9
1
mirror of https://github.com/QuantumPackage/qp2.git synced 2024-11-09 15:03:37 +01:00
qp2/plugins/local/tc_scf/fock_three_hermit.irp.f

772 lines
30 KiB
Fortran
Raw Normal View History

2023-06-22 18:26:52 +02:00
! ---
BEGIN_PROVIDER [ double precision, fock_3_mat, (mo_num, mo_num)]
2023-06-22 18:26:52 +02:00
implicit none
integer :: i,j
double precision :: contrib
2023-06-22 18:26:52 +02:00
fock_3_mat = 0.d0
2023-06-22 18:26:52 +02:00
if(.not.bi_ortho .and. three_body_h_tc) then
call give_fock_ia_three_e_total(1, 1, contrib)
!! !$OMP PARALLEL &
!! !$OMP DEFAULT (NONE) &
!! !$OMP PRIVATE (i,j,m,integral) &
!! !$OMP SHARED (mo_num,three_body_3_index)
!! !$OMP DO SCHEDULE (guided) COLLAPSE(3)
do i = 1, mo_num
do j = 1, mo_num
call give_fock_ia_three_e_total(j,i,contrib)
fock_3_mat(j,i) = -contrib
enddo
enddo
!else if(bi_ortho.and.three_body_h_tc) then
!! !$OMP END DO
!! !$OMP END PARALLEL
!! do i = 1, mo_num
!! do j = 1, i-1
!! mat_three(j,i) = mat_three(i,j)
!! enddo
!! enddo
endif
END_PROVIDER
subroutine give_fock_ia_three_e_total(i,a,contrib)
implicit none
BEGIN_DOC
! contrib is the TOTAL (same spins / opposite spins) contribution from the three body term to the Fock operator
!
END_DOC
integer, intent(in) :: i,a
double precision, intent(out) :: contrib
double precision :: int_1, int_2, int_3
double precision :: mos_i, mos_a, w_ia
double precision :: mos_ia, weight
integer :: mm, ipoint,k,l
int_1 = 0.d0
int_2 = 0.d0
int_3 = 0.d0
do mm = 1, 3
do ipoint = 1, n_points_final_grid
weight = final_weight_at_r_vector(ipoint)
mos_i = mos_in_r_array_transp(ipoint,i)
mos_a = mos_in_r_array_transp(ipoint,a)
mos_ia = mos_a * mos_i
w_ia = x_W_ij_erf_rk(ipoint,mm,i,a)
int_1 += weight * fock_3_w_kk_sum(ipoint,mm) * (4.d0 * fock_3_rho_beta(ipoint) * w_ia &
+ 2.0d0 * mos_ia * fock_3_w_kk_sum(ipoint,mm) &
- 2.0d0 * fock_3_w_ki_mos_k(ipoint,mm,i) * mos_a &
- 2.0d0 * fock_3_w_ki_mos_k(ipoint,mm,a) * mos_i )
int_2 += weight * (-1.d0) * ( 2.0d0 * fock_3_w_kl_mo_k_mo_l(ipoint,mm) * w_ia &
+ 2.0d0 * fock_3_rho_beta(ipoint) * fock_3_w_ki_wk_a(ipoint,mm,i,a) &
+ 1.0d0 * mos_ia * fock_3_trace_w_tilde(ipoint,mm) )
int_3 += weight * 1.d0 * (fock_3_w_kl_wla_phi_k(ipoint,mm,i) * mos_a + fock_3_w_kl_wla_phi_k(ipoint,mm,a) * mos_i &
+fock_3_w_ki_mos_k(ipoint,mm,i) * fock_3_w_ki_mos_k(ipoint,mm,a) )
enddo
enddo
contrib = int_1 + int_2 + int_3
end
! ---
BEGIN_PROVIDER [double precision, diag_three_elem_hf]
implicit none
2023-09-18 12:03:29 +02:00
integer :: i, j, k, ipoint, mm
double precision :: contrib, weight, four_third, one_third, two_third, exchange_int_231
double precision :: integral_aaa, hthree, integral_aab, integral_abb, integral_bbb
double precision, allocatable :: tmp(:)
double precision, allocatable :: tmp_L(:,:), tmp_R(:,:)
double precision, allocatable :: tmp_M(:,:), tmp_S(:), tmp_O(:), tmp_J(:,:)
double precision, allocatable :: tmp_M_priv(:,:), tmp_S_priv(:), tmp_O_priv(:), tmp_J_priv(:,:)
PROVIDE mo_l_coef mo_r_coef
!print *, ' providing diag_three_elem_hf'
if(.not. three_body_h_tc) then
2023-09-23 11:26:25 +02:00
