10
0
mirror of https://github.com/QuantumPackage/qp2.git synced 2024-11-05 21:53:57 +01:00
QuantumPackage/plugins/local/tc_scf/fock_three_bi_ortho.irp.f

470 lines
13 KiB
Fortran

! ---
BEGIN_PROVIDER [double precision, fock_a_tot_3e_bi_orth, (mo_num, mo_num)]
BEGIN_DOC
!
! Alpha part of the Fock matrix from three-electron terms
!
! WARNING :: non hermitian if bi-ortho MOS used
!
! This calculation becomes the dominant part one the integrals are provided
!
END_DOC
implicit none
integer :: i, a
double precision :: t0, t1
!print*, ' Providing fock_a_tot_3e_bi_orth ...'
!call wall_time(t0)
PROVIDE mo_l_coef mo_r_coef
PROVIDE fock_cs_3e_bi_orth fock_a_tmp1_bi_ortho fock_a_tmp2_bi_ortho
fock_a_tot_3e_bi_orth = 0.d0
do i = 1, mo_num
do a = 1, mo_num
fock_a_tot_3e_bi_orth(a,i) += fock_cs_3e_bi_orth (a,i)
fock_a_tot_3e_bi_orth(a,i) += fock_a_tmp1_bi_ortho(a,i)
fock_a_tot_3e_bi_orth(a,i) += fock_a_tmp2_bi_ortho(a,i)
enddo
enddo
!call wall_time(t1)
!print*, ' Wall time for fock_a_tot_3e_bi_orth =', t1 - t0
END_PROVIDER
! ---
BEGIN_PROVIDER [double precision, fock_b_tot_3e_bi_orth, (mo_num, mo_num)]
BEGIN_DOC
!
! Beta part of the Fock matrix from three-electron terms
!
! WARNING :: non hermitian if bi-ortho MOS used
!
! This calculation becomes the dominant part one the integrals are provided
!
END_DOC
implicit none
integer :: i, a
PROVIDE mo_l_coef mo_r_coef
fock_b_tot_3e_bi_orth = 0.d0
do i = 1, mo_num
do a = 1, mo_num
fock_b_tot_3e_bi_orth(a,i) += fock_cs_3e_bi_orth (a,i)
fock_b_tot_3e_bi_orth(a,i) += fock_b_tmp2_bi_ortho(a,i)
fock_b_tot_3e_bi_orth(a,i) += fock_b_tmp1_bi_ortho(a,i)
enddo
enddo
END_PROVIDER
! ---
BEGIN_PROVIDER [double precision, fock_cs_3e_bi_orth, (mo_num, mo_num)]
implicit none
integer :: i, a, j, k
double precision :: contrib_sss, contrib_sos, contrib_soo, contrib
double precision :: direct_int, exch_13_int, exch_23_int, exch_12_int, c_3_int, c_minus_3_int
double precision :: t0, t1
double precision, allocatable :: tmp(:,:)
!print*, ' Providing fock_cs_3e_bi_orth ...'
!call wall_time(t0)
PROVIDE mo_l_coef mo_r_coef
! to PROVIDE stuffs
call give_integrals_3_body_bi_ort(1, 1, 1, 1, 1, 1, contrib)
fock_cs_3e_bi_orth = 0.d0
!$OMP PARALLEL DEFAULT (NONE) &
!$OMP PRIVATE (i, a, j, k, direct_int, c_3_int, c_minus_3_int, exch_13_int, exch_23_int, exch_12_int, tmp) &
!$OMP SHARED (mo_num, elec_beta_num, fock_cs_3e_bi_orth)
allocate(tmp(mo_num,mo_num))
tmp = 0.d0
!$OMP DO
do i = 1, mo_num
do a = 1, mo_num
do j = 1, elec_beta_num
do k = 1, elec_beta_num
!!call contrib_3e_sss(a,i,j,k,contrib_sss)
!!call contrib_3e_soo(a,i,j,k,contrib_soo)
!!call contrib_3e_sos(a,i,j,k,contrib_sos)
!!contrib = 0.5d0 * (contrib_sss + contrib_soo) + contrib_sos
call give_integrals_3_body_bi_ort(a, k, j, i, k, j, direct_int )!!! < a k j | i k j >
call give_integrals_3_body_bi_ort(a, k, j, j, i, k, c_3_int) ! < a k j | j i k >
call give_integrals_3_body_bi_ort(a, k, j, k, j, i, c_minus_3_int)! < a k j | k j i >
