mirror of
https://github.com/LCPQ/quantum_package
synced 2024-11-03 20:54:00 +01:00
Merge branch 'master' of github.com:LCPQ/quantum_package
This commit is contained in:
commit
45d9331cd7
@ -17,8 +17,9 @@ cis_dressed
|
||||
|
||||
determinants
|
||||
n_states 1
|
||||
n_states_diag determinants_n_states
|
||||
n_det_max_jacobi 5000
|
||||
threshold_generators 0.999
|
||||
threshold_generators 0.995
|
||||
threshold_selectors 0.999
|
||||
read_wf False
|
||||
|
||||
|
@ -45,8 +45,8 @@ make -C ocaml qp_create_ezfio_from_xyz.native
|
||||
if [[ $? -ne 0 ]]
|
||||
then
|
||||
scripts/fetch_from_web.py "https://raw.github.com/hcarty/ocamlbrew/master/ocamlbrew-install" ocamlbrew-install.sh
|
||||
cat < ocamlbrew-install.sh | env OCAMLBREW_FLAGS="-r" bash > ocaml_install.log
|
||||
grep "source " install.log | grep "etc/ocamlbrew.bashrc" >> quantum_package.rc
|
||||
cat < ocamlbrew-install.sh | env OCAMLBREW_FLAGS="-r" bash | tee ocaml_install.log
|
||||
grep "source " ocaml_install.log | grep "etc/ocamlbrew.bashrc" >> quantum_package.rc
|
||||
source quantum_package.rc
|
||||
echo Y | opam install core async
|
||||
fi
|
||||
|
@ -4,7 +4,7 @@ from generate_h_apply import *
|
||||
from perturbation import perturbations
|
||||
|
||||
s = H_apply("PT2",SingleRef=True)
|
||||
s.set_perturbation("epstein_nesbet_sc2_projected")
|
||||
s.set_perturbation("epstein_nesbet_sc2_no_projected")
|
||||
print s
|
||||
|
||||
s = H_apply("PT2_en_sc2",SingleRef=True)
|
||||
|
@ -12,8 +12,9 @@ program cisd_sc2_selected
|
||||
|
||||
pt2 = 1.d0
|
||||
perturbation = "epstein_nesbet_sc2_projected"
|
||||
|
||||
E_old(1) = HF_energy
|
||||
davidson_threshold = 1.d-8
|
||||
davidson_threshold = 1.d-10
|
||||
if (N_det > n_det_max_cisd_sc2) then
|
||||
call diagonalize_CI_SC2
|
||||
call save_wavefunction
|
||||
@ -31,6 +32,8 @@ program cisd_sc2_selected
|
||||
print *, '-----'
|
||||
endif
|
||||
|
||||
integer :: i_count
|
||||
i_count = 0
|
||||
do while (N_det < n_det_max_cisd_sc2.and.maxval(abs(pt2(1:N_st))) > pt2_max)
|
||||
print*,'----'
|
||||
print*,''
|
||||
@ -49,6 +52,13 @@ program cisd_sc2_selected
|
||||
E_old(i) = CI_SC2_energy(i)
|
||||
enddo
|
||||
! print *, 'E corr = ', (E_old(1)) - HF_energy
|
||||
if(dabs(E_old(i) - CI_SC2_energy(i) ).le.1.d-12)then
|
||||
i_count += 1
|
||||
selection_criterion_factor = selection_criterion_factor * 0.5d0
|
||||
if(i_count > 5)then
|
||||
exit
|
||||
endif
|
||||
endif
|
||||
if (abort_all) then
|
||||
exit
|
||||
endif
|
||||
@ -81,7 +91,7 @@ program cisd_sc2_selected
|
||||
print *, 'PT2(SC2) = ', pt2(i)
|
||||
print *, 'E(SC2) = ', CI_SC2_energy(i)
|
||||
print *, 'E_before(SC2)+PT2(SC2) = ', CI_SC2_energy(i)+pt2(i)
|
||||
print *, 'E_before(SC2)+PT2(SC2)_new = ', CI_SC2_energy(i)+pt2(i)*(1.d0+norm_pert)
|
||||
print *, 'E_before(SC2)+PT2(SC2)_new = ', CI_SC2_energy(i)+pt2(i)* (1.d0 + norm_pert) - H_pert_diag(i)
|
||||
|
||||
print*,'greater coeficient of the state : ',dabs(psi_coef(imax,i))
|
