9
1
mirror of https://github.com/QuantumPackage/qp2.git synced 2024-11-19 20:02:20 +01:00
qp2/plugins/local/non_h_ints_mu/tc_integ.irp.f
2024-01-15 23:35:26 +01:00

606 lines
22 KiB
Fortran

BEGIN_PROVIDER [double precision, int2_grad1_u12_ao, (ao_num, ao_num, n_points_final_grid, 3)]
BEGIN_DOC
!
! int2_grad1_u12_ao(i,j,ipoint,:) = \int dr2 [-1 * \grad_r1 J(r1,r2)] \phi_i(r2) \phi_j(r2)
!
! where r1 = r(ipoint)
!
END_DOC
implicit none
integer :: ipoint, i, j, m, jpoint
double precision :: time0, time1
double precision :: x, y, z, r2
double precision :: dx, dy, dz
double precision :: tmp_ct
double precision :: tmp0, tmp1, tmp2
double precision :: tmp0_x, tmp0_y, tmp0_z
double precision :: tmp1_x, tmp1_y, tmp1_z
PROVIDE j2e_type
PROVIDE j1e_type
call wall_time(time0)
print*, ' providing int2_grad1_u12_ao ...'
if(read_tc_integ) then
print*, ' Reading int2_grad1_u12_ao from ', trim(ezfio_filename) // '/work/int2_grad1_u12_ao'
open(unit=11, form="unformatted", file=trim(ezfio_filename)//'/work/int2_grad1_u12_ao', action="read")
read(11) int2_grad1_u12_ao
close(11)
else
if(tc_integ_type .eq. "analytic") then
write(*, '(A, A, A)') ' Error: The integration type ', trim(tc_integ_type), ' has not been implemented yet.'
stop
elseif(tc_integ_type .eq. "numeric") then
print *, ' Numerical integration over r1 and r2 will be performed'
! TODO combine 1shot & int2_grad1_u12_ao_num
PROVIDE int2_grad1_u12_ao_num
int2_grad1_u12_ao = int2_grad1_u12_ao_num
!PROVIDE int2_grad1_u12_ao_num_1shot
!int2_grad1_u12_ao = int2_grad1_u12_ao_num_1shot
elseif(tc_integ_type .eq. "semi-analytic") then
print*, ' Numerical integration over r1, with analytical integration over r2'
! ---
if((j2e_type .eq. "rs-dft") .and. (env_type .eq. "none")) then
PROVIDE v_ij_erf_rk_cst_mu x_v_ij_erf_rk_cst_mu
int2_grad1_u12_ao = 0.d0
!$OMP PARALLEL &
!$OMP DEFAULT (NONE) &
!$OMP PRIVATE (ipoint, i, j, x, y, z, tmp1) &
!$OMP SHARED ( ao_num, n_points_final_grid, final_grid_points &
!$OMP , v_ij_erf_rk_cst_mu, x_v_ij_erf_rk_cst_mu, int2_grad1_u12_ao)
!$OMP DO SCHEDULE (static)
do ipoint = 1, n_points_final_grid
x = final_grid_points(1,ipoint)
y = final_grid_points(2,ipoint)
z = final_grid_points(3,ipoint)
do j = 1, ao_num
do i = 1, ao_num
tmp1 = v_ij_erf_rk_cst_mu(i,j,ipoint)
int2_grad1_u12_ao(i,j,ipoint,1) = 0.5d0 * (tmp1 * x - x_v_ij_erf_rk_cst_mu(i,j,ipoint,1))
int2_grad1_u12_ao(i,j,ipoint,2) = 0.5d0 * (tmp1 * y - x_v_ij_erf_rk_cst_mu(i,j,ipoint,2))
int2_grad1_u12_ao(i,j,ipoint,3) = 0.5d0 * (tmp1 * z - x_v_ij_erf_rk_cst_mu(i,j,ipoint,3))
enddo
enddo
enddo
!$OMP END DO
!$OMP END PARALLEL
elseif((j2e_type .eq. "rs-dft") .and. (env_type .eq. "prod-gauss")) then
PROVIDE env_type env_val env_grad