! negative terms :: exchange contrib
call give_integrals_3_body_bi_ort(a, k, j, j, k, i, exch_13_int)!!! < a k j | j k i > : E_13
call give_integrals_3_body_bi_ort(a, k, j, i, j, k, exch_23_int)!!! < a k j | i j k > : E_23
call give_integrals_3_body_bi_ort(a, k, j, k, i, j, exch_12_int)!!! < a k j | k i j > : E_12
tmp(a,i) += 2.d0 * direct_int + 0.5d0 * (c_3_int + c_minus_3_int - exch_12_int) -1.5d0 * exch_13_int - exch_23_int
enddo
enddo
enddo
enddo
!$OMP END DO NOWAIT
!$OMP CRITICAL
do i = 1, mo_num
do a = 1, mo_num
fock_cs_3e_bi_orth(a,i) += tmp(a,i)
enddo
enddo
!$OMP END CRITICAL
deallocate(tmp)
!$OMP END PARALLEL
fock_cs_3e_bi_orth = - fock_cs_3e_bi_orth
!call wall_time(t1)
!print*, ' Wall time for fock_cs_3e_bi_orth =', t1-t0
END_PROVIDER
! ---
BEGIN_PROVIDER [double precision, fock_a_tmp1_bi_ortho, (mo_num, mo_num)]
implicit none
integer :: i, a, j, k, ee
double precision :: contrib_sss, contrib_sos, contrib_soo, contrib
double precision :: direct_int, exch_13_int, exch_23_int, exch_12_int, c_3_int, c_minus_3_int
double precision :: t0, t1
double precision, allocatable :: tmp(:,:)
!print*, ' Providing fock_a_tmp1_bi_ortho ...'
!call wall_time(t0)
PROVIDE mo_l_coef mo_r_coef
! to PROVIDE stuffs
call give_integrals_3_body_bi_ort(1, 1, 1, 1, 1, 1, contrib)
ee = elec_beta_num + 1
fock_a_tmp1_bi_ortho = 0.d0
!$OMP PARALLEL DEFAULT (NONE) &
!$OMP PRIVATE (i, a, j, k, direct_int, c_3_int, c_minus_3_int, exch_13_int, exch_23_int, exch_12_int, tmp) &
!$OMP SHARED (mo_num, elec_alpha_num, elec_beta_num, ee, fock_a_tmp1_bi_ortho)
allocate(tmp(mo_num,mo_num))
tmp = 0.d0
!$OMP DO
do i = 1, mo_num
do a = 1, mo_num
do j = ee, elec_alpha_num
do k = 1, elec_beta_num
call give_integrals_3_body_bi_ort(a, k, j, i, k, j, direct_int )!!! < a k j | i k j >
call give_integrals_3_body_bi_ort(a, k, j, j, i, k, c_3_int) ! < a k j | j i k >
call give_integrals_3_body_bi_ort(a, k, j, k, j, i, c_minus_3_int)! < a k j | k j i >
call give_integrals_3_body_bi_ort(a, k, j, j, k, i, exch_13_int)!!! < a k j | j k i > : E_13
call give_integrals_3_body_bi_ort(a, k, j, i, j, k, exch_23_int)!!! < a k j | i j k > : E_23
call give_integrals_3_body_bi_ort(a, k, j, k, i, j, exch_12_int)!!! < a k j | k i j > : E_12
tmp(a,i) += 1.5d0 * (direct_int - exch_13_int) + 0.5d0 * (c_3_int + c_minus_3_int - exch_23_int - exch_12_int)
enddo
enddo
enddo
enddo
!$OMP END DO NOWAIT
!$OMP CRITICAL
do i = 1, mo_num
do a = 1, mo_num
fock_a_tmp1_bi_ortho(a,i) += tmp(a,i)
enddo
enddo
!$OMP END CRITICAL
deallocate(tmp)
!$OMP END PARALLEL
fock_a_tmp1_bi_ortho = - fock_a_tmp1_bi_ortho
!call wall_time(t1)
!print*, ' Wall time for fock_a_tmp1_bi_ortho =', t1-t0
END_PROVIDER
! ---
BEGIN_PROVIDER [double precision, fock_a_tmp2_bi_ortho, (mo_num, mo_num)]
implicit none
integer :: i, a, j, k, ee
double precision :: contrib_sss
double precision :: t0, t1
double precision, allocatable :: tmp(:,:)
!print*, ' Providing fock_a_tmp2_bi_ortho ...'
