10
0
mirror of https://github.com/QuantumPackage/qp2.git synced 2024-06-24 14:12:24 +02:00

Merge pull request #120 from kgasperich/dev-real-kpts

Merge dev real kpts
This commit is contained in:
Kevin Gasperich 2020-07-14 19:23:01 -05:00 committed by GitHub
commit 418c30c2ec
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
25 changed files with 1487 additions and 93 deletions

View File

@ -5,3 +5,19 @@
! END_DOC
! ao_num_per_kpt = ao_num/kpt_num
!END_PROVIDER
subroutine get_kpt_idx_ao(idx_full,k,i)
implicit none
BEGIN_DOC
! idx_full is ao index in full range (up to ao_num)
! k is index of the k-point for this ao
! i is index of this ao within k-point k
! this assumes that all kpts have the same number of aos
END_DOC
integer, intent(in) :: idx_full
integer, intent(out) :: i,k
i = mod(idx_full-1,ao_num_per_kpt)+1
k = (idx_full-1)/ao_num_per_kpt+1
ASSERT (k <= kpt_num)
end

View File

@ -122,6 +122,22 @@ BEGIN_PROVIDER [ complex*16, ao_overlap_kpts, (ao_num_per_kpt, ao_num_per_kpt, k
END_PROVIDER
BEGIN_PROVIDER [ double precision, ao_overlap_kpts_real, (ao_num_per_kpt, ao_num_per_kpt, kpt_num) ]
implicit none
BEGIN_DOC
! Overlap for complex AOs
END_DOC
integer :: i,j,k
do k=1,kpt_num
do j=1,ao_num_per_kpt
do i=1,ao_num_per_kpt
ao_overlap_kpts_real(i,j,k) = dble(ao_overlap_kpts(i,j,k))
enddo
enddo
enddo
END_PROVIDER
BEGIN_PROVIDER [ double precision, ao_overlap_abs,(ao_num,ao_num) ]

View File

@ -2918,14 +2918,10 @@ subroutine get_d1_kpts(gen, phasemask, bannedOrb, banned, mat, mask, h, p, sp, c
hfix = h(1,ma)
p1 = p(1,ma)
p2 = p(2,ma)
kputi = (puti-1)/mo_num_per_kpt + 1
khfix = (hfix-1)/mo_num_per_kpt + 1
kp1 = (p1-1)/mo_num_per_kpt + 1
kp2 = (p2-1)/mo_num_per_kpt + 1
iputi = mod(puti-1,mo_num_per_kpt) + 1
ihfix = mod(hfix-1,mo_num_per_kpt) + 1
ip1 = mod(p1-1, mo_num_per_kpt) + 1
ip2 = mod(p2-1, mo_num_per_kpt) + 1
call get_kpt_idx_mo(puti,kputi,iputi)
call get_kpt_idx_mo(hfix,khfix,ihfix)
call get_kpt_idx_mo(p1,kp1,ip1)
call get_kpt_idx_mo(p2,kp2,ip2)
if(.not. bannedOrb(puti, mi)) then
!==================
@ -3059,8 +3055,7 @@ subroutine get_d1_kpts(gen, phasemask, bannedOrb, banned, mat, mask, h, p, sp, c
!MOVE MI
pfix = p(1,mi)
kpfix = (pfix-1)/mo_num_per_kpt + 1
ipfix = mod(pfix-1,mo_num_per_kpt) + 1
call get_kpt_idx_mo(pfix,kpfix,ipfix)
tmp_row = (0.d0,0.d0)
tmp_row2 = (0.d0,0.d0)
!tmp_row_kpts = (0.d0,0.d0)
@ -3270,14 +3265,10 @@ subroutine get_d1_kpts(gen, phasemask, bannedOrb, banned, mat, mask, h, p, sp, c
puti = p(i, ma)
p1 = p(turn3(1,i), ma)
p2 = p(turn3(2,i), ma)
kputi = (puti-1)/mo_num_per_kpt + 1
khfix = (hfix-1)/mo_num_per_kpt + 1
kp1 = (p1-1)/mo_num_per_kpt + 1
kp2 = (p2-1)/mo_num_per_kpt + 1
iputi = mod(puti-1,mo_num_per_kpt) + 1
ihfix = mod(hfix-1,mo_num_per_kpt) + 1
ip1 = mod(p1-1, mo_num_per_kpt) + 1
ip2 = mod(p2-1, mo_num_per_kpt) + 1
call get_kpt_idx_mo(puti,kputi,iputi)
call get_kpt_idx_mo(hfix,khfix,ihfix)
call get_kpt_idx_mo(p1,kp1,ip1)
call get_kpt_idx_mo(p2,kp2,ip2)
call get_mo_two_e_integrals_complex(hfix,p1,p2,mo_num,hij_cache(1,1),mo_integrals_map,mo_integrals_map_2)
call get_mo_two_e_integrals_complex(hfix,p2,p1,mo_num,hij_cache(1,2),mo_integrals_map,mo_integrals_map_2)
call get_mo_two_e_integrals_kpts(hfix,ihfix,khfix,p1,ip1,kp1,p2,ip2,kp2,mo_num_per_kpt,hij_cache2(1,1),mo_integrals_map,mo_integrals_map_2)
@ -3425,14 +3416,10 @@ subroutine get_d1_kpts(gen, phasemask, bannedOrb, banned, mat, mask, h, p, sp, c
pfix = p(1,mi)
p1 = p(1,ma)
p2 = p(2,ma)
kpfix = (pfix-1)/mo_num_per_kpt + 1
khfix = (hfix-1)/mo_num_per_kpt + 1
kp1 = (p1-1)/mo_num_per_kpt + 1
kp2 = (p2-1)/mo_num_per_kpt + 1
ipfix = mod(pfix-1,mo_num_per_kpt) + 1
ihfix = mod(hfix-1,mo_num_per_kpt) + 1
ip1 = mod(p1-1, mo_num_per_kpt) + 1
ip2 = mod(p2-1, mo_num_per_kpt) + 1
call get_kpt_idx_mo(pfix,kpfix,ipfix)
call get_kpt_idx_mo(hfix,khfix,ihfix)
call get_kpt_idx_mo(p1,kp1,ip1)
call get_kpt_idx_mo(p2,kp2,ip2)
tmp_row = (0.d0,0.d0)
tmp_row2 = (0.d0,0.d0)
!tmp_row_kpts = (0.d0,0.d0)

View File

@ -56,10 +56,14 @@ subroutine run
double precision :: cisdq(N_states), delta_e
double precision,external :: diag_h_mat_elem
if(pseudo_sym)then
call H_apply_cisd_sym
if (is_complex) then
call H_apply_cisd_kpts
else
call H_apply_cisd
if(pseudo_sym)then
call H_apply_cisd_sym
else
call H_apply_cisd
endif
endif
if (is_complex) then
psi_coef_complex = ci_eigenvectors_complex

