph_x%ng, &
495 : 3 : nqlwl, qlwl, comm)
496 : : else
497 : : call vcp%init(gsph_c, cryst, pp_mesh, kmesh, dtset%gw_rcut, dtset%gw_icutcoul, dtset%vcutgeo, dtset%ecuteps, gsph_c%ng, &
498 : 0 : nqlwl, qlwl, comm)
499 : : end if
500 [ - + ]: 3 : ABI_FREE(qlwl)
501 : :
502 [ + - ]: 3 : if (my_rank == master) then
503 : 3 : call kmesh%print(units, header="K-mesh for wavefunctions", prtvol=dtset%prtvol)
504 : 3 : call pp_mesh%print(units, header="P-mesh for Screening", prtvol=dtset%prtvol)
505 : 3 : call gsph_x%print(units, dtset%prtvol, header="G-sphere for exchange")
506 : 3 : call gsph_c%print(units, dtset%prtvol, header="G-sphere for correlation")
507 : 3 : call vcp%print(units, prtvol=dtset%prtvol)
508 : : end if
509 : 3 : call kmesh%free()
510 : :
511 [ - + ]: 3 : ABI_CHECK_IGE(npw_x, 1, "npw_x <= 1")
512 [ - + ]: 3 : ABI_CHECK_IGE(npw_c, 1, "npw_c <= 1")
513 [ - + ]: 3 : ABI_CHECK_IGEQ(npw_x, npw_c, "npw_x < npw_c is not supported")
514 : :
515 : : ! Initialize the wave function descriptor.
516 : : ! Each node has all k-points and spins and all the nbsum bands: the Sternheimer solver needs the full set
517 : : ! of bands in stern%cgq, only the sum over bands is distributed inside bsum_comm (see my_bsum_start below).
518 : : ! TODO: One can exploit qq, kk and pp parallelism to find the wavevectors in the IBZ
519 : : ! that will be needed in the loops below and allocate only these wavevectors so that memory scales.
520 : :
521 [ + + + + ]: 9 : nbsum = dtset%mband; my_bsum_start = 1; my_bsum_stop = nbsum
522 : : ! FIXME: This term is needed in metals.
523 [ + + + + ]: 27 : fermie1_idir_ipert = zero
524 : :
525 : 3 : nkpt = wfk_hdr%nkpt
526 [ - + + - : 12 : ABI_MALLOC(nband, (nkpt, nsppol))
- + + - +
- + - - +
- + ]
527 [ - + + - : 15 : ABI_MALLOC(bks_mask, (dtset%mband, nkpt, nsppol))
- + + - -
+ + - + -
+ - - + -
+ ]
528 [ - + + - : 12 : ABI_MALLOC(keep_ur, (dtset%mband, nkpt, nsppol))
- + + - -
+ + - - +
- + ]
529 [ + + + + : 493 : nband = dtset%mband; bks_mask = .False.; keep_ur = .False.
+ + + + +
+ + + + +
+ + ]
530 : :
531 : : ! Distribute wavefunctions according to the set of kk, qq and pp wavevectors treated by this MPI proc.
532 : : ! Also, compute mpw and gmax including the additional pp
533 : : ! This is the maximum number of PWs for all possible k+q treated.
534 : : call gstore%fill_bks_mask_pp_mesh(dtset%ecut, dtset%mband, nkpt, nsppol, my_pp_start_spin, my_pp_stop_spin, pp_mesh, &
535 : 3 : my_bsum_start, my_bsum_stop, bks_mask, mpw, gmax)
536 : :
537 : : ! Distribute the sum over bands inside bsum_comm using block distribution.
538 : : ! Procs in the same pert_comm have the same bsum rank, hence the same bands (xmpi_sum inside the ib_sum loop).
539 [ + + + + ]: 9 : my_bsum_start = 1; my_bsum_stop = 0
540 [ + + ]: 6 : do my_is=1,gstore%my_nspins
541 : 3 : spin = gstore%my_spins(my_is); gqk => gstore%gqk(my_is)
542 : 6 : call xmpi_split_work(nbsum, gqk%bsum_comm%value, my_bsum_start(spin), my_bsum_stop(spin))
543 : : end do ! my_is
544 [ + + ]: 6 : my_nbsum = my_bsum_stop - my_bsum_start + 1
545 : :
546 : : ! Init work_ngfft
547 [ + + ]: 12 : gmax = gmax + 4 ! FIXME: this is to account for umklapp, should also consider Gamma-only and istwfk
548 [ + + ]: 12 : gmax = 2*gmax + 1
549 : : !gmax = 12*gmax + 1
550 : :
551 : 3 : call ngfft_seq(work_ngfft, gmax)
552 : : !print *, "gmax:", gmax
553 : : !write(std_out,*)"work_ngfft(1:3): ",work_ngfft(1:3)
554 [ - + + - : 15 : ABI_MALLOC(work, (2, work_ngfft(4), work_ngfft(5), work_ngfft(6)))
- + + - -
+ + - + -
+ - - + -
+ ]
555 : :
556 : : !if (dtset%userie == 124) then
557 : : ! ! Debugging section have all states on each MPI rank.
558 : : ! bks_mask = .True.; call wrtout(std_out, " Storing all bands for debugging purposes.")
559 : : !end if
560 : :
561 : : ! Impose istwfk=1 for all k points. This is also done in respfn (see inkpts)
562 : : ! wfd_read_wfk will handle a possible conversion if WFK contains istwfk /= 1.
563 [ - + + - : 9 : ABI_MALLOC(wfd_istwfk, (nkpt))
+ - + - -
+ - + ]
564 [ + + ]: 22 : wfd_istwfk = 1
565 : :
566 : : call wfd%init(cryst, pawtab, psps, keep_ur, dtset%mband, nband, nkpt, nsppol, bks_mask,&
567 : : nspden, nspinor, dtset%ecut, dtset%ecutsm, dtset%dilatmx, wfd_istwfk, ebands%kptns, ngfft,&
568 : 3 : dtset%nloalg, dtset%prtvol, dtset%pawprtvol, comm)
569 : :
570 [ - + ]: 3 : ABI_FREE(nband)
571 [ - + ]: 3 : ABI_FREE(bks_mask)
572 [ - + ]: 3 : ABI_FREE(keep_ur)
573 [ - + ]: 3 : ABI_FREE(wfd_istwfk)
574 : 3 : call pstat_proc%print(_PSTAT_ARGS_)
575 : :
576 : : ! Read wavefunctions.
577 : 3 : call wfd%read_wfk(wfk0_path, iomode_from_fname(wfk0_path))
578 : :
579 : : ! FFT meshes from input file, not necessarily equal to the ones found in the external files.
580 : : ! NB: ur arrays are always allocated with nfft and not with product(ngfft(4:6)).
581 [ + + ]: 12 : nfftf = product(ngfftf(1:3)); mgfftf = maxval(ngfftf(1:3))
582 [ + + + + ]: 21 : nfft = product(ngfft(1:3)) ; mgfft = maxval(ngfft(1:3))
583 : 3 : n1 = ngfft(1); n2 = ngfft(2); n3 = ngfft(3); n4 = ngfft(4); n5 = ngfft(5); n6 = ngfft(6)
584 : :
585 [ + + ]: 6 : call print_ngfft([std_out], ngfft, header="FFT mesh")
586 : :
587 : : ! Set the FFT mesh
588 : 3 : call wfd%change_ngfft(cryst, psps, ngfft)
589 : 3 : call wfd%print(units, header="Wavefunctions for GWPT calculation.")
590 : :
591 : 3 : usecprj = dtset%usepaw
592 [ - + + - : 12 : ABI_MALLOC(cwaveprj0, (natom, nspinor*usecprj))
+ - - - +
- - + - +
- + - + -
+ - - - -
- - - - ]
593 [ - + + - : 12 : ABI_MALLOC(cwaveprj, (natom, nspinor*usecprj))
+ - - - +
- - + - +
- + - + -
+ - - - -
- - - - ]
594 [ + - - + : 12 : ABI_MALLOC(gbound_k, (2*mgfft+8, 2))
+ - + - +
- + - - +
- + ]
595 [ - + + - : 6 : ABI_MALLOC(gbound_kq, (2*mgfft+8, 2))
- + - + ]
596 [ - + + - : 6 : ABI_MALLOC(gbound_kmp, (2*mgfft+8, 2))
- + - + ]
597 [ - + + - : 6 : ABI_MALLOC(gbound_kqmp, (2*mgfft+8, 2))
- + - + ]
598 [ - + + - : 6 : ABI_MALLOC(gbound_c, (2*mgfft+8, 2))
- + - + ]
599 [ - + + - : 6 : ABI_MALLOC(gbound_x, (2*mgfft+8, 2))
- + - + ]
600 [ - + + - : 9 : ABI_MALLOC(cg_work, (2, mpw*nspinor))
+ - + - -
+ - + ]
601 [ - + + - : 9 : ABI_MALLOC(full_ur1_kqmp, (nfft*nspinor))
+ - + - -
+ - + ]
602 [ - + + - : 6 : ABI_MALLOC(full_ur1_star_kmp, (nfft*nspinor))
- + - + ]
603 [ - + ]: 3 : if (dtset%userra /= zero) then
604 : : ! Real-space buffer for the auxiliary state chi_{n'k-p} entering the dG0/de term at q=Gamma (Appendix A).
605 [ # # # # : 0 : ABI_MALLOC(chi_ur_kqmp, (nfft*nspinor))
# # # # ]
606 : 0 : call wrtout(std_out, " Including the dG0/de term at q=Gamma (Appendix A) with the degenerate-block auxiliary state chi.")
607 : : end if
608 : :
609 : : ! ============================
610 : : ! Compute v_nk matrix elements
611 : : ! ============================
612 : : ! Diagonal elements of velocity operator in cartesian coordinates for all kk in the IBZ.
613 : : ! Use ndone to understand if velocities have been already compured in a previous run.
614 [ + - + - ]: 3 : if (gstore%with_vk /= 0 .and. ndone == 0) then
615 : 3 : call gstore%compute_and_write_vk(mpw, wfd, ebands, psps, pawtab, root_ncid)
616 : : end if
617 : :
618 : : ! Radius of sphere with volume equivalent to the micro zone.
619 : : !q0rad = two_pi * (three / (four_pi * cryst%ucvol * gstore%nqbz)) ** third
620 : : !bz_vol = two_pi**3 / cryst%ucvol
621 : :
622 : : ! Open the DVDB file with first-order potentials and drhodb with the first-order densities.
623 : 3 : call dvdb%open_read(ngfftf, xmpi_comm_self)
624 : 3 : call drhodb%open_read(ngfftf, xmpi_comm_self)
625 [ - + ]: 3 : ABI_CHECK(dvdb%has_fields("pot1", msg), msg)
626 [ - + ]: 3 : ABI_CHECK(drhodb%has_fields("den1", msg), msg)
627 : :
628 : : ! Make sure that dvdb and drhodb have the same q-points.
629 [ - + ]: 3 : ABI_CHECK_IEQ(dvdb%nqpt, drhodb%nqpt, "Different number of q-points in DVDB and DRHODB")
630 : 3 : ierr = 0
631 [ + + ]: 22 : do ii=1,dvdb%nqpt
632 [ + + + - : 79 : if (any(abs(dvdb%qpts(:, ii) - drhodb%qpts(:, ii)) > tol12)) then
- + ]
633 : 0 : ierr = ierr + 1; call wrtout(std_out, sjoin(ktoa(dvdb%qpts(:, ii)), " /= ", ktoa(drhodb%qpts(:, ii))))
634 : : end if
635 : : end do
636 [ - + ]: 3 : ABI_CHECK(ierr == 0, "Found different q-points in DVDB and DRHODB. See messages above!")
637 : :
638 : : ! Check if the q-points are present in the DVDB
639 : : ! qmap_symrec gives the mapping gstore%ibz --> dvdb%ibz
640 : : !call dvdb%need_ftinterp(gstore%nqibz, gstore%qibz, gstore%qptopt, qmap_symrec, need_ftinterp)
641 : : !ABI_FREE(qmap_symrec)
642 : : !need_ftinterp = .True.
643 : :
644 : : !if (.not. need_ftinterp .and. dtset%eph_need_ftinterp /= 0) then
645 : : ! ABI_WARNING("Enforcing FT interpolation for q-points even if it's not strictly needed.")
646 : : ! need_ftinterp = .True.
647 : : !end if
648 : :
649 : : !if (need_ftinterp) then
650 : : ! call wrtout(units, " Cannot find all q-points in the DVDB --> Activating Fourier interpolation.")
651 : : ! call dvdb%ftinterp_setup(dtset%ddb_ngqpt, qptopt, 1, dtset%ddb_shiftq, nfftf, ngfftf, xmpi_comm_self)
652 : : !else
653 : : ! call wrtout(units, " DVDB file contains all q-points along the path --> Reading DFPT potentials from file.")
654 : : !end if
655 : :
656 : : ! Activate parallelism over perturbations at the level of the DVDB, my_npert is output
657 : 3 : call gstore%set_perts_distrib(cryst, dvdb, my_npert)
658 : 3 : call gstore%set_perts_distrib(cryst, drhodb, my_npert)
659 : : !print *, "Treating my_npert", my_npert
660 : :
661 : : ! This part is taken from dfpt_vtorho
662 : : !==== Initialize most of the Hamiltonian (and derivative) ====
663 : : ! 1) Allocate all arrays and initialize quantities that do not depend on k and spin.
664 : : ! 2) Perform the setup needed for the non-local factors:
665 : : !
666 : : ! Norm-conserving: Constant kleimann-Bylander energies are copied from psps to gs_ham_kqmp.
667 : : ! PAW: Initialize the overlap coefficients and allocate the Dij coefficients.
668 : :
669 : : ! Get one-dimensional structure factor information on the coarse grid.
670 [ + - - + : 9 : ABI_MALLOC(ph1d, (2,3*(2*mgfft+1)*natom))
+ - + - +
- - + - +
- + ]
671 : 3 : call getph(cryst%atindx, natom, n1, n2, n3, ph1d, cryst%xred)
672 : :
673 : : call gs_ham_kqmp%init(psps, pawtab, nspinor, nsppol, nspden, natom, &
674 : : dtset%typat, cryst%xred, nfft, mgfft, ngfft, cryst%rprimd, dtset%nloalg, &
675 : : comm_atom=mpi_enreg%comm_atom, mpi_atmtab=mpi_enreg%my_atmtab, mpi_spintab=mpi_enreg%my_isppoltab, &
676 : 3 : usecprj=usecprj, ph1d=ph1d, nucdipmom=dtset%nucdipmom, gpu_option=dtset%gpu_option)
677 : :
678 : : call gs_ham_kmp%init(psps, pawtab, nspinor, nsppol, nspden, natom, &
679 : : dtset%typat, cryst%xred, nfft, mgfft, ngfft, cryst%rprimd, dtset%nloalg, &
680 : : comm_atom=mpi_enreg%comm_atom, mpi_atmtab=mpi_enreg%my_atmtab, mpi_spintab=mpi_enreg%my_isppoltab, &
681 : 3 : usecprj=usecprj, ph1d=ph1d, nucdipmom=dtset%nucdipmom, gpu_option=dtset%gpu_option)
682 : :
683 : 3 : matblk = gs_ham_kmp%matblk
684 : :
685 : : ! Allocate workspace arrays.
686 : : ! vtrial and vlocal are required for Sternheimer (H0). DFPT routines do not need it.
687 : : ! Note nvloc in vlocal (we will select one/four spin components afterwards)
688 : 3 : nvloc = gs_ham_kqmp%nvloc
689 [ - + + - : 14039 : ABI_CALLOC(vtrial, (nfftf, nspden))
- + + - +
- + - - +
- + + + +
+ ]
690 [ - + + - : 16703 : ABI_CALLOC(vlocal, (n4, n5, n6, nvloc))
- + + - -
+ + - - +
+ - + - +
- - + - +
- + + + +
+ + + +
+ ]
691 : :
692 : : ! Read the GS potential (vtrial) from input POT file
693 : : ! In principle one may store GS vtrial in the DVDB but getpot_filepath is simpler to implement.
694 : 3 : call wrtout(units, sjoin(" Reading KS GS potential for Sternheimer from: ", dtfil%filpotin))
695 : : call read_rhor(dtfil%filpotin, cplex1, nspden, nfftf, ngfftf, pawread0, mpi_enreg, vtrial, pot_hdr, pot_pawrhoij, comm, &
696 : 3 : allow_interp=.True., want_varname="vtrial")
697 [ - + - + : 3 : pot_cryst = pot_hdr%get_crystal()
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- + ]
698 [ - + ]: 3 : if (cryst%compare(pot_cryst, header=" Comparing input crystal with POT crystal") /= 0) then
699 : 0 : ABI_ERROR("Crystal structure from WFK and POT do not agree! Check messages above!")
700 : : end if
701 : 3 : call pot_cryst%free(); call pot_hdr%free()
702 : :
703 : : ! Find correspondence IBZ --> set of q-points in DVDB.
704 : : ! need_ftinterp selects whether DFPT potentials should be read from the DVDB or Fourier-interpolated on the fly.
705 : : ! Activate FT interpolation automatically if required q-points in the IBZ are not found in the DVDB.
706 : :
707 : : ! qibz2dvdb gives the mapping gstore%ibz --> dvdb%ibz
708 : : ! TODO: Make sure we have the same ibz in rho1%ibz
709 : 3 : need_ftinterp = .False.
710 [ - + + - : 9 : ABI_MALLOC(qibz2dvdb, (gstore%nqibz))
+ - + - -
+ - + ]
711 [ - + ]: 3 : if (dvdb%find_qpts(gstore%nqibz, gstore%qibz, qibz2dvdb, comm) /= 0) then
712 : 0 : call wrtout(units, " Cannot find eph_ngqpt_fine q-points in DVDB --> Activating Fourier interpolation.")
713 : 0 : need_ftinterp = .True.
714 : : else
715 : 3 : call wrtout(units, " DVDB file contains all q-points in the IBZ --> Reading DFPT potentials from file.")
716 : 3 : need_ftinterp = .False.
717 : : end if
718 : : !need_ftinterp = .True.
719 : :
720 : : ! Distribute DFPT potentials (IBZ q-points) inside qpt_comm.
721 : : ! Note that we distribute IBZ instead of the full BZ or the IBZ_k inside the loop over ikcalc.
722 : : ! This means that the load won't be equally distributed but memory will scale with qpt_comm%nproc.
723 : : ! To reduce load imbalance, we sort the qibz points by norm and use cyclic distribution inside qpt_comm
724 : :
725 : : ! itreat_qibz(nqibz)
726 : : ! Table used to distribute potentials over q-points in the IBZ.
727 : : ! The loop over qpts in the IBZ(k) is MPI distributed inside qpt_comm according to this table.
728 : : ! 0 if this IBZ point is not treated by this proc.
729 : : ! 1 if this IBZ is treated.
730 : :
731 : : ! TODO: Recheck this part
732 [ - + + - : 28 : ABI_ICALLOC(itreat_qibz, (gstore%nqibz))
+ - - + -
+ + + ]
733 [ + + ]: 22 : itreat_qibz = 1
734 : : !call sort_rpts(gstore%nqibz, gstore%qibz, cryst%gmet, iperm)
735 : : !do ii=1,gstore%nqibz
736 : : ! iq_ibz = iperm(ii)
737 : : ! do my_spin=1,gstore%my_nspins
738 : : ! gqk => gstore%gqk(my_spin)
739 : : ! if (mod(ii, gqk%qpt_comm%nproc) == gqk%qpt_comm%me) itreat_qibz(iq_ibz) = 1
740 : : ! end do
741 : : !end do
742 : : !ABI_FREE(iperm)
743 : :
744 [ + + + - ]: 22 : call wrtout(std_out, sjoin("P Number of q-points in the IBZ treated by this proc: " ,itoa(count(itreat_qibz == 1))))
745 : :
746 [ - + ]: 3 : if (need_ftinterp) then
747 : : ! Use ddb_ngqpt q-mesh to compute the real-space representation of DFPT v1scf_qq potentials to prepare Fourier interpolation.
