mirror of
https://github.com/LCPQ/quantum_package
synced 2024-11-09 07:33:53 +01:00
173 lines
5.6 KiB
Fortran
173 lines
5.6 KiB
Fortran
BEGIN_PROVIDER [ double precision, ao_ortho_mono_elec_integral_dressing, (ao_num,ao_num) ]
|
|
implicit none
|
|
BEGIN_DOC
|
|
! Dressing of the core hamiltonian in the orthogonal AO basis set
|
|
END_DOC
|
|
|
|
integer :: i,j,k
|
|
integer :: mu, nu, lambda, A
|
|
double precision :: tmp
|
|
|
|
ao_ortho_mono_elec_integral_dressing = 0.d0
|
|
i = idx_dressing
|
|
do mu=1,ao_num
|
|
if (dabs(mo_coef_in_ao_ortho_basis(mu,i)) > 1.d-5) then
|
|
do A=1,nucl_num
|
|
tmp = 0.d0
|
|
do nu=1,ao_num
|
|
tmp += AO_orthoSlaOverlap_matrix(nu,A) * ao_ortho_mono_elec_integral(mu,nu)
|
|
enddo
|
|
ao_ortho_mono_elec_integral_dressing(mu,mu) += cusp_C(A,i) * (AO_orthoSlaH_matrix(mu,A) - tmp)
|
|
enddo
|
|
ao_ortho_mono_elec_integral_dressing(mu,mu) *= 1.d0/mo_coef_in_ao_ortho_basis(mu,i)
|
|
endif
|
|
enddo
|
|
END_PROVIDER
|
|
|
|
|
|
BEGIN_PROVIDER [ double precision, ao_ortho_mono_elec_integral, (ao_num, ao_num) ]
|
|
implicit none
|
|
BEGIN_DOC
|
|
! h core in orthogonal AO basis
|
|
END_DOC
|
|
double precision, allocatable :: T(:,:)
|
|
allocate(T(ao_num,ao_num))
|
|
call dgemm('T','N',ao_num,ao_num,ao_num,1.d0, &
|
|
ao_ortho_canonical_coef, size(ao_ortho_canonical_coef,1), &
|
|
ao_mono_elec_integral, size(ao_mono_elec_integral,1), &
|
|
0.d0, T, size(T,1))
|
|
call dgemm('N','N',ao_num,ao_num,ao_num,1.d0, &
|
|
T, size(T,1), &
|
|
ao_ortho_canonical_coef, size(ao_ortho_canonical_coef,1), &
|
|
0.d0, ao_ortho_mono_elec_integral, size(ao_ortho_mono_elec_integral,1))
|
|
deallocate(T)
|
|
END_PROVIDER
|
|
|
|
|
|
BEGIN_PROVIDER [ double precision, ao_mono_elec_integral_dressing, (ao_num,ao_num) ]
|
|
implicit none
|
|
BEGIN_DOC
|
|
! Dressing of the core hamiltonian in the AO basis set
|
|
END_DOC
|
|
call ao_ortho_cano_to_ao(ao_ortho_mono_elec_integral_dressing,size(ao_ortho_mono_elec_integral_dressing,1),&
|
|
ao_mono_elec_integral_dressing,size(ao_mono_elec_integral_dressing,1))
|
|
END_PROVIDER
|
|
|
|
BEGIN_PROVIDER [ double precision, mo_mono_elec_integral_dressing, (mo_tot_num,mo_tot_num) ]
|
|
implicit none
|
|
BEGIN_DOC
|
|
! Dressing of the core hamiltonian in the MO basis set
|
|
END_DOC
|
|
|
|
call ao_to_mo(ao_mono_elec_integral_dressing,size(ao_mono_elec_integral_dressing,1),&
|
|
mo_mono_elec_integral_dressing,size(mo_mono_elec_integral_dressing,1))
|
|
END_PROVIDER
|
|
|
|
|
|
BEGIN_PROVIDER [ integer, idx_dressing ]
|
|
implicit none
|
|
BEGIN_DOC
|
|
! Index of the MO which is being dressed
|
|
END_DOC
|
|
idx_dressing = 1
|
|
END_PROVIDER
|
|
|
|
|
|
BEGIN_PROVIDER [ double precision, cusp_corrected_mos, (ao_num,mo_tot_num) ]
|
|
implicit none
|
|
BEGIN_DOC
|
|
! Dressing core hamiltonian in the AO basis set
|
|
END_DOC
|
|
integer :: i,j
|
|
double precision, allocatable :: F(:,:), M(:,:)
|
|
allocate(F(mo_tot_num,mo_tot_num),M(ao_num,mo_tot_num))
|
|
|
|
logical :: oneshot
|
|
|
|
! oneshot = .True.