666
src/cisd/kpts_cisd.irp.f Normal file
View File

@ -0,0 +1,666 @@
subroutine H_apply_cisd_kpts_diexc(key_in, key_prev, hole_1,particl_1, hole_2, particl_2, fock_diag_tmp, i_generator, iproc_in )
implicit none
integer(bit_kind), intent(in) :: key_in(N_int, 2), hole_1(N_int, 2), hole_2(N_int, 2)
integer(bit_kind), intent(in) :: particl_1(N_int, 2), particl_2(N_int, 2)
integer(bit_kind) :: p1_mask(N_int, 2), p2_mask(N_int, 2), tmp
integer,intent(in) :: i_generator,iproc_in
integer :: status(N_int*bit_kind_size, 2)
integer :: highest, p1,p2,sp,ni,i,mi,nt,ns,k
double precision, intent(in) :: fock_diag_tmp(2,mo_num+1)
integer(bit_kind), intent(in) :: key_prev(N_int, 2, *)
PROVIDE N_int
PROVIDE N_det
highest = 0
do k=1,N_int*bit_kind_size
status(k,1) = 0
status(k,2) = 0
enddo
do sp=1,2
do ni=1,N_int
do i=1,bit_kind_size
if(iand(1_bit_kind,shiftr(key_in(ni, sp), (i-1))) == 0) then
cycle
end if
mi = (ni-1)*bit_kind_size+i
status(mi, sp) = int(iand(1_bit_kind,shiftr(hole_1(ni,sp),(i-1))),4)
status(mi, sp) = status(mi, sp) + 2*int(iand(1_bit_kind,shiftr(hole_2(ni,sp),(i-1))),4)
if(status(mi, sp) /= 0 .and. mi > highest) then
highest = mi
end if
end do
end do
end do
do sp=1,2
do p1=1,highest
if(status(p1, sp) == 0) then
cycle
end if
do p2=1,highest
if(status(p2, sp) == 0) then
cycle
end if
if((status(p1, sp) == 1 .and. status(p2, sp) > 1) .or. &
(status(p1, sp) == 2 .and. status(p2, sp) == 3) .or. &
(status(p1, sp) == 3 .and. status(p2, sp) == 3 .and. p2 > p1)) then
call H_apply_cisd_kpts_diexcP(key_in, sp, p1, particl_1, sp, p2, particl_2, fock_diag_tmp, i_generator, iproc_in )
end if
end do
end do
end do
do p1=1,highest
if(status(p1, 1) == 0) then
cycle
end if
do p2=1,highest
if(status(p2, 2) == 0) then
cycle
end if
if((status(p1, 1) == 3) .or. &
(status(p1, 1) == 1 .and. status(p2, 2) >= 2) .or. &
(status(p1, 1) == 2 .and. status(p2, 2) /= 2)) then
call H_apply_cisd_kpts_diexcP(key_in, 1, p1, particl_1, 2, p2, particl_2, fock_diag_tmp, i_generator, iproc_in )
end if
end do
end do
end subroutine
subroutine H_apply_cisd_kpts_diexcP(key_in, fs1, fh1, particl_1, fs2, fh2, particl_2, fock_diag_tmp, i_generator, iproc_in )
implicit none
integer(bit_kind), intent(in) :: key_in(N_int, 2), particl_1(N_int, 2), particl_2(N_int, 2)
double precision, intent(in) :: fock_diag_tmp(2,mo_num+1)
integer(bit_kind) :: p1_mask(N_int, 2), p2_mask(N_int, 2), key_mask(N_int, 2)
integer,intent(in) :: fs1,fs2,i_generator,iproc_in, fh1,fh2
integer(bit_kind) :: miniList(N_int, 2, N_det)
integer :: n_minilist, n_alpha, n_beta, deg(2), i, ni, k
integer(bit_kind), parameter :: one = 1_bit_kind
do k=1,N_int
p1_mask(k,1) = 0_bit_kind
p1_mask(k,2) = 0_bit_kind
p2_mask(k,1) = 0_bit_kind
p2_mask(k,2) = 0_bit_kind
enddo
p1_mask(shiftr(fh1-1,bit_kind_shift) + 1, fs1) = shiftl(one,iand(fh1-1,bit_kind_size-1))
p2_mask(shiftr(fh2-1,bit_kind_shift) + 1, fs2) = shiftl(one,iand(fh2-1,bit_kind_size-1))
do k=1,N_int
key_mask(k,1) = key_in(k,1)
key_mask(k,2) = key_in(k,2)
enddo
key_mask(shiftr(fh1-1,bit_kind_shift) + 1, fs1) -= shiftl(one,iand(fh1-1,bit_kind_size-1))
key_mask(shiftr(fh2-1,bit_kind_shift) + 1, fs2) -= shiftl(one,iand(fh2-1,bit_kind_size-1))
call H_apply_cisd_kpts_diexcOrg(key_in, key_mask, p1_mask, particl_1, p2_mask, particl_2, fock_diag_tmp, i_generator, iproc_in )
end subroutine
subroutine H_apply_cisd_kpts_diexcOrg(key_in,key_mask,hole_1,particl_1,hole_2, particl_2, fock_diag_tmp, i_generator, iproc_in )
use omp_lib
use bitmasks
implicit none
BEGIN_DOC
! Generate all double excitations of key_in using the bit masks of holes and
! particles.
! Assume N_int is already provided.
END_DOC
integer,parameter :: size_max = 8192
integer ,intent(in) :: i_generator
integer(bit_kind),intent(in) :: key_in(N_int,2), key_mask(N_int, 2)
integer(bit_kind),allocatable :: keys_out(:,:,:)
integer(bit_kind), intent(in) :: hole_1(N_int,2), particl_1(N_int,2)
integer(bit_kind), intent(in) :: hole_2(N_int,2), particl_2(N_int,2)
integer, intent(in) :: iproc_in
double precision, intent(in) :: fock_diag_tmp(2,mo_num+1)
integer(bit_kind), allocatable :: hole_save(:,:)
integer(bit_kind), allocatable :: key(:,:),hole(:,:), particle(:,:)
integer(bit_kind), allocatable :: hole_tmp(:,:), particle_tmp(:,:)
integer(bit_kind), allocatable :: key_union_hole_part(:)
integer :: ii,i,jj,j,k,ispin,l
integer, allocatable :: occ_particle(:,:), occ_hole(:,:)
integer, allocatable :: occ_particle_tmp(:,:), occ_hole_tmp(:,:)
integer :: kk,pp,other_spin,key_idx
integer :: N_elec_in_key_hole_1(2),N_elec_in_key_part_1(2)
integer :: N_elec_in_key_hole_2(2),N_elec_in_key_part_2(2)
double precision :: mo_two_e_integral
logical :: is_a_two_holes_two_particles
integer, allocatable :: ia_ja_pairs(:,:,:)
integer, allocatable :: ib_jb_pairs(:,:)
double precision :: diag_H_mat_elem
integer :: iproc
integer :: jtest_vvvv
logical :: check_double_excitation
logical :: is_a_1h1p
logical :: is_a_1h2p
logical :: is_a_1h
logical :: is_a_1p
logical :: is_a_2p
logical :: is_a_2h1p
logical :: is_a_2h
logical :: b_cycle
logical :: yes_no
check_double_excitation = .True.
iproc = iproc_in
!$ iproc = omp_get_thread_num()
allocate (keys_out(N_int,2,size_max), hole_save(N_int,2), &
key(N_int,2),hole(N_int,2), particle(N_int,2), hole_tmp(N_int,2),&
particle_tmp(N_int,2), occ_particle(N_int*bit_kind_size,2), &
occ_hole(N_int*bit_kind_size,2), occ_particle_tmp(N_int*bit_kind_size,2),&
occ_hole_tmp(N_int*bit_kind_size,2),key_union_hole_part(N_int))
!!!! First couple hole particle
do j = 1, N_int
hole(j,1) = iand(hole_1(j,1),key_in(j,1))
hole(j,2) = iand(hole_1(j,2),key_in(j,2))
particle(j,1) = iand(xor(particl_1(j,1),key_in(j,1)),particl_1(j,1))
particle(j,2) = iand(xor(particl_1(j,2),key_in(j,2)),particl_1(j,2))
enddo
call bitstring_to_list_ab(particle,occ_particle,N_elec_in_key_part_1,N_int)
call bitstring_to_list_ab(hole,occ_hole,N_elec_in_key_hole_1,N_int)
allocate (ia_ja_pairs(2,0:(elec_alpha_num)*mo_num,2), &
ib_jb_pairs(2,0:(elec_alpha_num)*mo_num))
do ispin=1,2
i=0
do ii=N_elec_in_key_hole_1(ispin),1,-1 ! hole
i_a = occ_hole(ii,ispin)
ASSERT (i_a > 0)
ASSERT (i_a <= mo_num)
do jj=1,N_elec_in_key_part_1(ispin) !particle
j_a = occ_particle(jj,ispin)
ASSERT (j_a > 0)
ASSERT (j_a <= mo_num)
i += 1
ia_ja_pairs(1,i,ispin) = i_a
ia_ja_pairs(2,i,ispin) = j_a
enddo
enddo
ia_ja_pairs(1,0,ispin) = i
enddo
key_idx = 0
integer :: i_a,j_a,i_b,j_b,k_a,l_a,k_b,l_b
integer(bit_kind) :: test(N_int,2)
double precision :: accu
logical, allocatable :: array_pairs(:,:)
allocate(array_pairs(mo_num,mo_num))
accu = 0.d0
do ispin=1,2
other_spin = iand(ispin,1)+1
do ii=1,ia_ja_pairs(1,0,ispin)
i_a = ia_ja_pairs(1,ii,ispin)
ASSERT (i_a > 0)
ASSERT (i_a <= mo_num)
j_a = ia_ja_pairs(2,ii,ispin)
ASSERT (j_a > 0)
ASSERT (j_a <= mo_num)
hole = key_in
k = shiftr(i_a-1,bit_kind_shift)+1
j = i_a-shiftl(k-1,bit_kind_shift)-1
hole(k,ispin) = ibclr(hole(k,ispin),j)
k_a = shiftr(j_a-1,bit_kind_shift)+1
l_a = j_a-shiftl(k_a-1,bit_kind_shift)-1
hole(k_a,ispin) = ibset(hole(k_a,ispin),l_a)
!!!! Second couple hole particle
do j = 1, N_int
hole_tmp(j,1) = iand(hole_2(j,1),hole(j,1))
hole_tmp(j,2) = iand(hole_2(j,2),hole(j,2))
particle_tmp(j,1) = iand(xor(particl_2(j,1),hole(j,1)),particl_2(j,1))
particle_tmp(j,2) = iand(xor(particl_2(j,2),hole(j,2)),particl_2(j,2))
enddo
call bitstring_to_list_ab(particle_tmp,occ_particle_tmp,N_elec_in_key_part_2,N_int)
call bitstring_to_list_ab(hole_tmp,occ_hole_tmp,N_elec_in_key_hole_2,N_int)
! hole = a^(+)_j_a(ispin) a_i_a(ispin)|key_in> : single exc :: orb(i_a,ispin) --> orb(j_a,ispin)
hole_save = hole
! Build array of the non-zero integrals of second excitation
array_pairs = .True.
if (ispin == 1) then
integer :: jjj
i=0
do kk = 1,N_elec_in_key_hole_2(other_spin)
i_b = occ_hole_tmp(kk,other_spin)
ASSERT (i_b > 0)
ASSERT (i_b <= mo_num)
do jjj=1,N_elec_in_key_part_2(other_spin) ! particle
j_b = occ_particle_tmp(jjj,other_spin)
ASSERT (j_b > 0)
ASSERT (j_b <= mo_num)
if (array_pairs(i_b,j_b)) then
i+= 1
ib_jb_pairs(1,i) = i_b
ib_jb_pairs(2,i) = j_b
endif
enddo
enddo
ib_jb_pairs(1,0) = i
do kk = 1,ib_jb_pairs(1,0)
hole = hole_save
i_b = ib_jb_pairs(1,kk)
j_b = ib_jb_pairs(2,kk)
k = shiftr(i_b-1,bit_kind_shift)+1
j = i_b-shiftl(k-1,bit_kind_shift)-1
hole(k,other_spin) = ibclr(hole(k,other_spin),j)
key = hole
k = shiftr(j_b-1,bit_kind_shift)+1
l = j_b-shiftl(k-1,bit_kind_shift)-1
key(k,other_spin) = ibset(key(k,other_spin),l)
key_idx += 1
do k=1,N_int
keys_out(k,1,key_idx) = key(k,1)
keys_out(k,2,key_idx) = key(k,2)
enddo
ASSERT (key_idx <= size_max)
if (key_idx == size_max) then
call fill_H_apply_buffer_no_selection(key_idx,keys_out,N_int,iproc)
key_idx = 0
endif
enddo
endif
! does all the single excitations of the same spin
i=0
do kk = 1,N_elec_in_key_hole_2(ispin)
i_b = occ_hole_tmp(kk,ispin)
if (i_b <= i_a.or.i_b == j_a) cycle
ASSERT (i_b > 0)