PROVIDE v_ij_erf_rk_cst_mu_env v_ij_u_cst_mu_env_an x_v_ij_erf_rk_cst_mu_env
int2_grad1_u12_ao = 0.d0
!$OMP PARALLEL &
!$OMP DEFAULT (NONE) &
!$OMP PRIVATE (ipoint, i, j, x, y, z, tmp0, tmp1, tmp2, tmp0_x, tmp0_y, tmp0_z) &
!$OMP SHARED (ao_num, n_points_final_grid, final_grid_points, env_val, env_grad, &
!$OMP v_ij_erf_rk_cst_mu_env, v_ij_u_cst_mu_env_an, x_v_ij_erf_rk_cst_mu_env, int2_grad1_u12_ao)
!$OMP DO SCHEDULE (static)
do ipoint = 1, n_points_final_grid
x = final_grid_points(1,ipoint)
y = final_grid_points(2,ipoint)
z = final_grid_points(3,ipoint)
tmp0 = 0.5d0 * env_val(ipoint)
tmp0_x = env_grad(1,ipoint)
tmp0_y = env_grad(2,ipoint)
tmp0_z = env_grad(3,ipoint)
do j = 1, ao_num
do i = 1, ao_num
tmp1 = tmp0 * v_ij_erf_rk_cst_mu_env(i,j,ipoint)
tmp2 = v_ij_u_cst_mu_env_an(i,j,ipoint)
int2_grad1_u12_ao(i,j,ipoint,1) = tmp1 * x - tmp0 * x_v_ij_erf_rk_cst_mu_env(i,j,ipoint,1) - tmp2 * tmp0_x
int2_grad1_u12_ao(i,j,ipoint,2) = tmp1 * y - tmp0 * x_v_ij_erf_rk_cst_mu_env(i,j,ipoint,2) - tmp2 * tmp0_y
int2_grad1_u12_ao(i,j,ipoint,3) = tmp1 * z - tmp0 * x_v_ij_erf_rk_cst_mu_env(i,j,ipoint,3) - tmp2 * tmp0_z
enddo
enddo
enddo
!$OMP END DO
!$OMP END PARALLEL
elseif((j2e_type .eq. "rs-dft") .and. (env_type .eq. "sum-gauss")) then
PROVIDE mu_erf
PROVIDE env_type env_val env_grad
PROVIDE Ir2_rsdft_long_Du_0 Ir2_rsdft_long_Du_x Ir2_rsdft_long_Du_y Ir2_rsdft_long_Du_z Ir2_rsdft_long_Du_2
PROVIDE Ir2_rsdft_gauss_Du
tmp_ct = 0.5d0 / (dsqrt(dacos(-1.d0)) * mu_erf)
int2_grad1_u12_ao = 0.d0
!$OMP PARALLEL &
!$OMP DEFAULT (NONE) &
!$OMP PRIVATE (ipoint, i, j, x, y, z, r2, dx, dy, dz, tmp1, tmp2, &
!$OMP tmp0_x, tmp0_y, tmp0_z, tmp1_x, tmp1_y, tmp1_z) &
!$OMP SHARED (ao_num, n_points_final_grid, final_grid_points, &
!$OMP tmp_ct, env_val, env_grad, Ir2_rsdft_long_Du_0, &
!$OMP Ir2_rsdft_long_Du_x, Ir2_rsdft_long_Du_y, &
!$OMP Ir2_rsdft_long_Du_z, Ir2_rsdft_gauss_Du, &
!$OMP Ir2_rsdft_long_Du_2, int2_grad1_u12_ao)
!$OMP DO SCHEDULE (static)
do ipoint = 1, n_points_final_grid
x = final_grid_points(1,ipoint)
y = final_grid_points(2,ipoint)
z = final_grid_points(3,ipoint)
r2 = x*x + y*y + z*z
dx = env_grad(1,ipoint)
dy = env_grad(2,ipoint)
dz = env_grad(3,ipoint)
tmp0_x = 0.5d0 * (env_val(ipoint) * x + r2 * dx)
tmp0_y = 0.5d0 * (env_val(ipoint) * y + r2 * dy)
tmp0_z = 0.5d0 * (env_val(ipoint) * z + r2 * dz)
tmp1 = 0.5d0 * env_val(ipoint)
tmp1_x = tmp_ct * dx
tmp1_y = tmp_ct * dy
tmp1_z = tmp_ct * dz
do j = 1, ao_num
do i = 1, ao_num
tmp2 = 0.5d0 * Ir2_rsdft_long_Du_2(i,j,ipoint) - x * Ir2_rsdft_long_Du_x(i,j,ipoint) - y * Ir2_rsdft_long_Du_y(i,j,ipoint) - z * Ir2_rsdft_long_Du_z(i,j,ipoint)
int2_grad1_u12_ao(i,j,ipoint,1) = -Ir2_rsdft_long_Du_0(i,j,ipoint) * tmp0_x + tmp1 * Ir2_rsdft_long_Du_x(i,j,ipoint) - dx * tmp2 + tmp1_x * Ir2_rsdft_gauss_Du(i,j,ipoint)