if(noL_standard) then
PROVIDE noL_0e
diag_three_elem_hf = noL_0e
else
diag_three_elem_hf = 0.d0
endif
else
if(.not. bi_ortho) then
! ---
one_third = 1.d0/3.d0
two_third = 2.d0/3.d0
four_third = 4.d0/3.d0
diag_three_elem_hf = 0.d0
do i = 1, elec_beta_num
do j = 1, elec_beta_num
do k = 1, elec_beta_num
2023-06-24 18:54:32 +02:00
call give_integrals_3_body(k, j, i, j, i, k, exchange_int_231)
diag_three_elem_hf += two_third * exchange_int_231
enddo
enddo
enddo
do mm = 1, 3
do ipoint = 1, n_points_final_grid
weight = final_weight_at_r_vector(ipoint)
contrib = 3.d0 * fock_3_w_kk_sum(ipoint,mm) * fock_3_rho_beta(ipoint) * fock_3_w_kk_sum(ipoint,mm) &
- 2.d0 * fock_3_w_kl_mo_k_mo_l(ipoint,mm) * fock_3_w_kk_sum(ipoint,mm) &
- 1.d0 * fock_3_rho_beta(ipoint) * fock_3_w_kl_w_kl(ipoint,mm)
contrib *= four_third
contrib += -two_third * fock_3_rho_beta(ipoint) * fock_3_w_kl_w_kl(ipoint,mm) &
-four_third * fock_3_w_kk_sum(ipoint,mm) * fock_3_w_kl_mo_k_mo_l(ipoint,mm)
diag_three_elem_hf += weight * contrib
enddo
enddo
diag_three_elem_hf = - diag_three_elem_hf
! ---
else
2023-09-18 12:03:29 +02:00
! ------------
! SLOW VERSION
! ------------
!call give_aaa_contrib(integral_aaa)
!call give_aab_contrib(integral_aab)
!call give_abb_contrib(integral_abb)
!call give_bbb_contrib(integral_bbb)
!diag_three_elem_hf = integral_aaa + integral_aab + integral_abb + integral_bbb
! ------------
! ------------
PROVIDE int2_grad1_u12_bimo_t
PROVIDE mos_l_in_r_array_transp
PROVIDE mos_r_in_r_array_transp
if(elec_alpha_num .eq. elec_beta_num) then
allocate(tmp(elec_beta_num))
allocate(tmp_L(n_points_final_grid,3), tmp_R(n_points_final_grid,3))
!$OMP PARALLEL &
!$OMP DEFAULT(NONE) &
!$OMP PRIVATE(j, i, ipoint, tmp_L, tmp_R) &
!$OMP SHARED(elec_beta_num, n_points_final_grid, &
!$OMP mos_l_in_r_array_transp, mos_r_in_r_array_transp, &
!$OMP int2_grad1_u12_bimo_t, tmp, final_weight_at_r_vector)
!$OMP DO
do j = 1, elec_beta_num
tmp_L = 0.d0
tmp_R = 0.d0
do i = 1, elec_beta_num
do ipoint = 1, n_points_final_grid
tmp_L(ipoint,1) = tmp_L(ipoint,1) + int2_grad1_u12_bimo_t(ipoint,1,j,i) * mos_l_in_r_array_transp(ipoint,i)
tmp_L(ipoint,2) = tmp_L(ipoint,2) + int2_grad1_u12_bimo_t(ipoint,2,j,i) * mos_l_in_r_array_transp(ipoint,i)
tmp_L(ipoint,3) = tmp_L(ipoint,3) + int2_grad1_u12_bimo_t(ipoint,3,j,i) * mos_l_in_r_array_transp(ipoint,i)
tmp_R(ipoint,1) = tmp_R(ipoint,1) + int2_grad1_u12_bimo_t(ipoint,1,i,j) * mos_r_in_r_array_transp(ipoint,i)
tmp_R(ipoint,2) = tmp_R(ipoint,2) + int2_grad1_u12_bimo_t(ipoint,2,i,j) * mos_r_in_r_array_transp(ipoint,i)
tmp_R(ipoint,3) = tmp_R(ipoint,3) + int2_grad1_u12_bimo_t(ipoint,3,i,j) * mos_r_in_r_array_transp(ipoint,i)
enddo
enddo
tmp(j) = 0.d0
do ipoint = 1, n_points_final_grid
tmp(j) = tmp(j) + final_weight_at_r_vector(ipoint) * (tmp_L(ipoint,1)*tmp_R(ipoint,1) + tmp_L(ipoint,2)*tmp_R(ipoint,2) + tmp_L(ipoint,3)*tmp_R(ipoint,3))
enddo
enddo ! j
!$OMP END DO
!$OMP END PARALLEL
diag_three_elem_hf = -2.d0 * sum(tmp)
deallocate(tmp)
deallocate(tmp_L, tmp_R)
! ---
allocate(tmp_O(n_points_final_grid), tmp_J(n_points_final_grid,3))