||||
call get_excitation_degree(ref_bitmask,psi_det(1,1,imax),degree,N_int)
|
||||
|
@ -175,7 +175,7 @@ subroutine CISD_SC2(dets_in,u_in,energies,dim_in,sze,N_st,Nint,convergence)
|
||||
enddo
|
||||
call lapack_diag(eigenvalues,eigenvectors, &
|
||||
H_matrix_tmp,size(H_matrix_all_dets,1),sze)
|
||||
do j=1,min(N_states,sze)
|
||||
do j=1,min(N_states_diag,sze)
|
||||
do i=1,sze
|
||||
u_in(i,j) = eigenvectors(i,j)
|
||||
enddo
|
||||
|
@ -12,7 +12,7 @@ BEGIN_PROVIDER [ integer, davidson_sze_max ]
|
||||
! Max number of Davidson sizes
|
||||
END_DOC
|
||||
ASSERT (davidson_sze_max <= davidson_iter_max)
|
||||
davidson_sze_max = 8*N_states
|
||||
davidson_sze_max = 8*N_states_diag
|
||||
END_PROVIDER
|
||||
|
||||
subroutine davidson_diag(dets_in,u_in,energies,dim_in,sze,N_st,Nint,iunit)
|
||||
|
@ -4,6 +4,7 @@ determinants
|
||||
mo_label character*(64)
|
||||
n_det integer
|
||||
n_states integer
|
||||
n_states_diag integer
|
||||
psi_coef double precision (determinants_n_det,determinants_n_states)
|
||||
psi_det integer*8 (determinants_n_int*determinants_bit_kind/8,2,determinants_n_det)
|
||||
n_det_max_jacobi integer
|
||||
@ -13,4 +14,5 @@ determinants
|
||||
det_occ integer (electrons_elec_alpha_num,determinants_det_num,2)
|
||||
det_coef double precision (determinants_det_num)
|
||||
read_wf logical
|
||||
expected_s2 double precision
|
||||
|
||||
|
@ -150,7 +150,7 @@ subroutine read_dets(det,Nint,Ndet)
|
||||
end
|
||||
|
||||
|
||||
BEGIN_PROVIDER [ double precision, psi_coef, (psi_det_size,N_states) ]
|
||||
BEGIN_PROVIDER [ double precision, psi_coef, (psi_det_size,N_states_diag) ]
|
||||
implicit none
|
||||
BEGIN_DOC
|
||||
! The wave function coefficients. Initialized with Hartree-Fock if the EZFIO file
|
||||
@ -162,6 +162,11 @@ BEGIN_PROVIDER [ double precision, psi_coef, (psi_det_size,N_states) ]
|
||||
double precision, allocatable :: psi_coef_read(:,:)
|
||||
character*(64) :: label
|
||||
|
||||
psi_coef = 0.d0
|
||||
do i=1,N_states_diag
|
||||
psi_coef(i,i) = 1.d0
|
||||
enddo
|
||||
|
||||
if (read_wf) then
|
||||
call ezfio_has_determinants_psi_coef(exists)
|
||||
if (exists) then
|
||||
@ -183,22 +188,8 @@ BEGIN_PROVIDER [ double precision, psi_coef, (psi_det_size,N_states) ]
|
||||
enddo
|
||||
deallocate(psi_coef_read)
|
||||
|
||||
else
|
||||
|
||||
psi_coef = 0.d0
|
||||
do i=1,N_states
|
||||
psi_coef(i,i) = 1.d0
|
||||
enddo
|
||||
|
||||
endif
|
||||
|
||||
else
|
||||
|
||||
psi_coef = 0.d0
|
||||
do i=1,N_states
|
||||
psi_coef(i,i) = 1.d0
|
||||
enddo
|
||||
|
||||
endif
|
||||
|
||||
|
||||
|
@ -9,13 +9,13 @@ BEGIN_PROVIDER [ character*(64), diag_algorithm ]
|
||||
diag_algorithm = "Lapack"
|
||||
endif
|
||||
|
||||
if (N_det < N_states) then
|
||||
if (N_det < N_states_diag) then
|
||||
diag_algorithm = "Lapack"
|
||||
endif
|
||||
|
||||
END_PROVIDER
|
||||
|
||||
BEGIN_PROVIDER [ double precision, CI_energy, (N_states) ]
|
||||
BEGIN_PROVIDER [ double precision, CI_energy, (N_states_diag) ]