748 : : ! R-points are distributed inside comm_rpt
749 : : ! Note that when R-points are distributed inside qpt_comm we cannot interpolate potentials on-the-fly
750 : : ! inside the loop over q-points.
751 : : ! In this case, indeed, the interpolation must be done in gwpt_setup_qloop once we know the q-points contributing
752 : : ! to the integral and the potentials must be cached.
753 : : !FIXME: qpt_comm is buggy.
754 : : !if (gwpt%imag_only) comm_rpt = xmpi_comm_self
755 : : !comm_rpt = gwpt%bsum_comm%value
756 : 0 : comm_rpt = xmpi_comm_self
757 [ # # ]: 0 : qptopt = ebands%kptopt; if (dtset%qptopt /= 0) qptopt = dtset%qptopt
758 : 0 : call dvdb%ftinterp_setup(dtset%ddb_ngqpt, qptopt, 1, dtset%ddb_shiftq, nfftf, ngfftf, comm_rpt)
759 : 0 : call drhodb%ftinterp_setup(dtset%ddb_ngqpt, qptopt, 1, dtset%ddb_shiftq, nfftf, ngfftf, comm_rpt)
760 : : end if
761 : :
762 [ + + ]: 6 : call dvdb%print([std_out], "DVDB file", dtset%prtvol)
763 [ + + ]: 6 : call drhodb%print([std_out], "DRHODB file", dtset%prtvol)
764 : 3 : call pstat_proc%print(_PSTAT_ARGS_)
765 : :
766 [ + - ]: 3 : if (.not. need_ftinterp) then
767 : : ! Need to translate itreat_qibz into itreatq_dvdb.
768 [ - + + - : 28 : ABI_ICALLOC(itreatq_dvdb, (dvdb%nqpt))
+ - - + -
+ + + ]
769 [ + + ]: 22 : do iq_ibz=1,gstore%nqibz
770 [ - + ]: 19 : if (itreat_qibz(iq_ibz) == 0) cycle
771 : 19 : db_iqpt = qibz2dvdb(iq_ibz)
772 [ - + ]: 19 : ABI_CHECK(db_iqpt /= -1, sjoin("Could not find IBZ q-point:", ktoa(gstore%qibz(:, iq_ibz)), "in the DVDB file."))
773 : 22 : itreatq_dvdb(db_iqpt) = 1
774 : : end do
775 [ - + ]: 3 : ABI_FREE(itreatq_dvdb)
776 : : end if
777 : :
778 [ - + ]: 3 : ABI_FREE(itreat_qibz)
779 [ - + ]: 3 : ABI_FREE(qibz2dvdb)
780 : :
781 : : ! Read the GS density needed to compute vxc1_qq.
782 [ - + + - : 14039 : ABI_CALLOC(rhor, (nfftf, nspden))
- + + - +
- + - - +
- + + + +
+ ]
783 : : call read_rhor(dtfil%fildensin, cplex1, nspden, nfftf, ngfftf, pawread0, mpi_enreg, rhor, den_hdr, den_pawrhoij, comm, &
784 : 3 : allow_interp=.True., want_varname="density")
785 [ - + - + : 3 : den_cryst = den_hdr%get_crystal()
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- + ]
786 [ - + ]: 3 : if (cryst%compare(den_cryst, header=" Comparing input crystal with DEN crystal") /= 0) then
787 : 0 : ABI_ERROR("Crystal structures from WFK and DEN do not agree! Check messages above!")
788 : : end if
789 : 3 : call den_cryst%free(); call den_hdr%free()
790 : :
791 : : ! Initialize plasmon-pole object.
792 : 3 : mqmem = pp_mesh%nibz
793 : 3 : drude_plsmf = sqrt(four_pi * ebands%nelect / cryst%ucvol)
794 [ - + ]: 3 : my_plsmf = drude_plsmf; if (dtset%ppmfrq > tol6) my_plsmf = dtset%ppmfrq
795 : 3 : call ppm%init(mqmem, pp_mesh%nibz, npw_c, dtset%ppmodel, my_plsmf, dtset%gw_invalid_freq)
796 [ + - ]: 3 : if (my_rank == master) call ppm%print(units)
797 : :
798 : : ! Completeness corrections for the states above nbsum: extrapolar (gwcomp 1) [[cite:Bruneval2008]] or
799 : : ! dynamic remainder (gwcomp -2). All the states above nbsum are given a common energy E_high.
800 : : ! With frozen screening, the closure term does not depend on the wavefunctions and its derivative vanishes:
801 : : ! only the partial sum over the nbsum states survives, i.e. the PPM term of an empty state at E_high with
802 : : ! the first-order wavefunctions, with a minus sign. The first-order change of E_high at q = Gamma is neglected.
803 : 3 : en_high = zero; dyn_shift = zero
804 [ - + ]: 3 : if (dtset%gwcomp == 1) then
805 : : ! Global E_high, as in m_sigc.
806 [ # # # # : 0 : en_high = maxval(qp_ene(nbsum,:,:)) + dtset%gwencomp
# # # # #
# # # ]
807 : 0 : call wrtout(units, sjoin(" Activating extrapolar technique (gwcomp 1) with E_high:", ftoa(en_high*Ha_eV), "[eV]"))
808 : : end if
809 [ - + ]: 3 : if (dtset%gwcomp == -2) then
810 : : ! E_high = average of e_{nbsum} at k-p and k+q-p + shift, the same for the +q and -q terms so that the derivative
811 : : ! of the closure term vanishes. If gwencomp < 0, use 1.7 times the Drude plasma frequency (see m_sigc).
812 [ # # ]: 0 : dyn_shift = dtset%gwencomp; if (dyn_shift < zero) dyn_shift = 1.7_dp * drude_plsmf
813 : : call wrtout(units, sjoin(" Activating COH dynamic remainder technique (gwcomp -2) with energy shift:", &
814 : 0 : ftoa(dyn_shift*Ha_eV), "[eV]"))
815 : : end if
816 : :
817 : : ! Read symmetrized em1 from file and build ppmodel parameters.
818 : : ! TODO: MPI-shared memory + compute only my set of pp-vectors in ppm%new_setup
819 : 3 : scr_iomode = iomode_from_fname(screen_filepath)
820 [ - + + - : 15 : ABI_MALLOC(epsm1_ggw, (npw_c, npw_c, hscr%nomega))
- + + - -
+ + - + -
+ - - + -
+ ]
821 : :
822 [ - + ]: 3 : if (dtset%gwcomp == 2) then
823 : : ! Allocate memory to store static screening in the IBZ and workspace array for the value at pp_bz in the BZ.
824 : : ! TODO: Store only the ip_ibz needed by this MPI rank to save memory.
825 : 0 : call wrtout(units, " Activating COH remainder technique")
826 [ # # # # : 0 : ABI_MALLOC(wc0_pibz, (pp_mesh%nibz))
# # # # #
# # # # #
# # # # ]
827 [ # # # # : 0 : ABI_MALLOC(wc0_pbz, (npw_c, npw_c))
# # # # #
# # # # #
# # # # ]
828 : : end if
829 : :
830 [ + + ]: 22 : do iq_ibz=1,pp_mesh%nibz
831 : : call read_screening("inverse_dielectric_function", screen_filepath, &
832 : 19 : npw_c, 1, hscr%nomega, epsm1_ggw, scr_iomode, comm, iqiA=iq_ibz)
833 : :
834 : 19 : call ppm%new_setup(iq_ibz, cryst, pp_mesh, npw_c, hscr%nomega, hscr%omega, epsm1_ggw, nfftf, gsph_c%gvec, ngfftf, rhor(:,1))
835 : :
836 [ - + ]: 22 : if (dtset%gwcomp == 2) then
837 : : ! Keep static limit of Wc in memory.
838 [ # # ]: 0 : do ii=1,npw_c
839 : 0 : epsm1_ggw(ii,ii,1) = epsm1_ggw(ii,ii,1) - one
840 : : end do
841 [ # # # # : 0 : ABI_MALLOC(wc0_pibz(iq_ibz)%vals, (npw_c, npw_c))
# # # # #
# # # # #
# # # # ]
842 [ # # # # : 0 : wc0_pibz(iq_ibz)%vals = epsm1_ggw(:,:,1)
# # # # #
# # # ]
843 : : end if
844 : : end do ! iq_ibz
845 : :
846 [ - + ]: 3 : ABI_FREE(epsm1_ggw)
847 : 3 : call hscr%free()
848 : :
849 : : ! Allocate g-vectors centered on k, k+q, k-p, and k+q-p.
850 [ - + + - : 9 : ABI_MALLOC(kg_k, (3, mpw))
+ - + - -
+ - + -
+ ]
851 [ - + + - : 6 : ABI_MALLOC(kg_kq, (3, mpw))
- + - + -
+ ]
852 [ - + + - : 6 : ABI_MALLOC(kg_kmp, (3, mpw))
- + - + -
+ ]
853 [ - + + - : 6 : ABI_MALLOC(kg_kqmp, (3, mpw))
- + - + -
+ ]
854 : :
855 : : ! GS wavefunctions
856 [ - + + - : 9 : ABI_MALLOC(ur_star_kmp, (nfft*nspinor))
+ - + - -
+ - + ]
857 [ - + + - : 6 : ABI_MALLOC(ur_star_kqmp, (nfft*nspinor))
- + - + ]
858 [ - + + - : 6 : ABI_MALLOC(cwork_ur, (nfft*nspinor))
- + - + ]
859 [ - + + - : 9 : ABI_MALLOC(cg_kmp, (2, mpw*nspinor))
+ - + - -
+ - + ]
860 [ - + + - : 6 : ABI_MALLOC(cg_kqmp, (2, mpw*nspinor))
- + - + ]
861 : : ! First order change (full term including the active space).
862 [ - + + - : 6 : ABI_MALLOC(cg1_kqmp, (2, mpw*nspinor))
- + - + ]
863 [ - + + - : 6 : ABI_MALLOC(cg1_kmp, (2, mpw*nspinor))
- + - + ]
864 : :
865 : : !stern_use_cache = merge(.True., .False., dtset%eph_stern == 1)
866 : 3 : stern_use_cache = .False.
867 : :
868 [ + - + + ]: 6 : if (my_rank == master) call gstore%print([std_out])
869 : 3 : call pstat_proc%print(_PSTAT_ARGS_)
870 : :
871 : : ! TODO
872 : : !if (dtset%gstore_iv1p_comm /= 0) then
873 : : ! call gstore%compute_and_write_commutator(mpw, gmax, ngfft, ngfftf, dtset, cryst, pawfgr, psps, &
874 : : ! wfd, mpi_enreg, kg_k, ebands, dvdb, gs_ham_kq, root_ncid)
875 : : !end if
876 : :
877 : : ! This parameter defines the size of the q-buffer used to store the g(k, q) e-ph matrix elements
878 : : ! for all the k-point treated by this MPI rank.
879 : : ! Increasing the buffer size increases the memory requirements
880 : : ! but it leads to better performance as the number of IO operations is decreased.
881 : : ! TODO: Should compute it on the basis of my_nkpt and my_nqpt
882 : : !
883 : : ! The q-buffer flush logic has ONLY been validated for qbuf_size == 1.
884 : : ! dump_my_gbuf() resets iqbuf_cnt and iq_buf, so the flush after the my_iq loop only writes q-points
885 : : ! that are still in the buffer. With qbuf_size > 1 there is another problem:
886 : : ! dump_my_gbuf() writes a block of iqbuf_cnt q-slices starting at iq_buf(1,1), i.e. it ASSUMES
887 : : ! the buffered q are consecutive. A `cycle` at the q-level of the my_iq loop (eph_restart done_qbz_spin, userib)
888 : : ! breaks that assumption and the slices would be written to the WRONG file offsets => silent data corruption
889 : : ! in gvals/gvals_ks. The k-filters (gstore_use_lgk, gstore_use_lgq) cycle the my_ik loop only and are not affected.
890 : : ! m_gstore handles non-contiguous buffers by writing slice by slice: do the same before using qbuf_size > 1.
891 : 3 : qbuf_size = 1
892 : : !qbuf_size = 16
893 : 3 : call wrtout(std_out, sjoin(" Begin computation of GWPT e-ph matrix elements with qbuf_size:", itoa(qbuf_size)), pre_newlines=1)
894 : :
895 : : ! If True, only k-points in the IBZ and q-points in the IBZ_k are computed.
896 : : ! Matrix elements in full BZs are then reconstructed by symmetry at the end of the run by calling
897 : : ! gstore_symmetrize.
898 [ - + - - : 3 : symmetrize = (dtset%gstore_kzone == "bz" .and. dtset%gstore_qzone == "bz" .and. dtset%gstore_sym > 0)
- - ]
899 : :
900 : : ! Compute kxc needed for vxc1.
901 : : ! A similar piece of code is used in m_respfn_driver.
902 : : ! option 2 for xc and kxc (no paramagnetic part if xcdata%nspden=1).
903 : : ! Note usage of dum_xccc3d to ignore non-linear core correction.
904 [ + - ]: 3 : nkxc = 2*min(dtset%nspden,2)-1; if (dtset%xclevel==2) nkxc = 12*min(dtset%nspden,2)-5
905 : 3 : call xcdata_init(xcdata, dtset=dtset)
906 [ - + - - ]: 3 : non_magnetic_xc = (dtset%usepaw==1.and.mod(abs(dtset%usepawu),10)==4)
907 : :
908 : 3 : call check_kxc(dtset%ixc, dtset%optdriver)
909 [ - + + - : 12 : ABI_MALLOC(kxc, (nfft, nkxc))
- + + - +
- + - - +
- + ]
910 [ - + + - : 12 : ABI_MALLOC(vxc, (nfft, dtset%nspden))
- + + - +
- + - - +
- + ]
911 : :
912 : 3 : nk3xc=1; option=2; usexcnhat=0
913 : : call rhotoxc(bigexc, bigsxc, kxc, mpi_enreg, nfft, ngfft, &
914 : : dum_nhat, 0, dum_nhat, 0, nkxc, nk3xc, non_magnetic_xc, n3xccc0, option, rhor, &
915 : 3 : cryst%rprimd, usexcnhat, vxc, vxcavg, dum_xccc3d, xcdata)
916 : :
917 : : ! Here we decide if the q-points can be reduced to the IBZ(k)
918 [ - + - - ]: 3 : use_lgk = (dtset%gstore_use_lgk /= 0 .or. dtset%gstore_sym == 2)
919 : : if (use_lgk) then
920 : 3 : call wrtout(units, " Only q-points in the IBZ_k will be computed.")
921 [ # # ]: 0 : else if (dtset%gstore_use_lgq /= 0) then
922 : 0 : call wrtout(units, " Only k-points in the IBZ_q will be computed.")
923 : : else
924 : 0 : call wrtout(units, " Little group operations won't be used")
925 : : end if
926 : :
927 : : ! ===================================================
928 : : ! Loop over MPI distributed spins in Sigma (gqk%comm)
929 : : ! ===================================================
930 : 3 : call pstat_proc%print(_PSTAT_ARGS_)
931 : :
932 : 3 : stern_qq_ierr = 0; stern_mq_ierr = 0
933 : :
934 [ + + ]: 6 : do my_is=1,gstore%my_nspins
935 : 3 : spin = gstore%my_spins(my_is); gqk => gstore%gqk(my_is); my_npert = gqk%my_npert
936 : : ABI_CHECK_IEQ(my_npert, gqk%my_npert, "my_npert")
937 [ - + ]: 3 : ABI_CHECK_IEQ(gqk%bstart_kq, 1, "bstart_qk should be 1")
938 [ - + ]: 3 : ABI_CHECK_IGEQ(nbsum, gqk%bstop_kq, "nband must be greater than the max band in the e-ph matrix elements")
939 : :
940 [ - + ]: 3 : NCF_CHECK(nf90_inq_ncid(root_ncid, strcat("gqk", "_spin", itoa(spin)), spin_ncid))
941 : : ! SC: these nctk_prepare_mpiio are not needed anymore: gstore_prepare_mpiio has accessed all the variables after the open,
942 : : ! and netcdf reads the metadata of a variable only at the first access, so these calls would do nothing.
943 : : !NCF_CHECK(nctk_prepare_mpiio(root_ncid, "gstore_done_qbz_spin"))
944 : : !NCF_CHECK(nctk_prepare_mpiio(spin_ncid, "gvals"))
945 : : !NCF_CHECK(nctk_prepare_mpiio(spin_ncid, "gvals_ks"))
946 : :
947 : : ! Note the possibility of specifying different number of states for the incoming and the intermediate states.
948 : 3 : nb_k = gqk%nb_k; bstart_k = gqk%bstart_k; bstop_k = gqk%bstop_k
949 : 3 : nb_kq = gqk%nb_kq; bstart_kq = gqk%bstart_kq; bstop_kq = gqk%bstop_kq
950 : :
951 [ - + + - : 9 : ABI_MALLOC(iq_buf, (2, qbuf_size))
+ - + - -
+ - + -
+ ]
952 [ + + + + ]: 12 : iq_buf = 0
953 [ - + + - : 12 : ABI_MALLOC(state_kq, (gqk%my_nk, qbuf_size))
- + + - +
- + - - +
- + - + ]
954 [ + + + + ]: 11 : state_kq = GSTORE_KQ_MISSING
955 [ - + + - : 15 : ABI_MALLOC(gsig_atm, (2, nb_kq, nb_k, natom3))
- + + - -
+ + - + -
+ - - + -
+ - + ]
956 [ - + + - : 12 : ABI_MALLOC(gks_atm, (2, nb_kq, nb_k, natom3))
- + + - -
+ + - - +
- + - + ]
957 [ - + + - : 12 : ABI_MALLOC(gks_atm2, (2, nb_kq, nb_k, natom3))
- + + - -
+ + - - +
- + - + ]
958 [ - + + - : 12 : ABI_MALLOC(gxc_atm, (2, nb_kq, nb_k, natom3))
- + + - -
+ + - - +
- + - + ]
959 : :
960 : : ! Arrays used to compare gwpt with gw
961 : : if (test_sigma) then
962 : : ABI_CALLOC(vxc_nk, (nb_k, gqk%glob_nk))
963 : : ABI_CALLOC(sigx_nk, (nb_k, gqk%glob_nk))
964 : : ABI_CALLOC(sigce0_nk, (nb_k, gqk%glob_nk))
965 : : end if
966 : :
967 [ - + + - : 12 : ABI_MALLOC(ur_nk, (nfft*nspinor, bstart_k:bstop_k))
- + + - +
- + - - +
- + ]
968 [ - + + - : 12 : ABI_MALLOC(ur_mkq, (nfft*nspinor, bstart_kq:bstop_kq))
- + + - +
- + - - +
- + ]
969 : :
970 : : ! Inside the loops, we compute gsig_atm(2, nb_kq, nb_k, natom3)
971 [ - + + - : 24 : ABI_MALLOC(my_gbuf, (gqk%cplex, nb_kq, nb_k, natom3, gqk%my_nk, qbuf_size))
- + + - -
+ + - - +
+ - - + +
- - + + -
+ - + - -
+ - + -
+ ]
972 [ - + + - : 24 : ABI_MALLOC(my_gbuf_ks, (gqk%cplex, nb_kq, nb_k, natom3, gqk%my_nk, qbuf_size))
- + + - -
+ + - - +
+ - - + +
- - + + -
+ - + - -
+ - + -
+ ]
973 : :
974 : : ! Allocate memory to deal with frequencies in Sigma(w).
975 : :
976 : 5 : select case (dtset%gwpt_wmode)
977 : : case (1)
978 : : ! Prepare list of omegas: first e_nk then e_mkq for all m indices.