|
|
oneshot = .False.
|
|
|
|
if (oneshot) then
|
|
cusp_corrected_mos(1:ao_num,1:mo_tot_num) = mo_coef(1:ao_num,1:mo_tot_num)
|
|
slater_coef(1:nucl_num,1:mo_tot_num) = cusp_C(1:nucl_num,1:mo_tot_num)
|
|
return
|
|
|
|
else
|
|
|
|
|
|
do idx_dressing=1,mo_tot_num
|
|
|
|
if (idx_dressing>1) then
|
|
TOUCH idx_dressing
|
|
endif
|
|
|
|
do j=1,mo_tot_num
|
|
do i=1,mo_tot_num
|
|
F(i,j) = Fock_matrix_mo(i,j)
|
|
enddo
|
|
enddo
|
|
|
|
do j=1,mo_tot_num
|
|
do i=1,ao_num
|
|
M(i,j) = mo_coef(i,j)
|
|
enddo
|
|
enddo
|
|
|
|
integer :: it
|
|
do it=1,128
|
|
|
|
! print *, 'X', ao_ortho_canonical_coef(1:ao_num,1:ao_num)
|
|
! print *, 'C', mo_coef(1:ao_num,1:mo_tot_num)
|
|
! print *, 'Cp', mo_coef_in_ao_ortho_basis(1:ao_num,1:mo_tot_num)
|
|
! print *, 'cAi', cusp_C(1:nucl_num,1:mo_tot_num)
|
|
! print *, 'FmuA', AO_orthoSlaH_matrix(1:ao_num,1:nucl_num)
|
|
! print *, 'Fock:', Fock_matrix_ao(1:ao_num,1:ao_num)
|
|
! print *, 'Diag Dressing:', ao_ortho_mono_elec_integral_dressing(1:ao_num,1:ao_num)
|
|
! print *, 'Dressing:', ao_mono_elec_integral_dressing(1:ao_num,1:ao_num)
|
|
! print *, 'Dressed Fock:', Fock_matrix_ao(1:ao_num,1:ao_num) + ao_mono_elec_integral_dressing(1:ao_num,1:ao_num)
|
|
! print *, 'AO_orthoSlaOverlap_matrix', AO_orthoSlaOverlap_matrix(1:ao_num,1:nucl_num)
|
|
! print *, 'AO_orthoSlaH_matrix', AO_orthoSlaH_matrix(1:ao_num,1:nucl_num)
|
|
! print *, 'ao_ortho_mono_elec_integral', ao_ortho_mono_elec_integral(1:ao_num,1:ao_num)
|
|
! print *, 'Fock MO:', Fock_matrix_mo(1:mo_tot_num,1:mo_tot_num)
|
|
do j=1,mo_tot_num
|
|
do i=1,mo_tot_num
|
|
Fock_matrix_mo(i,j) += mo_mono_elec_integral_dressing(i,j)
|
|
enddo
|
|
enddo
|
|
do i=1,mo_tot_num
|
|
Fock_matrix_diag_mo(i) = Fock_matrix_mo(i,i)
|
|
enddo
|
|
! print *, 'Dressed Fock MO:', Fock_matrix_mo(1:mo_tot_num,1:mo_tot_num)
|
|
double precision :: conv
|
|
conv = 0.d0
|
|
do j=1,mo_tot_num
|
|
do i=1,mo_tot_num
|
|
if (i==j) cycle
|
|
conv = max(conv,Fock_matrix_mo(i,j))
|
|
enddo
|
|
enddo
|
|
TOUCH Fock_matrix_mo Fock_matrix_diag_mo
|
|
|
|
mo_coef(1:ao_num,1:mo_tot_num) = eigenvectors_fock_matrix_mo(1:ao_num,1:mo_tot_num)
|
|
TOUCH mo_coef
|
|
!print *, 'C', mo_coef(1:ao_num,1:mo_tot_num)
|
|
!print *, '-----'
|
|
print *, idx_dressing, it, real(mo_coef(1,idx_dressing)), real(conv)
|
|
if (conv < 1.d-5) exit
|
|
!stop
|
|
|
|
enddo
|
|
cusp_corrected_mos(1:ao_num,idx_dressing) = mo_coef(1:ao_num,idx_dressing)
|
|
slater_coef(1:nucl_num,idx_dressing) = cusp_C(1:nucl_num,idx_dressing)
|
|
enddo
|
|
|
|
idx_dressing = 1
|
|
mo_coef(1:ao_num,1:mo_tot_num) = M(1:ao_num,1:mo_tot_num)
|
|
soft_TOUCH mo_coef idx_dressing slater_coef
|
|
|
|
endif
|
|
|
|
END_PROVIDER
|
|
|
|
|