|
||||
implicit none
|
||||
BEGIN_DOC
|
||||
! N_states lowest eigenvalues of the CI matrix
|
||||
@ -24,23 +24,25 @@ BEGIN_PROVIDER [ double precision, CI_energy, (N_states) ]
|
||||
integer :: j
|
||||
character*(8) :: st
|
||||
call write_time(output_Dets)
|
||||
do j=1,N_states
|
||||
do j=1,N_states_diag
|
||||
CI_energy(j) = CI_electronic_energy(j) + nuclear_repulsion
|
||||
write(st,'(I4)') j
|
||||
call write_double(output_Dets,CI_energy(j),'Energy of state '//trim(st))
|
||||
call write_double(output_Dets,CI_eigenvectors_s2(j),'S^2 of state '//trim(st))
|
||||
enddo
|
||||
|
||||
END_PROVIDER
|
||||
|
||||
BEGIN_PROVIDER [ double precision, CI_electronic_energy, (N_states) ]
|
||||
&BEGIN_PROVIDER [ double precision, CI_eigenvectors, (N_det,N_states) ]
|
||||
BEGIN_PROVIDER [ double precision, CI_electronic_energy, (N_states_diag) ]
|
||||
&BEGIN_PROVIDER [ double precision, CI_eigenvectors, (N_det,N_states_diag) ]
|
||||
&BEGIN_PROVIDER [ double precision, CI_eigenvectors_s2, (N_states_diag) ]
|
||||
implicit none
|
||||
BEGIN_DOC
|
||||
! Eigenvectors/values of the CI matrix
|
||||
END_DOC
|
||||
integer :: i,j
|
||||
|
||||
do j=1,N_states
|
||||
do j=1,N_states_diag
|
||||
do i=1,N_det
|
||||
CI_eigenvectors(i,j) = psi_coef(i,j)
|
||||
enddo
|
||||
@ -49,7 +51,7 @@ END_PROVIDER
|
||||
if (diag_algorithm == "Davidson") then
|
||||
|
||||
call davidson_diag(psi_det,CI_eigenvectors,CI_electronic_energy, &
|
||||
size(CI_eigenvectors,1),N_det,N_states,N_int,output_Dets)
|
||||
size(CI_eigenvectors,1),N_det,N_states_diag,N_int,output_Dets)
|
||||
|
||||
else if (diag_algorithm == "Lapack") then
|
||||
|
||||
@ -59,11 +61,21 @@ END_PROVIDER
|
||||
call lapack_diag(eigenvalues,eigenvectors, &
|
||||
H_matrix_all_dets,size(H_matrix_all_dets,1),N_det)
|
||||
CI_electronic_energy(:) = 0.d0
|
||||
do j=1,min(N_states,N_det)
|
||||
integer :: i_state
|
||||
double precision :: s2
|
||||
j=0
|
||||
i_state = 0
|
||||
do while(i_state.lt.min(N_states_diag,N_det))
|
||||
j+=1
|
||||
call get_s2_u0(psi_det,eigenvectors(1,j),N_det,N_det,s2)
|
||||
if(dabs(s2-expected_s2).le.0.1d0)then
|
||||
i_state += 1
|
||||
do i=1,N_det
|
||||
CI_eigenvectors(i,j) = eigenvectors(i,j)
|
||||
CI_eigenvectors(i,i_state) = eigenvectors(i,j)
|
||||
enddo
|
||||
CI_electronic_energy(j) = eigenvalues(j)
|
||||
CI_electronic_energy(i_state) = eigenvalues(j)
|
||||
CI_eigenvectors_s2(i_state) = s2
|
||||
endif
|
||||
enddo
|
||||
deallocate(eigenvectors,eigenvalues)
|
||||
endif
|
||||
@ -77,10 +89,10 @@ subroutine diagonalize_CI
|
||||
! eigenstates of the CI matrix
|
||||
END_DOC
|
||||
integer :: i,j
|
||||
do j=1,N_states
|
||||
do j=1,N_states_diag
|
||||
do i=1,N_det
|
||||
psi_coef(i,j) = CI_eigenvectors(i,j)
|
||||
enddo
|
||||
enddo
|
||||
SOFT_TOUCH psi_coef CI_electronic_energy CI_energy CI_eigenvectors
|
||||
SOFT_TOUCH psi_coef CI_electronic_energy CI_energy CI_eigenvectors CI_eigenvectors_s2
|
||||
end
|
||||
|
@ -1,13 +1,13 @@