!call wall_time(t0)
PROVIDE mo_l_coef mo_r_coef
! to PROVIDE stuffs
call contrib_3e_sss(1, 1, 1, 1, contrib_sss)
ee = elec_beta_num + 1
fock_a_tmp2_bi_ortho = 0.d0
!$OMP PARALLEL DEFAULT (NONE) &
!$OMP PRIVATE (i, a, j, k, contrib_sss, tmp) &
!$OMP SHARED (mo_num, elec_alpha_num, ee, fock_a_tmp2_bi_ortho)
allocate(tmp(mo_num,mo_num))
tmp = 0.d0
!$OMP DO
do i = 1, mo_num
do a = 1, mo_num
do j = 1, elec_alpha_num
do k = ee, elec_alpha_num
call contrib_3e_sss(a, i, j, k, contrib_sss)
tmp(a,i) += 0.5d0 * contrib_sss
enddo
enddo
enddo
enddo
!$OMP END DO NOWAIT
!$OMP CRITICAL
do i = 1, mo_num
do a = 1, mo_num
fock_a_tmp2_bi_ortho(a,i) += tmp(a,i)
enddo
enddo
!$OMP END CRITICAL
deallocate(tmp)
!$OMP END PARALLEL
!call wall_time(t1)
!print*, ' Wall time for fock_a_tmp2_bi_ortho =', t1-t0
END_PROVIDER
! ---
BEGIN_PROVIDER [double precision, fock_b_tmp1_bi_ortho, (mo_num, mo_num)]
implicit none
integer :: i, a, j, k, ee
double precision :: direct_int, exch_13_int, exch_23_int, exch_12_int
double precision :: t0, t1
double precision, allocatable :: tmp(:,:)
!print*, ' Providing fock_b_tmp1_bi_ortho ...'
!call wall_time(t0)
PROVIDE mo_l_coef mo_r_coef
! to PROVIDE stuffs
call give_integrals_3_body_bi_ort(1, 1, 1, 1, 1, 1, direct_int)
ee = elec_beta_num + 1
fock_b_tmp1_bi_ortho = 0.d0
!$OMP PARALLEL DEFAULT (NONE) &
!$OMP PRIVATE (i, a, j, k, direct_int, exch_13_int, exch_23_int, tmp) &
!$OMP SHARED (mo_num, elec_beta_num, elec_alpha_num, ee, fock_b_tmp1_bi_ortho)
allocate(tmp(mo_num,mo_num))
tmp = 0.d0
!$OMP DO
do i = 1, mo_num
do a = 1, mo_num
do j = 1, elec_beta_num
do k = ee, elec_alpha_num
call give_integrals_3_body_bi_ort(a, k, j, i, k, j, direct_int )!!! < a k j | i k j >
call give_integrals_3_body_bi_ort(a, k, j, j, k, i, exch_13_int)!!! < a k j | j k i > : E_13
call give_integrals_3_body_bi_ort(a, k, j, i, j, k, exch_23_int)!!! < a k j | i j k > : E_23
tmp(a,i) += 1.5d0 * direct_int - 0.5d0 * exch_23_int - exch_13_int
enddo
enddo
enddo
enddo
!$OMP END DO NOWAIT
!$OMP CRITICAL
do i = 1, mo_num
do a = 1, mo_num
fock_b_tmp1_bi_ortho(a,i) += tmp(a,i)
enddo
enddo
!$OMP END CRITICAL
deallocate(tmp)
!$OMP END PARALLEL
fock_b_tmp1_bi_ortho = - fock_b_tmp1_bi_ortho
!call wall_time(t1)
!print*, ' Wall time for fock_b_tmp1_bi_ortho =', t1-t0
END_PROVIDER
! ---
BEGIN_PROVIDER [double precision, fock_b_tmp2_bi_ortho, (mo_num, mo_num)]
implicit none
integer :: i, a, j, k, ee
double precision :: contrib_soo
double precision :: t0, t1
double precision, allocatable :: tmp(:,:)
!print*, ' Providing fock_b_tmp2_bi_ortho ...'