tmp_O = 0.d0
tmp_J = 0.d0
!$OMP PARALLEL &
!$OMP DEFAULT(NONE) &
!$OMP PRIVATE(i, ipoint, tmp_O_priv, tmp_J_priv) &
!$OMP SHARED(elec_beta_num, n_points_final_grid, &
!$OMP mos_l_in_r_array_transp, mos_r_in_r_array_transp, &
!$OMP int2_grad1_u12_bimo_t, tmp_O, tmp_J)
allocate(tmp_O_priv(n_points_final_grid), tmp_J_priv(n_points_final_grid,3))
tmp_O_priv = 0.d0
tmp_J_priv = 0.d0
!$OMP DO
do i = 1, elec_beta_num
do ipoint = 1, n_points_final_grid
tmp_O_priv(ipoint) = tmp_O_priv(ipoint) + mos_l_in_r_array_transp(ipoint,i) * mos_r_in_r_array_transp(ipoint,i)
tmp_J_priv(ipoint,1) = tmp_J_priv(ipoint,1) + int2_grad1_u12_bimo_t(ipoint,1,i,i)
tmp_J_priv(ipoint,2) = tmp_J_priv(ipoint,2) + int2_grad1_u12_bimo_t(ipoint,2,i,i)
tmp_J_priv(ipoint,3) = tmp_J_priv(ipoint,3) + int2_grad1_u12_bimo_t(ipoint,3,i,i)
enddo
enddo
!$OMP END DO NOWAIT
!$OMP CRITICAL
tmp_O = tmp_O + tmp_O_priv
tmp_J = tmp_J + tmp_J_priv
!$OMP END CRITICAL
deallocate(tmp_O_priv, tmp_J_priv)
!$OMP END PARALLEL
allocate(tmp_M(n_points_final_grid,3), tmp_S(n_points_final_grid))
tmp_M = 0.d0
tmp_S = 0.d0
!$OMP PARALLEL &
!$OMP DEFAULT(NONE) &
!$OMP PRIVATE(i, j, ipoint, tmp_M_priv, tmp_S_priv) &
!$OMP SHARED(elec_beta_num, n_points_final_grid, &
!$OMP mos_l_in_r_array_transp, mos_r_in_r_array_transp, &
!$OMP int2_grad1_u12_bimo_t, tmp_M, tmp_S)
allocate(tmp_M_priv(n_points_final_grid,3), tmp_S_priv(n_points_final_grid))
tmp_M_priv = 0.d0
tmp_S_priv = 0.d0
!$OMP DO COLLAPSE(2)
do i = 1, elec_beta_num
do j = 1, elec_beta_num
do ipoint = 1, n_points_final_grid
tmp_M_priv(ipoint,1) = tmp_M_priv(ipoint,1) + int2_grad1_u12_bimo_t(ipoint,1,j,i) * mos_l_in_r_array_transp(ipoint,i) * mos_r_in_r_array_transp(ipoint,j)
tmp_M_priv(ipoint,2) = tmp_M_priv(ipoint,2) + int2_grad1_u12_bimo_t(ipoint,2,j,i) * mos_l_in_r_array_transp(ipoint,i) * mos_r_in_r_array_transp(ipoint,j)
tmp_M_priv(ipoint,3) = tmp_M_priv(ipoint,3) + int2_grad1_u12_bimo_t(ipoint,3,j,i) * mos_l_in_r_array_transp(ipoint,i) * mos_r_in_r_array_transp(ipoint,j)
tmp_S_priv(ipoint) = tmp_S_priv(ipoint) + int2_grad1_u12_bimo_t(ipoint,1,i,j) * int2_grad1_u12_bimo_t(ipoint,1,j,i) &
+ int2_grad1_u12_bimo_t(ipoint,2,i,j) * int2_grad1_u12_bimo_t(ipoint,2,j,i) &
+ int2_grad1_u12_bimo_t(ipoint,3,i,j) * int2_grad1_u12_bimo_t(ipoint,3,j,i)
enddo
enddo
enddo
!$OMP END DO NOWAIT
!$OMP CRITICAL
tmp_M = tmp_M + tmp_M_priv
tmp_S = tmp_S + tmp_S_priv
!$OMP END CRITICAL
deallocate(tmp_M_priv, tmp_S_priv)
!$OMP END PARALLEL
allocate(tmp(n_points_final_grid))
do ipoint = 1, n_points_final_grid
tmp_S(ipoint) = 2.d0 * (tmp_J(ipoint,1)*tmp_J(ipoint,1) + tmp_J(ipoint,2)*tmp_J(ipoint,2) + tmp_J(ipoint,3)*tmp_J(ipoint,3)) - tmp_S(ipoint)
tmp(ipoint) = final_weight_at_r_vector(ipoint) * ( tmp_O(ipoint) * tmp_S(ipoint) &
- 2.d0 * ( tmp_J(ipoint,1) * tmp_M(ipoint,1) &
+ tmp_J(ipoint,2) * tmp_M(ipoint,2) &
+ tmp_J(ipoint,3) * tmp_M(ipoint,3)))
enddo
diag_three_elem_hf = diag_three_elem_hf -2.d0 * (sum(tmp))
deallocate(tmp)
else
allocate(tmp(elec_alpha_num))
allocate(tmp_L(n_points_final_grid,3), tmp_R(n_points_final_grid,3))
!$OMP PARALLEL &