979 : : ! Note that nw_nk depends on bstart_kq and bstop_kq so nw_nk depends on bands at _kq.
980 : 2 : nw_nk = 1 + (bstop_kq - bstart_kq + 1)
981 : 2 : nw_mkq = 1 + (bstop_k - bstart_k + 1)
982 : 2 : call wrtout(units, " GWPT frequency convolution is evaluated at E_nk and E_mkq$ and the average is taken.")
983 : : case (2)
984 : 1 : nw_nk = 1
985 : 1 : nw_mkq = (bstop_k - bstart_k + 1)
986 : 1 : call wrtout(units, " GWPT frequency convolution is evaluated at E_nk.")
987 : : case default
988 [ + + - ]: 3 : ABI_ERROR(sjoin("Invalid gwpt_wmode:", itoa(dtset%gwpt_wmode)))
989 : : end select
990 : :
991 [ - + + - : 9 : ABI_MALLOC(omegame0i_nk, (nw_nk))
+ - + - -
+ - + ]
992 [ - + + - : 9 : ABI_MALLOC(omegame0i_mkq, (nw_mkq))
+ - + - -
+ - + ]
993 [ - + + - : 6 : ABI_MALLOC(omegas_nk, (nw_nk))
- + - + ]
994 [ - + + - : 6 : ABI_MALLOC(omegas_mkq, (nw_mkq))
- + - + ]
995 [ - + + - : 9 : ABI_MALLOC(sigcme_nk, (nw_nk))
+ - + - -
+ - + ]
996 [ - + + - : 9 : ABI_MALLOC(sigcme_mkq, (nw_mkq))
+ - + - -
+ - + ]
997 : :
998 : : ! Correlated part
999 [ - + + - : 15 : ABI_MALLOC(vec_gwc_nk, (npw_c*nspinor, nw_nk, bstart_k:bstop_k))
- + + - -
+ + - + -
+ - - + -
+ - + ]
1000 [ - + + - : 15 : ABI_MALLOC(vec_gwc_mkq, (npw_c*nspinor, nw_mkq, bstart_kq:bstop_kq))
- + + - -
+ + - + -
+ - - + -
+ - + ]
1001 [ - + ]: 3 : if (any(dtset%gwcomp == [1, -2])) then
1002 [ # # # # : 0 : ABI_MALLOC(vec_comp_nk, (npw_c*nspinor, nw_nk))
# # # # #
# # # # #
# # # # ]
1003 [ # # # # : 0 : ABI_MALLOC(vec_comp_mkq, (npw_c*nspinor, nw_mkq))
# # # # #
# # # # #
# # # # ]
1004 : : end if
1005 : : ! Exchange part
1006 [ - + + - : 12 : ABI_MALLOC(vec_gx_nk, (npw_x*nspinor, bstart_k:bstop_k))
- + + - +
- + - - +
- + - + ]
1007 [ - + + - : 12 : ABI_MALLOC(vec_gx_mkq, (npw_x*nspinor, bstart_kq:bstop_kq))
- + + - +
- + - - +
- + - + ]
1008 : :
1009 [ - + ]: 3 : if (dtset%userra /= zero) then
1010 : : ! Additional dG0/de term at q=Gamma (Appendix A): dWtilde/de contracted with the n-side oscillators.
1011 [ # # # # : 0 : ABI_MALLOC(vec_dgwc_nk, (npw_c*nspinor, nw_nk, bstart_k:bstop_k))
# # # # #
# # # # #
# # # # ]
1012 : : end if
1013 : :
1014 [ - + ]: 3 : if (dtset%gwcomp == 2) then
1015 [ # # # # : 0 : ABI_MALLOC(vec_coh_nk, (npw_c, bstart_k:bstop_k))
# # # # #
# # # # #
# # # # ]
1016 [ # # # # : 0 : ABI_MALLOC(vec_coh_mkq, (npw_c, bstart_kq:bstop_kq))
# # # # #
# # # # #
# # # # ]
1017 : : end if
1018 : :
1019 : : ! Compute the little group of the k-point so that we can compute g(k,q) only for q in the IBZ_k
1020 [ + - ]: 3 : if (use_lgk) then
1021 : 3 : timrev_k = kpts_timrev_from_kptopt(ebands%kptopt)
1022 [ - + + - : 14 : ABI_MALLOC(lg_myk, (gqk%my_nk))
+ - + - -
+ - + - +
+ + + - -
+ - + - +
- + - + -
+ - + ]
1023 [ + + ]: 8 : do my_ik=1,gqk%my_nk
1024 [ + + ]: 20 : kk = gqk%my_kpts(:, my_ik)
1025 : 8 : call lg_myk(my_ik)%init(cryst, kk, timrev_k, gstore%nqbz, gstore%qbz, gstore%nqibz, gstore%qibz, xmpi_comm_self)
1026 : : end do
1027 : : end if
1028 : :
1029 : : ! ============================================================
1030 : : ! Loop over MPI distributed q-points in Sigma_q (gqk%qpt_comm)
1031 : : ! ============================================================
1032 : : ! the loop of my_iq can be cycled, so it is good to set iqbuf_cnt to 0 before entering the loop
1033 : 3 : iqbuf_cnt = 0
1034 [ + + ]: 83 : do my_iq=1,gqk%my_nq
1035 : 80 : call gqk%myqpt(my_iq, gstore, weight_q, qq_bz)
1036 : :
1037 [ + + + + : 115 : if (dtset%userib /= 0 .and. (any(abs(qq_bz - [0.5, 0.0, 0.0]) > tol14))) cycle
- + - - ]
1038 : :
1039 [ + + ]: 320 : iq_bz = gqk%my_q2bz(my_iq); qq_is_gamma = sum(qq_bz**2) < tol14; qq_bz_string = ktoa(qq_bz)
1040 [ + + + - ]: 80 : use_dg0 = qq_is_gamma .and. dtset%userra /= zero
1041 : :
1042 : : ! Handle possible restart.
1043 [ - + ]: 80 : if (done_qbz_spin(iq_bz, spin) == 1) then
1044 : 0 : call wrtout(std_out, sjoin(" iq_bz:", itoa(iq_bz), ", spin: ", itoa(spin), " already computed --> skipping iteration"), do_flush=.True.)
1045 : 0 : cycle
1046 : : end if
1047 : :
1048 : : ! Compute the little group of the q-point so that we can compute g(k,q) only for k in the IBZ_q
1049 [ - + ]: 80 : if (dtset%gstore_use_lgq /= 0) then
1050 : 0 : timrev_q = kpts_timrev_from_kptopt(gstore%qptopt)
1051 [ # # # # : 0 : call lg_myq%init(cryst, qq_bz, timrev_q, gstore%nkbz, gstore%kbz, gstore%nkibz, gstore%kibz, xmpi_comm_self)
# # # # #
# ]
1052 : : end if
1053 : :
1054 : : ! Position of this q-point in the q-buffer (set at the q-level so that it is also defined when my_nk == 0).
1055 : 80 : iqbuf_cnt = 1 + mod(my_iq - 1, qbuf_size)
1056 [ + + ]: 240 : iq_buf(:, iqbuf_cnt) = [my_iq, iq_bz]
1057 : :
1058 : : ! Note symrec conventions here as needed to symmetrize the DFPT potentials.
1059 : 80 : iq_ibz = gqk%my_q2ibz(1, my_iq) ; isym_q = gqk%my_q2ibz(2, my_iq)
1060 : 80 : trev_q = gqk%my_q2ibz(6, my_iq) ; g0_q = gqk%my_q2ibz(3:5, my_iq)
1061 : : isirr_q = (isym_q == 1 .and. trev_q == 0 .and. all(g0_q == 0))
1062 : : !tsign_q = 1; if (trev_q == 1) tsign_q = -1
1063 [ + + ]: 320 : qq_ibz = gstore%qibz(:, iq_ibz)
1064 [ + + ]: 560 : mapc_qq = gqk%my_q2ibz(:, my_iq)
1065 : :
1066 : 80 : print_time_qq = my_rank == 0 .and. (my_iq <= LOG_MODQ .or. mod(my_iq, LOG_MODQ) == 0)
1067 [ + - ]: 80 : if (print_time_qq) then
1068 : 80 : call cwtime(cpu_qq, wall_qq, gflops_qq, "start")
1069 : 80 : call inds2str(0, sjoin(" Computing g^Sigma(k, q) for qq_bz:", qq_bz_string), my_iq, gqk%my_nq, gqk%glob_nq, msg)
1070 : 80 : call wrtout(std_out, sjoin(msg, ", and spin:", itoa(spin)), pre_newlines=1)
1071 : : !print *, "iq_ibz:", iq_ibz, "qq_bz:", qq_bz, "qq_ibz:", qq_ibz
1072 : : end if
1073 : :
1074 : : ! ==================================================
1075 : : ! Get DFPT potentials and densities for this q-point
1076 : : ! ==================================================
1077 : : ! After this branch we know `cplex` and we have allocated:
1078 : : !
1079 : : ! v1scf_qq(cplex, nfftf, nspden, my_npert))
1080 : : ! vxc1_qq(cplex, nfft, nspden, my_npert)
1081 : : !
1082 : : ! Important: vxc1_qq does not include the contribution due to the model core charge (if any).
1083 : :
1084 [ - + ]: 80 : if (need_ftinterp) then
1085 : : ! Use Fourier interpolation to get DFPT potentials and DFPT densities for this qpt.
1086 [ # # ]: 0 : call dvdb%get_ftqbz(qq_bz, cplex, nfftf, ngfftf, v1scf_qq, gqk%pert_comm%value)
1087 : :
1088 : : call drhodb%get_vxc1_ftqbz(dtset, cryst, qq_bz, drho_cplex, nfftf, ngfftf, nkxc, kxc, &
1089 [ # # ]: 0 : vxc1_qq, non_magnetic_xc, usexcnhat, gqk%pert_comm%value)
1090 : : else
1091 : : ! Read and reconstruct the dvscf potentials and the densities for this qq_bz and my_npert perturbations.
1092 : 80 : db_iqpt = dvdb%findq(qq_ibz)
1093 [ - + ]: 80 : ABI_CHECK(db_iqpt /= -1, sjoin("Could not find symmetric of q-point:", ktoa(qq_bz), "in DVDB file."))
1094 : : ! The first entry in mapc_qq2dvdb gives the index in dvdb%qpts. The other entries in mapc_qq are OK as they refer to symmetries.
1095 : 80 : mapc_qq2dvdb = mapc_qq; mapc_qq2dvdb(1) = db_iqpt
1096 [ - + ]: 80 : call dvdb%readsym_qbz(cryst, qq_bz, mapc_qq2dvdb, cplex, nfftf, ngfftf, v1scf_qq, gqk%pert_comm%value)
1097 : :
1098 : 80 : db_iqpt = drhodb%findq(qq_ibz)
1099 [ - + ]: 80 : ABI_CHECK(db_iqpt /= -1, sjoin("Could not find symmetric of q-point:", ktoa(qq_bz), "in DRHODB file."))
1100 : 80 : mapc_qq2dvdb = mapc_qq; mapc_qq2dvdb(1) = db_iqpt
1101 : : call drhodb%read_vxc1_qbz(dtset, cryst, qq_bz, mapc_qq2dvdb, drho_cplex, nfftf, ngfftf, nkxc, kxc, &
1102 [ - + ]: 80 : vxc1_qq, non_magnetic_xc, usexcnhat, gqk%pert_comm%value)
1103 : : end if
1104 : :
1105 [ - + ]: 80 : ABI_CHECK_IEQ(cplex, drho_cplex, "Different values of cplex for v1 and rho1!")
1106 [ + + + + ]: 311 : cvxc1_qq_ptr => null(); if (cplex == 2) call c_f_pointer(c_loc(vxc1_qq), cvxc1_qq_ptr, [nfft, nspden, my_npert])
1107 : :
1108 : : ! Allocate vlocal1_qq with correct cplex and nvloc.
1109 [ - + + - : 560 : ABI_MALLOC(vlocal1_qq, (cplex*n4, n5, n6, nvloc, my_npert))
- + + - -
+ + - - +
+ - - + +
- + - + -
- + - + ]
1110 [ - + + - : 480 : ABI_MALLOC(vlocal1_mqq, (cplex*n4, n5, n6, nvloc, my_npert))
- + + - -
+ + - - +
+ - - + +
- - + -
+ ]
1111 : :
1112 : : ! Build DFPT potential at -qq by taking CC.
1113 [ + + + + : 10364596 : v1scf_mq = v1scf_qq; if (cplex == 2) v1scf_mq(2,:,:,:) = -v1scf_mq(2,:,:,:)
+ - + - -
+ + - + -
+ - + - +
+ + + + +
+ + + + +
+ + + + +
+ + ]
1114 : :
1115 : : ! =============================================================
1116 : : ! Loop over k-points in the e-ph matrix elements (gqk%kpt_comm)
1117 : : ! =============================================================
1118 [ + + ]: 176 : do my_ik=1,gqk%my_nk
1119 [ + + ]: 384 : kk = gqk%my_kpts(:, my_ik); kk_string = ktoa(kk); ik_glob = gqk%my_k2glob(my_ik)
1120 : :
1121 : : ! NB: All procs in gqk%pert_comm and gqk%bsum_comm and gqk%pp_sum_comm enter this section.
1122 : : ! Set entry to zero BEFORE the cycle instructions below, otherwise cycled (filtered)
1123 : : ! k-points may keep stale data in my_gbuf and write them to disk.
1124 [ + + + + : 90912 : my_gbuf(:,:,:,:, my_ik, iqbuf_cnt) = zero
+ + + + ]
1125 [ + + + + : 90912 : my_gbuf_ks(:,:,:,:, my_ik, iqbuf_cnt) = zero
+ + + + ]
1126 : :
1127 : : ! The k-point and the symmetries relating the BZ k-point to the IBZ.
1128 : 96 : ik_ibz = gqk%my_k2ibz(1, my_ik) ; isym_k = gqk%my_k2ibz(2, my_ik)
1129 [ + + ]: 384 : trev_k = gqk%my_k2ibz(6, my_ik); g0_k = gqk%my_k2ibz(3:5,my_ik)
1130 [ + + + - : 384 : isirr_k = (isym_k == 1 .and. trev_k == 0 .and. all(g0_k == 0))
+ - - + ]
1131 [ + + ]: 672 : mapl_k = gqk%my_k2ibz(:, my_ik)
1132 : :
1133 [ + + ]: 384 : kk_ibz = ebands%kptns(:,ik_ibz)
1134 : :
1135 : : ! If we are going to reconstruct g(k,q) in the full BZ by symmetry at the end of the run,
1136 : : ! only k-points in the IBZ need to be computed here.
1137 [ - + ]: 96 : if (symmetrize .and. .not. isirr_k) then
1138 : 0 : state_kq(my_ik, iqbuf_cnt) = GSTORE_KQ_MISSING; cycle
1139 : : end if
1140 : :
1141 [ - + ]: 96 : if (dtset%userib /= 0) then
1142 [ # # # # : 0 : if (any(abs(gqk%my_kpts(:, my_ik) - [0.25, 0.0, 0.0]) > tol14) .and. &
# # # # #
# ]
1143 : : any(abs(gqk%my_kpts(:, my_ik) - [-0.25, 0.0, 0.0]) > tol14)) then
1144 : 0 : state_kq(my_ik, iqbuf_cnt) = GSTORE_KQ_MISSING; cycle
1145 : : end if
1146 : : end if
1147 : :
1148 : : ! Here we skip points if little group tricks are activated.
1149 [ + - ]: 96 : if (use_lgk) then
1150 [ + + ]: 96 : if (lg_myk(my_ik)%findq_ibzk(qq_bz) == -1) then
1151 : 65 : state_kq(my_ik, iqbuf_cnt) = GSTORE_KQ_MISSING; cycle
1152 : : end if
1153 : : end if
1154 : :
1155 [ - + ]: 31 : if (dtset%gstore_use_lgq /= 0) then
1156 [ # # ]: 0 : if (lg_myq%findq_ibzk(kk) == -1) then
1157 : 0 : state_kq(my_ik, iqbuf_cnt) = GSTORE_KQ_MISSING; cycle
1158 : : end if
1159 : : end if
1160 : :
1161 [ + + + + : 36805 : gks_atm = zero
+ + + + ]
1162 [ + + + + : 36805 : gks_atm2 = zero
+ + + + ]
1163 : :
1164 : 31 : call inds2str(0, sjoin(" Computing g^Sigma(k,q) for kpt:", kk_string), my_ik, gqk%my_nk, gqk%glob_nk, msg)
1165 : 31 : call wrtout(std_out, sjoin(msg, ", for spin:", itoa(spin)), pre_newlines=1)
1166 : 31 : call timab(1940, 1, tsec)
1167 : :
1168 : 31 : istwf_k_ibz = wfd%istwfk(ik_ibz); npw_k_ibz = wfd%npwarr(ik_ibz)
1169 : :
1170 [ + - - + : 31 : print_time_kk = my_rank == 0 .and. (my_ik <= LOG_MODK .or. mod(my_ik, LOG_MODK) == 0)
- - ]
1171 : 31 : if (print_time_kk) call cwtime(cpu_kk, wall_kk, gflops_kk, "start")
1172 : :
1173 : : ! Get npw_k, kg_k for kk
1174 : : call wfd%get_gvec_gbound(cryst%gmet, dtset%ecut, kk, ik_ibz, isirr_k, dtset%nloalg, & ! in
1175 [ - + ]: 31 : istwf_k, npw_k, kg_k, nkpg_k, kpg_k, gbound_k) ! out
1176 [ - + ]: 31 : ABI_CHECK_ILEQ(npw_k, mpw, "npw_k > mpw!")
1177 : :
1178 : : ! Find k + q in the extended zone and extract symmetry info.
1179 : : ! Be careful here because there are two umklapp vectors to be considered as:
1180 : : !
1181 : : ! k + q = k_bz + g0_bz = IS(k_ibz) + g0_ibz + g0_bz
1182 : : !
1183 [ + + ]: 124 : kq = kk + qq_bz
1184 : :
1185 [ - + ]: 31 : if (kpts_map("symrel", ebands%kptopt, cryst, gstore%krank_ibz, 1, kq, mapl_kq) /= 0) then
1186 [ # # ]: 0 : write(msg, '(4a)' )"k-mesh is not closed!",ch10, "k+q could not be generated from a symmetrical one.",trim(ltoa(kq))
1187 : 0 : ABI_ERROR(msg)
1188 : : end if
1189 [ + + ]: 124 : ikq_ibz = mapl_kq(1); isym_kq = mapl_kq(2); trev_kq = mapl_kq(6); g0_kq = mapl_kq(3:5)
1190 [ + + + + : 99 : isirr_kq = (isym_kq == 1 .and. trev_kq == 0 .and. all(g0_kq == 0))
+ + + + ]
1191 [ + + ]: 124 : kq_ibz = ebands%kptns(:, ikq_ibz)
1192 : 31 : istwf_kq_ibz = wfd%istwfk(ikq_ibz); npw_kq_ibz = wfd%npwarr(ikq_ibz)
1193 : :
1194 : : ! Get npw_kq, kg_kq for k+q.
1195 : : call wfd%get_gvec_gbound(cryst%gmet, dtset%ecut, kq, ikq_ibz, isirr_kq, dtset%nloalg, & ! in
1196 [ - + ]: 31 : istwf_kq, npw_kq, kg_kq, nkpg_kq, kpg_kq, gbound_kq) ! out
1197 [ - + ]: 31 : ABI_CHECK_ILEQ(npw_kq, mpw, "npw_kq > mpw!")