|
||||
BEGIN_PROVIDER [ double precision, CI_SC2_energy, (N_states) ]
|
||||
BEGIN_PROVIDER [ double precision, CI_SC2_energy, (N_states_diag) ]
|
||||
implicit none
|
||||
BEGIN_DOC
|
||||
! N_states lowest eigenvalues of the CI matrix
|
||||
! N_states_diag lowest eigenvalues of the CI matrix
|
||||
END_DOC
|
||||
|
||||
integer :: j
|
||||
character*(8) :: st
|
||||
call write_time(output_Dets)
|
||||
do j=1,N_states
|
||||
do j=1,N_states_diag
|
||||
CI_SC2_energy(j) = CI_SC2_electronic_energy(j) + nuclear_repulsion
|
||||
write(st,'(I4)') j
|
||||
call write_double(output_Dets,CI_SC2_energy(j),'Energy of state '//trim(st))
|
||||
@ -20,38 +20,38 @@ END_PROVIDER
|
||||
BEGIN_DOC
|
||||
! convergence of the correlation energy of SC2 iterations
|
||||
END_DOC
|
||||
threshold_convergence_SC2 = 1.d-8
|
||||
threshold_convergence_SC2 = 1.d-10
|
||||
|
||||
END_PROVIDER
|
||||
BEGIN_PROVIDER [ double precision, CI_SC2_electronic_energy, (N_states) ]
|
||||
&BEGIN_PROVIDER [ double precision, CI_SC2_eigenvectors, (N_det,N_states) ]
|
||||
BEGIN_PROVIDER [ double precision, CI_SC2_electronic_energy, (N_states_diag) ]
|
||||
&BEGIN_PROVIDER [ double precision, CI_SC2_eigenvectors, (N_det,N_states_diag) ]
|
||||
implicit none
|
||||
BEGIN_DOC
|
||||
! Eigenvectors/values of the CI matrix
|
||||
END_DOC
|
||||
integer :: i,j
|
||||
|
||||
do j=1,N_states
|
||||
do j=1,N_states_diag
|
||||
do i=1,N_det
|
||||
! CI_SC2_eigenvectors(i,j) = psi_coef(i,j)
|
||||
CI_SC2_eigenvectors(i,j) = CI_eigenvectors(i,j)
|
||||
CI_SC2_eigenvectors(i,j) = psi_coef(i,j)
|
||||
! CI_SC2_eigenvectors(i,j) = CI_eigenvectors(i,j)
|
||||
enddo
|
||||
CI_SC2_electronic_energy(j) = CI_electronic_energy(j)
|
||||
enddo
|
||||
|
||||
double precision :: convergence
|
||||
call CISD_SC2(psi_det,CI_SC2_eigenvectors,CI_SC2_electronic_energy, &
|
||||
size(CI_SC2_eigenvectors,1),N_det,N_states,N_int,threshold_convergence_SC2)
|
||||
size(CI_SC2_eigenvectors,1),N_det,N_states_diag,N_int,threshold_convergence_SC2)
|
||||
END_PROVIDER
|
||||
|
||||
subroutine diagonalize_CI_SC2
|
||||
implicit none
|
||||
BEGIN_DOC
|
||||
! Replace the coefficients of the CI states by the coefficients of the
|
||||
! Replace the coefficients of the CI states_diag by the coefficients of the
|
||||
! eigenstates of the CI matrix
|
||||
END_DOC
|
||||
integer :: i,j
|
||||
do j=1,N_states
|
||||
do j=1,N_states_diag
|
||||
do i=1,N_det
|
||||
psi_coef(i,j) = CI_SC2_eigenvectors(i,j)
|
||||
enddo
|
||||
|
@ -9,6 +9,16 @@ T.set_ezfio_name( "N_states" )
|
||||
T.set_output ( "output_dets" )
|
||||
print T
|
||||
|
||||
|
||||
T.set_type ( "integer" )
|
||||
T.set_name ( "N_states_diag" )
|
||||
T.set_doc ( "Number of states to consider for the diagonalization " )
|
||||
T.set_ezfio_dir ( "determinants" )
|
||||
T.set_ezfio_name( "N_states_diag" )
|
||||
T.set_output ( "output_dets" )
|
||||
print T
|
||||
|
||||
|
||||
T.set_name ( "N_det_max_jacobi" )
|
||||