ASSERT (i_b <= mo_num)
do jjj=1,N_elec_in_key_part_2(ispin) ! particule
j_b = occ_particle_tmp(jjj,ispin)
ASSERT (j_b > 0)
ASSERT (j_b <= mo_num)
if (j_b <= j_a) cycle
if (array_pairs(i_b,j_b)) then
i+= 1
ib_jb_pairs(1,i) = i_b
ib_jb_pairs(2,i) = j_b
endif
enddo
enddo
ib_jb_pairs(1,0) = i
do kk = 1,ib_jb_pairs(1,0)
hole = hole_save
i_b = ib_jb_pairs(1,kk)
j_b = ib_jb_pairs(2,kk)
k = shiftr(i_b-1,bit_kind_shift)+1
j = i_b-shiftl(k-1,bit_kind_shift)-1
hole(k,ispin) = ibclr(hole(k,ispin),j)
key = hole
k = shiftr(j_b-1,bit_kind_shift)+1
l = j_b-shiftl(k-1,bit_kind_shift)-1
key(k,ispin) = ibset(key(k,ispin),l)
key_idx += 1
do k=1,N_int
keys_out(k,1,key_idx) = key(k,1)
keys_out(k,2,key_idx) = key(k,2)
enddo
ASSERT (key_idx <= size_max)
if (key_idx == size_max) then
call fill_H_apply_buffer_no_selection(key_idx,keys_out,N_int,iproc)
key_idx = 0
endif
enddo ! kk
enddo ! ii
enddo ! ispin
call fill_h_apply_buffer_no_selection(key_idx,keys_out,N_int,iproc)
deallocate (ia_ja_pairs, ib_jb_pairs, &
keys_out, hole_save, &
key,hole, particle, hole_tmp, &
particle_tmp, occ_particle, &
occ_hole, occ_particle_tmp, &
occ_hole_tmp,array_pairs,key_union_hole_part)
end
subroutine H_apply_cisd_kpts_monoexc(key_in, hole_1,particl_1,fock_diag_tmp,i_generator,iproc_in )
use omp_lib
use bitmasks
implicit none
BEGIN_DOC
! Generate all single excitations of key_in using the bit masks of holes and
! particles.
! Assume N_int is already provided.
END_DOC
integer,parameter :: size_max = 8192
integer ,intent(in) :: i_generator
integer(bit_kind),intent(in) :: key_in(N_int,2)
integer(bit_kind),intent(in) :: hole_1(N_int,2), particl_1(N_int,2)
integer, intent(in) :: iproc_in
double precision, intent(in) :: fock_diag_tmp(2,mo_num+1)
integer(bit_kind),allocatable :: keys_out(:,:,:)
integer(bit_kind),allocatable :: hole_save(:,:)
integer(bit_kind),allocatable :: key(:,:),hole(:,:), particle(:,:)
integer(bit_kind),allocatable :: hole_tmp(:,:), particle_tmp(:,:)
integer(bit_kind),allocatable :: hole_2(:,:), particl_2(:,:)
integer :: ii,i,jj,j,k,ispin,l
integer,allocatable :: occ_particle(:,:), occ_hole(:,:)
integer,allocatable :: occ_particle_tmp(:,:), occ_hole_tmp(:,:)
integer,allocatable :: ib_jb_pairs(:,:)
integer :: kk,pp,other_spin,key_idx
integer :: N_elec_in_key_hole_1(2),N_elec_in_key_part_1(2)
integer :: N_elec_in_key_hole_2(2),N_elec_in_key_part_2(2)
logical :: is_a_two_holes_two_particles
integer(bit_kind), allocatable :: key_union_hole_part(:)
integer, allocatable :: ia_ja_pairs(:,:,:)
logical, allocatable :: array_pairs(:,:)
double precision :: diag_H_mat_elem
integer :: iproc
integer(bit_kind) :: key_mask(N_int, 2)
logical :: check_double_excitation
logical :: is_a_2h1p
logical :: is_a_2h
logical :: is_a_1h1p
logical :: is_a_1h2p
logical :: is_a_1h
logical :: is_a_1p
logical :: is_a_2p
logical :: yes_no
do k=1,N_int
key_mask(k,1) = 0_bit_kind
key_mask(k,2) = 0_bit_kind
enddo
iproc = iproc_in
check_double_excitation = .True.
!$ iproc = omp_get_thread_num()
allocate (keys_out(N_int,2,size_max), hole_save(N_int,2), &
key(N_int,2),hole(N_int,2), particle(N_int,2), hole_tmp(N_int,2),&
particle_tmp(N_int,2), occ_particle(N_int*bit_kind_size,2), &
occ_hole(N_int*bit_kind_size,2), occ_particle_tmp(N_int*bit_kind_size,2),&
occ_hole_tmp(N_int*bit_kind_size,2),key_union_hole_part(N_int))
!!!! First couple hole particle
do j = 1, N_int
hole(j,1) = iand(hole_1(j,1),key_in(j,1))
hole(j,2) = iand(hole_1(j,2),key_in(j,2))
particle(j,1) = iand(xor(particl_1(j,1),key_in(j,1)),particl_1(j,1))
particle(j,2) = iand(xor(particl_1(j,2),key_in(j,2)),particl_1(j,2))
enddo
call bitstring_to_list_ab(particle,occ_particle,N_elec_in_key_part_1,N_int)
call bitstring_to_list_ab(hole,occ_hole,N_elec_in_key_hole_1,N_int)
allocate (ia_ja_pairs(2,0:(elec_alpha_num)*mo_num,2))
do ispin=1,2
i=0
do ii=N_elec_in_key_hole_1(ispin),1,-1 ! hole
i_a = occ_hole(ii,ispin)
do jj=1,N_elec_in_key_part_1(ispin) !particule
j_a = occ_particle(jj,ispin)
i += 1
ia_ja_pairs(1,i,ispin) = i_a
ia_ja_pairs(2,i,ispin) = j_a
enddo
enddo
ia_ja_pairs(1,0,ispin) = i
enddo
key_idx = 0
integer :: i_a,j_a,i_b,j_b,k_a,l_a,k_b,l_b
integer(bit_kind) :: test(N_int,2)
double precision :: accu
accu = 0.d0
do ispin=1,2
other_spin = iand(ispin,1)+1
do ii=1,ia_ja_pairs(1,0,ispin)
i_a = ia_ja_pairs(1,ii,ispin)
j_a = ia_ja_pairs(2,ii,ispin)
hole = key_in
k = shiftr(i_a-1,bit_kind_shift)+1
j = i_a-shiftl(k-1,bit_kind_shift)-1
hole(k,ispin) = ibclr(hole(k,ispin),j)
k_a = shiftr(j_a-1,bit_kind_shift)+1
l_a = j_a-shiftl(k_a-1,bit_kind_shift)-1
hole(k_a,ispin) = ibset(hole(k_a,ispin),l_a)
key_idx += 1
do k=1,N_int
keys_out(k,1,key_idx) = hole(k,1)
keys_out(k,2,key_idx) = hole(k,2)
enddo
if (key_idx == size_max) then
call fill_H_apply_buffer_no_selection(key_idx,keys_out,N_int,iproc)
key_idx = 0
endif
enddo ! ii
enddo ! ispin
call fill_H_apply_buffer_no_selection(key_idx,keys_out,N_int,iproc)
deallocate (ia_ja_pairs, &
keys_out, hole_save, &
key,hole, particle, hole_tmp,&
particle_tmp, occ_particle, &
occ_hole, occ_particle_tmp,&
occ_hole_tmp,key_union_hole_part)
end
subroutine H_apply_cisd_kpts()
implicit none
use omp_lib
use bitmasks
BEGIN_DOC
! Calls H_apply on the |HF| determinant and selects all connected single and double
! excitations (of the same symmetry). Auto-generated by the ``generate_h_apply`` script.
END_DOC
integer :: i_generator
double precision :: wall_0, wall_1
integer(bit_kind), allocatable :: mask(:,:,:)
integer :: ispin, k
integer :: iproc
double precision, allocatable :: fock_diag_tmp(:,:)
integer :: kk,kh1,kh2,kp1,kp2
integer(bit_kind), allocatable :: mask_kpts(:,:,:,:)
if (is_complex) then
PROVIDE H_apply_buffer_allocated mo_two_e_integrals_in_map psi_det_generators psi_coef_generators_complex
else
PROVIDE H_apply_buffer_allocated mo_two_e_integrals_in_map psi_det_generators psi_coef_generators
endif
call wall_time(wall_0)
iproc = 0
!allocate( mask(N_int,2,6), fock_diag_tmp(2,mo_num+1) )
allocate( mask_kpts(N_int,2,6,kpt_num), fock_diag_tmp(2,mo_num+1) )
do i_generator=1,N_det_generators
! Compute diagonal of the Fock matrix
call build_fock_tmp(fock_diag_tmp,psi_det_generators(1,1,i_generator),N_int)
! Create bit masks for holes and particles
do kk=1,kpt_num
do ispin=1,2
do k=1,N_int
mask_kpts(k,ispin,s_hole,kk) = &
iand(generators_bitmask_kpts(k,ispin,s_hole,kk), &
psi_det_generators(k,ispin,i_generator) )
mask_kpts(k,ispin,s_part,kk) = &
iand(generators_bitmask_kpts(k,ispin,s_part,kk), &
not(psi_det_generators(k,ispin,i_generator)) )
mask_kpts(k,ispin,d_hole1,kk) = &
iand(generators_bitmask_kpts(k,ispin,d_hole1,kk), &
psi_det_generators(k,ispin,i_generator) )
mask_kpts(k,ispin,d_part1,kk) = &
iand(generators_bitmask_kpts(k,ispin,d_part1,kk), &
not(psi_det_generators(k,ispin,i_generator)) )
mask_kpts(k,ispin,d_hole2,kk) = &
iand(generators_bitmask_kpts(k,ispin,d_hole2,kk), &
psi_det_generators(k,ispin,i_generator) )
mask_kpts(k,ispin,d_part2,kk) = &
iand(generators_bitmask_kpts(k,ispin,d_part2,kk), &
not(psi_det_generators(k,ispin,i_generator)) )
enddo
enddo
enddo
if(.True.)then
do kh1=1,kpt_num
do kh2=1,kpt_num
do kp1=1,kpt_num
kp2=kconserv(kh1,kh2,kp1)
!print*,'kh1h2p1p1',kh1,kh2,kp1,kp2
!print*,'size_before: ',h_apply_buffer(iproc)%n_det
call H_apply_cisd_kpts_diexc(psi_det_generators(1,1,i_generator), &
psi_det_generators(1,1,1), &
mask_kpts(1,1,d_hole1,kh1), mask_kpts(1,1,d_part1,kp1), &
mask_kpts(1,1,d_hole2,kh2), mask_kpts(1,1,d_part2,kp2), &
fock_diag_tmp, i_generator, iproc )
!print*,'size_after: ',h_apply_buffer(iproc)%n_det
enddo
enddo
enddo
endif
if(.True.)then
do kk=1,kpt_num
call H_apply_cisd_kpts_monoexc(psi_det_generators(1,1,i_generator), &
mask_kpts(1,1,s_hole,kk), mask_kpts(1,1,s_part,kk), &
fock_diag_tmp, i_generator, iproc )
enddo
endif
call wall_time(wall_1)
if (wall_1 - wall_0 > 2.d0) then
write(6,*) &
100.*float(i_generator)/float(N_det_generators), '% in ', wall_1-wall_0, 's'
wall_0 = wall_1
endif
enddo
!deallocate( mask, fock_diag_tmp )
deallocate( mask_kpts, fock_diag_tmp )
call copy_H_apply_buffer_to_wf
if (s2_eig) then
call make_s2_eigenfunction
endif
if (is_complex) then
SOFT_TOUCH psi_det psi_coef_complex N_det
else
SOFT_TOUCH psi_det psi_coef N_det
endif
! Sort H_jj to find the N_states lowest states
integer :: i
integer, allocatable :: iorder(:)
double precision, allocatable :: H_jj(:)
double precision, external :: diag_h_mat_elem
allocate(H_jj(N_det),iorder(N_det))
!$OMP PARALLEL DEFAULT(NONE) &
!$OMP SHARED(psi_det,N_int,H_jj,iorder,N_det) &
!$OMP PRIVATE(i)
!$OMP DO
do i = 1, N_det
H_jj(i) = diag_h_mat_elem(psi_det(1,1,i),N_int)
iorder(i) = i
enddo
!$OMP END DO
!$OMP END PARALLEL
call dsort(H_jj,iorder,N_det)
if (is_complex) then
do k=1,N_states
psi_coef_complex(iorder(k),k) = (1.d0,0.d0)
enddo
else
do k=1,N_states
psi_coef(iorder(k),k) = 1.d0
enddo
endif
deallocate(H_jj,iorder)
end

