9
1
mirror of https://github.com/QuantumPackage/qp2.git synced 2024-11-18 11:23:38 +01:00
qp2/src/scf_utils/fock_matrix.irp.f
2021-04-08 20:37:17 +02:00

186 lines
5.6 KiB
Fortran

BEGIN_PROVIDER [ double precision, Fock_matrix_mo, (mo_num,mo_num) ]
&BEGIN_PROVIDER [ double precision, Fock_matrix_diag_mo, (mo_num)]
implicit none
BEGIN_DOC
! Fock matrix on the MO basis.
! For open shells, the ROHF Fock Matrix is ::
!
! | F-K | F + K/2 | F |
! |---------------------------------|
! | F + K/2 | F | F - K/2 |
! |---------------------------------|
! | F | F - K/2 | F + K |
!
!
! F = 1/2 (Fa + Fb)
!
! K = Fb - Fa
!
END_DOC
integer :: i,j,n
if (elec_alpha_num == elec_beta_num) then
Fock_matrix_mo = Fock_matrix_mo_alpha
else
do j=1,elec_beta_num
! F-K
do i=1,elec_beta_num !CC
Fock_matrix_mo(i,j) = 0.5d0*(Fock_matrix_mo_alpha(i,j)+Fock_matrix_mo_beta(i,j))&
- (Fock_matrix_mo_beta(i,j) - Fock_matrix_mo_alpha(i,j))
enddo
! F+K/2
do i=elec_beta_num+1,elec_alpha_num !CA
Fock_matrix_mo(i,j) = 0.5d0*(Fock_matrix_mo_alpha(i,j)+Fock_matrix_mo_beta(i,j))&
+ 0.5d0*(Fock_matrix_mo_beta(i,j) - Fock_matrix_mo_alpha(i,j))
enddo
! F
do i=elec_alpha_num+1, mo_num !CV
Fock_matrix_mo(i,j) = 0.5d0*(Fock_matrix_mo_alpha(i,j)+Fock_matrix_mo_beta(i,j))
enddo
enddo
do j=elec_beta_num+1,elec_alpha_num
! F+K/2
do i=1,elec_beta_num !AC
Fock_matrix_mo(i,j) = 0.5d0*(Fock_matrix_mo_alpha(i,j)+Fock_matrix_mo_beta(i,j))&
+ 0.5d0*(Fock_matrix_mo_beta(i,j) - Fock_matrix_mo_alpha(i,j))
enddo
! F
do i=elec_beta_num+1,elec_alpha_num !AA
Fock_matrix_mo(i,j) = 0.5d0*(Fock_matrix_mo_alpha(i,j)+Fock_matrix_mo_beta(i,j))
enddo
! F-K/2
do i=elec_alpha_num+1, mo_num !AV
Fock_matrix_mo(i,j) = 0.5d0*(Fock_matrix_mo_alpha(i,j)+Fock_matrix_mo_beta(i,j))&
- 0.5d0*(Fock_matrix_mo_beta(i,j) - Fock_matrix_mo_alpha(i,j))
enddo
enddo
do j=elec_alpha_num+1, mo_num
! F
do i=1,elec_beta_num !VC
Fock_matrix_mo(i,j) = 0.5d0*(Fock_matrix_mo_alpha(i,j)+Fock_matrix_mo_beta(i,j))
enddo
! F-K/2
do i=elec_beta_num+1,elec_alpha_num !VA
Fock_matrix_mo(i,j) = 0.5d0*(Fock_matrix_mo_alpha(i,j)+Fock_matrix_mo_beta(i,j))&
- 0.5d0*(Fock_matrix_mo_beta(i,j) - Fock_matrix_mo_alpha(i,j))
enddo
! F+K
do i=elec_alpha_num+1,mo_num !VV
Fock_matrix_mo(i,j) = 0.5d0*(Fock_matrix_mo_alpha(i,j)+Fock_matrix_mo_beta(i,j)) &
+ (Fock_matrix_mo_beta(i,j) - Fock_matrix_mo_alpha(i,j))
enddo
enddo
endif
do i = 1, mo_num
Fock_matrix_diag_mo(i) = Fock_matrix_mo(i,i)
enddo
if(frozen_orb_scf)then
integer :: iorb,jorb
do i = 1, n_core_orb
iorb = list_core(i)
do j = 1, n_act_orb
jorb = list_act(j)
Fock_matrix_mo(iorb,jorb) = 0.d0
Fock_matrix_mo(jorb,iorb) = 0.d0
enddo
enddo
endif
if(no_oa_or_av_opt)then
do i = 1, n_act_orb
iorb = list_act(i)
do j = 1, n_inact_orb
jorb = list_inact(j)
Fock_matrix_mo(iorb,jorb) = 0.d0
Fock_matrix_mo(jorb,iorb) = 0.d0
enddo
do j = 1, n_virt_orb
jorb = list_virt(j)
Fock_matrix_mo(iorb,jorb) = 0.d0
Fock_matrix_mo(jorb,iorb) = 0.d0
enddo
do j = 1, n_core_orb
jorb = list_core(j)
Fock_matrix_mo(iorb,jorb) = 0.d0
Fock_matrix_mo(jorb,iorb) = 0.d0
enddo
enddo
endif
END_PROVIDER
BEGIN_PROVIDER [ double precision, Fock_matrix_mo_alpha, (mo_num,mo_num) ]
implicit none
BEGIN_DOC
! Fock matrix on the MO basis
END_DOC
call ao_to_mo(Fock_matrix_ao_alpha,size(Fock_matrix_ao_alpha,1), &
Fock_matrix_mo_alpha,size(Fock_matrix_mo_alpha,1))
END_PROVIDER
BEGIN_PROVIDER [ double precision, Fock_matrix_mo_beta, (mo_num,mo_num) ]
implicit none
BEGIN_DOC
! Fock matrix on the MO basis
END_DOC
call ao_to_mo(Fock_matrix_ao_beta,size(Fock_matrix_ao_beta,1), &
Fock_matrix_mo_beta,size(Fock_matrix_mo_beta,1))
END_PROVIDER
BEGIN_PROVIDER [ double precision, Fock_matrix_ao, (ao_num, ao_num) ]
implicit none
BEGIN_DOC
! Fock matrix in AO basis set
END_DOC
if(frozen_orb_scf)then
call mo_to_ao(Fock_matrix_mo,size(Fock_matrix_mo,1), &
Fock_matrix_ao,size(Fock_matrix_ao,1))
else
if ( (elec_alpha_num == elec_beta_num).and. &
(level_shift == 0.) ) &
then
integer :: i,j
do j=1,ao_num
do i=1,ao_num
Fock_matrix_ao(i,j) = Fock_matrix_ao_alpha(i,j)
enddo
enddo
else
call mo_to_ao(Fock_matrix_mo,size(Fock_matrix_mo,1), &
Fock_matrix_ao,size(Fock_matrix_ao,1))
endif
endif
END_PROVIDER
BEGIN_PROVIDER [ double precision, SCF_energy ]
implicit none
BEGIN_DOC
! Hartree-Fock energy
END_DOC
SCF_energy = nuclear_repulsion
integer :: i,j
do j=1,ao_num
do i=1,ao_num
SCF_energy += 0.5d0 * ( &
(ao_one_e_integrals(i,j) + Fock_matrix_ao_alpha(i,j) ) * SCF_density_matrix_ao_alpha(i,j) +&
(ao_one_e_integrals(i,j) + Fock_matrix_ao_beta (i,j) ) * SCF_density_matrix_ao_beta (i,j) )
enddo
enddo
SCF_energy += extra_e_contrib_density
END_PROVIDER