1198 : :
1199 [ - + + - : 104952 : ABI_CALLOC(ug_k, (2, npw_k*nspinor))
+ - + - -
+ - + + +
+ + ]
1200 [ - + + - : 106581 : ABI_CALLOC(ug_kq, (2, npw_kq*nspinor))
+ - + - -
+ - + + +
+ + ]
1201 : :
1202 : : ! Precompute ur_nk and ur_mkq for all m and n band indices treated.
1203 : : ! TODO: Can distribute operations inside gqk%pert_comm
1204 : :
1205 [ + + ]: 201 : do n_k=bstart_k, bstop_k
1206 : 170 : in_k = n_k - bstart_k + 1 !; if (gqk%pert_ppsum_comm%skip(in_k)) cycle ! MPI parallelism inside pert_comm
1207 : : call wfd%rotate_cg(n_k, ndat1, spin, kk_ibz, npw_k, kg_k, istwf_k, &
1208 : 170 : cryst, mapl_k, gbound_k, work_ngfft, work, ug_k, urs_kbz=ur_nk(:,n_k))
1209 : 31 : if (test_sigma) vxc_nk(in_k, ik_glob) = dot_product(ur_nk(:,n_k), vxc(:, spin) * ur_nk(:,n_k)) / nfftf
1210 : : end do
1211 : : !call xmpi_sum(ur_nk, gqk%pert_ppsum_comm%value, ierr)
1212 : :
1213 [ + + ]: 387 : do m_kq=bstart_kq, bstop_kq
1214 : 356 : im_kq = m_kq - bstart_kq + 1 !; if (gqk%pert_ppsum_comm%skip(im_kq)) cycle ! MPI parallelism inside pert_comm
1215 : : call wfd%rotate_cg(m_kq, ndat1, spin, kq_ibz, npw_kq, kg_kq, istwf_kq, &
1216 : 387 : cryst, mapl_kq, gbound_kq, work_ngfft, work, ug_kq, urs_kbz=ur_mkq(:,m_kq))
1217 : : end do
1218 : : !call xmpi_sum(ur_mkq, gqk%pert_ppsum_comm%value, ierr)
1219 : :
1220 : : ! ===========================
1221 : : ! Compute <m,k+q|vxc1_qq|n,k>
1222 : : ! ===========================
1223 : :
1224 [ + + + + : 36805 : gxc_atm = czero; cnt = 0
+ + + + ]
1225 [ + + ]: 387 : do m_kq=bstart_kq, bstop_kq
1226 : 356 : im_kq = m_kq - bstart_kq + 1
1227 [ + + ]: 2363 : do n_k=bstart_k, bstop_k
1228 : 1976 : in_k = n_k - bstart_k + 1
1229 : 1976 : cnt = cnt + 1
1230 [ + - ]: 1976 : if (gqk%pp_sum_comm%skip(cnt)) cycle ! MPI parallelism inside pp_sum_comm
1231 [ + + ]: 14188 : do imyp=1,gqk%my_npert
1232 [ + + ]: 11856 : if (cplex == 1) then
1233 [ + + ]: 8479536 : ctmp_gwpc = sum(GWPC_CONJG(ur_mkq(:,m_kq)) * ur_nk(:,n_k) * vxc1_qq(1,:,spin,imyp)) / nfftf
1234 : : else
1235 [ + + ]: 43427616 : ctmp_gwpc = sum(GWPC_CONJG(ur_mkq(:,m_kq)) * ur_nk(:,n_k) * cvxc1_qq_ptr(:,spin,imyp)) / nfftf
1236 : : end if
1237 : 11856 : ipc = gqk%my_pertcases(imyp)
1238 [ + + ]: 37544 : gxc_atm(:, im_kq, in_k, ipc) = [real(ctmp_gwpc), aimag(ctmp_gwpc)]
1239 : : end do ! imyp
1240 : : end do ! n_k
1241 : : end do ! m_kq
1242 : :
1243 : : ! TODO: this is an all_gatherv but oh well.
1244 : 31 : call xmpi_sum(gxc_atm, gqk%pert_ppsum_comm%value, ierr)
1245 : 31 : call timab(1940, 2, tsec)
1246 : :
1247 : : ! ==========================================================
1248 : : ! MPI sum over the pp momenta in the full BZ gqk%pp_sum_comm
1249 : : ! ==========================================================
1250 : : !
1251 : : ! Be careful here because pp should run over the list of wavevectors in the screening matrix!
1252 : : ! as pp_mesh%bz is not necessarily equivalent to the k-mesh for the wavefunctions.
1253 : : ! Moreover we have to use the ipp_bz index to symmetrize W(pp_bz) from W(pp_ibz).
1254 : : !
1255 : : ! TODO: Should order nbz in shells so that one can reduce the memory required
1256 : : ! to store W(pp) if pp_parallelism is activated.
1257 [ + + + + : 36805 : gsig_atm = zero
+ + + + ]
1258 : :
1259 [ + + ]: 727 : do ipp_bz=my_pp_start_spin(spin), my_pp_stop_spin(spin)
1260 : : ! All procs in gqk%pert_comm and gqk%bsum_com enter this section.
1261 : 696 : call timab(1941, 1, tsec)
1262 : :
1263 : 696 : my_ipp = ipp_bz - my_pp_start_spin(spin) + 1
1264 [ + - + + : 696 : print_time_pp = my_rank == 0 .and. (my_ipp <= LOG_MODP .or. mod(my_ipp, LOG_MODP) == 0)
+ + ]
1265 : 267 : if (print_time_pp) call cwtime(cpu_pp, wall_pp, gflops_pp, "start")
1266 : :
1267 [ + + + + ]: 4872 : pp = pp_mesh%bz(:,ipp_bz); pp_is_gamma = sum(pp**2) < tol14; pp_string = ktoa(pp)
1268 : :
1269 : : ! Debug, include only pp=Gamma
1270 [ - + ]: 696 : if (dtset%userie > 0) then
1271 [ # # ]: 0 : if (.not. pp_is_gamma) cycle
1272 : : end if
1273 : :
1274 [ + - ]: 696 : qkp_string = sjoin("While treating qq_bz: ", qq_bz_string, "kpt:", kk_string, "pp:", pp_string, ch10)
1275 : :
1276 : : ! Symmetry tables and g-sphere centered on k-p.
1277 [ + + ]: 2784 : kmp = kk - pp
1278 [ - + ]: 696 : if (kpts_map("symrel", ebands%kptopt, cryst, gstore%krank_ibz, 1, kmp, mapl_kmp) /= 0) then
1279 : 0 : write(msg, '(4a)' )"k-mesh is not closed!",ch10, &
1280 [ # # ]: 0 : "k-p could not be generated from a symmetrical one.",trim(ltoa(kmp))
1281 : 0 : ABI_ERROR(msg)
1282 : : end if
1283 [ + + ]: 2784 : ikmp_ibz = mapl_kmp(1); isym_kmp = mapl_kmp(2); trev_kmp = mapl_kmp(6); g0_kmp = mapl_kmp(3:5)
1284 [ + + + + : 2349 : isirr_kmp = (isym_kmp == 1 .and. trev_kmp == 0 .and. all(g0_kmp == 0))
+ + + + ]
1285 [ + + ]: 2784 : kmp_ibz = ebands%kptns(:, ikmp_ibz)
1286 : :
1287 : : ! Get npw_kmp, kg_kmp for k-p.
1288 : : call wfd%get_gvec_gbound(cryst%gmet, dtset%ecut, kmp, ikmp_ibz, isirr_kmp, dtset%nloalg, & ! in
1289 [ - + ]: 696 : istwf_kmp, npw_kmp, kg_kmp, nkpg_kmp, kpg_kmp, gbound_kmp) ! out
1290 [ - + ]: 696 : ABI_CHECK_ILEQ(npw_kmp, mpw, "npw_kmp > mpw!")
1291 : :
1292 [ - + + - : 2784 : ABI_MALLOC(ph3d_kmp, (2, npw_kmp, matblk))
- + + - +
- + - - +
- + - + ]
1293 [ - + + - : 2088 : ABI_MALLOC(kinpw_kmp, (npw_kmp))
+ - + - -
+ - + -
+ ]
1294 : 696 : call mkkin(dtset%ecut, dtset%ecutsm, dtset%effmass_free, cryst%gmet, kg_kmp, kinpw_kmp, kmp, npw_kmp, 0, 0)
1295 : :
1296 : : ! Compute nonlocal form factors ffnl_kmp at (k-p+G).
1297 [ - + + - : 3480 : ABI_MALLOC(ffnl_kmp, (npw_kmp, 1, psps%lmnmax, psps%ntypat))
- + + - -
+ + - + -
+ - - + -
+ - + ]
1298 : :
1299 [ - + + - : 2784 : ABI_MALLOC(ylm_kmp, (npw_kmp, psps%mpsang**2 * psps%useylm))
+ - - - +
- - + - +
- + - + ]
1300 [ - + ]: 696 : if (psps%useylm == 1) then
1301 : 0 : call initylmg_k(npw_kmp, psps%mpsang, optder0, cryst%rprimd, cryst%gprimd, kmp, kg_kmp, ylm_kmp, ylmgr_dum)
1302 : : end if
1303 : :
1304 : : call mkffnl_objs(cryst, psps, 1, ffnl_kmp, ider0, idir0, kg_kmp, kpg_kmp, kmp, nkpg_kmp, &
1305 : 696 : npw_kmp, ylm_kmp, ylmgr_dum) !, comm=gqk%pert_comm%value, request=ffnl_kmp_request)
1306 : :
1307 : : ! Symmetry tables and g-sphere centered on k+q-p.
1308 [ + + ]: 2784 : kqmp = kq - pp
1309 [ - + ]: 696 : if (kpts_map("symrel", ebands%kptopt, cryst, gstore%krank_ibz, 1, kqmp, mapl_kqmp) /= 0) then
1310 [ # # ]: 0 : ABI_ERROR(sjoin("k-mesh not closed!",ch10, "k+q-p could not be generated from a symmetrical one:",trim(ltoa(kqmp))))
1311 : : end if
1312 [ + + ]: 2784 : ikqmp_ibz = mapl_kqmp(1); isym_kqmp = mapl_kqmp(2); trev_kqmp = mapl_kqmp(6); g0_kqmp = mapl_kqmp(3:5)
1313 [ + + + + : 1832 : isirr_kqmp = (isym_kqmp == 1 .and. trev_kqmp == 0 .and. all(g0_kqmp == 0))
+ + + + ]
1314 [ + + ]: 2784 : kqmp_ibz = ebands%kptns(:, ikqmp_ibz)
1315 : 696 : istwf_kqmp_ibz = wfd%istwfk(ikqmp_ibz); npw_kqmp_ibz = wfd%npwarr(ikqmp_ibz)
1316 : :
1317 : : ! Get npw_kqmp, kg_kqmp for k+q-p.
1318 : : call wfd%get_gvec_gbound(cryst%gmet, dtset%ecut, kqmp, ikqmp_ibz, isirr_kqmp, dtset%nloalg, & ! in
1319 [ - + ]: 696 : istwf_kqmp, npw_kqmp, kg_kqmp, nkpg_kqmp, kpg_kqmp, gbound_kqmp) ! out
1320 [ - + ]: 696 : ABI_CHECK_ILEQ(npw_kqmp, mpw, "npw_kqmp > mpw!")
1321 : :
1322 [ - + + - : 2088 : ABI_MALLOC(full_cg1_kqmp, (2, npw_kqmp*nspinor))
+ - + - -
+ - + ]
1323 [ - + + - : 2088 : ABI_MALLOC(full_cg1_kmp, (2, npw_kmp*nspinor))
+ - + - -
+ - + ]
1324 : :
1325 [ - + + - : 2784 : ABI_MALLOC(ph3d_kqmp, (2, npw_kqmp, matblk))
- + + - +
- + - - +
- + - + ]
1326 [ - + + - : 2088 : ABI_MALLOC(kinpw_kqmp, (npw_kqmp))
+ - + - -
+ - + -
+ ]
1327 : 696 : call mkkin(dtset%ecut, dtset%ecutsm, dtset%effmass_free, cryst%gmet, kg_kqmp, kinpw_kqmp, kqmp, npw_kqmp, 0, 0)
1328 : :
1329 : : ! Compute nonlocal form factors ffnl_kqmp at (k+q-p+G).
1330 [ - + + - : 3480 : ABI_MALLOC(ffnl_kqmp, (npw_kqmp, 1, psps%lmnmax, psps%ntypat))
- + + - -
+ + - + -
+ - - + -
+ - + ]
1331 : :
1332 [ - + + - : 2784 : ABI_MALLOC(ylm_kqmp, (npw_kqmp, psps%mpsang**2 * psps%useylm))
+ - - - +
- - + - +
- + - + ]
1333 [ - + ]: 696 : if (psps%useylm == 1) then
1334 : 0 : call initylmg_k(npw_kqmp, psps%mpsang, optder0, cryst%rprimd, cryst%gprimd, kqmp, kg_kqmp, ylm_kqmp, ylmgr_dum)
1335 : : end if
1336 : :
1337 : : call mkffnl_objs(cryst, psps, 1, ffnl_kqmp, ider0, idir0, kg_kqmp, kpg_kqmp, kqmp, nkpg_kqmp, &
1338 : 696 : npw_kqmp, ylm_kqmp, ylmgr_dum) ! , comm=gqk%pert_comm%value, request=ffnl_kqmp_request)
1339 : :
1340 : : ! =============================================
1341 : : ! Load the k dependent parts of the Hamiltonian
1342 : : ! =============================================
1343 : :
1344 : : ! (k+q-p, k-p)
1345 : : call gs_ham_kqmp%load_k(kpt_k=kmp, npw_k=npw_kmp, istwf_k=istwf_kmp, kg_k=kg_kmp, kpg_k=kpg_kmp, kinpw_k=kinpw_kmp, &
1346 : 696 : ph3d_k=ph3d_kmp, ffnl_k=ffnl_kmp, compute_ph3d=.true., compute_gbound=.true.)
1347 : :
1348 : : call gs_ham_kqmp%load_kprime(kpt_kp=kqmp, npw_kp=npw_kqmp, istwf_kp=istwf_kqmp, kg_kp=kg_kqmp, kpg_kp=kpg_kqmp, &
1349 : 696 : kinpw_kp=kinpw_kqmp, ph3d_kp=ph3d_kqmp, ffnl_kp=ffnl_kqmp, compute_ph3d=.true., compute_gbound=.true.)
1350 : :
1351 : : ! (k-p, k+q-p)
1352 : : call gs_ham_kmp%load_k(kpt_k=kqmp, npw_k=npw_kqmp, istwf_k=istwf_kqmp, kg_k=kg_kqmp, kpg_k=kpg_kqmp, kinpw_k=kinpw_kqmp, &
1353 : 696 : ph3d_k=ph3d_kqmp, ffnl_k=ffnl_kqmp, compute_ph3d=.true., compute_gbound=.true.)
1354 : :
1355 : : call gs_ham_kmp%load_kprime(kpt_kp=kmp, npw_kp=npw_kmp, istwf_kp=istwf_kmp, kg_kp=kg_kmp, kpg_kp=kpg_kmp, &
1356 : 696 : kinpw_kp=kinpw_kmp, ph3d_kp=ph3d_kmp, ffnl_kp=ffnl_kmp, compute_ph3d=.true., compute_gbound=.true.)
1357 : :
1358 : : ! ====================================
1359 : : ! This is the g-sphere for W_{gg'}(pp)
1360 : : ! ====================================
1361 : : ! Note that in this case, the spheres are always Gamma-centered i.e. it does not depend on the pp wavevector.
1362 : 696 : kg_c => gsph_c%gvec(:, 1:npw_c)
1363 : 696 : kg_x => gsph_x%gvec(:, 1:npw_x)
1364 : 696 : call sphereboundary(gbound_c, istwfk1, kg_c, mgfft, npw_c)
1365 : 696 : call sphereboundary(gbound_x, istwfk1, kg_x, mgfft, npw_x)
1366 : :
1367 [ - + + - : 2088 : ABI_MALLOC(rhotwg_c, (npw_c*nspinor))
+ - + - -
+ - + -
+ ]
1368 [ - + + - : 2088 : ABI_MALLOC(rhotwg_x, (npw_x*nspinor))
+ - + - -
+ - + ]
1369 [ - + + - : 2088 : ABI_MALLOC(vc_sqrt_gx, (npw_x))
+ - + - -
+ - + ]
1370 [ - + ]: 696 : if (dtset%userra /= zero) then
1371 : : ! Auxiliary state chi_{n'k-p} on the k+q-p g-sphere and its oscillator (dG0/de term at q=Gamma).
1372 [ # # # # : 0 : ABI_MALLOC(chi_ug_kqmp, (npw_kqmp*nspinor))
# # # # #
# # # #
# ]
1373 [ # # # # : 0 : ABI_MALLOC(rhotwg_chi, (npw_c*nspinor))
# # # # #
# ]
1374 : : end if
1375 : :
1376 : : ! We need two stern_t objects to compute the first order change of the wavefunctions at k-p and k+q-p.
1377 : : ! Clearly, we should not duplicate the work when pp = 0.
1378 : : ! When pp == 0, we also get the gks matrix elements after stern_solve.
1379 : : ! Alternatively, one can solve the Sternheimer in the IBZ(kappa, alpha), store the results on disk
1380 : : ! and then use symmetries to reconstruct delta_u in the full BZ on the fly assuming spatial inversion or TR.
1381 : : ! Also, one should handle more carefully the integration in g_sigma around pp = Gamma in the case of semiconductors.
1382 : :
1383 : 696 : nband_me = nbsum; stern_comm = xmpi_comm_self
1384 : :
1385 : : ! =======================
1386 : : ! Delta_{qka} psi_{n'k-p}
1387 : : ! =======================
1388 : : ! (k+q-p, k-p)
1389 : : call stern_kmp%init(dtset, npw_kmp, npw_kqmp, nspinor, nbsum, nband_me, fermie1_idir_ipert, &
1390 : 696 : stern_use_cache, work_ngfft, mpi_enreg, stern_comm)
1391 : :
1392 : : ! Use symmetries to reconstruct GS wavefunctions at k+q-p and store them in stern_kmp%cgq.
1393 : : ! NB: all the nbsum bands are needed here, also when the sum over bands is distributed.
1394 [ + + ]: 8024 : do ib_sum=1,nbsum
1395 : : call wfd%rotate_cg(ib_sum, ndat1, spin, kqmp_ibz, npw_kqmp, kg_kqmp, istwf_kqmp, &
1396 : 7328 : cryst, mapl_kqmp, gbound_kqmp, work_ngfft, work, cg_kqmp)
1397 : :
1398 : : ! NB: cg_kqmp is dimensioned with mpw --> have to slice cg_kqmp
1399 [ + + + + ]: 27033572 : stern_kmp%cgq(:,:,ib_sum) = cg_kqmp(:,1:npw_kqmp*nspinor)
1400 : : end do ! ib_sum
1401 : :
1402 : : ! ==========================
1403 : : ! Delta_{-qka} psi_{n'k+q-p}
1404 : : ! ==========================
1405 : : ! (k-p, k+q-p)
1406 : : call stern_kqmp%init(dtset, npw_kqmp, npw_kmp, nspinor, nbsum, nband_me, fermie1_idir_ipert, &
1407 : 696 : stern_use_cache, work_ngfft, mpi_enreg, stern_comm)
1408 : :
1409 : : ! Use symmetries to reconstruct GS wavefunctions at k-p and store them in stern_kqmp%cgq.
1410 [ + + ]: 8024 : do ib_sum=1,nbsum
1411 : : call wfd%rotate_cg(ib_sum, ndat1, spin, kmp_ibz, npw_kmp, kg_kmp, istwf_kmp, &
1412 : 7328 : cryst, mapl_kmp, gbound_kmp, work_ngfft, work, cg_kmp)
1413 : :
1414 : : ! NB: cg_kmp is dimensioned with mpw --> have to slice cg_kmp
1415 [ + + + + ]: 27033572 : stern_kqmp%cgq(:,:,ib_sum) = cg_kmp(:,1:npw_kmp*nspinor)
1416 : : end do ! ib_sum
1417 : :
1418 : : ! Get PPM parameters at pp_bz to applying W_{gg'}(pp_bz).
