9
1
mirror of https://github.com/QuantumPackage/qp2.git synced 2024-10-04 15:15:58 +02:00
qp2/src/ao_one_e_ints/aos_cosgtos.irp.f
Anthony Scemama 77186e0560
Some checks failed
continuous-integration/drone/push Build is failing
Cleaned cosgtos
2023-06-01 11:11:29 +02:00

211 lines
6.3 KiB
Fortran

! ---
BEGIN_PROVIDER [ double precision, ao_coef_norm_ord_transp_cosgtos, (ao_prim_num_max, ao_num) ]
implicit none
integer :: i, j
do j = 1, ao_num
do i = 1, ao_prim_num_max
ao_coef_norm_ord_transp_cosgtos(i,j) = ao_coef_norm_ord_cosgtos(j,i)
enddo
enddo
END_PROVIDER
! ---
BEGIN_PROVIDER [ complex*16, ao_expo_ord_transp_cosgtos, (ao_prim_num_max, ao_num) ]
implicit none
integer :: i, j
do j = 1, ao_num
do i = 1, ao_prim_num_max
ao_expo_ord_transp_cosgtos(i,j) = ao_expo_ord_cosgtos(j,i)
enddo
enddo
END_PROVIDER
! ---
BEGIN_PROVIDER [ double precision, ao_coef_norm_cosgtos, (ao_num, ao_prim_num_max) ]
implicit none
integer :: i, j, powA(3), nz
double precision :: norm
complex*16 :: overlap_x, overlap_y, overlap_z, C_A(3)
complex*16 :: integ1, integ2, expo
nz = 100
C_A(1) = (0.d0, 0.d0)
C_A(2) = (0.d0, 0.d0)
C_A(3) = (0.d0, 0.d0)
ao_coef_norm_cosgtos = 0.d0
do i = 1, ao_num
powA(1) = ao_power(i,1)
powA(2) = ao_power(i,2)
powA(3) = ao_power(i,3)
! Normalization of the primitives
if(primitives_normalized) then
do j = 1, ao_prim_num(i)
expo = ao_expo(i,j) + (0.d0, 1.d0) * ao_expoim_cosgtos(i,j)
call overlap_cgaussian_xyz(C_A, C_A, expo, expo, powA, powA, overlap_x, overlap_y, overlap_z, integ1, nz)
call overlap_cgaussian_xyz(C_A, C_A, conjg(expo), expo, powA, powA, overlap_x, overlap_y, overlap_z, integ2, nz)
norm = 2.d0 * real( integ1 + integ2 )
ao_coef_norm_cosgtos(i,j) = ao_coef(i,j) / dsqrt(norm)
enddo
else
do j = 1, ao_prim_num(i)
ao_coef_norm_cosgtos(i,j) = ao_coef(i,j)
enddo
endif
enddo
END_PROVIDER
! ---
BEGIN_PROVIDER [ double precision, ao_coef_norm_ord_cosgtos, (ao_num, ao_prim_num_max) ]
&BEGIN_PROVIDER [ complex*16 , ao_expo_ord_cosgtos, (ao_num, ao_prim_num_max) ]
implicit none
integer :: i, j
integer :: iorder(ao_prim_num_max)
double precision :: d(ao_prim_num_max,3)
d = 0.d0
do i = 1, ao_num
do j = 1, ao_prim_num(i)
iorder(j) = j
d(j,1) = ao_expo(i,j)
d(j,2) = ao_coef_norm_cosgtos(i,j)
d(j,3) = ao_expoim_cosgtos(i,j)
enddo
call dsort (d(1,1), iorder, ao_prim_num(i))
call dset_order(d(1,2), iorder, ao_prim_num(i))
call dset_order(d(1,3), iorder, ao_prim_num(i))
do j = 1, ao_prim_num(i)
ao_expo_ord_cosgtos (i,j) = d(j,1) + (0.d0, 1.d0) * d(j,3)
ao_coef_norm_ord_cosgtos(i,j) = d(j,2)
enddo
enddo
END_PROVIDER
! ---
BEGIN_PROVIDER [ double precision, ao_overlap_cosgtos, (ao_num, ao_num) ]
&BEGIN_PROVIDER [ double precision, ao_overlap_cosgtos_x, (ao_num, ao_num) ]
&BEGIN_PROVIDER [ double precision, ao_overlap_cosgtos_y, (ao_num, ao_num) ]