int2_grad1_u12_ao(i,j,ipoint,2) = -Ir2_rsdft_long_Du_0(i,j,ipoint) * tmp0_y + tmp1 * Ir2_rsdft_long_Du_y(i,j,ipoint) - dy * tmp2 + tmp1_y * Ir2_rsdft_gauss_Du(i,j,ipoint)
int2_grad1_u12_ao(i,j,ipoint,3) = -Ir2_rsdft_long_Du_0(i,j,ipoint) * tmp0_z + tmp1 * Ir2_rsdft_long_Du_z(i,j,ipoint) - dz * tmp2 + tmp1_z * Ir2_rsdft_gauss_Du(i,j,ipoint)
enddo
enddo
enddo
!$OMP END DO
!$OMP END PARALLEL
else
print *, ' Error in int2_grad1_u12_ao: Unknown Jastrow'
stop
endif ! j2e_type
! ---
if(j1e_type .ne. "none") then
PROVIDE elec_num
PROVIDE ao_overlap
PROVIDE j1e_gradx j1e_grady j1e_gradz
tmp_ct = 1.d0 / (dble(elec_num) - 1.d0)
!$OMP PARALLEL &
!$OMP DEFAULT (NONE) &
!$OMP PRIVATE (ipoint, i, j, tmp0_x, tmp0_y, tmp0_z) &
!$OMP SHARED (ao_num, n_points_final_grid, tmp_ct, &
!$OMP j1e_gradx, j1e_grady, j1e_gradz, ao_overlap, int2_grad1_u12_ao)
!$OMP DO SCHEDULE (static)
do ipoint = 1, n_points_final_grid
tmp0_x = tmp_ct * j1e_gradx(ipoint)
tmp0_y = tmp_ct * j1e_grady(ipoint)
tmp0_z = tmp_ct * j1e_gradz(ipoint)
do j = 1, ao_num
do i = 1, ao_num
int2_grad1_u12_ao(i,j,ipoint,1) = int2_grad1_u12_ao(i,j,ipoint,1) + tmp0_x * ao_overlap(i,j)
int2_grad1_u12_ao(i,j,ipoint,2) = int2_grad1_u12_ao(i,j,ipoint,2) + tmp0_y * ao_overlap(i,j)
int2_grad1_u12_ao(i,j,ipoint,3) = int2_grad1_u12_ao(i,j,ipoint,3) + tmp0_z * ao_overlap(i,j)
enddo
enddo
enddo
!$OMP END DO
!$OMP END PARALLEL
else
if((j2e_type .eq. "rs-dft") .and. (env_type .eq. "none")) then
FREE v_ij_erf_rk_cst_mu x_v_ij_erf_rk_cst_mu
elseif((j2e_type .eq. "rs-dft") .and. (env_type .eq. "prod-gauss")) then
FREE v_ij_erf_rk_cst_mu_env v_ij_u_cst_mu_env_an x_v_ij_erf_rk_cst_mu_env
elseif((j2e_type .eq. "rs-dft") .and. (env_type .eq. "sum-gauss")) then
FREE Ir2_rsdft_long_Du_0 Ir2_rsdft_long_Du_x Ir2_rsdft_long_Du_y Ir2_rsdft_long_Du_z Ir2_rsdft_gauss_Du Ir2_rsdft_long_Du_2
endif
endif ! j1e_type
! ---
else
write(*, '(A, A, A)') ' Error: The integration type ', trim(tc_integ_type), ' has not been implemented yet'
stop
endif ! tc_integ_type
endif ! read_tc_integ
if(write_tc_integ .and. mpi_master) then
print*, ' Writing int2_grad1_u12_ao in ', trim(ezfio_filename) // '/work/int2_grad1_u12_ao'
open(unit=11, form="unformatted", file=trim(ezfio_filename)//'/work/int2_grad1_u12_ao', action="write")
call ezfio_set_work_empty(.False.)
write(11) int2_grad1_u12_ao
close(11)
call ezfio_set_tc_keywords_io_tc_integ('Read')
endif
call wall_time(time1)
print*, ' wall time for int2_grad1_u12_ao (min) =', (time1-time0)/60.d0
call print_memory_usage()
END_PROVIDER
! ---
BEGIN_PROVIDER [double precision, int2_grad1_u12_square_ao, (ao_num, ao_num, n_points_final_grid)]
BEGIN_DOC
!