!$OMP DEFAULT(NONE) &
!$OMP PRIVATE(j, i, ipoint, tmp_L, tmp_R) &
!$OMP SHARED(elec_beta_num, elec_alpha_num, n_points_final_grid, &
!$OMP mos_l_in_r_array_transp, mos_r_in_r_array_transp, &
!$OMP int2_grad1_u12_bimo_t, tmp, final_weight_at_r_vector)
!$OMP DO
do j = 1, elec_beta_num
tmp_L = 0.d0
tmp_R = 0.d0
do i = elec_beta_num+1, elec_alpha_num
do ipoint = 1, n_points_final_grid
tmp_L(ipoint,1) = tmp_L(ipoint,1) + 0.5d0 * int2_grad1_u12_bimo_t(ipoint,1,j,i) * mos_l_in_r_array_transp(ipoint,i)
tmp_L(ipoint,2) = tmp_L(ipoint,2) + 0.5d0 * int2_grad1_u12_bimo_t(ipoint,2,j,i) * mos_l_in_r_array_transp(ipoint,i)
tmp_L(ipoint,3) = tmp_L(ipoint,3) + 0.5d0 * int2_grad1_u12_bimo_t(ipoint,3,j,i) * mos_l_in_r_array_transp(ipoint,i)
tmp_R(ipoint,1) = tmp_R(ipoint,1) + 0.5d0 * int2_grad1_u12_bimo_t(ipoint,1,i,j) * mos_r_in_r_array_transp(ipoint,i)
tmp_R(ipoint,2) = tmp_R(ipoint,2) + 0.5d0 * int2_grad1_u12_bimo_t(ipoint,2,i,j) * mos_r_in_r_array_transp(ipoint,i)
tmp_R(ipoint,3) = tmp_R(ipoint,3) + 0.5d0 * int2_grad1_u12_bimo_t(ipoint,3,i,j) * mos_r_in_r_array_transp(ipoint,i)
enddo
enddo
tmp(j) = 0.d0
do ipoint = 1, n_points_final_grid
tmp(j) = tmp(j) + final_weight_at_r_vector(ipoint) * (tmp_L(ipoint,1)*tmp_R(ipoint,1) + tmp_L(ipoint,2)*tmp_R(ipoint,2) + tmp_L(ipoint,3)*tmp_R(ipoint,3))
enddo
do i = 1, elec_beta_num
do ipoint = 1, n_points_final_grid
tmp_L(ipoint,1) = tmp_L(ipoint,1) + int2_grad1_u12_bimo_t(ipoint,1,j,i) * mos_l_in_r_array_transp(ipoint,i)
tmp_L(ipoint,2) = tmp_L(ipoint,2) + int2_grad1_u12_bimo_t(ipoint,2,j,i) * mos_l_in_r_array_transp(ipoint,i)
tmp_L(ipoint,3) = tmp_L(ipoint,3) + int2_grad1_u12_bimo_t(ipoint,3,j,i) * mos_l_in_r_array_transp(ipoint,i)
tmp_R(ipoint,1) = tmp_R(ipoint,1) + int2_grad1_u12_bimo_t(ipoint,1,i,j) * mos_r_in_r_array_transp(ipoint,i)
tmp_R(ipoint,2) = tmp_R(ipoint,2) + int2_grad1_u12_bimo_t(ipoint,2,i,j) * mos_r_in_r_array_transp(ipoint,i)
tmp_R(ipoint,3) = tmp_R(ipoint,3) + int2_grad1_u12_bimo_t(ipoint,3,i,j) * mos_r_in_r_array_transp(ipoint,i)
enddo
enddo
do ipoint = 1, n_points_final_grid
tmp(j) = tmp(j) + final_weight_at_r_vector(ipoint) * (tmp_L(ipoint,1)*tmp_R(ipoint,1) + tmp_L(ipoint,2)*tmp_R(ipoint,2) + tmp_L(ipoint,3)*tmp_R(ipoint,3))
enddo
enddo ! j
!$OMP END DO
!$OMP END PARALLEL
! ---
!$OMP PARALLEL &
!$OMP DEFAULT(NONE) &
!$OMP PRIVATE(j, i, ipoint, tmp_L, tmp_R) &
!$OMP SHARED(elec_beta_num, elec_alpha_num, n_points_final_grid, &
!$OMP mos_l_in_r_array_transp, mos_r_in_r_array_transp, &
!$OMP int2_grad1_u12_bimo_t, tmp, final_weight_at_r_vector)
!$OMP DO
do j = elec_beta_num+1, elec_alpha_num
tmp_L = 0.d0
tmp_R = 0.d0
do i = 1, elec_alpha_num
do ipoint = 1, n_points_final_grid
tmp_L(ipoint,1) = tmp_L(ipoint,1) + int2_grad1_u12_bimo_t(ipoint,1,j,i) * mos_l_in_r_array_transp(ipoint,i)
tmp_L(ipoint,2) = tmp_L(ipoint,2) + int2_grad1_u12_bimo_t(ipoint,2,j,i) * mos_l_in_r_array_transp(ipoint,i)
tmp_L(ipoint,3) = tmp_L(ipoint,3) + int2_grad1_u12_bimo_t(ipoint,3,j,i) * mos_l_in_r_array_transp(ipoint,i)
tmp_R(ipoint,1) = tmp_R(ipoint,1) + int2_grad1_u12_bimo_t(ipoint,1,i,j) * mos_r_in_r_array_transp(ipoint,i)