View File

@ -1070,9 +1070,10 @@ subroutine davidson_diag_hjj_sjj_complex(dets_in,u_in,H_jj,s2_out,energies,dim_i
r1 = dsqrt(-2.d0*dlog(r1))
r2 = dtwo_pi*r2
!todo: real or complex? rescale for complex? sqrt(2)?
!u_in(i,k) = dcmplx(r1*dcos(r2),0.d0)
u_in(i,k) = dcmplx(r1*dcos(r2),r1*dsin(r2))
u_in(i,k) = dcmplx(r1*dcos(r2),0.d0)
!u_in(i,k) = dcmplx(r1*dcos(r2),r1*dsin(r2))
enddo
u_in(k,k) = (10.d0,0.d0)
enddo
do k=1,N_st_diag
call normalize_complex(u_in(1,k),sze)

View File

@ -411,6 +411,15 @@ END_PROVIDER
END_PROVIDER
subroutine diagonalize_ci
implicit none
if (is_complex) then
call diagonalize_ci_complex
else
call diagonalize_ci_real
endif
end
subroutine diagonalize_CI_complex
implicit none
BEGIN_DOC
@ -429,7 +438,7 @@ subroutine diagonalize_CI_complex
SOFT_TOUCH psi_coef_complex CI_electronic_energy_complex ci_energy CI_eigenvectors_complex CI_s2_complex psi_energy psi_s2
end
subroutine diagonalize_CI
subroutine diagonalize_CI_real
implicit none
BEGIN_DOC
! Replace the coefficients of the |CI| states by the coefficients of the

View File

@ -495,10 +495,8 @@ END_PROVIDER
call decode_exc_spin(exc,h1,p1,h2,p2)
! h1 occ in k
! p1 occ in l
ih1 = mod(h1-1,mo_num_per_kpt)+1
ip1 = mod(p1-1,mo_num_per_kpt)+1
kh1 = (h1-1)/mo_num_per_kpt + 1
kp1 = (p1-1)/mo_num_per_kpt + 1
call get_kpt_idx_mo(h1,kh1,ih1)
call get_kpt_idx_mo(p1,kp1,ip1)
if (kh1.ne.kp1) then
print *,'problem in: ',irp_here,'a'
print *,' h1 = ',h1
@ -577,10 +575,8 @@ END_PROVIDER
exc = 0
call get_single_excitation_spin(tmp_det(1,2),tmp_det2,exc,phase,N_int)
call decode_exc_spin(exc,h1,p1,h2,p2)
ih1 = mod(h1-1,mo_num_per_kpt)+1
ip1 = mod(p1-1,mo_num_per_kpt)+1
kh1 = (h1-1)/mo_num_per_kpt + 1
kp1 = (p1-1)/mo_num_per_kpt + 1
call get_kpt_idx_mo(h1,kh1,ih1)
call get_kpt_idx_mo(p1,kp1,ip1)
if (kh1.ne.kp1) then
print *,'problem in: ',irp_here,'b'
print *,' h1 = ',h1

View File

@ -449,12 +449,14 @@ subroutine get_single_excitation_from_fock_kpts(det_1,det_2,ih,ip,spin,phase,hij
integer :: occ_partcl(N_int*bit_kind_size,2)
integer :: n_occ_ab_hole(2),n_occ_ab_partcl(2)
integer :: i0,i,h,p
integer :: ki,khp
integer :: ki,khp,kh
complex*16 :: buffer_c(mo_num_per_kpt),buffer_x(mo_num_per_kpt)
khp = (ip-1)/mo_num_per_kpt+1
p = mod(ip-1,mo_num_per_kpt)+1
h = mod(ih-1,mo_num_per_kpt)+1
call get_kpt_idx_mo(ip,khp,p)
call get_kpt_idx_mo(ih,kh,h)
ASSERT (kh==khp)
!todo: omp kpts
hij = fock_op_cshell_ref_bitmask_kpts(h,p,khp)
do ki=1,kpt_num
do i=1, mo_num_per_kpt
!<hi|pi>
@ -476,7 +478,6 @@ subroutine get_single_excitation_from_fock_kpts(det_1,det_2,ih,ip,spin,phase,hij
enddo
call bitstring_to_list_ab(hole, occ_hole, n_occ_ab_hole, N_int)
call bitstring_to_list_ab(partcl, occ_partcl, n_occ_ab_partcl, N_int)
hij = fock_op_cshell_ref_bitmask_kpts(h,p,khp)
! holes :: direct terms
do i0 = 1, n_occ_ab_hole(1)
i = occ_hole(i0,1) - (ki-1)*mo_num_per_kpt