! int2_grad1_u12_square_ao = -(1/2) x int dr2 chi_l(r2) chi_j(r2) [grad_1 u(r1,r2)]^2
!
END_DOC
implicit none
integer :: ipoint, i, j, m, jpoint
double precision :: x, y, z, r2
double precision :: dx, dy, dz, dr2
double precision :: dx1, dy1, dz1, dx2, dy2, dz2, dr12
double precision :: tmp_ct, tmp_ct1, tmp_ct2
double precision :: tmp0, tmp1, tmp2
double precision :: tmp3, tmp4, tmp5, tmp6
double precision :: tmp0_x, tmp0_y, tmp0_z
double precision :: tmp1_x, tmp1_y, tmp1_z
double precision :: time0, time1
PROVIDE j2e_type
PROVIDE j1e_type
PROVIDE tc_integ_type
call wall_time(time0)
print*, ' providing int2_grad1_u12_square_ao ...'
if(tc_integ_type .eq. "analytic") then
write(*, '(A, A, A)') ' Error: The integration type ', trim(tc_integ_type), ' has not been implemented yet.'
stop
elseif(tc_integ_type .eq. "numeric") then
print *, ' Numerical integration over r1 and r2 will be performed'
! TODO combine 1shot & int2_grad1_u12_square_ao_num
PROVIDE int2_grad1_u12_square_ao_num
int2_grad1_u12_square_ao = int2_grad1_u12_square_ao_num
!PROVIDE int2_grad1_u12_square_ao_num_1shot
!int2_grad1_u12_square_ao = int2_grad1_u12_square_ao_num_1shot
elseif(tc_integ_type .eq. "semi-analytic") then
print*, ' Numerical integration over r1, with analytical integration over r2'
! ---
if((j2e_type .eq. "rs-dft") .and. (env_type .eq. "none")) then
PROVIDE int2_grad1u2_grad2u2
int2_grad1_u12_square_ao = 0.d0
!$OMP PARALLEL &
!$OMP DEFAULT (NONE) &
!$OMP PRIVATE (i, j, ipoint) &
!$OMP SHARED (int2_grad1_u12_square_ao, ao_num, n_points_final_grid, int2_grad1u2_grad2u2)
!$OMP DO SCHEDULE (static)
do ipoint = 1, n_points_final_grid
do j = 1, ao_num
do i = 1, ao_num
int2_grad1_u12_square_ao(i,j,ipoint) = int2_grad1u2_grad2u2(i,j,ipoint)
enddo
enddo
enddo
!$OMP END DO
!$OMP END PARALLEL
FREE int2_grad1u2_grad2u2
elseif((j2e_type .eq. "rs-dft") .and. (env_type .eq. "prod-gauss")) then
PROVIDE mu_erf
PROVIDE env_val env_grad
if(use_ipp) then
! the term u12_grad1_u12_env_grad1_env is added directly for performance
PROVIDE u12sq_envsq grad12_j12
int2_grad1_u12_square_ao = 0.d0
!$OMP PARALLEL &
!$OMP DEFAULT (NONE) &
!$OMP PRIVATE (i, j, ipoint) &
!$OMP SHARED (int2_grad1_u12_square_ao, ao_num, n_points_final_grid, u12sq_envsq, grad12_j12)
!$OMP DO SCHEDULE (static)
do ipoint = 1, n_points_final_grid
do j = 1, ao_num
do i = 1, ao_num
int2_grad1_u12_square_ao(i,j,ipoint) = u12sq_envsq(i,j,ipoint) + 0.5d0 * grad12_j12(i,j,ipoint)
enddo
enddo
enddo
!$OMP END DO
!$OMP END PARALLEL
FREE u12sq_envsq grad12_j12
else
PROVIDE u12sq_envsq u12_grad1_u12_env_grad1_env grad12_j12
int2_grad1_u12_square_ao = 0.d0