tmp_R(ipoint,2) = tmp_R(ipoint,2) + int2_grad1_u12_bimo_t(ipoint,2,i,j) * mos_r_in_r_array_transp(ipoint,i)
tmp_R(ipoint,3) = tmp_R(ipoint,3) + int2_grad1_u12_bimo_t(ipoint,3,i,j) * mos_r_in_r_array_transp(ipoint,i)
enddo
enddo
tmp(j) = 0.d0
do ipoint = 1, n_points_final_grid
tmp(j) = tmp(j) + 0.5d0 * final_weight_at_r_vector(ipoint) * (tmp_L(ipoint,1)*tmp_R(ipoint,1) + tmp_L(ipoint,2)*tmp_R(ipoint,2) + tmp_L(ipoint,3)*tmp_R(ipoint,3))
enddo
enddo ! j
!$OMP END DO
!$OMP END PARALLEL
diag_three_elem_hf = -2.d0 * sum(tmp)
deallocate(tmp)
deallocate(tmp_L, tmp_R)
! ---
allocate(tmp_O(n_points_final_grid), tmp_J(n_points_final_grid,3))
tmp_O = 0.d0
tmp_J = 0.d0
!$OMP PARALLEL &
!$OMP DEFAULT(NONE) &
!$OMP PRIVATE(i, ipoint, tmp_O_priv, tmp_J_priv) &
!$OMP SHARED(elec_beta_num, elec_alpha_num, n_points_final_grid, &
!$OMP mos_l_in_r_array_transp, mos_r_in_r_array_transp, &
!$OMP int2_grad1_u12_bimo_t, tmp_O, tmp_J)
allocate(tmp_O_priv(n_points_final_grid), tmp_J_priv(n_points_final_grid,3))
tmp_O_priv = 0.d0
tmp_J_priv = 0.d0
!$OMP DO
do i = 1, elec_beta_num
do ipoint = 1, n_points_final_grid
tmp_O_priv(ipoint) = tmp_O_priv(ipoint) + mos_l_in_r_array_transp(ipoint,i) * mos_r_in_r_array_transp(ipoint,i)
tmp_J_priv(ipoint,1) = tmp_J_priv(ipoint,1) + int2_grad1_u12_bimo_t(ipoint,1,i,i)
tmp_J_priv(ipoint,2) = tmp_J_priv(ipoint,2) + int2_grad1_u12_bimo_t(ipoint,2,i,i)
tmp_J_priv(ipoint,3) = tmp_J_priv(ipoint,3) + int2_grad1_u12_bimo_t(ipoint,3,i,i)
enddo
enddo
!$OMP END DO NOWAIT
!$OMP DO
do i = elec_beta_num+1, elec_alpha_num
do ipoint = 1, n_points_final_grid
tmp_O_priv(ipoint) = tmp_O_priv(ipoint) + 0.5d0 * mos_l_in_r_array_transp(ipoint,i) * mos_r_in_r_array_transp(ipoint,i)
tmp_J_priv(ipoint,1) = tmp_J_priv(ipoint,1) + 0.5d0 * int2_grad1_u12_bimo_t(ipoint,1,i,i)
tmp_J_priv(ipoint,2) = tmp_J_priv(ipoint,2) + 0.5d0 * int2_grad1_u12_bimo_t(ipoint,2,i,i)
tmp_J_priv(ipoint,3) = tmp_J_priv(ipoint,3) + 0.5d0 * int2_grad1_u12_bimo_t(ipoint,3,i,i)
enddo
enddo
!$OMP END DO NOWAIT
!$OMP CRITICAL
tmp_O = tmp_O + tmp_O_priv
tmp_J = tmp_J + tmp_J_priv
!$OMP END CRITICAL
deallocate(tmp_O_priv, tmp_J_priv)
!$OMP END PARALLEL
! ---
allocate(tmp_M(n_points_final_grid,3), tmp_S(n_points_final_grid))
tmp_M = 0.d0
tmp_S = 0.d0
!$OMP PARALLEL &
!$OMP DEFAULT(NONE) &
!$OMP PRIVATE(i, j, ipoint, tmp_M_priv, tmp_S_priv) &
!$OMP SHARED(elec_beta_num, elec_alpha_num, n_points_final_grid, &
!$OMP mos_l_in_r_array_transp, mos_r_in_r_array_transp, &
!$OMP int2_grad1_u12_bimo_t, tmp_M, tmp_S)
allocate(tmp_M_priv(n_points_final_grid,3), tmp_S_priv(n_points_final_grid))
tmp_M_priv = 0.d0
tmp_S_priv = 0.d0
!$OMP DO COLLAPSE(2)
do i = 1, elec_beta_num
do j = 1, elec_beta_num
do ipoint = 1, n_points_final_grid
tmp_M_priv(ipoint,1) = tmp_M_priv(ipoint,1) + int2_grad1_u12_bimo_t(ipoint,1,j,i) * mos_l_in_r_array_transp(ipoint,i) * mos_r_in_r_array_transp(ipoint,j)
tmp_M_priv(ipoint,2) = tmp_M_priv(ipoint,2) + int2_grad1_u12_bimo_t(ipoint,2,j,i) * mos_l_in_r_array_transp(ipoint,i) * mos_r_in_r_array_transp(ipoint,j)
