mirror of
https://github.com/QuantumPackage/qp2.git
synced 2024-12-21 11:03:29 +01:00
220 lines
6.4 KiB
Fortran
220 lines
6.4 KiB
Fortran
|
|
! ---
|
|
|
|
BEGIN_PROVIDER [double precision, ao_non_hermit_term_chemist, (ao_num, ao_num, ao_num, ao_num)]
|
|
|
|
BEGIN_DOC
|
|
! 1 1 2 2 1 2 1 2
|
|
!
|
|
! ao_non_hermit_term_chemist(k,i,l,j) = < k l | [erf( mu r12) - 1] d/d_r12 | i j > on the AO basis
|
|
!
|
|
END_DOC
|
|
|
|
implicit none
|
|
integer :: i, j, k, l, ipoint, m
|
|
double precision :: weight1, r(3)
|
|
double precision :: wall1, wall0
|
|
double precision, allocatable :: b_mat(:,:,:,:), ac_mat(:,:,:,:)
|
|
|
|
print*, ' providing ao_non_hermit_term_chemist ...'
|
|
call wall_time(wall0)
|
|
|
|
provide v_ij_erf_rk_cst_mu x_v_ij_erf_rk_cst_mu
|
|
|
|
allocate(b_mat(n_points_final_grid,ao_num,ao_num,3), ac_mat(ao_num,ao_num,ao_num,ao_num))
|
|
|
|
!$OMP PARALLEL &
|
|
!$OMP DEFAULT (NONE) &
|
|
!$OMP PRIVATE (i,k,m,ipoint,r,weight1) &
|
|
!$OMP SHARED (aos_in_r_array_transp,aos_grad_in_r_array_transp_bis,b_mat)&
|
|
!$OMP SHARED (ao_num,n_points_final_grid,final_grid_points,final_weight_at_r_vector)
|
|
!$OMP DO SCHEDULE (static)
|
|
do m = 1, 3
|
|
do i = 1, ao_num
|
|
do k = 1, ao_num
|
|
do ipoint = 1, n_points_final_grid
|
|
r(1) = final_grid_points(1,ipoint)
|
|
r(2) = final_grid_points(2,ipoint)
|
|
r(3) = final_grid_points(3,ipoint)
|
|
weight1 = final_weight_at_r_vector(ipoint)
|
|
b_mat(ipoint,k,i,m) = 0.5d0 * aos_in_r_array_transp(ipoint,k) * r(m) * weight1 * aos_grad_in_r_array_transp_bis(ipoint,i,m)
|
|
enddo
|
|
enddo
|
|
enddo
|
|
enddo
|
|
!$OMP END DO
|
|
!$OMP END PARALLEL
|
|
|
|
! (A) b_mat(ipoint,k,i,m) X v_ij_erf_rk_cst_mu(j,l,r1)
|
|
! 1/2 \int dr1 x1 phi_k(1) d/dx1 phi_i(1) \int dr2 (1 - erf(mu_r12))/r12 phi_j(2) phi_l(2)
|
|
ac_mat = 0.d0
|
|
do m = 1, 3
|
|
! A B^T dim(A,1) dim(B,2) dim(A,2) alpha * A LDA
|
|
|
|
call dgemm( "N", "N", ao_num*ao_num, ao_num*ao_num, n_points_final_grid, 1.d0 &
|
|
, v_ij_erf_rk_cst_mu(1,1,1), ao_num*ao_num, b_mat(1,1,1,m), n_points_final_grid &
|
|
, 1.d0, ac_mat, ao_num*ao_num)
|
|
|
|
enddo
|
|
|
|
!$OMP PARALLEL &
|
|
!$OMP DEFAULT (NONE) &
|
|
!$OMP PRIVATE (i,k,m,ipoint,weight1) &
|
|
!$OMP SHARED (aos_in_r_array_transp,aos_grad_in_r_array_transp_bis,b_mat,ao_num,n_points_final_grid,final_weight_at_r_vector)
|
|
!$OMP DO SCHEDULE (static)
|
|
do m = 1, 3
|
|
do i = 1, ao_num
|
|
do k = 1, ao_num
|
|
do ipoint = 1, n_points_final_grid
|
|
weight1 = final_weight_at_r_vector(ipoint)
|
|
b_mat(ipoint,k,i,m) = 0.5d0 * aos_in_r_array_transp(ipoint,k) * weight1 * aos_grad_in_r_array_transp_bis(ipoint,i,m)
|
|
enddo
|
|
enddo
|
|
enddo
|
|
enddo
|
|
!$OMP END DO
|
|
!$OMP END PARALLEL
|
|
|
|
! (B) b_mat(ipoint,k,i,m) X x_v_ij_erf_rk_cst_mu(j,l,r1,m)
|
|
! 1/2 \int dr1 phi_k(1) d/dx1 phi_i(1) \int dr2 x2(1 - erf(mu_r12))/r12 phi_j(2) phi_l(2)
|
|
do m = 1, 3
|
|
! A B^T dim(A,1) dim(B,2) dim(A,2) alpha * A LDA
|
|
|
|
call dgemm( "N", "N", ao_num*ao_num, ao_num*ao_num, n_points_final_grid, -1.d0 &
|
|
, x_v_ij_erf_rk_cst_mu(1,1,1,m), ao_num*ao_num, b_mat(1,1,1,m), n_points_final_grid &
|
|
, 1.d0, ac_mat, ao_num*ao_num)
|
|
enddo
|
|
|
|
!$OMP PARALLEL &
|
|
!$OMP DEFAULT (NONE) &
|
|
!$OMP PRIVATE (i,k,j,l) &
|
|
!$OMP SHARED (ac_mat,ao_non_hermit_term_chemist,ao_num)
|
|
!$OMP DO SCHEDULE (static)
|
|
do j = 1, ao_num
|
|
do l = 1, ao_num
|
|
do i = 1, ao_num
|
|
do k = 1, ao_num
|
|
! (ki|lj) (ki|lj) (lj|ki)
|
|
ao_non_hermit_term_chemist(k,i,l,j) = ac_mat(k,i,l,j) + ac_mat(l,j,k,i)
|
|
enddo
|
|
enddo
|
|
enddo
|
|
enddo
|
|
!$OMP END DO
|
|
!$OMP END PARALLEL
|
|
|
|
call wall_time(wall1)
|
|
print *, ' wall time for ao_non_hermit_term_chemist ', wall1 - wall0
|
|
|
|
END_PROVIDER
|
|
|
|
! ---
|
|
|
|
! TODO :: optimization :: transform into DGEM
|
|
|
|
BEGIN_PROVIDER [double precision, mo_non_hermit_term_chemist, (mo_num, mo_num, mo_num, mo_num)]
|
|
|
|
BEGIN_DOC
|
|
! 1 1 2 2 1 2 1 2
|
|
!