View File

@ -2443,18 +2443,19 @@ subroutine i_H_j_complex(key_i,key_j,Nint,hij)
if (exc(0,1,1) == 1) then
call double_allowed_mo_kpts(exc(1,1,1),exc(1,1,2),exc(1,2,1),exc(1,2,2),is_allowed)
if (.not.is_allowed) then
! excitation doesn't conserve momentum
hij = (0.d0,0.d0)
return
endif
! Single alpha, single beta
if(exc(1,1,1) == exc(1,2,2) )then
ih1 = mod(exc(1,1,1)-1,mo_num_per_kpt)+1
ih2 = mod(exc(1,1,2)-1,mo_num_per_kpt)+1
kh1 = (exc(1,1,1)-1)/mo_num_per_kpt+1
kh2 = (exc(1,1,2)-1)/mo_num_per_kpt+1
ip1 = mod(exc(1,2,1)-1,mo_num_per_kpt)+1
kp1 = (exc(1,2,1)-1)/mo_num_per_kpt+1
!h1(a) = p2(b)
call get_kpt_idx_mo(exc(1,1,1),kh1,ih1)
call get_kpt_idx_mo(exc(1,1,2),kh2,ih2)
call get_kpt_idx_mo(exc(1,2,1),kp1,ip1)
if(kp1.ne.kh2) then
!if h1==p2 then kp1==kh2
print*,'problem with hij kpts: ',irp_here
print*,is_allowed
print*,exc(1,1,1),exc(1,1,2),exc(1,2,1),exc(1,2,2)
@ -2464,12 +2465,10 @@ subroutine i_H_j_complex(key_i,key_j,Nint,hij)
hij = phase * big_array_exchange_integrals_kpts(ih1,kh1,ih2,ip1,kp1)
!hij = phase * big_array_exchange_integrals_complex(exc(1,1,1),exc(1,1,2),exc(1,2,1))
else if (exc(1,2,1) ==exc(1,1,2))then
ih1 = mod(exc(1,1,1)-1,mo_num_per_kpt)+1
kh1 = (exc(1,1,1)-1)/mo_num_per_kpt+1
ip1 = mod(exc(1,2,1)-1,mo_num_per_kpt)+1
kp1 = (exc(1,2,1)-1)/mo_num_per_kpt+1
ip2 = mod(exc(1,2,2)-1,mo_num_per_kpt)+1
kp2 = (exc(1,2,2)-1)/mo_num_per_kpt+1
!p1(a)==h2(b)
call get_kpt_idx_mo(exc(1,1,1),kh1,ih1)
call get_kpt_idx_mo(exc(1,2,1),kp1,ip1)
call get_kpt_idx_mo(exc(1,2,2),kp2,ip2)
if(kp2.ne.kh1) then
print*,'problem with hij kpts: ',irp_here
print*,is_allowed

View File

@ -0,0 +1,92 @@
program scf_k_real
BEGIN_DOC
!
! The :ref:`scf` program performs *Restricted* Hartree-Fock
! calculations (the spatial part of the |MOs| is common for alpha and beta
! spinorbitals).
!
! It performs the following actions:
!
! #. Compute/Read all the one- and two-electron integrals, and store them
! in memory
! #. Check in the |EZFIO| database if there is a set of |MOs|.
! If there is, it will read them as initial guess. Otherwise, it will
! create a guess.
! #. Perform the |SCF| iterations
!
! For the keywords related to the |SCF| procedure, see the ``scf_utils``
! directory where you will find all options.
!
! At each iteration, the |MOs| are saved in the |EZFIO| database. Hence,
! if the calculation crashes for any unexpected reason, the calculation
! can be restarted by running again the |SCF| with the same |EZFIO|
! database.
!
! To start again a fresh |SCF| calculation, the |MOs| can be reset by
! running the :ref:`qp_reset` command.
!
! The `DIIS`_ algorithm is implemented, as well as the `level-shifting`_
! method. If the |SCF| does not converge, try again with a higher value of
! :option:`level_shift`.
!
! .. _DIIS: https://en.wikipedia.org/w/index.php?title=DIIS
! .. _level-shifting: https://doi.org/10.1002/qua.560070407
!
END_DOC
call create_guess_k_real
call orthonormalize_mos_k_real
call run_k_real
end
subroutine create_guess_k_real
implicit none
BEGIN_DOC
! Create a MO guess if no MOs are present in the EZFIO directory
END_DOC
logical :: exists
PROVIDE ezfio_filename
call ezfio_has_mo_basis_mo_coef_kpts(exists)
if (.not.exists) then
if (mo_guess_type == "HCore") then
!mo_coef_complex = ao_ortho_lowdin_coef_complex
mo_coef_kpts = ao_ortho_lowdin_coef_kpts_real
TOUCH mo_coef_kpts
mo_label = 'Guess'
!call mo_as_eigvectors_of_mo_matrix_complex(mo_one_e_integrals_kpts, &
call mo_as_eigvectors_of_mo_matrix_kpts_real(mo_one_e_integrals_kpts_real, &
size(mo_one_e_integrals_kpts_real,1), &
size(mo_one_e_integrals_kpts_real,2), &
size(mo_one_e_integrals_kpts_real,3), &
mo_label,1,.false.)
SOFT_TOUCH mo_coef_kpts mo_label
else if (mo_guess_type == "Huckel") then
call huckel_guess_kpts_real
else
print *, 'Unrecognized MO guess type : '//mo_guess_type
stop 1
endif
endif
end
subroutine run_k_real
BEGIN_DOC
! Run SCF calculation
END_DOC
use bitmasks
implicit none
integer :: i_it, i, j, k
mo_label = "Orthonormalized"
call roothaan_hall_scf_kpts_real
call ezfio_set_hartree_fock_energy(SCF_energy)
print*,'hf 1e,2e,total energy'
print*,hf_one_electron_energy
print*,hf_two_electron_energy
print*,hf_energy
end

View File

@ -80,6 +80,22 @@ BEGIN_PROVIDER [ integer, mo_num_per_kpt ]
END_PROVIDER
subroutine get_kpt_idx_mo(idx_full,k,i)
implicit none
BEGIN_DOC
! idx_full is mo index in full range (up to mo_num)
! k is index of the k-point for this mo
! i is index of this mo within k-point k
! this assumes that all kpts have the same number of mos
END_DOC
integer, intent(in) :: idx_full
integer, intent(out) :: i,k
i = mod(idx_full-1,mo_num_per_kpt)+1
k = (idx_full-1)/mo_num_per_kpt+1
ASSERT (k <= kpt_num)
end
BEGIN_PROVIDER [ double precision, mo_coef, (ao_num,mo_num) ]
implicit none

View File

@ -327,6 +327,85 @@ subroutine mo_as_eigvectors_of_mo_matrix_kpts(matrix,n,m,nk,label,sign,output)
endif
end
subroutine mo_as_eigvectors_of_mo_matrix_kpts_real(matrix,n,m,nk,label,sign,output)
!TODO: test this
implicit none
integer,intent(in) :: n,m,nk, sign
character*(64), intent(in) :: label
double precision, intent(in) :: matrix(n,m,nk)
logical, intent(in) :: output
integer :: i,j,k
double precision, allocatable :: eigvalues(:)
!complex*16, allocatable :: mo_coef_new(:,:)
double precision, allocatable :: mo_coef_new(:,:),mo_coef_tmp(:,:),R(:,:), A(:,:)
!DIR$ ATTRIBUTES ALIGN : $IRP_ALIGN :: mo_coef_new, R
call write_time(6)
if (m /= mo_num_per_kpt) then
print *, irp_here, ': Error : m/= mo_num_per_kpt'
stop 1
endif
if (nk /= kpt_num) then
print *, irp_here, ': Error : nk/= kpt_num'
stop 1
endif
allocate(A(n,m),R(n,m),mo_coef_tmp(ao_num_per_kpt,m),mo_coef_new(ao_num_per_kpt,m),eigvalues(m))
do k=1,nk
if (sign == -1) then
do j=1,m
do i=1,n
A(i,j) = -matrix(i,j,k)
enddo
enddo
else
do j=1,m
do i=1,n
A(i,j) = matrix(i,j,k)
enddo
enddo
endif
mo_coef_new = dble(mo_coef_kpts(:,:,k))
call lapack_diag(eigvalues,R,A,n,m)
if (sign == -1) then
do i=1,m
eigvalues(i) = -eigvalues(i)
enddo
endif
if (output) then
do i=1,m
write (6,'(2(I8),1X,F16.10)') k,i,eigvalues(i)
enddo
write (6,'(A)') '======== ================'
write (6,'(A)') ''
!write (6,'(A)') 'Fock Matrix'
!write (6,'(A)') '-----------'
!do i=1,n
! write(*,'(200(E24.15))') A(i,:)
!enddo
endif
call dgemm('N','N',ao_num_per_kpt,m,m,1.d0, &
mo_coef_new,size(mo_coef_new,1),R,size(R,1),0.d0, &
mo_coef_tmp,size(mo_coef_tmp,1))
call zlacp2('N',ao_num_per_kpt,m,mo_coef_tmp,size(mo_coef_tmp,1), &
mo_coef_kpts(:,:,k),size(mo_coef_kpts,1))
enddo
deallocate(A,mo_coef_new,mo_coef_tmp,R,eigvalues)
call write_time(6)
mo_label = label
if (output) then
write (6,'(A)') 'MOs are now **'//trim(label)//'**'
write (6,'(A)') ''
write (6,'(A)') 'Eigenvalues'
write (6,'(A)') '-----------'
write (6,'(A)') ''
write (6,'(A)') '======== ================'
endif
end
subroutine mo_as_svd_vectors_of_mo_matrix_kpts(matrix,lda,m,n,label)
!TODO: implement
print *, irp_here, ' not implemented for kpts'