tmp_M_priv(ipoint,3) = tmp_M_priv(ipoint,3) + int2_grad1_u12_bimo_t(ipoint,3,j,i) * mos_l_in_r_array_transp(ipoint,i) * mos_r_in_r_array_transp(ipoint,j)
tmp_S_priv(ipoint) = tmp_S_priv(ipoint) + int2_grad1_u12_bimo_t(ipoint,1,i,j) * int2_grad1_u12_bimo_t(ipoint,1,j,i) &
+ int2_grad1_u12_bimo_t(ipoint,2,i,j) * int2_grad1_u12_bimo_t(ipoint,2,j,i) &
+ int2_grad1_u12_bimo_t(ipoint,3,i,j) * int2_grad1_u12_bimo_t(ipoint,3,j,i)
enddo
enddo
enddo
!$OMP END DO NOWAIT
!$OMP DO COLLAPSE(2)
do i = elec_beta_num+1, elec_alpha_num
do j = 1, elec_beta_num
do ipoint = 1, n_points_final_grid
tmp_M_priv(ipoint,1) = tmp_M_priv(ipoint,1) + 0.5d0 * int2_grad1_u12_bimo_t(ipoint,1,j,i) * mos_l_in_r_array_transp(ipoint,i) * mos_r_in_r_array_transp(ipoint,j)
tmp_M_priv(ipoint,2) = tmp_M_priv(ipoint,2) + 0.5d0 * int2_grad1_u12_bimo_t(ipoint,2,j,i) * mos_l_in_r_array_transp(ipoint,i) * mos_r_in_r_array_transp(ipoint,j)
tmp_M_priv(ipoint,3) = tmp_M_priv(ipoint,3) + 0.5d0 * int2_grad1_u12_bimo_t(ipoint,3,j,i) * mos_l_in_r_array_transp(ipoint,i) * mos_r_in_r_array_transp(ipoint,j)
tmp_M_priv(ipoint,1) = tmp_M_priv(ipoint,1) + 0.5d0 * int2_grad1_u12_bimo_t(ipoint,1,i,j) * mos_l_in_r_array_transp(ipoint,j) * mos_r_in_r_array_transp(ipoint,i)
tmp_M_priv(ipoint,2) = tmp_M_priv(ipoint,2) + 0.5d0 * int2_grad1_u12_bimo_t(ipoint,2,i,j) * mos_l_in_r_array_transp(ipoint,j) * mos_r_in_r_array_transp(ipoint,i)
tmp_M_priv(ipoint,3) = tmp_M_priv(ipoint,3) + 0.5d0 * int2_grad1_u12_bimo_t(ipoint,3,i,j) * mos_l_in_r_array_transp(ipoint,j) * mos_r_in_r_array_transp(ipoint,i)
tmp_S_priv(ipoint) = tmp_S_priv(ipoint) + int2_grad1_u12_bimo_t(ipoint,1,i,j) * int2_grad1_u12_bimo_t(ipoint,1,j,i) &
+ int2_grad1_u12_bimo_t(ipoint,2,i,j) * int2_grad1_u12_bimo_t(ipoint,2,j,i) &
+ int2_grad1_u12_bimo_t(ipoint,3,i,j) * int2_grad1_u12_bimo_t(ipoint,3,j,i)
enddo
enddo
enddo
!$OMP END DO NOWAIT
!$OMP DO COLLAPSE(2)
do i = elec_beta_num+1, elec_alpha_num
do j = elec_beta_num+1, elec_alpha_num
do ipoint = 1, n_points_final_grid
tmp_M_priv(ipoint,1) = tmp_M_priv(ipoint,1) + 0.5d0 * int2_grad1_u12_bimo_t(ipoint,1,j,i) * mos_l_in_r_array_transp(ipoint,i) * mos_r_in_r_array_transp(ipoint,j)
tmp_M_priv(ipoint,2) = tmp_M_priv(ipoint,2) + 0.5d0 * int2_grad1_u12_bimo_t(ipoint,2,j,i) * mos_l_in_r_array_transp(ipoint,i) * mos_r_in_r_array_transp(ipoint,j)
tmp_M_priv(ipoint,3) = tmp_M_priv(ipoint,3) + 0.5d0 * int2_grad1_u12_bimo_t(ipoint,3,j,i) * mos_l_in_r_array_transp(ipoint,i) * mos_r_in_r_array_transp(ipoint,j)
tmp_S_priv(ipoint) = tmp_S_priv(ipoint) + 0.5d0 * int2_grad1_u12_bimo_t(ipoint,1,i,j) * int2_grad1_u12_bimo_t(ipoint,1,j,i) &
+ 0.5d0 * int2_grad1_u12_bimo_t(ipoint,2,i,j) * int2_grad1_u12_bimo_t(ipoint,2,j,i) &
+ 0.5d0 * int2_grad1_u12_bimo_t(ipoint,3,i,j) * int2_grad1_u12_bimo_t(ipoint,3,j,i)
enddo
enddo
enddo
!$OMP END DO NOWAIT
!$OMP CRITICAL
tmp_M = tmp_M + tmp_M_priv
tmp_S = tmp_S + tmp_S_priv
!$OMP END CRITICAL
deallocate(tmp_M_priv, tmp_S_priv)
!$OMP END PARALLEL
allocate(tmp(n_points_final_grid))
do ipoint = 1, n_points_final_grid