!$OMP PARALLEL &
!$OMP DEFAULT (NONE) &
!$OMP PRIVATE (i, j, ipoint) &
!$OMP SHARED (int2_grad1_u12_square_ao, ao_num, n_points_final_grid, u12sq_envsq, grad12_j12, u12_grad1_u12_env_grad1_env)
!$OMP DO SCHEDULE (static)
do ipoint = 1, n_points_final_grid
do j = 1, ao_num
do i = 1, ao_num
int2_grad1_u12_square_ao(i,j,ipoint) = u12sq_envsq(i,j,ipoint) + u12_grad1_u12_env_grad1_env(i,j,ipoint) + 0.5d0 * grad12_j12(i,j,ipoint)
enddo
enddo
enddo
!$OMP END DO
!$OMP END PARALLEL
FREE u12sq_envsq u12_grad1_u12_env_grad1_env grad12_j12
endif ! use_ipp
elseif((j2e_type .eq. "rs-dft") .and. (env_type .eq. "sum-gauss")) then
PROVIDE mu_erf
PROVIDE env_type env_val env_grad
if(use_ipp) then
! do not free int2_u2_env2 here
PROVIDE int2_u2_env2
PROVIDE int2_grad1u2_grad2u2_env2
int2_grad1_u12_square_ao = 0.d0
!$OMP PARALLEL &
!$OMP DEFAULT (NONE) &
!$OMP PRIVATE (i, j, ipoint, tmp0_x, tmp0_y, tmp0_z, tmp1, tmp2) &
!$OMP SHARED (int2_grad1_u12_square_ao, ao_num, n_points_final_grid, &
!$OMP env_val, env_grad, int2_u2_env2, int2_grad1u2_grad2u2_env2)
!$OMP DO SCHEDULE (static)
do ipoint = 1, n_points_final_grid
tmp0_x = env_grad(1,ipoint)
tmp0_y = env_grad(2,ipoint)
tmp0_z = env_grad(3,ipoint)
tmp1 = -0.5d0 * (tmp0_x * tmp0_x + tmp0_y * tmp0_y + tmp0_z * tmp0_z)
tmp2 = 0.5d0 * env_val(ipoint) * env_val(ipoint)
do j = 1, ao_num
do i = 1, ao_num
int2_grad1_u12_square_ao(i,j,ipoint) = tmp1 * int2_u2_env2(i,j,ipoint) + tmp2 * int2_grad1u2_grad2u2_env2(i,j,ipoint)
enddo
enddo
enddo
!$OMP END DO
!$OMP END PARALLEL
FREE int2_grad1u2_grad2u2_env2
else
PROVIDE u12sq_envsq u12_grad1_u12_env_grad1_env grad12_j12
int2_grad1_u12_square_ao = 0.d0
!$OMP PARALLEL &
!$OMP DEFAULT (NONE) &
!$OMP PRIVATE (i, j, ipoint) &
!$OMP SHARED (int2_grad1_u12_square_ao, ao_num, n_points_final_grid, u12sq_envsq, grad12_j12, u12_grad1_u12_env_grad1_env)
!$OMP DO SCHEDULE (static)
do ipoint = 1, n_points_final_grid
do j = 1, ao_num
do i = 1, ao_num
int2_grad1_u12_square_ao(i,j,ipoint) = u12sq_envsq(i,j,ipoint) + u12_grad1_u12_env_grad1_env(i,j,ipoint) + 0.5d0 * grad12_j12(i,j,ipoint)
enddo
enddo
enddo
!$OMP END DO
!$OMP END PARALLEL
FREE u12sq_envsq u12_grad1_u12_env_grad1_env grad12_j12
endif ! use_ipp
! elseif((j2e_type .eq. "rs-dft") .and. (env_type .eq. "sum-gauss")) then
!
! PROVIDE mu_erf
! PROVIDE env_val env_grad
! PROVIDE Ir2_rsdft_short_Du2_0 Ir2_rsdft_short_Du2_x Ir2_rsdft_short_Du2_y Ir2_rsdft_short_Du2_z Ir2_rsdft_short_Du2_2
! PROVIDE Ir2_rsdft_long_Du2_0 Ir2_rsdft_long_Du2_x Ir2_rsdft_long_Du2_y Ir2_rsdft_long_Du2_z Ir2_rsdft_long_Du2_2
! PROVIDE Ir2_rsdft_gauss_Du2
!