T.set_doc ( "Maximum number of determinants diagonalized by Jacobi" )
|
||||
T.set_ezfio_name( "N_det_max_jacobi" )
|
||||
@ -20,6 +30,7 @@ T.set_doc ( "If true, read the wave function from the EZFIO file" )
|
||||
T.set_ezfio_name( "read_wf" )
|
||||
T.set_output ( "output_dets" )
|
||||
print T
|
||||
|
||||
END_SHELL
|
||||
|
||||
|
||||
|
@ -42,6 +42,21 @@ BEGIN_PROVIDER [ double precision, S_z ]
|
||||
|
||||
END_PROVIDER
|
||||
|
||||
BEGIN_PROVIDER [ double precision, expected_s2]
|
||||
implicit none
|
||||
PROVIDE ezfio_filename
|
||||
logical :: has_expected_s2
|
||||
|
||||
call ezfio_has_determinants_expected_s2(has_expected_s2)
|
||||
if (has_expected_s2) then
|
||||
call ezfio_get_determinants_expected_s2(expected_s2)
|
||||
else
|
||||
expected_s2 = elec_alpha_num - elec_beta_num + 0.5d0 * ((elec_alpha_num - elec_beta_num)**2*0.5d0 - (elec_alpha_num-elec_beta_num))
|
||||
! call ezfio_set_determinants_expected_s2(expected_s2)
|
||||
endif
|
||||
|
||||
END_PROVIDER
|
||||
|
||||
|
||||
subroutine get_s2_u0(psi_keys_tmp,psi_coefs_tmp,n,nmax,s2)
|
||||
implicit none
|
||||
|
@ -72,8 +72,8 @@ subroutine pt2_epstein_nesbet_SC2_projected(det_pert,c_pert,e_2_pert,H_pert_diag
|
||||
enddo
|
||||
if(degree==4)then
|
||||
! <psi|delta_H|psi>
|
||||
H_pert_diag(1) = e_2_pert(1)
|
||||
e_2_pert_fonda = H_pert_diag(1)
|
||||
e_2_pert_fonda = e_2_pert(1)
|
||||
H_pert_diag(1) = e_2_pert(1) * c_pert(1) * c_pert(1)
|
||||
do i = 1, N_st
|
||||
do j = 1, idx_repeat(0)
|
||||
e_2_pert(i) += e_2_pert_fonda * psi_selectors_coef(idx_repeat(j),i) * psi_selectors_coef(idx_repeat(j),i)
|
||||
@ -83,6 +83,76 @@ subroutine pt2_epstein_nesbet_SC2_projected(det_pert,c_pert,e_2_pert,H_pert_diag
|
||||
|
||||
end
|
||||
|
||||
|
||||
subroutine pt2_epstein_nesbet_SC2_no_projected(det_pert,c_pert,e_2_pert,H_pert_diag,Nint,ndet,N_st)
|
||||
use bitmasks
|
||||
implicit none
|
||||
integer, intent(in) :: Nint,ndet,N_st
|
||||
integer(bit_kind), intent(in) :: det_pert(Nint,2)
|
||||
double precision , intent(out) :: c_pert(N_st),e_2_pert(N_st),H_pert_diag(N_st)
|
||||
double precision :: i_H_psi_array(N_st)
|
||||
integer :: idx_repeat(0:ndet)
|
||||
|
||||
BEGIN_DOC
|
||||
! compute the Epstein-Nesbet perturbative first order coefficient and second order energetic contribution
|
||||
!
|
||||
! for the various N_st states,
|
||||
!
|
||||
! but with the correction in the denominator
|
||||
!
|
||||
! comming from the interaction of that determinant with all the others determinants
|
||||
!
|
||||
! that can be repeated by repeating all the double excitations
|
||||
!
|
||||
! : you repeat all the correlation energy already taken into account in CI_electronic_energy(1)
|
||||
!
|
||||
! that could be repeated to this determinant.
|
||||
!
|
||||
! In addition, for the perturbative energetic contribution you have the standard second order
|
||||
!
|
||||
! e_2_pert = <psi_i|H|det_pert>^2/(Delta_E)
|
||||
!
|
||||
! and also the purely projected contribution
|
||||
!