tmp_S(ipoint) = 2.d0 * (tmp_J(ipoint,1)*tmp_J(ipoint,1) + tmp_J(ipoint,2)*tmp_J(ipoint,2) + tmp_J(ipoint,3)*tmp_J(ipoint,3)) - tmp_S(ipoint)
tmp(ipoint) = final_weight_at_r_vector(ipoint) * ( tmp_O(ipoint) * tmp_S(ipoint) &
- 2.d0 * ( tmp_J(ipoint,1) * tmp_M(ipoint,1) &
+ tmp_J(ipoint,2) * tmp_M(ipoint,2) &
+ tmp_J(ipoint,3) * tmp_M(ipoint,3)))
enddo
diag_three_elem_hf = diag_three_elem_hf - 2.d0 * (sum(tmp))
deallocate(tmp)
endif
endif
endif
END_PROVIDER
! ---
BEGIN_PROVIDER [ double precision, fock_3_mat_a_op_sh, (mo_num, mo_num)]
implicit none
integer :: h,p,i,j
double precision :: direct_int, exch_int, exchange_int_231, exchange_int_312
double precision :: exchange_int_23, exchange_int_12, exchange_int_13
fock_3_mat_a_op_sh = 0.d0
do h = 1, mo_num
do p = 1, mo_num
!F_a^{ab}(h,p)
do i = 1, elec_beta_num ! beta
do j = elec_beta_num+1, elec_alpha_num ! alpha
call give_integrals_3_body(h,j,i,p,j,i,direct_int) ! <hji|pji>
call give_integrals_3_body(h,j,i,j,p,i,exch_int)
fock_3_mat_a_op_sh(h,p) -= direct_int - exch_int
enddo
enddo
!F_a^{aa}(h,p)
do i = 1, elec_beta_num ! alpha
do j = elec_beta_num+1, elec_alpha_num ! alpha
call give_integrals_3_body(h,j,i,p,j,i,direct_int)
call give_integrals_3_body(h,j,i,i,p,j,exchange_int_231)
call give_integrals_3_body(h,j,i,j,i,p,exchange_int_312)
call give_integrals_3_body(h,j,i,p,i,j,exchange_int_23)
call give_integrals_3_body(h,j,i,i,j,p,exchange_int_12)
call give_integrals_3_body(h,j,i,j,p,i,exchange_int_13)
fock_3_mat_a_op_sh(h,p) -= ( direct_int + exchange_int_231 + exchange_int_312 &
- exchange_int_23 & ! i <-> j
- exchange_int_12 & ! p <-> j
- exchange_int_13 )! p <-> i
enddo
enddo
enddo
enddo
! symmetrized
! do p = 1, elec_beta_num
! do h = elec_alpha_num +1, mo_num
! fock_3_mat_a_op_sh(h,p) = fock_3_mat_a_op_sh(p,h)
! enddo
! enddo
! do h = elec_beta_num+1, elec_alpha_num
! do p = elec_alpha_num +1, mo_num
! !F_a^{bb}(h,p)
! do i = 1, elec_beta_num
! do j = i+1, elec_beta_num
! call give_integrals_3_body(h,j,i,p,j,i,direct_int)
! call give_integrals_3_body(h,j,i,p,i,j,exch_int)
! fock_3_mat_a_op_sh(h,p) -= direct_int - exch_int
! enddo
! enddo
! enddo
! enddo
END_PROVIDER
BEGIN_PROVIDER [ double precision, fock_3_mat_b_op_sh, (mo_num, mo_num)]
implicit none
integer :: h,p,i,j
double precision :: direct_int, exch_int
fock_3_mat_b_op_sh = 0.d0
do h = 1, elec_beta_num
do p = elec_alpha_num +1, mo_num
!F_b^{aa}(h,p)
do i = 1, elec_beta_num
do j = elec_beta_num+1, elec_alpha_num
call give_integrals_3_body(h,j,i,p,j,i,direct_int)
call give_integrals_3_body(h,j,i,p,i,j,exch_int)
fock_3_mat_b_op_sh(h,p) += direct_int - exch_int
enddo
enddo
!F_b^{ab}(h,p)
do i = elec_beta_num+1, elec_beta_num
do j = 1, elec_beta_num
call give_integrals_3_body(h,j,i,p,j,i,direct_int)
call give_integrals_3_body(h,j,i,j,p,i,exch_int)
fock_3_mat_b_op_sh(h,p) += direct_int - exch_int
enddo
enddo
enddo
enddo
END_PROVIDER
2023-02-07 16:45:10 +01:00
BEGIN_PROVIDER [ double precision, fock_3_w_kk_sum, (n_points_final_grid,3)]
implicit none
integer :: mm, ipoint,k
double precision :: w_kk
fock_3_w_kk_sum = 0.d0
do k = 1, elec_beta_num