|
|
! mo_non_hermit_term_chemist(k,i,l,j) = < k l | [erf( mu r12) - 1] d/d_r12 | i j > on the MO basis
|
|
END_DOC
|
|
|
|
implicit none
|
|
integer :: i, j, k, l, m, n, p, q
|
|
double precision, allocatable :: mo_tmp_1(:,:,:,:), mo_tmp_2(:,:,:,:)
|
|
|
|
allocate(mo_tmp_1(mo_num,ao_num,ao_num,ao_num))
|
|
mo_tmp_1 = 0.d0
|
|
|
|
do m = 1, ao_num
|
|
do p = 1, ao_num
|
|
do n = 1, ao_num
|
|
do q = 1, ao_num
|
|
do k = 1, mo_num
|
|
! (k n|p m) = sum_q c_qk * (q n|p m)
|
|
mo_tmp_1(k,n,p,m) += mo_coef_transp(k,q) * ao_non_hermit_term_chemist(q,n,p,m)
|
|
enddo
|
|
enddo
|
|
enddo
|
|
enddo
|
|
enddo
|
|
free ao_non_hermit_term_chemist
|
|
|
|
allocate(mo_tmp_2(mo_num,mo_num,ao_num,ao_num))
|
|
mo_tmp_2 = 0.d0
|
|
|
|
do m = 1, ao_num
|
|
do p = 1, ao_num
|
|
do n = 1, ao_num
|
|
do i = 1, mo_num
|
|
do k = 1, mo_num
|
|
! (k i|p m) = sum_n c_ni * (k n|p m)
|
|
mo_tmp_2(k,i,p,m) += mo_coef_transp(i,n) * mo_tmp_1(k,n,p,m)
|
|
enddo
|
|
enddo
|
|
enddo
|
|
enddo
|
|
enddo
|
|
deallocate(mo_tmp_1)
|
|
|
|
allocate(mo_tmp_1(mo_num,mo_num,mo_num,ao_num))
|
|
mo_tmp_1 = 0.d0
|
|
|
|
do m = 1, ao_num
|
|
do p = 1, ao_num
|
|
do l = 1, mo_num
|
|
do i = 1, mo_num
|
|
do k = 1, mo_num
|
|
mo_tmp_1(k,i,l,m) += mo_coef_transp(l,p) * mo_tmp_2(k,i,p,m)
|
|
enddo
|
|
enddo
|
|
enddo
|
|
enddo
|
|
enddo
|
|
deallocate(mo_tmp_2)
|
|
|
|
mo_non_hermit_term_chemist = 0.d0
|
|
do m = 1, ao_num
|
|
do j = 1, mo_num
|
|
do l = 1, mo_num
|
|
do i = 1, mo_num
|
|
do k = 1, mo_num
|
|
mo_non_hermit_term_chemist(k,i,l,j) += mo_coef_transp(j,m) * mo_tmp_1(k,i,l,m)
|
|
enddo
|
|
enddo
|
|
enddo
|
|
enddo
|
|
enddo
|
|
deallocate(mo_tmp_1)
|
|
|
|
END_PROVIDER
|
|
|
|
! ---
|
|
|
|
BEGIN_PROVIDER [double precision, mo_non_hermit_term, (mo_num, mo_num, mo_num, mo_num)]
|
|
|
|
BEGIN_DOC
|
|
! 1 2 1 2 1 2 1 2
|
|
!
|
|
! mo_non_hermit_term(k,l,i,j) = < k l | [erf( mu r12) - 1] d/d_r12 | i j > on the MO basis
|
|
END_DOC
|
|
|
|
implicit none
|
|
integer :: i, j, k, l
|
|
|
|
do j = 1, mo_num
|
|
do i = 1, mo_num
|
|
do l = 1, mo_num
|
|
do k = 1, mo_num
|
|
mo_non_hermit_term(k,l,i,j) = mo_non_hermit_term_chemist(k,i,l,j)
|
|
enddo
|
|
enddo
|
|
enddo
|
|
enddo
|
|
|
|
END_PROVIDER
|
|
|
|
! ---
|
|
|