1419 : : ! Note: Sq_ibz = q_bz + G0 with non-zero G0 is not supported here
1420 [ - + - + : 696 : call ppm%get_qbz(gsph_c, pp_mesh, ipp_bz, botsq_pbz, otq_pbz, dmeig_pbz)
- + ]
1421 : :
1422 : : ! Need transpose of PPM matrices when summing over the G index.
1423 [ + - ]: 696 : if (ppm%dm2_botsq /= 0) then
1424 [ - + ]: 696 : call alloc_copy(botsq_pbz, trans_botsq_pbz)
1425 : 696 : call sqmat_itranspose(npw_c, trans_botsq_pbz)
1426 : : end if
1427 [ + - ]: 696 : if (ppm%dm2_otq /= 0) then
1428 [ - + ]: 696 : call alloc_copy(otq_pbz, trans_otq_pbz)
1429 : 696 : call sqmat_itranspose(npw_c, trans_otq_pbz)
1430 : : end if
1431 [ + - ]: 696 : if (ppm%dm_eig /= 0) then
1432 [ - + ]: 696 : call alloc_copy(dmeig_pbz, trans_dmeig_pbz)
1433 : 696 : call sqmat_itranspose(ppm%dm_eig, trans_dmeig_pbz)
1434 : : end if
1435 : :
1436 : : ! Find the corresponding irred pp-point in the pp_mesh.
1437 : 696 : call pp_mesh%get_bz_item(ipp_bz, pp, ipp_ibz, isym_pp, itim_pp)
1438 : :
1439 : : ! Get Fourier components of the Coulomb interaction in the BZ
1440 : : ! In 3D systems, neglecting umklapp: vc(Sq,sG) = vc(q,G) = 4pi/|q+G|**2
1441 : : ! The same relation holds for 0-D systems, but not in 1-D or 2D systems. It depends on S.
1442 : : ! NOTE: vc_sqrt_gx is dimensioned with npw_x --> use rottb table from gsph_x.
1443 [ + + ]: 396144 : do ig=1,npw_x
1444 : 396144 : vc_sqrt_gx(gsph_x%rottb(ig, itim_pp, isym_pp)) = vcp%vc_sqrt(ig, ipp_ibz)
1445 : : end do
1446 : :
1447 [ - + ]: 696 : if (dtset%gwcomp == 2) then
1448 : : ! Compute static limit at pp_bz from the symmetrical image in the IBZ
1449 : 0 : call em1_symmetrize_op(ipp_bz, npw_c, 1, gsph_c, pp_mesh, wc0_pibz(ipp_ibz)%vals, wc0_pbz)
1450 : : end if
1451 : 696 : call timab(1941, 2, tsec)
1452 : :
1453 : : ! ===========================================
1454 : : ! Sum over bands (n' index in equations), distributed inside bsum_comm.
1455 : : ! All procs in gqk%pert_comm enter this part.
1456 : : ! ===========================================
1457 : :
1458 [ + + ]: 8024 : do ib_sum=my_bsum_start(spin), my_bsum_stop(spin)
1459 : 7328 : call timab(1942, 1, tsec)
1460 : : ! Get u_{n',k-p}(r)^*, stored in ur_star_kmp.
1461 : : call wfd%rotate_cg(ib_sum, ndat1, spin, kmp_ibz, npw_kmp, kg_kmp, istwf_kmp, &
1462 : 7328 : cryst, mapl_kmp, gbound_kmp, work_ngfft, work, cg_kmp, urs_kbz=ur_star_kmp)
1463 : :
1464 [ + + ]: 38911136 : ur_star_kmp = GWPC_CONJG(ur_star_kmp)
1465 : 7328 : ebsum_kmp = ebands%eig(ib_sum, ikmp_ibz, spin)
1466 : 7328 : ebsum_kqmp = ebands%eig(ib_sum, ikqmp_ibz, spin)
1467 : :
1468 : : ! Energy of the states above nbsum for gwcomp 1 and -2.
1469 : 7328 : e_high = en_high
1470 [ - + ]: 7328 : if (dtset%gwcomp == -2) e_high = half * (qp_ene(nbsum, ikmp_ibz, spin) + qp_ene(nbsum, ikqmp_ibz, spin)) + dyn_shift
1471 : :
1472 : : ! =====================================
1473 : : ! Precompute oscillator matrix elements
1474 : : ! =====================================
1475 : : ! These terms do not depend on (idir, ipert) and can be reused in the loop over perturbations below.
1476 : : ! If the n' bands in the sum are distributed, one should transmit the (m, n) indices.
1477 : :
1478 : 7328 : theta_mu_minus_e0i = fact_spin * qp_occ(ib_sum, ikmp_ibz, spin)
1479 : 7328 : need_x_kmp = (abs(theta_mu_minus_e0i / fact_spin) >= tol_empty) ! allow negative occ numbers
1480 : : !need_x_kmp = .True.
1481 : : !print *, "kmp, ib_sum, theta_mu_minus_e0i", ib_sum, theta_mu_minus_e0i
1482 : :
1483 : : ! Contract immediately over g' with the frequency convolution:
1484 : : !
1485 : : ! \int de' Wc_{gg'}(pp, e') / (omega - e_{bsum, kmp) - e')
1486 : : !
1487 : : ! vec_gwc_nk(:,:,n_k) stores:
1488 : : !
1489 : : ! sum_g' \int de' Wc_{gg'}(pp, e') / (omega - e_{bsum, kmp) - e') <bsum,k-p|e^{-i(p+g')}r|n,k>
1490 : : !
1491 [ - + ]: 7328 : if (gqk%pert_comm%nproc > 1) then
1492 [ # # # # : 0 : vec_gwc_nk = zero
# # ]
1493 [ # # # # ]: 0 : vec_gx_nk = zero
1494 [ # # # # : 0 : if (use_dg0) vec_dgwc_nk = zero
# # # # ]
1495 [ # # # # : 0 : if (dtset%gwcomp == 2) vec_coh_nk = zero
# # ]
1496 : : end if
1497 : :
1498 [ + + ]: 41056 : do n_k=bstart_k, bstop_k
1499 [ + - ]: 33728 : in_k = n_k - bstart_k + 1; if (gqk%pert_comm%skip(in_k)) cycle ! MPI parallelism inside pert_comm
1500 : :
1501 : : ! Compute <bsum,k-p|e^{-i(p+G')}r|n,k> * vc_sqrt(p,G')
1502 [ + - - + : 173770624 : cwork_ur = ur_star_kmp * ur_nk(:,n_k)
- - + + ]
1503 : :
1504 [ + + ]: 33728 : if (need_x_kmp) then
1505 : 20800 : call fft_ur(npw_x, nfft, nspinor, ndat1, mgfft, ngfft, istwfk1, kg_x, gbound_x, cwork_ur, rhotwg_x)
1506 : 20800 : call sigtk_multiply_by_vc_sqrt("N", npw_x, nspinor, ndat1, vc_sqrt_gx, rhotwg_x)
1507 [ + + ]: 20800 : if (pp_is_gamma) then
1508 [ + + ]: 808 : rhotwg_x(1) = czero_gw; if (ib_sum == n_k) rhotwg_x(1) = cmplx(sqrt(vcp%i_sz),0.0_gwp)
1509 : : end if
1510 : :
1511 [ + + ]: 12579456 : vec_gx_nk(:,n_k) = rhotwg_x(1:npw_x*nspinor)
1512 [ - + - - ]: 20800 : if (dtset%userid /= 0) vec_gx_nk(:,n_k) = zero
1513 : : ! FIXME: This is wrong if nspinor == 2
1514 [ + + ]: 553600 : rhotwg_c(:) = rhotwg_x(1:npw_c*nspinor)
1515 : :
1516 : : if (qq_is_gamma .and. test_sigma) then
1517 : : sigx_nk(in_k, ik_glob) = sigx_nk(in_k, ik_glob) + dot_product(rhotwg_x, rhotwg_x)
1518 : : end if
1519 : :
1520 : : else
1521 : 12928 : call fft_ur(npw_c, nfft, nspinor, ndat1, mgfft, ngfft, istwfk1, kg_c, gbound_c, cwork_ur, rhotwg_c)
1522 : 12928 : call sigtk_multiply_by_vc_sqrt("N", npw_c, nspinor, ndat1, vc_sqrt_gx, rhotwg_c)
1523 [ + + ]: 12928 : if (pp_is_gamma) then
1524 [ + + ]: 1168 : rhotwg_c(1) = czero_gw; if (ib_sum == n_k) rhotwg_c(1) = cmplx(sqrt(vcp%i_sz),0.0_gwp)
1525 : : end if
1526 : : end if
1527 : :
1528 : : ! Prepare list of omegas: first e_nk then e_mkq for all m indices.
1529 : 33728 : omegas_nk(1) = qp_ene(n_k, ik_ibz, spin); cnt = 1
1530 : :
1531 [ + + ]: 33728 : if (dtset%gwpt_wmode == 1) then
1532 [ + + ]: 172224 : do m_kq=bstart_kq, bstop_kq
1533 : 172224 : cnt = cnt + 1; omegas_nk(cnt) = qp_ene(m_kq, ikq_ibz, spin)
1534 : : end do
1535 : : end if
1536 [ - + + + ]: 260160 : omegame0i_nk = omegas_nk - qp_ene(ib_sum, ikmp_ibz, spin)
1537 : :
1538 : : ! Note that the i/two_pi factor in Sigma(w) is included in calc_sigc
1539 [ + + + + ]: 11728192 : vec_gwc_nk(:,:,n_k) = zero
1540 : : call ppm%calc_sigc(nspinor, npw_c, nw_nk, rhotwg_c, botsq_pbz, otq_pbz, &
1541 : : omegame0i_nk, dtset%zcut, theta_mu_minus_e0i, dmeig_pbz, npw_c, &
1542 : 33728 : vec_gwc_nk(:,:,n_k), sigcme_nk)
1543 : :
1544 [ - + ]: 33728 : if (use_dg0) then
1545 : : ! Same contraction but with the omega-derivative of the PPM kernel (dWtilde/de).
1546 : : ! Contracted with the oscillator of the auxiliary state chi_{bsum,k-p} in the perturbation loop below.
1547 [ # # # # ]: 0 : vec_dgwc_nk(:,:,n_k) = zero
1548 : : call ppm%calc_dsigc(nspinor, npw_c, nw_nk, rhotwg_c, botsq_pbz, otq_pbz, &
1549 : 0 : omegame0i_nk, dtset%zcut, theta_mu_minus_e0i, npw_c, vec_dgwc_nk(:,:,n_k))
1550 [ # # # # : 0 : if (dtset%useric /= 0) vec_dgwc_nk(:,:,n_k) = zero
# # ]
1551 : : end if
1552 : :
1553 [ - + ]: 33728 : if (dtset%gwcomp == 2) then
1554 [ # # # # : 0 : vec_coh_nk(:, n_k) = matmul(wc0_pbz, rhotwg_c)
# # # # #
# # # #
# ]
1555 : : !call xgemv("N", npw_c, npw_c, cone_gw, wc0_pbz, npw_c, rhotwg_c, 1, czero_gw, vec_coh_nk(:, n_k), 1)
1556 : : end if
1557 : :
1558 : : if (qq_is_gamma .and. test_sigma) then
1559 : : sigce0_nk(in_k, ik_glob) = sigce0_nk(in_k, ik_glob) + dot_product(rhotwg_c, vec_gwc_nk(:,1,n_k))
1560 : : end if
1561 : :
1562 [ - + ]: 33728 : if (any(dtset%gwcomp == [1, -2])) then
1563 : : ! Subtract the PPM term of the same state moved to E_high (empty, theta = 0).
1564 : : ! NB: vec_comp_nk must be zeroed because calc_sigc multiplies the whole ket by 1/2.
1565 [ # # # # ]: 0 : omegame0i_nk = omegas_nk - e_high
1566 [ # # # # ]: 0 : vec_comp_nk = zero
1567 : : call ppm%calc_sigc(nspinor, npw_c, nw_nk, rhotwg_c, botsq_pbz, otq_pbz, &
1568 : 0 : omegame0i_nk, dtset%zcut, zero, dmeig_pbz, npw_c, vec_comp_nk, sigcme_nk)
1569 [ # # # # ]: 0 : vec_gwc_nk(:,:,n_k) = vec_gwc_nk(:,:,n_k) - vec_comp_nk
1570 : : end if
1571 : :
1572 [ - + - - : 41056 : if (dtset%useric /= 0) vec_gwc_nk(:,:,n_k) = zero
- - ]
1573 : : end do ! n_k
1574 : :
1575 : : ! TODO: this is an all_gatherv but oh well.
1576 [ - + ]: 7328 : if (gqk%pert_comm%nproc > 1) then
1577 : 0 : call xmpi_sum(vec_gx_nk, gqk%pert_comm%value, ierr)
1578 : 0 : call xmpi_sum(vec_gwc_nk, gqk%pert_comm%value, ierr)
1579 [ # # ]: 0 : if (use_dg0) call xmpi_sum(vec_dgwc_nk, gqk%pert_comm%value, ierr)
1580 [ # # ]: 0 : if (dtset%gwcomp == 2) call xmpi_sum(vec_coh_nk, gqk%pert_comm%value, ierr)
1581 : : end if
1582 : :
1583 : 7328 : call timab(1942, 2, tsec)
1584 : 7328 : call timab(1943, 1, tsec)
1585 : :
1586 : : ! Get u_{n',k+q-p}(r), stored in ur_star_kqmp
1587 : : call wfd%rotate_cg(ib_sum, ndat1, spin, kqmp_ibz, npw_kqmp, kg_kqmp, istwf_kqmp, &
1588 : 7328 : cryst, mapl_kqmp, gbound_kqmp, work_ngfft, work, cg_kqmp, urs_kbz=ur_star_kqmp)
1589 : :
1590 [ + + ]: 38911136 : ur_star_kqmp = GWPC_CONJG(ur_star_kqmp)
1591 : 7328 : theta_mu_minus_e0i = fact_spin * qp_occ(ib_sum, ikqmp_ibz, spin)
1592 : :
1593 : 7328 : need_x_kqmp = (abs(theta_mu_minus_e0i / fact_spin) >= tol_empty) ! allow negative occ numbers
1594 : : !print *, "kqmp, ib_sum, theta_mu_minus_e0i", ib_sum, theta_mu_minus_e0i
1595 : : !need_x_kqmp = .True.
1596 : :
1597 : : ! Contract immediately over g with the frequency convolution:
1598 : : !
1599 : : ! \int de' Wc_{gg'}(pp, e') / (omega - e_{bsum, kqmp) - e')
1600 : : !
1601 : : ! Store results in vec_gwc_mkq(:,:,m_kq).
1602 : : !
1603 : : ! sum_g \int de' Wc_{gg'}(pp, e') / (omega - e_{bsum, kqmp) - e') <m,k+q|e^{+i(p+g)}r|bsum,k+q-p>
1604 : : !
1605 [ - + ]: 7328 : if (gqk%pert_comm%nproc > 1) then
1606 [ # # # # : 0 : vec_gwc_mkq = zero
# # ]
1607 [ # # # # : 0 : if (need_x_kqmp) vec_gx_mkq = zero
# # ]
1608 [ # # # # : 0 : if (dtset%gwcomp == 2) vec_coh_mkq = zero
# # ]
1609 : : end if
1610 : :
1611 [ + + ]: 85024 : do m_kq=bstart_kq, bstop_kq
1612 [ + - ]: 77696 : im_kq = m_kq - bstart_kq + 1; if (gqk%pert_comm%skip(im_kq)) cycle ! MPI parallelism inside pert_comm
1613 : :
1614 : : ! <m,k+q|e^{i(p+G)}r|bsum,k+q-p> * vc_sqrt(p,G).
1615 : : ! Exchange bra and ket and take the CC of the FFT in sigtk_multiply_by_vc_sqrt.
1616 [ + - - + : 407281408 : cwork_ur = ur_star_kqmp * ur_mkq(:,m_kq)
- - + + ]
1617 : :
1618 [ + + ]: 77696 : if (need_x_kqmp) then
1619 : 49792 : call fft_ur(npw_x, nfft, nspinor, ndat1, mgfft, ngfft, istwfk1, kg_x, gbound_x, cwork_ur, rhotwg_x)
1620 : 49792 : call sigtk_multiply_by_vc_sqrt("C", npw_x, nspinor, ndat1, vc_sqrt_gx, rhotwg_x)
1621 [ + + ]: 49792 : if (pp_is_gamma) then
1622 [ + + ]: 1744 : rhotwg_x(1) = czero_gw; if (ib_sum == m_kq) rhotwg_x(1) = cmplx(sqrt(vcp%i_sz),0.0_gwp)
1623 : : end if
1624 : :
1625 [ + + ]: 31302912 : vec_gx_mkq(:,m_kq) = rhotwg_x(1:npw_x*nspinor)
1626 [ - + - - ]: 49792 : if (dtset%userid /= 0) vec_gx_mkq(:,m_kq) = zero
1627 [ + + ]: 1238272 : rhotwg_c(:) = rhotwg_x(1:npw_c*nspinor)
1628 : :
1629 : : else
1630 : 27904 : call fft_ur(npw_c, nfft, nspinor, ndat1, mgfft, ngfft, istwfk1, kg_c, gbound_c, cwork_ur, rhotwg_c)
1631 : 27904 : call sigtk_multiply_by_vc_sqrt("C", npw_c, nspinor, ndat1, vc_sqrt_gx, rhotwg_c)
1632 [ + + ]: 27904 : if (pp_is_gamma) then
1633 [ + + ]: 2368 : rhotwg_c(1) = czero_gw; if (ib_sum == m_kq) rhotwg_c(1) = cmplx(sqrt(vcp%i_sz),0.0_gwp)
1634 : : end if
1635 : : end if
1636 : :
1637 [ + + ]: 77696 : if (dtset%gwpt_wmode == 1) then
1638 : : ! Prepare list of omegas: first e_mkq then e_nk for all n indices.
1639 : 26496 : omegas_mkq(1) = qp_ene(m_kq, ikq_ibz, spin); cnt = 1
1640 [ + + ]: 185472 : do n_k=bstart_k, bstop_k
1641 : 185472 : cnt = cnt + 1; omegas_mkq(cnt) = qp_ene(n_k, ik_ibz, spin)
1642 : : end do
1643 : : else
1644 : : ! Here we compute the convolution at e_nk for all n_k bands at fixed m_kq
1645 : 51200 : cnt = 0
1646 [ + + ]: 256000 : do n_k=bstart_k, bstop_k
1647 : 256000 : cnt = cnt + 1; omegas_mkq(cnt) = qp_ene(n_k, ik_ibz, spin)
1648 : : end do
1649 : : end if
1650 : :
1651 [ - + + + ]: 545664 : omegame0i_mkq = omegas_mkq - qp_ene(ib_sum, ikqmp_ibz, spin)
1652 : :
1653 : : ! Here we sum over G instead of G' so we have to pass the transpose of the PPM matrix elements.
1654 : : ! TODO: Generalize ppm%calc_sigc with BLAS-like API.
1655 [ + + + + ]: 15595648 : vec_gwc_mkq(:,:,m_kq) = zero
1656 : : call ppm%calc_sigc(nspinor, npw_c, nw_mkq, rhotwg_c, trans_botsq_pbz, trans_otq_pbz, &
1657 : : omegame0i_mkq, dtset%zcut, theta_mu_minus_e0i, trans_dmeig_pbz, npw_c, &
1658 : 77696 : vec_gwc_mkq(:,:,m_kq), sigcme_mkq)
1659 : :
1660 [ - + ]: 77696 : if (any(dtset%gwcomp == [1, -2])) then
1661 : : ! Subtract the PPM term of the same state moved to E_high (empty, theta = 0), transposed PPM matrices.