do mm = 1, 3
do ipoint = 1, n_points_final_grid
w_kk = x_W_ij_erf_rk(ipoint,mm,k,k)
fock_3_w_kk_sum(ipoint,mm) += w_kk
enddo
enddo
enddo
END_PROVIDER
BEGIN_PROVIDER [ double precision, fock_3_w_ki_mos_k, (n_points_final_grid,3,mo_num)]
implicit none
integer :: mm, ipoint,k,i
double precision :: w_ki, mo_k
fock_3_w_ki_mos_k = 0.d0
do i = 1, mo_num
do k = 1, elec_beta_num
do mm = 1, 3
do ipoint = 1, n_points_final_grid
w_ki = x_W_ij_erf_rk(ipoint,mm,k,i)
mo_k = mos_in_r_array(k,ipoint)
fock_3_w_ki_mos_k(ipoint,mm,i) += w_ki * mo_k
enddo
enddo
enddo
enddo
END_PROVIDER
BEGIN_PROVIDER [ double precision, fock_3_w_kl_w_kl, (n_points_final_grid,3)]
implicit none
integer :: k,j,ipoint,mm
double precision :: w_kj
fock_3_w_kl_w_kl = 0.d0
do j = 1, elec_beta_num
do k = 1, elec_beta_num
do mm = 1, 3
do ipoint = 1, n_points_final_grid
w_kj = x_W_ij_erf_rk(ipoint,mm,k,j)
fock_3_w_kl_w_kl(ipoint,mm) += w_kj * w_kj
enddo
enddo
enddo
enddo
END_PROVIDER
BEGIN_PROVIDER [ double precision, fock_3_rho_beta, (n_points_final_grid)]
implicit none
integer :: ipoint,k
fock_3_rho_beta = 0.d0
do ipoint = 1, n_points_final_grid
do k = 1, elec_beta_num
fock_3_rho_beta(ipoint) += mos_in_r_array(k,ipoint) * mos_in_r_array(k,ipoint)
enddo
enddo
END_PROVIDER
BEGIN_PROVIDER [ double precision, fock_3_w_kl_mo_k_mo_l, (n_points_final_grid,3)]
implicit none
integer :: ipoint,k,l,mm
double precision :: mos_k, mos_l, w_kl
fock_3_w_kl_mo_k_mo_l = 0.d0
do k = 1, elec_beta_num
do l = 1, elec_beta_num
do mm = 1, 3
do ipoint = 1, n_points_final_grid
mos_k = mos_in_r_array_transp(ipoint,k)
mos_l = mos_in_r_array_transp(ipoint,l)
w_kl = x_W_ij_erf_rk(ipoint,mm,l,k)
fock_3_w_kl_mo_k_mo_l(ipoint,mm) += w_kl * mos_k * mos_l
enddo
enddo
enddo
enddo
END_PROVIDER
BEGIN_PROVIDER [ double precision, fock_3_w_ki_wk_a, (n_points_final_grid,3,mo_num, mo_num)]
implicit none
integer :: ipoint,i,a,k,mm
double precision :: w_ki,w_ka
fock_3_w_ki_wk_a = 0.d0
do i = 1, mo_num
do a = 1, mo_num
do mm = 1, 3
do ipoint = 1, n_points_final_grid
do k = 1, elec_beta_num
w_ki = x_W_ij_erf_rk(ipoint,mm,k,i)
w_ka = x_W_ij_erf_rk(ipoint,mm,k,a)
fock_3_w_ki_wk_a(ipoint,mm,a,i) += w_ki * w_ka
enddo
enddo
enddo
enddo
enddo
END_PROVIDER
BEGIN_PROVIDER [ double precision, fock_3_trace_w_tilde, (n_points_final_grid,3)]
implicit none
integer :: ipoint,k,mm
fock_3_trace_w_tilde = 0.d0
do k = 1, elec_beta_num
do mm = 1, 3
do ipoint = 1, n_points_final_grid
fock_3_trace_w_tilde(ipoint,mm) += fock_3_w_ki_wk_a(ipoint,mm,k,k)
enddo
enddo
enddo
END_PROVIDER
BEGIN_PROVIDER [ double precision, fock_3_w_kl_wla_phi_k, (n_points_final_grid,3,mo_num)]
implicit none
integer :: ipoint,a,k,mm,l
double precision :: w_kl,w_la, mo_k
fock_3_w_kl_wla_phi_k = 0.d0
do a = 1, mo_num
do k = 1, elec_beta_num
do l = 1, elec_beta_num
do mm = 1, 3
do ipoint = 1, n_points_final_grid
w_kl = x_W_ij_erf_rk(ipoint,mm,l,k)
w_la = x_W_ij_erf_rk(ipoint,mm,l,a)
mo_k = mos_in_r_array_transp(ipoint,k)
fock_3_w_kl_wla_phi_k(ipoint,mm,a) += w_kl * w_la * mo_k
enddo
enddo
enddo
enddo
enddo
END_PROVIDER
2023-09-18 12:03:29 +02:00