View File

@ -107,3 +107,35 @@ BEGIN_PROVIDER [complex*16, ao_ortho_lowdin_overlap_kpts, (ao_num_per_kpt,ao_num
enddo
enddo
END_PROVIDER
!============================================!
! !
! kpts_real !
! !
!============================================!
BEGIN_PROVIDER [ double precision, ao_ortho_lowdin_coef_kpts_real, (ao_num_per_kpt,ao_num_per_kpt,kpt_num)]
implicit none
BEGIN_DOC
! matrix of the coefficients of the mos generated by the
! orthonormalization by the S^{-1/2} canonical transformation of the aos
! ao_ortho_lowdin_coef(i,j) = coefficient of the ith ao on the jth ao_ortho_lowdin orbital
END_DOC
integer :: i,j,k,l
double precision, allocatable :: tmp_matrix(:,:)
allocate (tmp_matrix(ao_num,ao_num))
do k=1,kpt_num
tmp_matrix(:,:) = 0.d0
do j=1, ao_num
tmp_matrix(j,j) = 1.d0
enddo
call ortho_lowdin(ao_overlap_kpts_real(:,:,k),ao_num_per_kpt,ao_num_per_kpt,tmp_matrix,ao_num_per_kpt,ao_num_per_kpt,lin_dep_cutoff)
do i=1, ao_num_per_kpt
do j=1, ao_num_per_kpt
ao_ortho_lowdin_coef_kpts_real(j,i,k) = tmp_matrix(i,j)
enddo
enddo
enddo
deallocate(tmp_matrix)
END_PROVIDER

View File

@ -59,3 +59,20 @@ BEGIN_PROVIDER [ complex*16, mo_one_e_integrals_kpts,(mo_num_per_kpt,mo_num_per_
print*,'Provided the one-electron integrals'
END_PROVIDER
BEGIN_PROVIDER [ double precision, mo_one_e_integrals_kpts_real,(mo_num_per_kpt,mo_num_per_kpt,kpt_num)]
implicit none
BEGIN_DOC
! array of the one-electron Hamiltonian on the |MO| basis :
! sum of the kinetic and nuclear electronic potentials (and pseudo potential if needed)
END_DOC
integer :: i,j,k
do k=1,kpt_num
do j=1,mo_num_per_kpt
do i=1,mo_num_per_kpt
mo_one_e_integrals_kpts_real(i,j,k) = dble(mo_one_e_integrals_kpts(i,j,k))
enddo
enddo
enddo
END_PROVIDER

View File

@ -128,3 +128,18 @@ BEGIN_PROVIDER [ complex*16, mo_overlap_kpts,(mo_num_per_kpt,mo_num_per_kpt,kpt_
endif
END_PROVIDER
BEGIN_PROVIDER [ double precision, mo_overlap_kpts_real, (mo_num_per_kpt, mo_num_per_kpt, kpt_num) ]
implicit none
BEGIN_DOC
! Overlap for complex MOs
END_DOC
integer :: i,j,k
do k=1,kpt_num
do j=1,mo_num_per_kpt
do i=1,mo_num_per_kpt
mo_overlap_kpts_real(i,j,k) = dble(mo_overlap_kpts(i,j,k))
enddo
enddo
enddo
END_PROVIDER

View File

@ -19,3 +19,23 @@ subroutine orthonormalize_mos
end
subroutine orthonormalize_mos_k_real
implicit none
integer :: m,p,s,k
double precision, allocatable :: mo_coef_tmp(:,:)
allocate(mo_coef_tmp(ao_num_per_kpt,mo_num_per_kpt))
do k=1,kpt_num
m = size(mo_coef_kpts,1)
p = size(mo_overlap_kpts,1)
mo_coef_tmp = dble(mo_coef_kpts(:,:,k))
call ortho_lowdin(mo_overlap_kpts_real(1,1,k),p,mo_num_per_kpt,mo_coef_tmp,m,ao_num_per_kpt,lin_dep_cutoff)
call zlacp2('X',ao_num_per_kpt,mo_num_per_kpt,mo_coef_tmp,size(mo_coef_tmp,1), &
mo_coef_kpts(1,1,k),size(mo_coef_kpts,1))
enddo
deallocate(mo_coef_tmp)
mo_label = 'Orthonormalized'
SOFT_TOUCH mo_coef_kpts mo_label
end

View File

@ -112,4 +112,71 @@ BEGIN_PROVIDER [ complex*16, eigenvectors_Fock_matrix_mo_kpts, (ao_num_per_kpt,m
deallocate(F, diag)
END_PROVIDER
BEGIN_PROVIDER [ complex*16, eigenvectors_Fock_matrix_mo_kpts_real, (ao_num_per_kpt,mo_num_per_kpt,kpt_num) ]
implicit none
BEGIN_DOC
! Eigenvectors of the Fock matrix in the |MO| basis obtained with level shift.
END_DOC
integer :: i,j,k
integer :: n
!complex*16, allocatable :: F(:,:)
double precision, allocatable :: F(:,:)
double precision, allocatable :: diag(:), mo_coef_tmp(:,:), eigvecs_tmp(:,:)
allocate( F(mo_num_per_kpt,mo_num_per_kpt) )
allocate (diag(mo_num_per_kpt) )
allocate (mo_coef_tmp(ao_num_per_kpt,mo_num_per_kpt) )
allocate (eigvecs_tmp(ao_num_per_kpt,mo_num_per_kpt) )
do k=1,kpt_num
do j=1,mo_num_per_kpt
do i=1,mo_num_per_kpt
!F(i,j) = fock_matrix_mo_complex(i,j)
F(i,j) = dble(fock_matrix_mo_kpts(i,j,k))
enddo
enddo
if(frozen_orb_scf)then
integer :: iorb,jorb
!todo: core/act per kpt
do i = 1, n_core_orb
iorb = list_core(i)
do j = 1, n_act_orb
jorb = list_act(j)
F(iorb,jorb) = 0.d0
F(jorb,iorb) = 0.d0
enddo
enddo
endif
! Insert level shift here
!todo: elec per kpt
do i = elec_beta_num_kpts(k)+1, elec_alpha_num_kpts(k)
F(i,i) += 0.5d0*level_shift
enddo
do i = elec_alpha_num_kpts(k)+1, mo_num_per_kpt
F(i,i) += level_shift
enddo
n = mo_num_per_kpt
call lapack_diagd_diag_in_place(diag,F,n,n)
mo_coef_tmp = dble(mo_coef_kpts(:,:,k))
call dgemm('N','N',ao_num_per_kpt,mo_num_per_kpt,mo_num_per_kpt, 1.d0, &
mo_coef_tmp, size(mo_coef_tmp,1), F, size(F,1), &
0.d0, eigvecs_tmp, size(eigvecs_tmp,1))
call zlacp2('X',ao_num_per_kpt,mo_num_per_kpt,eigvecs_tmp,size(eigvecs_tmp,1), &
eigenvectors_fock_matrix_mo_kpts_real(:,:,k), size(eigenvectors_Fock_matrix_mo_kpts_real,1))
! call zgemm('N','N',ao_num_per_kpt,mo_num_per_kpt,mo_num_per_kpt, (1.d0,0.d0), &
! mo_coef_kpts(:,:,k), size(mo_coef_kpts,1), F, size(F,1), &
! (0.d0,0.d0), eigenvectors_Fock_matrix_mo_kpts(:,:,k), size(eigenvectors_Fock_matrix_mo_kpts,1))
enddo
deallocate(F, diag,mo_coef_tmp,eigvecs_tmp)
END_PROVIDER

View File

@ -164,7 +164,7 @@ BEGIN_PROVIDER [complex*16, FPS_SPF_Matrix_AO_kpts, (AO_num_per_kpt, AO_num_per_
call zgemm('N','N',AO_num_per_kpt,AO_num_per_kpt,AO_num_per_kpt, &
(1.d0,0.d0), &
Fock_Matrix_AO_kpts(1,1,k),Size(Fock_Matrix_AO_kpts,1), &
SCF_Density_Matrix_AO_kpts(1,1,k),Size(SCF_Density_Matrix_AO_kpts,1), &
scf_density_matrix_ao_kpts(1,1,k),Size(SCF_Density_Matrix_AO_kpts,1), &
(0.d0,0.d0), &
scratch,Size(scratch,1))