&BEGIN_PROVIDER [ double precision, ao_overlap_cosgtos_z, (ao_num, ao_num) ]
implicit none
integer :: i, j, n, l, dim1, power_A(3), power_B(3)
double precision :: c, overlap, overlap_x, overlap_y, overlap_z
complex*16 :: alpha, beta, A_center(3), B_center(3)
complex*16 :: overlap1, overlap_x1, overlap_y1, overlap_z1
complex*16 :: overlap2, overlap_x2, overlap_y2, overlap_z2
ao_overlap_cosgtos = 0.d0
ao_overlap_cosgtos_x = 0.d0
ao_overlap_cosgtos_y = 0.d0
ao_overlap_cosgtos_z = 0.d0
dim1 = 100
!$OMP PARALLEL DO SCHEDULE(GUIDED) &
!$OMP DEFAULT(NONE) &
!$OMP PRIVATE( A_center, B_center, power_A, power_B, alpha, beta, i, j, n, l, c &
!$OMP , overlap_x , overlap_y , overlap_z , overlap &
!$OMP , overlap_x1, overlap_y1, overlap_z1, overlap1 &
!$OMP , overlap_x2, overlap_y2, overlap_z2, overlap2 ) &
!$OMP SHARED( nucl_coord, ao_power, ao_prim_num, ao_num, ao_nucl, dim1 &
!$OMP , ao_overlap_cosgtos_x, ao_overlap_cosgtos_y, ao_overlap_cosgtos_z, ao_overlap_cosgtos &
!$OMP , ao_coef_norm_ord_transp_cosgtos, ao_expo_ord_transp_cosgtos )
do j = 1, ao_num
A_center(1) = nucl_coord(ao_nucl(j),1) * (1.d0, 0.d0)
A_center(2) = nucl_coord(ao_nucl(j),2) * (1.d0, 0.d0)
A_center(3) = nucl_coord(ao_nucl(j),3) * (1.d0, 0.d0)
power_A(1) = ao_power(j,1)
power_A(2) = ao_power(j,2)
power_A(3) = ao_power(j,3)
do i = 1, ao_num
B_center(1) = nucl_coord(ao_nucl(i),1) * (1.d0, 0.d0)
B_center(2) = nucl_coord(ao_nucl(i),2) * (1.d0, 0.d0)
B_center(3) = nucl_coord(ao_nucl(i),3) * (1.d0, 0.d0)
power_B(1) = ao_power(i,1)
power_B(2) = ao_power(i,2)
power_B(3) = ao_power(i,3)
do n = 1, ao_prim_num(j)
alpha = ao_expo_ord_transp_cosgtos(n,j)
do l = 1, ao_prim_num(i)
c = ao_coef_norm_ord_transp_cosgtos(n,j) * ao_coef_norm_ord_transp_cosgtos(l,i)
beta = ao_expo_ord_transp_cosgtos(l,i)
call overlap_cgaussian_xyz( A_center, B_center, alpha, beta, power_A, power_B &
, overlap_x1, overlap_y1, overlap_z1, overlap1, dim1 )
call overlap_cgaussian_xyz( A_center, B_center, conjg(alpha), beta, power_A, power_B &
, overlap_x2, overlap_y2, overlap_z2, overlap2, dim1 )
overlap_x = 2.d0 * real( overlap_x1 + overlap_x2 )
overlap_y = 2.d0 * real( overlap_y1 + overlap_y2 )
overlap_z = 2.d0 * real( overlap_z1 + overlap_z2 )
overlap = 2.d0 * real( overlap1 + overlap2 )
ao_overlap_cosgtos(i,j) = ao_overlap_cosgtos(i,j) + c * overlap
if( isnan(ao_overlap_cosgtos(i,j)) ) then
print*,'i, j', i, j
print*,'l, n', l, n
print*,'c, overlap', c, overlap
print*, overlap_x, overlap_y, overlap_z
stop
endif
ao_overlap_cosgtos_x(i,j) = ao_overlap_cosgtos_x(i,j) + c * overlap_x
ao_overlap_cosgtos_y(i,j) = ao_overlap_cosgtos_y(i,j) + c * overlap_y
ao_overlap_cosgtos_z(i,j) = ao_overlap_cosgtos_z(i,j) + c * overlap_z
enddo
enddo
enddo
enddo
!$OMP END PARALLEL DO
END_PROVIDER
! ---