1662 [ # # # # ]: 0 : omegame0i_mkq = omegas_mkq - e_high
1663 [ # # # # ]: 0 : vec_comp_mkq = zero
1664 : : call ppm%calc_sigc(nspinor, npw_c, nw_mkq, rhotwg_c, trans_botsq_pbz, trans_otq_pbz, &
1665 : 0 : omegame0i_mkq, dtset%zcut, zero, trans_dmeig_pbz, npw_c, vec_comp_mkq, sigcme_mkq)
1666 [ # # # # ]: 0 : vec_gwc_mkq(:,:,m_kq) = vec_gwc_mkq(:,:,m_kq) - vec_comp_mkq
1667 : : end if
1668 : :
1669 [ - + ]: 77696 : if (dtset%gwcomp == 2) then
1670 : : ! Same contraction as the -q main term (transposed PPM matrices applied to rhotwg_c, no complex conjugate):
1671 : : ! the static limit of calc_sigc with trans_botsq_pbz is (1/2 - f) W^c(0)^T rhotwg_c.
1672 [ # # # # : 0 : vec_coh_mkq(:, m_kq) = matmul(rhotwg_c, wc0_pbz)
# # # # #
# # # # #
# # ]
1673 : : !call xgemv("H", npw_c, npw_c, cone_gw, wc0_pbz, npw_c, rhotwg_c, 1, czero_gw, vec_coh_mkq(:, m_kq), 1)
1674 : : end if
1675 : :
1676 [ - + - - : 85024 : if (dtset%useric /= 0) vec_gwc_mkq(:,:,m_kq) = zero
- - ]
1677 : : end do ! m_kq
1678 : :
1679 : : ! TODO: this is an all_gatherv but oh well.
1680 [ - + ]: 7328 : if (gqk%pert_comm%nproc > 1) then
1681 : 0 : call xmpi_sum(vec_gwc_mkq, gqk%pert_comm%value, ierr)
1682 [ # # ]: 0 : if (need_x_kqmp) call xmpi_sum(vec_gx_mkq, gqk%pert_comm%value, ierr)
1683 [ # # ]: 0 : if (dtset%gwcomp == 2) call xmpi_sum(vec_coh_mkq, gqk%pert_comm%value, ierr)
1684 : : end if
1685 : :
1686 : 7328 : call timab(1943, 2, tsec)
1687 : :
1688 : : ! ========================================
1689 : : ! Loop over my set of atomic perturbations
1690 : : ! ========================================
1691 : : ! For each perturbation:
1692 : : ! - setup H1 from vlocal1_qq or vlocal1_mqq.
1693 : : ! For each band in band_sum:
1694 : : ! - Solve the Sternheimer non-self-consistently and get the KS e-ph matrix elements.
1695 : : ! - Build the full first-order wavefunction including the active subspace.
1696 : :
1697 [ + + ]: 51296 : do imyp=1,gqk%my_npert
1698 : 43968 : call timab(1944, 1, tsec)
1699 : : ! NB: Only one proc enters this section. No MPI parallelism is allowed here.
1700 : 43968 : idir = dvdb%my_pinfo(1, imyp); ipert = dvdb%my_pinfo(2, imyp); ipc = dvdb%my_pinfo(3, imyp)
1701 : : !print *, "For kk, ", kk, "pp:", pp, "idir, ipert", idir, ipert
1702 : :
1703 : : ! Set up local potential vlocal1_qq with proper dimensioning, from vtrial1 taking into account the spin
1704 : : ! and prepare application of the NL part. Each MPI rank prepares its own potential.
1705 : : call rf_transgrid_and_pack(spin, nspden, psps%usepaw, cplex, nfftf, nfft, ngfft, nvloc, &
1706 : 43968 : pawfgr, mpi_enreg, vtrial, v1scf_qq(:,:,:,imyp), vlocal, vlocal1_qq(:,:,:,:,imyp))
1707 : :
1708 : 43968 : call gs_ham_kqmp%load_spin(spin, vlocal=vlocal, with_nonlocal=.true.)
1709 : :
1710 : 43968 : call rf_ham_kqmp%init(cplex, gs_ham_kqmp, ipert, has_e1kbsc=.true.)
1711 : 43968 : call rf_ham_kqmp%load_spin(spin, vlocal1=vlocal1_qq(:,:,:,:,imyp), with_nonlocal=.true.)
1712 : :
1713 : : ! ======================
1714 : : ! NSCF Sternheimer at qq
1715 : : ! ======================
1716 : : ! Compute Delta_{q,idir,ipert} \psi_{bsum, k-p}
1717 : : call wfd%rotate_cg(ib_sum, ndat1, spin, kmp_ibz, npw_kmp, kg_kmp, istwf_kmp, &
1718 : 43968 : cryst, mapl_kmp, gbound_kmp, work_ngfft, work, cg_kmp)
1719 : :
1720 [ + + ]: 510144 : stern_kmp%bands_treated_now(:) = 0; stern_kmp%bands_treated_now(ib_sum) = 1
1721 [ + + ]: 510144 : stern_kmp%rank_band = 0; u1_band = ib_sum; band_me = ib_sum
1722 : :
1723 : : ! (k+q-p, k-p)
1724 : 43968 : init_mode = "None"
1725 : : call stern_kmp%solve(u1_band, band_me, idir, ipert, qq_bz, gs_ham_kqmp, rf_ham_kqmp, &
1726 : : ebands%eig(:,ikmp_ibz,spin), ebands%eig(:,ikqmp_ibz,spin), &
1727 : : cg_kmp, cwaveprj0, cg1_kqmp, cwaveprj, msg, ierr, &
1728 : 43968 : full_cg1=full_cg1_kqmp, full_ur1=full_ur1_kqmp, init_mode=init_mode)
1729 : :
1730 : : ! The last bands may fail to converge with resid=-two. In this case we ignore the contribution.
1731 [ + + ]: 43968 : if (ierr /= 0) then
1732 [ + + + + : 1446230 : full_cg1_kqmp = zero; full_ur1_kqmp = zero; stern_qq_ierr = stern_qq_ierr + 1
+ + ]
1733 [ + + ]: 182 : if (stern_qq_ierr <= ENOUGH_STERN) then
1734 : 15 : ABI_WARNING(sjoin("Stern at +q", qkp_string, msg))
1735 : : end if
1736 : : end if
1737 : :
1738 : : ! Store KS e-ph matrix elements for this ipc perturbation.
1739 [ + + + + : 43968 : if (pp_is_gamma .and. (ib_sum >= gqk%bstart_k .and. ib_sum <= gqk%bstop_k)) then
+ + ]
1740 : 1020 : ii = ib_sum - bstart_k + 1
1741 [ + + + + ]: 36588 : gks_atm(:,:,ii,ipc) = stern_kmp%eig1_k(:, bstart_kq:bstop_kq, ib_sum)
1742 : : end if
1743 : :
1744 [ - + ]: 43968 : if (use_dg0) then
1745 : : ! -----------------------------------------------------------------------------------------
1746 : : ! Auxiliary state for the dG0/de term at q=Gamma (Appendix A, degenerate intermediate states):
1747 : : !
1748 : : ! |chi_{n'k-p}> = sum_{n''} eta^2/(Delta^2+eta^2) g^KS_{n''n'} |psi_{n''k-p}>,
1749 : : ! Delta = e_{n''} - e_{n'}.
1750 : : !
1751 : : ! The exact variation dG0 = G0 dV G0 gives, for each pair (n'', n'), the partial fractions
1752 : : ! [1/(e-e_{n''}) - 1/(e-e_{n'})] / Delta. The +q/-q dpsi terms reproduce them with 1/Delta
1753 : : ! regularized as Delta/(Delta^2+eta^2) (see full_active_wf1), i.e. only the fraction
1754 : : ! Delta^2/(Delta^2+eta^2). The missing Lorentzian-weighted remainder is approximated by the
1755 : : ! double pole 1/(e-e_{n'})^2 (exact for Delta=0, error O(Delta) damped by the weight) and
1756 : : ! contributes -<m|e^{i(p+G)r}|chi_{n'}> dWtilde_{n'}/de(G,G') <n'|e^{-i(p+G')r'}|n>.
1757 : : ! For a non-degenerate n' the weight collapses to delta_{n''n'} and chi -> (d eps_{n'}) psi_{n'}.
1758 : : ! Inside a degenerate manifold the full block P dV P enters, so the result is basis-invariant.
1759 : : ! n'' runs over the active space only: the Sternheimer complement is not regularized.
1760 : : ! -----------------------------------------------------------------------------------------
1761 [ # # ]: 0 : chi_ug_kqmp = czero_gw
1762 [ # # ]: 0 : do jb_sum=1,nbsum
1763 [ # # # # ]: 0 : if (dtset%userrb /= zero .and. jb_sum /= ib_sum) cycle ! debug: non-degenerate (diagonal) form
1764 : 0 : delta_E = ebsum_kmp - ebands%eig(jb_sum, ikqmp_ibz, spin)
1765 : 0 : wgt_dg0 = eta_sum ** 2 / (delta_E ** 2 + eta_sum ** 2)
1766 : 0 : ctmp_dp = wgt_dg0 * r2c(stern_kmp%eig1_k(:, jb_sum, ib_sum))
1767 : : chi_ug_kqmp = chi_ug_kqmp + cmplx(ctmp_dp, kind=gwp) * &
1768 [ # # # # ]: 0 : cmplx(stern_kmp%cgq(1,:,jb_sum), stern_kmp%cgq(2,:,jb_sum), kind=gwp)
1769 : : end do
1770 : : ! Store chi^*(r): the oscillator is computed with bra and ket exchanged, as for Delta_q psi below.
1771 : 0 : call fft_ug(npw_kqmp, nfft, nspinor, ndat1, mgfft, ngfft, istwfk1, kg_kqmp, gbound_kqmp, chi_ug_kqmp, chi_ur_kqmp)
1772 [ # # ]: 0 : chi_ur_kqmp = GWPC_CONJG(chi_ur_kqmp)
1773 : : end if
1774 : :
1775 : : ! Compute <m,k+q|e^{i(p+G)r}|Delta_q psi_{bsum,k-p}>
1776 : : ! Exchange bra and ket and take the CC of the FFT.
1777 [ + + ]: 233466816 : full_ur1_kqmp = GWPC_CONJG(full_ur1_kqmp)
1778 : :
1779 [ + + ]: 510144 : do m_kq=bstart_kq, bstop_kq
1780 : 466176 : im_kq = m_kq - bstart_kq + 1
1781 [ + - - + : 2443688448 : cwork_ur = full_ur1_kqmp * ur_mkq(:,m_kq)
- - + + ]
1782 : :
1783 : 466176 : ctmp_dp = r2c(stern_kmp%eig1_k(:, m_kq, ib_sum))
1784 : 466176 : e_mkq = ebands%eig(m_kq, ikq_ibz, spin)
1785 : 466176 : delta_E = ebsum_kmp - e_mkq
1786 : 466176 : inv_delta_E = delta_E / ( delta_E ** 2 + eta_sum ** 2)
1787 : 466176 : ctmp_dp = ctmp_dp * inv_delta_E * sqrt(vcp%i_sz)
1788 : : !ctmp_dp = zero
1789 : : !if (qq_is_gamma) ctmp_dp = zero
1790 : :
1791 : : ! SC: the bare-exchange occupation factor belongs to the intermediate state entering tilde_v
1792 : : ! i.e. f(ib_sum, k-p) for the +q term of Delta Sigma. See Eqs. (eq:sigma_g) & (tv_bare) in the paper.
1793 [ + + ]: 466176 : if (need_x_kmp) then
1794 : 298752 : call fft_ur(npw_x, nfft, nspinor, ndat1, mgfft, ngfft, istwfk1, kg_x, gbound_x, cwork_ur, rhotwg_x)
1795 : : !if (pp_is_gamma) print *, "rhotwg_x(1)", rhotwg_x(1), "vc(0)", vc_sqrt_gx(1)
1796 : 298752 : ctmp_gwpc = rhotwg_x(1)
1797 : 298752 : call sigtk_multiply_by_vc_sqrt("C", npw_x, nspinor, ndat1, vc_sqrt_gx, rhotwg_x)
1798 [ + + ]: 298752 : if (pp_is_gamma) then
1799 : : ! SC: the FFT gives the CC of the wanted oscillator (bra/ket exchanged, see above);
1800 : : ! mode "C" conjugates the body (G/=0), so the head (G=0) saved BEFORE that call
1801 : : ! must be conjugated here as well. The -q term keeps the derivative wfn in the
1802 : : ! bra, uses mode "N", and thus needs no conjugation.
1803 : 10464 : rhotwg_x(1) = GWPC_CONJG(ctmp_gwpc) * cmplx(sqrt(vcp%i_sz), 0.0_gwp)
1804 : : !print *, "ctmp_dp +qq", ctmp_gwpc
1805 : : !print *, "rhotwg_x(1)", rhotwg_x(1)
1806 : : !rhotwg_x(1) = ctmp_dp
1807 : : end if
1808 [ + + ]: 7429632 : rhotwg_c(:) = rhotwg_x(1:npw_c*nspinor)
1809 : : else
1810 : 167424 : call fft_ur(npw_c, nfft, nspinor, ndat1, mgfft, ngfft, istwfk1, kg_c, gbound_c, cwork_ur, rhotwg_c)
1811 : 167424 : ctmp_gwpc = rhotwg_c(1)
1812 : 167424 : call sigtk_multiply_by_vc_sqrt("C", npw_c, nspinor, ndat1, vc_sqrt_gx, rhotwg_c)
1813 [ + + ]: 167424 : if (pp_is_gamma) then
1814 : : ! SC: the FFT gives the CC of the wanted oscillator (bra/ket exchanged, see above);
1815 : : ! mode "C" conjugates the body (G/=0), so the head (G=0) saved BEFORE that call
1816 : : ! must be conjugated here as well. The -q term keeps the derivative wfn in the
1817 : : ! bra, uses mode "N", and thus needs no conjugation.
1818 : 14208 : rhotwg_c(1) = GWPC_CONJG(ctmp_gwpc) * cmplx(sqrt(vcp%i_sz), 0.0_gwp)
1819 : : !rhotwg_c(1) = ctmp_dp
1820 : : end if
1821 : : end if
1822 : :
1823 [ - + ]: 466176 : if (use_dg0) then
1824 : : ! <m,k+q| e^{i(p+G)r} |chi_{bsum,k-p}> * vc_sqrt(p,G): bra and ket exchanged as above, hence mode "C".
1825 : : ! Only the correlation sphere is needed since the bare v is frequency-independent.
1826 [ # # # # : 0 : cwork_ur = chi_ur_kqmp * ur_mkq(:,m_kq)
# # # # ]
1827 : 0 : call fft_ur(npw_c, nfft, nspinor, ndat1, mgfft, ngfft, istwfk1, kg_c, gbound_c, cwork_ur, rhotwg_chi)
1828 : 0 : call sigtk_multiply_by_vc_sqrt("C", npw_c, nspinor, ndat1, vc_sqrt_gx, rhotwg_chi)
1829 [ # # ]: 0 : if (pp_is_gamma) then
1830 : : ! Analytic head: <m,k|chi_{bsum,k}> = eta^2/(Delta^2+eta^2) g^KS_{m,bsum} since m is in the active space.
1831 : 0 : wgt_dg0 = eta_sum ** 2 / (delta_E ** 2 + eta_sum ** 2)
1832 [ # # # # ]: 0 : if (dtset%userrb /= zero) wgt_dg0 = merge(one, zero, m_kq == ib_sum) ! debug: diagonal form
1833 : 0 : rhotwg_chi(1) = cmplx(wgt_dg0 * r2c(stern_kmp%eig1_k(:, m_kq, ib_sum)) * sqrt(vcp%i_sz), kind=gwp)
1834 : : end if
1835 : : end if
1836 : :
1837 [ + + ]: 2692800 : do n_k=bstart_k, bstop_k
1838 : 2182656 : in_k = n_k - bstart_k + 1
1839 : :
1840 [ + + ]: 2182656 : if (dtset%gwpt_wmode == 1) then
1841 : : ! SC: +2 because omegas_nk(1) = e_nk and omegas_nk(2:cnt) = e_mkq for all m_kq bands
1842 : 953856 : iw_mkq = m_kq - bstart_kq + 2
1843 : : ! Take the average at e_nk and e_mkq
1844 [ + + ]: 62954496 : ctmp_gwpc = half * sum(rhotwg_c(:) * (vec_gwc_nk(:,1,n_k) + vec_gwc_nk(:,iw_mkq,n_k)))
1845 : : else
1846 : : ! Use the value at e_nk
1847 [ + + ]: 19660800 : ctmp_gwpc = sum(rhotwg_c(:) * (vec_gwc_nk(:,1,n_k)))
1848 : : end if
1849 : :
1850 [ - + ]: 2182656 : if (use_dg0) then
1851 : : ! Additional dG0/de term at q=Gamma (Appendix A):
1852 : : ! -<m,k| e^{i(p+G)r} |chi_{bsum,k-p}> dWtilde/de(G,G') <bsum,k-p| e^{-i(p+G')r'} |n,k>
1853 : : ! This term appears once (it is neither the +q nor the -q dpsi term), hence it is added only here.
1854 [ # # ]: 0 : if (dtset%gwpt_wmode == 1) then
1855 [ # # ]: 0 : dg0_tmp = half * sum(rhotwg_chi(:) * (vec_dgwc_nk(:,1,n_k) + vec_dgwc_nk(:,iw_mkq,n_k)))
1856 : : else
1857 [ # # ]: 0 : dg0_tmp = sum(rhotwg_chi(:) * vec_dgwc_nk(:,1,n_k))
1858 : : end if
1859 : 0 : ctmp_gwpc = ctmp_gwpc - dg0_tmp
1860 : : end if
1861 : :
1862 [ + + ]: 2182656 : if (need_x_kmp) then
1863 : 1300992 : xdot_tmp = - xdotu(npw_x*nspinor, rhotwg_x, 1, vec_gx_nk(:,n_k), 1)
1864 : 1300992 : ctmp_gwpc = ctmp_gwpc + xdot_tmp ! * theta_mu_minus_e0i ! theta_mu_minus_e0i is only needed for metals
1865 : : end if
1866 : :
1867 [ - + ]: 2182656 : if (dtset%gwcomp == 2) then
1868 [ # # ]: 0 : ctmp_gwpc = ctmp_gwpc - quarter * dot_product(conjg(rhotwg_c), vec_coh_nk(:, n_k))
1869 : : end if
1870 : :
1871 [ - + ]: 2648832 : if (dtset%userie < 0) then
1872 [ # # ]: 0 : if (pp_is_gamma) then
1873 : : gsig_atm(1, im_kq, in_k, ipc) = gsig_atm(1, im_kq, in_k, ipc)
1874 : : gsig_atm(2, im_kq, in_k, ipc) = gsig_atm(2, im_kq, in_k, ipc)
1875 : : else
1876 : 0 : gsig_atm(1, im_kq, in_k, ipc) = gsig_atm(1, im_kq, in_k, ipc) + real(ctmp_gwpc)
1877 : 0 : gsig_atm(2, im_kq, in_k, ipc) = gsig_atm(2, im_kq, in_k, ipc) + aimag(ctmp_gwpc)
1878 : : end if
1879 : : else
1880 : 2182656 : gsig_atm(1, im_kq, in_k, ipc) = gsig_atm(1, im_kq, in_k, ipc) + real(ctmp_gwpc)
1881 : 2182656 : gsig_atm(2, im_kq, in_k, ipc) = gsig_atm(2, im_kq, in_k, ipc) + aimag(ctmp_gwpc)
1882 : : end if
1883 : :
1884 : : end do ! n_k
1885 : : end do ! m_kq
1886 : :
1887 : : !if (.not. qq_is_gamma) then
1888 : : ! ==========================
1889 : : ! Same operations but for -q
1890 : : ! ==========================
1891 : : ! Set up local potential vlocal1_qq with proper dimensioning, from vtrial1 taking into account the spin
1892 : : ! and prepare application of the NL part. Each MPI rank prepares its own potential.