View File

@ -360,6 +360,24 @@ END_PROVIDER
END_PROVIDER
!============================================!
! !
! kpts_real !
! !
!============================================!
BEGIN_PROVIDER [ double precision, Fock_matrix_mo_kpts_real, (mo_num_per_kpt,mo_num_per_kpt,kpt_num) ]
implicit none
integer :: i,j,k
do k=1,kpt_num
do j=1,mo_num_per_kpt
do i=1,mo_num_per_kpt
fock_matrix_mo_kpts_real(i,j,k) = dble(fock_matrix_mo_kpts(i,j,k))
enddo
enddo
enddo
END_PROVIDER
!============================================!
! !
! kpts !
@ -593,14 +611,10 @@ END_PROVIDER
j = jj(k2)
k = kk(k2)
l = ll(k2)
kpt_i = (i-1)/ao_num_per_kpt +1
kpt_j = (j-1)/ao_num_per_kpt +1
kpt_k = (k-1)/ao_num_per_kpt +1
kpt_l = (l-1)/ao_num_per_kpt +1
idx_i = mod(i-1,ao_num_per_kpt)+1
idx_j = mod(j-1,ao_num_per_kpt)+1
idx_k = mod(k-1,ao_num_per_kpt)+1
idx_l = mod(l-1,ao_num_per_kpt)+1
call get_kpt_idx_ao(i,kpt_i,idx_i)
call get_kpt_idx_ao(j,kpt_j,idx_j)
call get_kpt_idx_ao(k,kpt_k,idx_k)
call get_kpt_idx_ao(l,kpt_l,idx_l)
integral = i_sign(k2)*values(k1) !for klij and lkji, take complex conjugate
!G_a(i,k) += D_{ab}(l,j)*(<ij|kl>)
@ -636,14 +650,10 @@ END_PROVIDER
j = jj(k2)
k = kk(k2)
l = ll(k2)
kpt_i = (i-1)/ao_num_per_kpt +1
kpt_j = (j-1)/ao_num_per_kpt +1
kpt_k = (k-1)/ao_num_per_kpt +1
kpt_l = (l-1)/ao_num_per_kpt +1
idx_i = mod(i-1,ao_num_per_kpt)+1
idx_j = mod(j-1,ao_num_per_kpt)+1
idx_k = mod(k-1,ao_num_per_kpt)+1
idx_l = mod(l-1,ao_num_per_kpt)+1
call get_kpt_idx_ao(i,kpt_i,idx_i)
call get_kpt_idx_ao(j,kpt_j,idx_j)
call get_kpt_idx_ao(k,kpt_k,idx_k)
call get_kpt_idx_ao(l,kpt_l,idx_l)
integral = values(k1)
if (kpt_l.eq.kpt_j) then
@ -714,14 +724,10 @@ END_PROVIDER
j = jj(k2)
k = kk(k2)
l = ll(k2)
kpt_i = (i-1)/ao_num_per_kpt +1
kpt_j = (j-1)/ao_num_per_kpt +1
kpt_k = (k-1)/ao_num_per_kpt +1
kpt_l = (l-1)/ao_num_per_kpt +1
idx_i = mod(i-1,ao_num_per_kpt)+1
idx_j = mod(j-1,ao_num_per_kpt)+1
idx_k = mod(k-1,ao_num_per_kpt)+1
idx_l = mod(l-1,ao_num_per_kpt)+1
call get_kpt_idx_ao(i,kpt_i,idx_i)
call get_kpt_idx_ao(j,kpt_j,idx_j)
call get_kpt_idx_ao(k,kpt_k,idx_k)
call get_kpt_idx_ao(l,kpt_l,idx_l)
integral = i_sign(k2)*values(k1) ! for klij and lkji, take conjugate
!G_a(i,k) += D_{ab}(l,j)*(<ij|kl>)
@ -757,14 +763,10 @@ END_PROVIDER
j = jj(k2)
k = kk(k2)
l = ll(k2)
kpt_i = (i-1)/ao_num_per_kpt +1
kpt_j = (j-1)/ao_num_per_kpt +1
kpt_k = (k-1)/ao_num_per_kpt +1
kpt_l = (l-1)/ao_num_per_kpt +1
idx_i = mod(i-1,ao_num_per_kpt)+1
idx_j = mod(j-1,ao_num_per_kpt)+1
idx_k = mod(k-1,ao_num_per_kpt)+1
idx_l = mod(l-1,ao_num_per_kpt)+1
call get_kpt_idx_ao(i,kpt_i,idx_i)
call get_kpt_idx_ao(j,kpt_j,idx_j)
call get_kpt_idx_ao(k,kpt_k,idx_k)
call get_kpt_idx_ao(l,kpt_l,idx_l)
integral = values(k1)
if (kpt_l.eq.kpt_j) then

View File

@ -89,3 +89,52 @@ subroutine huckel_guess_kpts
deallocate(A)
end
subroutine huckel_guess_kpts_real
implicit none
BEGIN_DOC
! Build the MOs using the extended Huckel model
END_DOC
integer :: i,j,k
double precision :: accu
double precision :: c
character*(64) :: label
!complex*16, allocatable :: A(:,:)
double precision, allocatable :: A(:,:)
label = "Guess"
c = 0.5d0 * 1.75d0
allocate (A(ao_num_per_kpt, ao_num_per_kpt))
do k=1,kpt_num
A = 0.d0
do j=1,ao_num_per_kpt
do i=1,ao_num_per_kpt
A(i,j) = c * ao_overlap_kpts_real(i,j,k) * (ao_one_e_integrals_diag_kpts(i,k) + ao_one_e_integrals_diag_kpts(j,k))
enddo
A(j,j) = ao_one_e_integrals_diag_kpts(j,k) + dble(ao_two_e_integral_alpha_kpts(j,j,k))
if (dabs(dimag(ao_two_e_integral_alpha_kpts(j,j,k))) .gt. 1.0d-10) then
stop 'diagonal elements of ao_two_e_integral_alpha should be real'
endif
enddo
! Fock_matrix_ao_alpha(1:ao_num,1:ao_num) = A(1:ao_num,1:ao_num)
! Fock_matrix_ao_beta (1:ao_num,1:ao_num) = A(1:ao_num,1:ao_num)
call zlacp2('X', ao_num_per_kpt, ao_num_per_kpt, A, size(A,1), &
Fock_matrix_ao_alpha_kpts(:,:,k), size(Fock_matrix_ao_alpha_kpts,1))
call zlacp2('X', ao_num_per_kpt, ao_num_per_kpt, A, size(A,1), &
Fock_matrix_ao_beta_kpts(:,:,k), size(Fock_matrix_ao_beta_kpts, 1))
!call zlacpy('X', ao_num_per_kpt, ao_num_per_kpt, A, size(A,1), &
! Fock_matrix_ao_alpha_kpts(:,:,k), size(Fock_matrix_ao_alpha_kpts,1))
!call zlacpy('X', ao_num_per_kpt, ao_num_per_kpt, A, size(A,1), &
! Fock_matrix_ao_beta_kpts(:,:,k), size(Fock_matrix_ao_beta_kpts, 1))
enddo
! TOUCH mo_coef
!TOUCH fock_matrix_ao_alpha_complex fock_matrix_ao_beta_kpts
TOUCH fock_matrix_ao_alpha_kpts fock_matrix_ao_beta_kpts
mo_coef_kpts = eigenvectors_fock_matrix_mo_kpts_real
SOFT_TOUCH mo_coef_kpts
call save_mos
deallocate(A)
end