!call wall_time(t0)
PROVIDE mo_l_coef mo_r_coef
! to PROVIDE stuffs
call contrib_3e_soo(1, 1, 1, 1, contrib_soo)
ee = elec_beta_num + 1
fock_b_tmp2_bi_ortho = 0.d0
!$OMP PARALLEL DEFAULT (NONE) &
!$OMP PRIVATE (i, a, j, k, contrib_soo, tmp) &
!$OMP SHARED (mo_num, elec_alpha_num, ee, fock_b_tmp2_bi_ortho)
allocate(tmp(mo_num,mo_num))
tmp = 0.d0
!$OMP DO
do i = 1, mo_num
do a = 1, mo_num
do j = ee, elec_alpha_num
do k = 1, elec_alpha_num
call contrib_3e_soo(a, i, j, k, contrib_soo)
tmp(a,i) += 0.5d0 * contrib_soo
enddo
enddo
enddo
enddo
!$OMP END DO NOWAIT
!$OMP CRITICAL
do i = 1, mo_num
do a = 1, mo_num
fock_b_tmp2_bi_ortho(a,i) += tmp(a,i)
enddo
enddo
!$OMP END CRITICAL
deallocate(tmp)
!$OMP END PARALLEL
!call wall_time(t1)
!print*, ' Wall time for fock_b_tmp2_bi_ortho =', t1-t0
END_PROVIDER
! ---
subroutine contrib_3e_sss(a, i, j, k, integral)
BEGIN_DOC
! returns the pure same spin contribution to F(a,i) from two orbitals j,k
END_DOC
implicit none
integer, intent(in) :: a, i, j, k
double precision, intent(out) :: integral
double precision :: direct_int, exch_13_int, exch_23_int, exch_12_int, c_3_int, c_minus_3_int
PROVIDE mo_l_coef mo_r_coef
call give_integrals_3_body_bi_ort(a, k, j, i, k, j, direct_int )!!! < a k j | i k j >
call give_integrals_3_body_bi_ort(a, k, j, j, i, k, c_3_int) ! < a k j | j i k >
call give_integrals_3_body_bi_ort(a, k, j, k, j, i, c_minus_3_int)! < a k j | k j i >
integral = direct_int + c_3_int + c_minus_3_int
! negative terms :: exchange contrib
call give_integrals_3_body_bi_ort(a, k, j, j, k, i, exch_13_int)!!! < a k j | j k i > : E_13
call give_integrals_3_body_bi_ort(a, k, j, i, j, k, exch_23_int)!!! < a k j | i j k > : E_23
call give_integrals_3_body_bi_ort(a, k, j, k, i, j, exch_12_int)!!! < a k j | k i j > : E_12
integral += - exch_13_int - exch_23_int - exch_12_int
integral = -integral
end
! ---
subroutine contrib_3e_soo(a,i,j,k,integral)
BEGIN_DOC
! returns the same spin / opposite spin / opposite spin contribution to F(a,i) from two orbitals j,k
END_DOC
implicit none
integer, intent(in) :: a, i, j, k
double precision, intent(out) :: integral
double precision :: direct_int, exch_23_int
PROVIDE mo_l_coef mo_r_coef
call give_integrals_3_body_bi_ort(a, k, j, i, k, j, direct_int) ! < a k j | i k j >
call give_integrals_3_body_bi_ort(a, k, j, i, j, k, exch_23_int)! < a k j | i j k > : E_23
integral = direct_int - exch_23_int
integral = -integral
end
! ---
subroutine contrib_3e_sos(a, i, j, k, integral)
BEGIN_DOC
! returns the same spin / opposite spin / same spin contribution to F(a,i) from two orbitals j,k
END_DOC
PROVIDE mo_l_coef mo_r_coef
implicit none
integer, intent(in) :: a, i, j, k
double precision, intent(out) :: integral
double precision :: direct_int, exch_13_int
call give_integrals_3_body_bi_ort(a, k, j, i, k, j, direct_int )! < a k j | i k j >
call give_integrals_3_body_bi_ort(a, k, j, j, k, i, exch_13_int)! < a k j | j k i > : E_13
integral = direct_int - exch_13_int
integral = -integral
end
! ---