1893 : : call rf_transgrid_and_pack(spin, nspden, psps%usepaw, cplex, nfftf, nfft, ngfft, nvloc, &
1894 : 43968 : pawfgr, mpi_enreg, vtrial, v1scf_mq(:,:,:,imyp), vlocal, vlocal1_mqq(:,:,:,:,imyp))
1895 : :
1896 : 43968 : call gs_ham_kmp%load_spin(spin, vlocal=vlocal, with_nonlocal=.true.)
1897 : :
1898 : 43968 : call rf_ham_kmp%init(cplex, gs_ham_kmp, ipert, has_e1kbsc=.true.)
1899 : 43968 : call rf_ham_kmp%load_spin(spin, vlocal1=vlocal1_mqq(:,:,:,:,imyp), with_nonlocal=.true.)
1900 : :
1901 : : ! ======================
1902 : : ! NSCF Sternheimer at -q
1903 : : ! ======================
1904 : : ! Compute Delta_{-q,idir,ipert} \psi_{bsum, k+q-p}
1905 : : call wfd%rotate_cg(ib_sum, ndat1, spin, kqmp_ibz, npw_kqmp, kg_kqmp, istwf_kqmp, &
1906 : 43968 : cryst, mapl_kqmp, gbound_kqmp, work_ngfft, work, cg_kqmp)
1907 : :
1908 [ + + ]: 510144 : stern_kqmp%bands_treated_now(:) = 0; stern_kqmp%bands_treated_now(ib_sum) = 1
1909 [ + + ]: 510144 : stern_kqmp%rank_band = 0; u1_band = ib_sum; band_me = ib_sum
1910 : :
1911 : : !init_mode = "input"
1912 : 43968 : init_mode = "None"
1913 [ - + ]: 43968 : if (init_mode == "input") then
1914 : : call cgtk_change_gsphere(nspinor, &
1915 : : npw_kqmp, istwfk1, kg_kqmp, cg1_kqmp, &
1916 : 0 : npw_kmp, istwfk1, kg_kmp, cg1_kmp, work_ngfft, work)
1917 : : end if
1918 : :
1919 : : ! (k-p, k+q-p)
1920 : : call stern_kqmp%solve(u1_band, band_me, idir, ipert, -qq_bz, gs_ham_kmp, rf_ham_kmp, &
1921 : : ebands%eig(:,ikqmp_ibz,spin), ebands%eig(:,ikmp_ibz,spin), &
1922 : : cg_kqmp, cwaveprj0, cg1_kmp, cwaveprj, msg, ierr, &
1923 [ + + ]: 175872 : full_cg1=full_cg1_kmp, full_ur1=full_ur1_star_kmp, init_mode=init_mode)
1924 : :
1925 [ + + ]: 233466816 : full_ur1_star_kmp = GWPC_CONJG(full_ur1_star_kmp)
1926 : :
1927 : : ! The last bands may fail to converge with resid=-two. In this case we ignore the contribution.
1928 [ + + ]: 43968 : if (ierr /= 0) then
1929 [ + + + + : 1446230 : full_cg1_kmp = zero; full_ur1_star_kmp = zero; stern_mq_ierr = stern_mq_ierr + 1
+ + ]
1930 [ + + ]: 182 : if (stern_mq_ierr <= ENOUGH_STERN) then
1931 : 15 : ABI_WARNING(sjoin("Stern at -q:", qkp_string, msg))
1932 : : end if
1933 : : end if
1934 : :
1935 : : ! For debug, gks_atm2 and gks_atm should be consistent
1936 [ + + + + : 43968 : if (pp_is_gamma .and. (ib_sum >= gqk%bstart_k .and. ib_sum <= gqk%bstop_k)) then
+ + ]
1937 : 1020 : in_k = ib_sum - bstart_k + 1
1938 [ + + + + ]: 36588 : gks_atm2(:,:,in_k,ipc) = stern_kqmp%eig1_k(:, bstart_kq:bstop_kq, ib_sum)
1939 : : end if
1940 : :
1941 : 43968 : call timab(1944, 2, tsec)
1942 : 43968 : call timab(1945, 1, tsec)
1943 [ + + ]: 246336 : do n_k=bstart_k, bstop_k
1944 : 202368 : in_k = n_k - bstart_k + 1
1945 : :
1946 : : ! <Delta_{-q} psi_{bsum,k+q-p}|e^{-i(p+G')r}|n,k>
1947 [ + - - + : 1042623744 : cwork_ur = full_ur1_star_kmp * ur_nk(:,n_k)
- - + + ]
1948 : :
1949 : : ! TODO: Recheck equations, may need a complex conjugate
1950 : 202368 : ctmp_dp = r2c(stern_kqmp%eig1_k(:, n_k, ib_sum))
1951 : 202368 : e_nk = ebands%eig(n_k, ik_ibz, spin)
1952 : 202368 : delta_E = ebsum_kqmp - e_nk
1953 : 202368 : inv_delta_E = delta_E / ( delta_E ** 2 + eta_sum ** 2)
1954 : 202368 : ctmp_dp = ctmp_dp * inv_delta_E * sqrt(vcp%i_sz)
1955 : : !ctmp_dp = zero
1956 : : !if (qq_is_gamma) ctmp_dp = zero
1957 : :
1958 : : ! SC: the bare-exchange occupation factor belongs to the intermediate state entering tilde_v
1959 : : ! i.e. f(ib_sum, k+q-p) for the -q term of Delta Sigma. See Eqs. (eq:sigma_g) & (tv_bare) in the paper.
1960 [ + + ]: 202368 : if (need_x_kqmp) then
1961 : 124800 : call fft_ur(npw_x, nfft, nspinor, ndat1, mgfft, ngfft, istwfk1, kg_x, gbound_x, cwork_ur, rhotwg_x)
1962 : 124800 : ctmp_gwpc = rhotwg_x(1)
1963 : 124800 : call sigtk_multiply_by_vc_sqrt("N", npw_x, nspinor, ndat1, vc_sqrt_gx, rhotwg_x)
1964 [ + + ]: 124800 : if (pp_is_gamma) then
1965 : 4848 : rhotwg_x(1) = ctmp_gwpc * cmplx(sqrt(vcp%i_sz),0.0_gwp)
1966 : : !print *, "ctmp_dp -qq", ctmp_dp
1967 : : !print *, "rhotwg_x(1)", rhotwg_x(1)
1968 : : !rhotwg_x(1) = ctmp_dp
1969 : : end if
1970 : :
1971 [ + + ]: 3321600 : rhotwg_c(:) = rhotwg_x(1:npw_c*nspinor)
1972 : : else
1973 : 77568 : call fft_ur(npw_c, nfft, nspinor, ndat1, mgfft, ngfft, istwfk1, kg_c, gbound_c, cwork_ur, rhotwg_c)
1974 : 77568 : ctmp_gwpc = rhotwg_c(1)
1975 : 77568 : call sigtk_multiply_by_vc_sqrt("N", npw_c, nspinor, ndat1, vc_sqrt_gx, rhotwg_c)
1976 [ + + ]: 77568 : if (pp_is_gamma) then
1977 : 7008 : rhotwg_c(1) = ctmp_gwpc * cmplx(sqrt(vcp%i_sz),0.0_gwp)
1978 : : !rhotwg_c(1) = ctmp_dp
1979 : : end if
1980 : : end if
1981 : :
1982 [ + + ]: 2428992 : do m_kq=bstart_kq, bstop_kq
1983 : 2182656 : im_kq = m_kq - bstart_kq + 1
1984 : :
1985 [ + + ]: 2182656 : if (dtset%gwpt_wmode == 1) then
1986 : : ! SC: +2 because omegas_mkq(1) = e_mkq and omegas_mkq(2:cnt) = e_nk for all n_k bands
1987 : 953856 : iw_nk = n_k - bstart_k + 2
1988 : : ! Take the average at e_nk and e_mkq
1989 [ + + ]: 62954496 : ctmp_gwpc = half * sum(rhotwg_c(:) * (vec_gwc_mkq(:,1,m_kq) + vec_gwc_mkq(:,iw_nk,m_kq)))
1990 : : else
1991 : : ! Use the value at e_nk. Note in_k index
1992 [ + + ]: 19660800 : ctmp_gwpc = sum(rhotwg_c(:) * vec_gwc_mkq(:,in_k,m_kq))
1993 : : end if
1994 : :
1995 [ + + ]: 2182656 : if (need_x_kqmp) then
1996 : : ! TODO recheck
1997 : 1300992 : xdot_tmp = - xdotu(npw_x*nspinor, vec_gx_mkq(:,m_kq), 1, rhotwg_x, 1)
1998 : 1300992 : ctmp_gwpc = ctmp_gwpc + xdot_tmp ! * theta_mu_minus_e0i ! theta_mu_minus_e0i is only needed for metals
1999 : : end if
2000 : :
2001 [ - + ]: 2182656 : if (dtset%gwcomp == 2) then
2002 [ # # ]: 0 : ctmp_gwpc = ctmp_gwpc - quarter * dot_product(conjg(rhotwg_c), vec_coh_mkq(:, m_kq))
2003 : : end if
2004 : :
2005 [ - + ]: 2385024 : if (dtset%userie < 0) then
2006 [ # # ]: 0 : if (pp_is_gamma) then
2007 : : gsig_atm(1, im_kq, in_k, ipc) = gsig_atm(1, im_kq, in_k, ipc)
2008 : : gsig_atm(2, im_kq, in_k, ipc) = gsig_atm(2, im_kq, in_k, ipc)
2009 : : else
2010 : 0 : gsig_atm(1, im_kq, in_k, ipc) = gsig_atm(1, im_kq, in_k, ipc) + real(ctmp_gwpc)
2011 : 0 : gsig_atm(2, im_kq, in_k, ipc) = gsig_atm(2, im_kq, in_k, ipc) + aimag(ctmp_gwpc)
2012 : : end if
2013 : : else
2014 : 2182656 : gsig_atm(1, im_kq, in_k, ipc) = gsig_atm(1, im_kq, in_k, ipc) + real(ctmp_gwpc)
2015 : 2182656 : gsig_atm(2, im_kq, in_k, ipc) = gsig_atm(2, im_kq, in_k, ipc) + aimag(ctmp_gwpc)
2016 : : end if
2017 : :
2018 : : ! DEBUG: sanity check that the +q and -q Sternheimer solutions agree when
2019 : : ! qq=pp=Gamma. Note: Disabled it in production as each sum(abs(stern%cgq - ...))
2020 : : ! is O(2*npw_kq*nbsum) and is executed inside the innermost m_kq/n_k/ib_sum/imyp loops,
2021 : : ! i.e. ~nbsum * my_npert * (bdgw_size)^2 times per (qq=Gamma, pp=Gamma) sub-task.
2022 : : ! For a typical run (nbsum~300, npw_kq~few*10^3, bdgw=8) this is ~TB of memory
2023 : : ! traffic and turns the Gamma-Gamma sub-task from ~minutes into ~hours.
2024 : : !
2025 : : !if (qq_is_gamma .and. pp_is_gamma) then
2026 : : ! !print '(A7, A7, A7, A7, A7, A7)', 'my_is', 'im_kq', 'in_k', 'ipp_bz', 'ib_sum', 'ipc'
2027 : : ! !print '(I7, I7, I7, I7, I7, I7)', my_is, im_kq, in_k, ipp_bz, ib_sum, ipc
2028 : : ! !print *, "gsig_atm(:, im_kq, in_k, ipc):", gsig_atm(:, im_kq, in_k, ipc)
2029 : : ! !print *, "gks_atm(:, im_kq, in_k, ipc):", gks_atm(:, im_kq, in_k, ipc)
2030 : : ! !print *, "gks_atm2(:, im_kq, in_k, ipc):", gks_atm2(:, im_kq, in_k, ipc)
2031 : : ! !print *, ' '
2032 : :
2033 : : ! rtmp = sum(abs(stern_kqmp%cgq - stern_kmp%cgq)) / size(stern_kqmp%cgq)
2034 : : ! if (rtmp > tol14) then
2035 : : ! print *, "Sternheimer cgq diff", rtmp; stop
2036 : : ! end if
2037 : :
2038 : : ! ii = npw_kqmp * nspinor
2039 : : ! rtmp = sum(abs(cg_kmp(:,1:ii) - cg_kqmp(:,1:ii))) / (two * ii)
2040 : : ! if (rtmp > tol14) then
2041 : : ! print *, "Sternheimer cg_kmp diff", rtmp; stop
2042 : : ! end if
2043 : :
2044 : : ! ii = npw_kqmp * nspinor
2045 : : ! rtmp = sum(abs(full_cg1_kmp(:,1:ii) - full_cg1_kqmp(:,1:ii))) / (two * ii)
2046 : : ! if (rtmp > tol14) then
2047 : : ! print *, "full_cg1_kmp", rtmp; stop
2048 : : ! end if
2049 : :
2050 : : ! rtmp = sum(abs(full_ur1_star_kmp - full_ur1_kqmp)) / (nfft * nspinor)
2051 : : ! if (rtmp > tol14) then
2052 : : ! print *, "full_ur1_kmp diff", rtmp; stop
2053 : : ! end if
2054 : : !end if ! qq == pp = gamma
2055 : :
2056 : : end do ! m_kq
2057 : : end do ! n_k
2058 : 51296 : call timab(1945, 2, tsec)
2059 : : !end if ! .not qq_is_gamma.
2060 : : end do ! imyp (my perturbations)
2061 : :
2062 : 8024 : call rf_ham_kqmp%free(); call rf_ham_kmp%free()
2063 : : end do ! ib_sum (sum over bands)
2064 : :
2065 : : ! Show progress of all pp points.
2066 [ + + ]: 696 : if (print_time_pp) then
2067 : 267 : call inds2str(1, " My pp-point:", my_ipp, my_npp(spin), pp_mesh%nbz, msg)
2068 : 267 : call cwtime_report(msg, cpu_pp, wall_pp, gflops_pp);
2069 [ + + ]: 267 : if (my_ipp == LOG_MODP) call wrtout(std_out, "...", do_flush=.True.)
2070 : : end if
2071 : :
2072 [ - + ]: 696 : ABI_FREE(kpg_kmp)
2073 [ - + ]: 696 : ABI_FREE(kpg_kqmp)
2074 [ - + ]: 696 : ABI_FREE(ffnl_kmp)
2075 [ - + ]: 696 : ABI_FREE(ffnl_kqmp)
2076 [ - + ]: 696 : ABI_FREE(full_cg1_kqmp)
2077 [ - + ]: 696 : ABI_FREE(full_cg1_kmp)
2078 [ - + ]: 696 : ABI_FREE(rhotwg_c)
2079 [ - + ]: 696 : ABI_FREE(rhotwg_x)
2080 [ - + ]: 696 : ABI_FREE(vc_sqrt_gx)
2081 [ - + ]: 696 : ABI_SFREE(chi_ug_kqmp)
2082 [ - + ]: 696 : ABI_SFREE(rhotwg_chi)
2083 [ - + ]: 696 : ABI_FREE(kinpw_kmp)
2084 [ - + ]: 696 : ABI_FREE(kinpw_kqmp)
2085 [ - + ]: 696 : ABI_FREE(ph3d_kmp)
2086 [ - + ]: 696 : ABI_FREE(ph3d_kqmp)
2087 : :
2088 [ + - ]: 696 : ABI_SFREE(ylm_kmp)
2089 [ + - ]: 696 : ABI_SFREE(ylm_kqmp)
2090 [ + - ]: 696 : ABI_SFREE(botsq_pbz)
2091 [ + - ]: 696 : ABI_SFREE(otq_pbz)
2092 [ + - ]: 696 : ABI_SFREE(dmeig_pbz)
2093 [ + - ]: 696 : ABI_SFREE(trans_botsq_pbz)
2094 [ + - ]: 696 : ABI_SFREE(trans_otq_pbz)
2095 [ + - ]: 696 : ABI_SFREE(trans_dmeig_pbz)
2096 : :
2097 : 727 : call stern_kmp%free(); call stern_kqmp%free()
2098 : : end do ! ipp_bz
2099 : :
2100 [ - + ]: 31 : ABI_FREE(kpg_k)
2101 [ - + ]: 31 : ABI_FREE(kpg_kq)
2102 [ - + ]: 31 : ABI_FREE(ug_k)
2103 [ - + ]: 31 : ABI_FREE(ug_kq)
2104 : :
2105 : : ! Here we are outside of the loop over pp_sum, band_sum and perturbations.
2106 : : ! Collect gsig_atm and gks_atm inside pert_ppsum_bsum_comm so that all procs can write the data.
2107 : : !call xmpi_sum_master(gsig_atm, master, gqk%pert_ppsum_bsum_comm%value, ierr)
2108 : : !call xmpi_sum_master(gks_atm , master, gqk%pert_ppsum_bsum_comm%value, ierr)
2109 : : !call xmpi_sum_master(gks_atm2, master, gqk%pert_ppsum_bsum_comm%value, ierr)
2110 : : ! SC: not sure why, but xmpi_sum_master sometimes causes gvals to be filled
2111 : : ! with zeros and none in the GSTORE.nc file when running in parallel (observed
2112 : : ! on my desktop, lemaitre4, and lucia).
2113 : 31 : call xmpi_sum(gsig_atm, gqk%pert_ppsum_bsum_comm%value, ierr)
2114 : 31 : call xmpi_sum(gks_atm, gqk%pert_ppsum_bsum_comm%value, ierr)
2115 : 31 : call xmpi_sum(gks_atm2, gqk%pert_ppsum_bsum_comm%value, ierr)
2116 : :
2117 [ + + + + : 36805 : gsig_atm = gsig_atm / (cryst%ucvol * pp_mesh%nbz)
+ + + + ]
2118 : :
2119 [ + - ]: 31 : if (dtset%useria == 0) then
2120 [ - + + + : 36836 : gsig_atm = gsig_atm + gks_atm - gxc_atm
+ + + + +
+ ]
2121 [ # # ]: 0 : else if (dtset%useria > 0) then
2122 [ # # # # : 0 : gsig_atm = gxc_atm
# # # # #
# # # # #
# # # # #
# # # # #
# # # # ]
2123 : : else if (dtset%useria < 0) then
2124 : : gsig_atm = gsig_atm
2125 : : end if
2126 : :
2127 : : ! Save e-ph matrix elements in the buffer.
2128 [ + + + + : 36805 : my_gbuf(:,:,:,:, my_ik, iqbuf_cnt) = gsig_atm
+ + + + ]
2129 [ + + + + : 36805 : my_gbuf_ks(:,:,:,:, my_ik, iqbuf_cnt) = gks_atm
+ + + + ]
2130 : 31 : state_kq(my_ik, iqbuf_cnt) = GSTORE_KQ_COMPUTED
2131 : :
2132 [ - + ]: 173 : if (print_time_kk) then
2133 : 31 : call inds2str(3, "My k-point", my_ik, gqk%my_nk, gqk%glob_nk, msg)
2134 [ + - ]: 31 : call cwtime_report(msg, cpu_kk, wall_kk, gflops_kk); if (my_ik == LOG_MODK) call wrtout(std_out, "...", do_flush=.True.)