View File

@ -653,3 +653,192 @@ END_DOC
endif
end
!============================================!
! !
! kpts_real !
! !
!============================================!
subroutine Roothaan_Hall_SCF_kpts_real
BEGIN_DOC
! Roothaan-Hall algorithm for SCF Hartree-Fock calculation
END_DOC
implicit none
double precision :: energy_SCF,energy_SCF_previous,Delta_energy_SCF
double precision :: max_error_DIIS,max_error_DIIS_alpha,max_error_DIIS_beta
complex*16, allocatable :: Fock_matrix_DIIS(:,:,:,:),error_matrix_DIIS(:,:,:,:)
integer :: iteration_SCF,dim_DIIS,index_dim_DIIS
integer :: i,j,k,kk
logical, external :: qp_stop
complex*16, allocatable :: mo_coef_save(:,:,:)
PROVIDE ao_md5 mo_occ_kpts level_shift
allocate(mo_coef_save(ao_num_per_kpt,mo_num_per_kpt,kpt_num), &
Fock_matrix_DIIS (ao_num_per_kpt,ao_num_per_kpt,max_dim_DIIS,kpt_num), &
error_matrix_DIIS(ao_num_per_kpt,ao_num_per_kpt,max_dim_DIIS,kpt_num) &
)
!todo: add kpt_num dim to diis mats? (3 or 4)
call write_time(6)
print*,'Energy of the guess = ',scf_energy
write(6,'(A4, 1X, A16, 1X, A16, 1X, A16, 1X, A16)') &
'====','================','================','================','================'
write(6,'(A4, 1X, A16, 1X, A16, 1X, A16, 1X, A16)') &
' N ', 'Energy ', 'Energy diff ', 'DIIS error ', 'Level shift '
write(6,'(A4, 1X, A16, 1X, A16, 1X, A16, 1X, A16)') &
'====','================','================','================','================'
! Initialize energies and density matrices
energy_SCF_previous = SCF_energy
Delta_energy_SCF = 1.d0
iteration_SCF = 0
dim_DIIS = 0
max_error_DIIS = 1.d0
!
! Start of main SCF loop
!
!PROVIDE fps_spf_matrix_ao_complex fock_matrix_ao_complex
PROVIDE fps_spf_matrix_ao_kpts fock_matrix_ao_kpts
do while ( &
( (max_error_DIIS > threshold_DIIS_nonzero) .or. &
(dabs(Delta_energy_SCF) > thresh_SCF) &
) .and. (iteration_SCF < n_it_SCF_max) )
! Increment cycle number
iteration_SCF += 1
if(frozen_orb_scf)then
call initialize_mo_coef_begin_iteration
endif
! Current size of the DIIS space
dim_DIIS = min(dim_DIIS+1,max_dim_DIIS)
if (scf_algorithm == 'DIIS') then
do kk=1,kpt_num
! Store Fock and error matrices at each iteration
do j=1,ao_num_per_kpt
do i=1,ao_num_per_kpt
index_dim_DIIS = mod(dim_DIIS-1,max_dim_DIIS)+1
Fock_matrix_DIIS (i,j,index_dim_DIIS,kk) = fock_matrix_ao_kpts(i,j,kk)
error_matrix_DIIS(i,j,index_dim_DIIS,kk) = fps_spf_matrix_ao_kpts(i,j,kk)
enddo
enddo
! Compute the extrapolated Fock matrix
call extrapolate_fock_matrix_kpts( &
error_matrix_DIIS(1,1,1,kk),Fock_matrix_DIIS(1,1,1,kk), &
Fock_matrix_AO_kpts(1,1,kk),size(Fock_matrix_AO_kpts,1), &
iteration_SCF,dim_DIIS &
)
enddo
Fock_matrix_AO_alpha_kpts = Fock_matrix_AO_kpts*0.5d0
Fock_matrix_AO_beta_kpts = Fock_matrix_AO_kpts*0.5d0
TOUCH Fock_matrix_AO_alpha_kpts Fock_matrix_AO_beta_kpts
endif
mo_coef_kpts = eigenvectors_fock_matrix_mo_kpts
if(frozen_orb_scf)then
call reorder_core_orb
call initialize_mo_coef_begin_iteration
endif
TOUCH mo_coef_kpts
! Calculate error vectors
max_error_DIIS = maxval(cdabs(FPS_SPF_Matrix_MO_kpts))
! SCF energy
! call print_debug_scf_complex
energy_SCF = scf_energy
Delta_Energy_SCF = energy_SCF - energy_SCF_previous
if ( (SCF_algorithm == 'DIIS').and.(Delta_Energy_SCF > 0.d0) ) then
do kk=1,kpt_num
Fock_matrix_AO_kpts(1:ao_num_per_kpt,1:ao_num_per_kpt,kk) = &
Fock_matrix_DIIS (1:ao_num_per_kpt,1:ao_num_per_kpt,index_dim_DIIS,kk)
enddo
Fock_matrix_AO_alpha_kpts = Fock_matrix_AO_kpts*0.5d0
Fock_matrix_AO_beta_kpts = Fock_matrix_AO_kpts*0.5d0
TOUCH fock_matrix_ao_alpha_kpts Fock_matrix_AO_beta_kpts
endif
double precision :: level_shift_save
level_shift_save = level_shift
mo_coef_save(1:ao_num_per_kpt,1:mo_num_per_kpt,1:kpt_num) = mo_coef_kpts(1:ao_num_per_kpt,1:mo_num_per_kpt,1:kpt_num)
do while (Delta_energy_SCF > 0.d0)
mo_coef_kpts(1:ao_num_per_kpt,1:mo_num_per_kpt,1:kpt_num) = mo_coef_save
if (level_shift <= .1d0) then
level_shift = 1.d0
else
level_shift = level_shift * 3.0d0
endif
TOUCH mo_coef_kpts level_shift
mo_coef_kpts(1:ao_num_per_kpt,1:mo_num_per_kpt,1:kpt_num) = &
eigenvectors_fock_matrix_mo_kpts_real(1:ao_num_per_kpt,1:mo_num_per_kpt,1:kpt_num)
if(frozen_orb_scf)then
call reorder_core_orb
call initialize_mo_coef_begin_iteration
endif
TOUCH mo_coef_kpts
Delta_Energy_SCF = SCF_energy - energy_SCF_previous
energy_SCF = SCF_energy
if (level_shift-level_shift_save > 40.d0) then
level_shift = level_shift_save * 4.d0
SOFT_TOUCH level_shift
exit
endif
dim_DIIS=0
enddo
level_shift = level_shift * 0.5d0
SOFT_TOUCH level_shift
energy_SCF_previous = energy_SCF
! Print results at the end of each iteration
write(6,'(I4, 1X, F16.10, 1X, F16.10, 1X, F16.10, 1X, F16.10, 1X, I3)') &
iteration_SCF, energy_scf, Delta_energy_SCF, max_error_DIIS, level_shift, dim_DIIS
if (Delta_energy_SCF < 0.d0) then
call save_mos
endif
if (qp_stop()) exit
enddo
if (iteration_SCF < n_it_SCF_max) then
mo_label = "Canonical"
endif
!
! End of Main SCF loop
!
write(6,'(A4, 1X, A16, 1X, A16, 1X, A16, 1X, A16)') &
'====','================','================','================','================'
write(6,*)
if(.not.frozen_orb_scf)then
call mo_as_eigvectors_of_mo_matrix_kpts_real(fock_matrix_mo_kpts_real,size(Fock_matrix_mo_kpts_real,1),size(Fock_matrix_mo_kpts_real,2),size(Fock_matrix_mo_kpts_real,3),mo_label,1,.true.)
call save_mos
endif
call write_double(6, Energy_SCF, 'SCF energy')
call write_time(6)
end

View File

@ -17,7 +17,11 @@ end
subroutine routine
implicit none
call diagonalize_CI
call diagonalize_ci
print*,'N_det = ',N_det
call save_wavefunction_general(N_det,N_states,psi_det_sorted,size(psi_coef_sorted,1),psi_coef_sorted)
if (is_complex) then
call save_wavefunction_general_complex(N_det,N_states,psi_det_sorted,size(psi_coef_sorted_complex,1),psi_coef_sorted_complex)
else
call save_wavefunction_general(N_det,N_states,psi_det_sorted,size(psi_coef_sorted,1),psi_coef_sorted)
endif
end

View File

@ -9,7 +9,11 @@ program print_hamiltonian
! psi_coef_sorted are the wave function stored in the |EZFIO| directory.
read_wf = .True.
touch read_wf
call run
if (is_complex) then
call run_complex
else
call run
endif
end
subroutine run
@ -27,3 +31,42 @@ subroutine run
enddo
end
subroutine run_complex
implicit none
integer :: i, j
complex*16 :: hij
double precision :: s2
print*,'i,j,Hij'
do j=1,N_det
do i=1,N_det
call i_h_j_complex(psi_det(1,1,i), psi_det(1,1,j), N_int, hij)
if (cdabs(hij) > 1.d-20) then
print *, i, j, dble(hij), dimag(hij)
endif
enddo
enddo
print*,'i,j,S2ij'
do j=1,N_det
do i=1,N_det
call get_s2(psi_det(1,1,i), psi_det(1,1,j), N_int, s2)
if (dabs(s2) > 1.d-20) then
print *, i, j, s2
endif
enddo
enddo
! use bitmasks
integer :: degree
print*,'i,j,degij'
do j=1,N_det
do i=1,N_det
call get_excitation_degree(psi_det(1,1,i), psi_det(1,1,j), degree, N_int)
if (degree.le.2) then
print *, i, j, degree
endif
enddo
enddo
end

View File

@ -1277,3 +1277,77 @@ subroutine lapack_diag(eigvalues,eigvectors,H,nmax,n)
deallocate(A,eigenvalues)
end
subroutine lapack_diagd_diag_in_place(eigvalues,eigvectors,nmax,n)
implicit none
BEGIN_DOC
! Diagonalize matrix H(complex)
!
! H is untouched between input and ouptut
!
! eigevalues(i) = ith lowest eigenvalue of the H matrix
!
! eigvectors(i,j) = <i|psi_j> where i is the basis function and psi_j is the j th eigenvector
!
END_DOC
integer, intent(in) :: n,nmax
double precision, intent(out) :: eigvectors(nmax,n)
! complex*16, intent(inout) :: eigvectors(nmax,n)
double precision, intent(out) :: eigvalues(n)
! double precision, intent(in) :: H(nmax,n)
double precision,allocatable :: work(:)
integer ,allocatable :: iwork(:)
! complex*16,allocatable :: A(:,:)
integer :: lwork, info, i,j,l,k, liwork
! print*,'Diagonalization by jacobi'
! print*,'n = ',n
lwork = 2*n*n + 6*n + 1
liwork = 5*n + 3
allocate (work(lwork),iwork(liwork))
lwork = -1
liwork = -1
! get optimal work size
call DSYEVD( 'V', 'U', n, eigvectors, nmax, eigvalues, work, lwork, &
iwork, liwork, info )
if (info < 0) then
print *, irp_here, ': DSYEVD: the ',-info,'-th argument had an illegal value'
stop 2
endif
lwork = int( real(work(1)))
liwork = iwork(1)
deallocate (work,iwork)
allocate (work(lwork),iwork(liwork))
call DSYEVD( 'V', 'U', n, eigvectors, nmax, eigvalues, work, lwork, &
iwork, liwork, info )
deallocate(work,iwork)
if (info < 0) then
print *, irp_here, ': DSYEVD: the ',-info,'-th argument had an illegal value'
stop 2
else if( info > 0 ) then
write(*,*)'DSYEVD Failed; calling DSYEV'
lwork = 3*n - 1
allocate(work(lwork))
lwork = -1
call DSYEV('V','L',n,eigvectors,nmax,eigvalues,work,lwork,info)
if (info < 0) then
print *, irp_here, ': DSYEV: the ',-info,'-th argument had an illegal value'
stop 2
endif
lwork = int(work(1))
deallocate(work)
allocate(work(lwork))
call DSYEV('V','L',n,eigvectors,nmax,eigvalues,work,lwork,info)
if (info /= 0 ) then
write(*,*)'DSYEV Failed'
stop 1
endif
deallocate(work)
end if
end