! tmp_ct = 1.d0 / (dsqrt(dacos(-1.d0)) * mu_erf)
! tmp_ct2 = tmp_ct * tmp_ct
!
! int2_grad1_u12_square_ao = 0.d0
!
! !$OMP PARALLEL &
! !$OMP DEFAULT (NONE) &
! !$OMP PRIVATE (ipoint, i, j, x, y, z, r2, dx, dy, dz, dr2, &
! !$OMP tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, &
! !$OMP tmp0_x, tmp0_y, tmp0_z, tmp1_x, tmp1_y, tmp1_z) &
! !$OMP SHARED (ao_num, n_points_final_grid, final_grid_points, &
! !$OMP tmp_ct, tmp_ct2, env_val, env_grad, &
! !$OMP Ir2_rsdft_long_Du2_0, Ir2_rsdft_long_Du2_x, &
! !$OMP Ir2_rsdft_long_Du2_y, Ir2_rsdft_long_Du2_z, &
! !$OMP Ir2_rsdft_gauss_Du2, Ir2_rsdft_long_Du2_2, &
! !$OMP Ir2_rsdft_short_Du2_0, Ir2_rsdft_short_Du2_x, &
! !$OMP Ir2_rsdft_short_Du2_y, Ir2_rsdft_short_Du2_z, &
! !$OMP Ir2_rsdft_short_Du2_2, int2_grad1_u12_square_ao)
! !$OMP DO SCHEDULE (static)
! do ipoint = 1, n_points_final_grid
!
! x = final_grid_points(1,ipoint)
! y = final_grid_points(2,ipoint)
! z = final_grid_points(3,ipoint)
! r2 = x*x + y*y + z*z
!
! dx = env_grad(1,ipoint)
! dy = env_grad(2,ipoint)
! dz = env_grad(3,ipoint)
! dr2 = dx*dx + dy*dy + dz*dz
!
! tmp0_x = 0.5d0 * (dr2 * x + env_val(ipoint) * dx)
! tmp0_y = 0.5d0 * (dr2 * y + env_val(ipoint) * dy)
! tmp0_z = 0.5d0 * (dr2 * z + env_val(ipoint) * dz)
!
! tmp1 = 0.25d0 * (env_val(ipoint)*env_val(ipoint) + r2*dr2 + 2.d0*env_val(ipoint)*(x*dx+y*dy+z*dz))
! tmp3 = 0.25d0 * dr2
! tmp4 = tmp3 * tmp_ct2
! tmp5 = 0.50d0 * tmp_ct * (r2*dr2 + env_val(ipoint)*(x*dx+y*dy+z*dz))
! tmp6 = 0.50d0 * tmp_ct * dr2
!
! tmp1_x = 0.5d0 * tmp_ct * (2.d0*dr2*x + env_val(ipoint)*dx)
! tmp1_y = 0.5d0 * tmp_ct * (2.d0*dr2*y + env_val(ipoint)*dy)
! tmp1_z = 0.5d0 * tmp_ct * (2.d0*dr2*z + env_val(ipoint)*dz)
!
! do j = 1, ao_num
! do i = 1, ao_num
!
! tmp2 = tmp1_x * Ir2_rsdft_long_Du2_x (i,j,ipoint) + tmp1_y * Ir2_rsdft_long_Du2_y (i,j,ipoint) + tmp1_z * Ir2_rsdft_long_Du2_z (i,j,ipoint) &
! - tmp0_x * Ir2_rsdft_short_Du2_x(i,j,ipoint) - tmp0_y * Ir2_rsdft_short_Du2_y(i,j,ipoint) - tmp0_z * Ir2_rsdft_short_Du2_z(i,j,ipoint)
!
! int2_grad1_u12_square_ao(i,j,ipoint) = tmp1 * Ir2_rsdft_short_Du2_0(i,j,ipoint) + tmp2 + tmp3 * Ir2_rsdft_short_Du2_2(i,j,ipoint) &
! + tmp4 * Ir2_rsdft_gauss_Du2(i,j,ipoint) - tmp5 * Ir2_rsdft_long_Du2_0(i,j,ipoint) &
! - tmp6 * Ir2_rsdft_long_Du2_2(i,j,ipoint)
! enddo
! enddo
! enddo
! !$OMP END DO
! !$OMP END PARALLEL
!