2135 : : end if
2136 : : end do ! my_ik
2137 : :
2138 : : ! Dump buffer inside the loop over my_iq (and outside of my_ik) otherwise when restarting GWPT some my_ik points will be missing
2139 [ + - ]: 80 : if (iqbuf_cnt == qbuf_size) call dump_my_gbuf()
2140 : :
2141 [ - + ]: 80 : ABI_FREE(v1scf_qq)
2142 [ - + ]: 80 : ABI_FREE(vlocal1_qq)
2143 [ - + ]: 80 : ABI_FREE(vlocal1_mqq)
2144 [ - + ]: 80 : ABI_FREE(vxc1_qq)
2145 : :
2146 [ + - + - ]: 80 : if (print_time_qq .and. wall_qq > tol3) then
2147 : 80 : call inds2str(2, "My q-point", my_iq, gqk%my_nq, gqk%glob_nq, msg)
2148 [ + + ]: 80 : call cwtime_report(msg, cpu_qq, wall_qq, gflops_qq); if (my_iq == LOG_MODQ) call wrtout(std_out, "...", do_flush=.True.)
2149 : : end if
2150 : 83 : call lg_myq%free()
2151 : : end do ! my_iq
2152 : :
2153 : : ! Dump the remainder.
2154 [ - + ]: 3 : if (iqbuf_cnt /= 0) call dump_my_gbuf()
2155 : :
2156 [ - + ]: 3 : ABI_FREE(ur_nk)
2157 [ - + ]: 3 : ABI_FREE(ur_mkq)
2158 [ - + ]: 3 : ABI_FREE(iq_buf)
2159 [ - + ]: 3 : ABI_FREE(state_kq)
2160 [ - + ]: 3 : ABI_FREE(my_gbuf)
2161 [ - + ]: 3 : ABI_FREE(my_gbuf_ks)
2162 [ - + ]: 3 : ABI_FREE(gsig_atm)
2163 [ - + ]: 3 : ABI_FREE(gks_atm)
2164 [ - + ]: 3 : ABI_FREE(gks_atm2)
2165 [ - + ]: 3 : ABI_FREE(gxc_atm)
2166 [ - + ]: 3 : ABI_FREE(omegas_nk)
2167 [ - + ]: 3 : ABI_FREE(omegas_mkq)
2168 [ - + ]: 3 : ABI_FREE(omegame0i_nk)
2169 [ - + ]: 3 : ABI_FREE(omegame0i_mkq)
2170 [ - + ]: 3 : ABI_FREE(vec_gwc_nk)
2171 [ - + ]: 3 : ABI_SFREE(vec_dgwc_nk)
2172 [ - + ]: 3 : ABI_FREE(vec_gwc_mkq)
2173 [ - + ]: 3 : ABI_SFREE(vec_comp_nk)
2174 [ - + ]: 3 : ABI_SFREE(vec_comp_mkq)
2175 [ - + ]: 3 : ABI_FREE(vec_gx_nk)
2176 [ - + ]: 3 : ABI_FREE(vec_gx_mkq)
2177 [ - + ]: 3 : ABI_FREE(sigcme_nk)
2178 [ - + ]: 3 : ABI_FREE(sigcme_mkq)
2179 [ - + ]: 3 : ABI_SFREE(vec_coh_nk)
2180 [ - + ]: 3 : ABI_SFREE(vec_coh_mkq)
2181 : :
2182 : : ! Collect self-energy matrix elements.
2183 : : if (test_sigma) then
2184 : : call xmpi_sum(vxc_nk, gqk%kpt_comm%value, ierr)
2185 : : call xmpi_sum(sigx_nk, gqk%kpt_comm%value, ierr)
2186 : : call xmpi_sum(sigce0_nk, gqk%kpt_comm%value, ierr)
2187 : : sigx_nk = -sigx_nk * (one / (cryst%ucvol * pp_mesh%nbz))
2188 : : sigce0_nk = sigce0_nk * (one / (cryst%ucvol * pp_mesh%nbz))
2189 : :
2190 : : if (gqk%comm%me == master) then
2191 : : write(ab_out, "(2a)") ch10, " Sigma^x_nk and Sigma^c_nk(E0) in eV:"
2192 : : do ik_glob=1, gqk%glob_nk
2193 : : ik_bz = gstore%kglob2bz(ik_glob, spin)
2194 : : ik_ibz = gstore%kbz2ibz(1, ik_bz)
2195 : : write(ab_out, "(2a)") "Band E0 <VxcDFT> SigX SigC(E0) for k-point:", trim(ktoa(gstore%kbz(:, ik_bz)))
2196 : : do band=gqk%bstart_k, gqk%bstop_k
2197 : : in_k = band - gqk%bstart_k + 1
2198 : : write(ab_out, "(i5, 4(f8.3))") &
2199 : : band, ebands%eig(band, ik_ibz, spin) * Ha_eV, &
2200 : : vxc_nk(in_k, ik_glob) * Ha_eV, &
2201 : : real(sigx_nk(in_k, ik_glob)) * Ha_eV, &
2202 : : real(sigce0_nk(in_k, ik_glob)) * Ha_eV
2203 : : end do
2204 : : end do
2205 : : end if
2206 : :
2207 : : ABI_SFREE(vxc_nk)
2208 : : ABI_SFREE(sigx_nk)
2209 : : ABI_SFREE(sigce0_nk)
2210 : : end if ! test_sigma
2211 : :
2212 [ + - ]: 6 : if (use_lgk) then
2213 [ + + ]: 8 : do my_ik=1,gqk%my_nk
2214 : 8 : call lg_myk(my_ik)%free()
2215 : : end do
2216 [ + - + + : 8 : ABI_FREE(lg_myk)
- + - + -
+ - + - +
- + - + -
+ ]
2217 : : end if
2218 : : end do ! my_is
2219 : :
2220 : 3 : call xmpi_sum(stern_qq_ierr, comm, ierr)
2221 : 3 : call xmpi_sum(stern_mq_ierr, comm, ierr)
2222 : 3 : call wrtout(units, sjoin(" Total number of failures in Sternheimer solvers at +q:", itoa(stern_qq_ierr)))
2223 : 3 : call wrtout(units, sjoin(" Total number of failures in Sternheimer solvers at -q:", itoa(stern_mq_ierr)))
2224 : :
2225 : 3 : call cwtime_report(" gwpt_eph full calculation", cpu_all, wall_all, gflops_all, end_str=ch10)
2226 : :
2227 : : ! Set gstore_completed to 1 so that we can easily check if restarted is needed.
2228 : : !if (my_rank == master) then
2229 [ - + ]: 3 : NCF_CHECK(nf90_put_var(root_ncid, root_vid("gstore_completed"), 1))
2230 : : ! 1 if the DFPT potentials and densities have been Fourier-interpolated (same convention as gstore_compute).
2231 [ - + ]: 3 : NCF_CHECK(nf90_put_var(root_ncid, root_vid("used_ftinterp"), merge(1, 0, need_ftinterp)))
2232 : : ! gwcomp = 2 means that the COH static remainder has been added to the GWPT matrix elements.
2233 [ - + ]: 3 : NCF_CHECK(nf90_put_var(root_ncid, root_vid("gwcomp"), dtset%gwcomp))
2234 : : !end if
2235 : : ! SC: not sure why, but nf90_sync sometimes can cause a deadlock, observed on lemaitre4
2236 : : ! Follow-up: This sync was not the cause. The deadlock took place because the procs were
2237 : : ! already out of step because of an earlier collective operation.
2238 [ - + ]: 3 : NCF_CHECK(nf90_sync(root_ncid))
2239 [ - + ]: 3 : NCF_CHECK(nf90_close(root_ncid))
2240 : 3 : call xmpi_barrier(comm)
2241 : :
2242 : : ! Reconstruct g(k,q) and g_KS(k,q) matrix elements in the full BZ by symmetry.
2243 [ - + ]: 3 : if (symmetrize) then
2244 : 0 : call gstore_symmetrize(gstore%path, wfk0_path, ngfft, dtset, dtfil, cryst, psps, pawtab, ebands, ifc, comm)
2245 : : end if
2246 : :
2247 : : ! Output some of the results to ab_out for testing purposes
2248 : 3 : call gstore%print_for_abitests(dtset, ebands, .True., with_ks=.True.)
2249 : :
2250 : : ! Free memory
2251 [ - + ]: 3 : ABI_FREE(kg_k)
2252 [ - + ]: 3 : ABI_FREE(kg_kq)
2253 [ - + ]: 3 : ABI_FREE(kg_kmp)
2254 [ - + ]: 3 : ABI_FREE(kg_kqmp)
2255 [ - + ]: 3 : ABI_FREE(cg_work)
2256 [ - + ]: 3 : ABI_FREE(ur_star_kmp)
2257 [ - + ]: 3 : ABI_FREE(ur_star_kqmp)
2258 [ - + ]: 3 : ABI_FREE(full_ur1_kqmp)
2259 [ - + ]: 3 : ABI_FREE(full_ur1_star_kmp)
2260 [ - + ]: 3 : ABI_SFREE(chi_ur_kqmp)
2261 [ - + ]: 3 : ABI_FREE(cwork_ur)
2262 [ - + ]: 3 : ABI_FREE(cg_kmp)
2263 [ - + ]: 3 : ABI_FREE(cg_kqmp)
2264 [ - + ]: 3 : ABI_FREE(cg1_kqmp)
2265 [ - + ]: 3 : ABI_FREE(cg1_kmp)
2266 [ - + ]: 3 : ABI_FREE(vtrial)
2267 [ - + ]: 3 : ABI_FREE(work)
2268 [ - + ]: 3 : ABI_FREE(ph1d)
2269 [ - + ]: 3 : ABI_FREE(vlocal)
2270 [ - + ]: 3 : ABI_FREE(gbound_k)
2271 [ - + ]: 3 : ABI_FREE(gbound_kq)
2272 [ - + ]: 3 : ABI_FREE(gbound_kmp)
2273 [ - + ]: 3 : ABI_FREE(gbound_kqmp)
2274 [ - + ]: 3 : ABI_FREE(gbound_c)
2275 [ - + ]: 3 : ABI_FREE(gbound_x)
2276 [ - + ]: 3 : ABI_FREE(done_qbz_spin)
2277 [ - + ]: 3 : ABI_FREE(rhor)
2278 [ - + ]: 3 : ABI_FREE(kxc)
2279 [ - + ]: 3 : ABI_FREE(vxc)
2280 : :
2281 [ - + ]: 3 : if (dtset%gwcomp == 2) then
2282 [ # # ]: 0 : ABI_FREE(wc0_pbz)
2283 [ # # ]: 0 : do ip_ibz=1,pp_mesh%nibz
2284 : 0 : call wc0_pibz(ip_ibz)%free()
2285 : : end do
2286 [ # # # # : 0 : ABI_FREE(wc0_pibz)
# # # # ]
2287 : : end if
2288 : :
2289 : 3 : call gs_ham_kqmp%free(); call gs_ham_kmp%free(); call wfd%free(); call vcp%free(); call ppm%free()
2290 : 3 : call pp_mesh%free(); call gsph_c%free(); call gsph_x%free(); call gstore%free()
2291 : 3 : call pawcprj_free(cwaveprj0); call pawcprj_free(cwaveprj)
2292 [ + - - + : 3 : ABI_FREE(cwaveprj0)
- - - - -
+ ]
2293 [ + - - + : 3 : ABI_FREE(cwaveprj)
- - - - -
+ ]
2294 : :
2295 : 3 : call xmpi_barrier(comm) ! This to make sure that the parallel output of GSTORE is completed
2296 [ - + - + : 27 : call cwtime_report(" gwpt_run: MPI barrier before returning.", cpu_all, wall_all, gflops_all, end_str=ch10, comm=comm)
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- - - - -
- - - - -
- - - - -
+ - - - -
- - - - -
- - - - -
- - - - -
- - - - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - - -
- - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + + -
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - - - -
- - - + -
- - - - -
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- - - - -
- - + - -
- - - - -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - -
- - - + -
- - - - +
- - - - -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - -
- - - - -
- - - - -
- - - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
- - - - -
- - - - -
- - - - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- - - - -
- - - - -
- - - - -
- - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- - - - -
- - - - -
- - - - -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - - - -
- - - - -
- - - - -
- - - - -
- - - - -
- - - - -
- - - - -
- - - - -
- - - - -
- - - - -
- - - - -
- - - - -
- - - - -
- - - - -
- - - - -
- - - - -
- - - - -
- - - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- - - - -
- - - - -
- - - - -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - - - -
- - - - -
- - - - -
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - +
- + - + -
+ - + - -
- - - - -
+ - - - -
- - - + -
+ - + - +
- + ]
2297 : :
2298 : : contains
2299 : :
2300 : 489 : subroutine inds2str(level, prefix, my_ik, my_nk, nk_tot, out_str)
2301 : : integer,intent(in) :: level, my_ik, my_nk, nk_tot
2302 : : character(len=*),intent(in) :: prefix
2303 : : character(len=*),intent(out) :: out_str
2304 : :
2305 [ + - + - ]: 489 : out_str = sjoin(prefix, itoa(my_ik), "/", itoa(my_nk), "[", itoa(nk_tot), "]")
2306 [ - + + + : 2569 : out_str = repeat(' ', 4 * level) // trim(out_str)
+ - ]
2307 : 489 : end subroutine inds2str
2308 : :
2309 : 80 : subroutine dump_my_gbuf()
2310 : :
2311 : : ! This function is called inside the double loop over (my_is, my_iq) or when we exit
2312 : : ! from the my_iq loop to dump the remainder that is still in the q-buffer,
2313 : : ! All the MPI procs in the (kpt_comm x pert_comm) grid shall call this contained routine
2314 : : ! as we have side-effects i.e. iqbuf_cnt set to 0.
2315 : :
2316 : : ! On disk we have the global arrays:
2317 : : !
2318 : : ! nctkarr_t("gvals", "dp", "two, nb_kq, nb_k, natom3, glob_nk, glob_nq")
2319 : : !
2320 : : ! while the local MPI buffers are dimensioned as follows:
2321 : : !
2322 : : ! my_gbuf(2, nb_kq, nb_k, natom3, gqk%my_nk, qbuf_size)
2323 : :
2324 : : ! If parallelism over perturbation is activated, only the procs treating the first perturbation
2325 : : ! i.e. the procs treating different k-points for this q are involved in IO
2326 : : ! as all the local buffers store results for all natom3 perturbations.
2327 : :
2328 : : ! NOTE: A similar routine is used in m_gstore. The two implementations should be kept in synch.
2329 : :
2330 : : integer :: ii, iq_bz, iq_glob, my_iq
2331 : :
2332 : : !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
2333 : : ! FIXME: Recheck this part as we have way more levels of parallelism in GWPT
2334 : : !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
2335 [ + - ]: 80 : if (gqk%coords_qkpb_sumbp(3) /= 0) goto 10 ! Yes, I'm very proud of this GOTO.
2336 [ - + ]: 80 : if (gqk%pert_ppsum_bsum_comm%me /= 0) goto 10 ! Yes, I'm very proud of this GOTO.
2337 : : ! SC: I commented out the above two GOTOs to avoid the deadlock issue on my desktop, lemaitre4 and lucia.
2338 : : ! Follow-up: The GOTOs were not the cause. The first access to the variables that were not prepared
2339 : : ! triggered a collective metadata read done by the writers only. All the variables are now prepared after
2340 : : ! the open (gstore_prepare_mpiio), so the GOTOs are restored. Only one proc of pert_ppsum_bsum_comm writes
2341 : : ! the data now.
2342 : :
2343 : : !iq_buf(:, iqbuf_cnt) = [my_iq, iq_bz]
2344 : 80 : my_iq = iq_buf(1, 1)
2345 : 80 : iq_glob = my_iq + gqk%my_qstart - 1
2346 : :
2347 : : !print *, "in dump_my_gbuf with start: ", [1, 1, 1, 1, gqk%my_kstart, iq_glob]
2348 : : !print *, " count; ", [2, gqk%nb_kq, gqk%nb_k, gqk%natom3, gqk%my_nk, iqbuf_cnt]
2349 : : !print *, "my_gbuf", my_gbuf(:,:,:,natom3,1,1)
2350 : :
2351 : : ! Output g^Sigma
2352 : : ! NB: this is an individual IO operation
2353 : : ncerr = nf90_put_var(spin_ncid, spin_vid("gvals"), my_gbuf, &
2354 : : start=[1, 1, 1, 1, gqk%my_kstart, iq_glob], &
2355 [ + + + + ]: 1040 : count=[2, gqk%nb_kq, gqk%nb_k, gqk%natom3, gqk%my_nk, iqbuf_cnt])
2356 [ - + ]: 80 : NCF_CHECK(ncerr)
2357 : :
2358 : : ! Output g^KS
2359 : : ncerr = nf90_put_var(spin_ncid, spin_vid("gvals_ks"), my_gbuf_ks, &
2360 : : start=[1, 1, 1, 1, gqk%my_kstart, iq_glob], &
2361 [ + + + + ]: 1040 : count=[2, gqk%nb_kq, gqk%nb_k, gqk%natom3, gqk%my_nk, iqbuf_cnt])
2362 [ - + ]: 80 : NCF_CHECK(ncerr)
2363 : :
2364 : : ncerr = nf90_put_var(root_ncid, root_vid("gstore_glob_state_kqs"), state_kq(:, 1:iqbuf_cnt), &
2365 : : start=[gqk%my_kstart, iq_glob, spin], &
2366 [ + + + + ]: 560 : count=[gqk%my_nk, iqbuf_cnt, 1])
2367 [ - + ]: 80 : NCF_CHECK(ncerr)
2368 : :
2369 : : ! Only one proc sets the entry in done_qbz_spin to 1 for all the q-points in the buffer.
2370 : : !if (all(gqk%coords_qkpb_sumbp(2:3) == [0, 0])) then
2371 [ + + ]: 160 : do ii=1,iqbuf_cnt
2372 : 80 : iq_bz = iq_buf(2, ii)
2373 [ + + - + : 320 : NCF_CHECK(nf90_put_var(root_ncid, root_vid("gstore_done_qbz_spin"), 1, start=[iq_bz, spin]))
- - ]
2374 : : end do
2375 : : !end if
2376 : :
2377 : : ! Zero the counter before returning, otherwise the flush after the my_iq loop dumps the last q-points again
2378 : : ! with state_kq already reset to MISSING (same logic as in m_gstore).
2379 : 80 : 10 iqbuf_cnt = 0
2380 [ + + + + ]: 320 : iq_buf = 0
2381 [ + + + + ]: 256 : state_kq = GSTORE_KQ_MISSING
2382 : :
2383 : : ! SC: these syncs can deadlock on lemaitre4; keep them disabled.
2384 : : ! SC: nf90_sync is collective on a file opened with comm (HDF5 flush of the metadata cache), so all the procs
2385 : : ! must call it the same number of times. dump_my_gbuf is called a number of times that depends on the proc
2386 : : ! (my_nq differs when nqbz is not a multiple of the number of q-blocks, q-points done in a previous run are
2387 : : ! skipped on restart), hence these syncs deadlock on the machine. The file is synced once by all the procs
2388 : : ! after the loops. The syncs are not needed for restart: the q-points written before a kill are in the file
2389 : : ! (checked with kill -9 followed by eph_restart 1).
2390 : : !NCF_CHECK(nf90_sync(spin_ncid))
2391 : : !NCF_CHECK(nf90_sync(root_ncid))
2392 : :
2393 : 80 : end subroutine dump_my_gbuf
2394 : :
2395 : 169 : integer function root_vid(var_name)
2396 : : character(len=*),intent(in) :: var_name
2397 [ + - ]: 169 : root_vid = nctk_idname(root_ncid, var_name)
2398 : : end function root_vid
2399 : :
2400 : 160 : integer function spin_vid(var_name)
2401 : : character(len=*),intent(in) :: var_name
2402 : 160 : spin_vid = nctk_idname(spin_ncid, var_name)
2403 : : end function spin_vid
2404 : :
2405 : : end subroutine gwpt_run
2406 : : !!***
2407 : :
2408 : : end module m_gwpt
2409 : : !!***
|