! int2_grad1_u12_square_ao = -0.5d0 * int2_grad1_u12_square_ao
else
print *, ' Error in int2_grad1_u12_square_ao: Unknown Jhastrow'
stop
endif ! j2e_type
! ---
if(j1e_type .ne. "none") then
PROVIDE elec_num
PROVIDE ao_overlap
PROVIDE j1e_gradx j1e_grady j1e_gradz
tmp_ct1 = 1.0d0 / (dsqrt(dacos(-1.d0)) * mu_erf)
tmp_ct2 = 1.0d0 / (dble(elec_num) - 1.d0)
!$OMP PARALLEL &
!$OMP DEFAULT (NONE) &
!$OMP PRIVATE (ipoint, i, j, x, y, z, r2, dx1, dy1, dz1, &
!$OMP dx2, dy2, dz2, dr12, tmp0, tmp1, tmp2, tmp3, tmp4, &
!$OMP tmp0_x, tmp0_y, tmp0_z) &
!$OMP SHARED (ao_num, n_points_final_grid, final_grid_points, &
!$OMP tmp_ct1, tmp_ct2, env_val, env_grad, &
!$OMP j1e_gradx, j1e_grady, j1e_gradz, &
!$OMP Ir2_rsdft_long_Du_0, Ir2_rsdft_long_Du_2, &
!$OMP Ir2_rsdft_long_Du_x, Ir2_rsdft_long_Du_y, &
!$OMP Ir2_rsdft_long_Du_z, Ir2_rsdft_gauss_Du, &
!$OMP ao_overlap, int2_grad1_u12_square_ao)
!$OMP DO SCHEDULE (static)
do ipoint = 1, n_points_final_grid
x = final_grid_points(1,ipoint)
y = final_grid_points(2,ipoint)
z = final_grid_points(3,ipoint)
r2 = x*x + y*y + z*z
dx1 = env_grad(1,ipoint)
dy1 = env_grad(2,ipoint)
dz1 = env_grad(3,ipoint)
dx2 = j1e_gradx(ipoint)
dy2 = j1e_grady(ipoint)
dz2 = j1e_gradz(ipoint)
dr12 = dx1*dx2 + dy1*dy2 + dz1*dz2
tmp0 = tmp_ct2 * (env_val(ipoint) * (dx2*x + dy2*y + dz2*z) + r2*dr12)
tmp1 = tmp_ct2 * dr12
tmp2 = tmp_ct1 * tmp_ct2 * dr12
tmp3 = tmp_ct2 * tmp_ct2 * (dx2*dx2 + dy2*dy2 + dz2*dz2)
tmp0_x = tmp_ct2 * (env_val(ipoint) * dx2 + 2.d0 * dr12 * x)
tmp0_y = tmp_ct2 * (env_val(ipoint) * dy2 + 2.d0 * dr12 * y)
tmp0_z = tmp_ct2 * (env_val(ipoint) * dz2 + 2.d0 * dr12 * z)
do j = 1, ao_num
do i = 1, ao_num
tmp4 = tmp0_x * Ir2_rsdft_long_Du_x(i,j,ipoint) + tmp0_y * Ir2_rsdft_long_Du_y(i,j,ipoint) + tmp0_z * Ir2_rsdft_long_Du_z(i,j,ipoint)
int2_grad1_u12_square_ao(i,j,ipoint) = int2_grad1_u12_square_ao(i,j,ipoint) &
+ tmp0 * Ir2_rsdft_long_Du_0(i,j,ipoint) - tmp4 + tmp1 * Ir2_rsdft_long_Du_2(i,j,ipoint) &
- tmp2 * Ir2_rsdft_gauss_Du(i,j,ipoint) &
+ tmp3 * ao_overlap(i,j)
enddo
enddo
enddo
!$OMP END DO
!$OMP END PARALLEL
FREE Ir2_rsdft_long_Du_0 Ir2_rsdft_long_Du_x Ir2_rsdft_long_Du_y Ir2_rsdft_long_Du_z Ir2_rsdft_gauss_Du Ir2_rsdft_long_Du_2
endif ! j1e_type
! ---
else
write(*, '(A, A, A)') ' Error: The integration type ', trim(tc_integ_type), ' has not been implemented yet'
stop
endif ! tc_integ_type
call wall_time(time1)
print*, ' wall time for int2_grad1_u12_square_ao (min) = ', (time1-time0) / 60.d0
call print_memory_usage()
END_PROVIDER
! ---