|
||||
! H_pert_diag = <HF|H|det_pert> c_pert
|
||||
END_DOC
|
||||
|
||||
integer :: i,j,degree,l
|
||||
double precision :: diag_H_mat_elem,accu_e_corr,hij,h0j,h,delta_E
|
||||
double precision :: repeat_all_e_corr,accu_e_corr_tmp,e_2_pert_fonda
|
||||
|
||||
ASSERT (Nint == N_int)
|
||||
ASSERT (Nint > 0)
|
||||
|
||||
call i_H_psi_SC2(det_pert,psi_selectors,psi_selectors_coef,Nint,N_det_selectors,psi_selectors_size,N_st,i_H_psi_array,idx_repeat)
|
||||
accu_e_corr = 0.d0
|
||||
!$IVDEP
|
||||
do i = 1, idx_repeat(0)
|
||||
accu_e_corr = accu_e_corr + E_corr_per_selectors(idx_repeat(i))
|
||||
enddo
|
||||
h = diag_H_mat_elem(det_pert,Nint) + accu_e_corr
|
||||
delta_E = 1.d0/(CI_SC2_electronic_energy(1) - h)
|
||||
|
||||
|
||||
c_pert(1) = i_H_psi_array(1) /(CI_SC2_electronic_energy(1) - h)
|
||||
e_2_pert(1) = i_H_psi_array(1) * c_pert(1)
|
||||
|
||||
do i =2,N_st
|
||||
H_pert_diag(i) = h
|
||||
if (dabs(CI_SC2_electronic_energy(i) - h) > 1.d-6) then
|
||||
c_pert(i) = i_H_psi_array(i) / (-dabs(CI_SC2_electronic_energy(i) - h))
|
||||
e_2_pert(i) = (c_pert(i) * i_H_psi_array(i))
|
||||
else
|
||||
c_pert(i) = i_H_psi_array(i)
|
||||
e_2_pert(i) = -dabs(i_H_psi_array(i))
|
||||
endif
|
||||
enddo
|
||||
end
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
double precision function repeat_all_e_corr(key_in)
|
||||
implicit none
|
||||
integer(bit_kind), intent(in) :: key_in(N_int,2)
|
||||
|
@ -307,6 +307,72 @@ subroutine lapack_diag(eigvalues,eigvectors,H,nmax,n)
|
||||
deallocate(A,eigenvalues)
|
||||
end
|
||||
|
||||
subroutine lapack_diag_s2(eigvalues,eigvectors,H,nmax,n)
|
||||
implicit none
|
||||
BEGIN_DOC
|
||||
! Diagonalize matrix H
|
||||
!
|
||||
! H is untouched between input and ouptut
|
||||
!
|
||||
! eigevalues(i) = ith lowest eigenvalue of the H matrix
|
||||
!
|
||||
! eigvectors(i,j) = <i|psi_j> where i is the basis function and psi_j is the j th eigenvector
|
||||
!
|
||||
END_DOC
|
||||
integer, intent(in) :: n,nmax
|
||||
double precision, intent(out) :: eigvectors(nmax,n)
|
||||
double precision, intent(out) :: eigvalues(n)
|
||||
double precision, intent(in) :: H(nmax,n)
|
||||
double precision,allocatable :: eigenvalues(:)
|
||||
double precision,allocatable :: work(:)
|
||||
double precision,allocatable :: A(:,:)
|
||||
integer :: lwork, info, i,j,l,k, liwork
|
||||
|
||||
allocate(A(nmax,n),eigenvalues(n))
|
||||
! print*,'Diagonalization by jacobi'
|
||||
! print*,'n = ',n
|
||||
|
||||
A=H
|
||||
lwork = 2*n*n + 6*n+ 1
|
||||
allocate (work(lwork))
|
||||
|
||||
lwork = -1
|
||||
call DSYEV( 'V', 'U', n, A, nmax, eigenvalues, work, lwork, &
|
||||
info )
|
||||
if (info < 0) then
|
||||
print *, irp_here, ': DSYEV: the ',-info,'-th argument had an illegal value'
|
||||
stop 2
|
||||
endif
|
||||
lwork = int( work( 1 ) )
|
||||
deallocate (work)
|
||||
|
||||
allocate (work(lwork))
|
||||
call DSYEV( 'V', 'U', n, A, nmax, eigenvalues, work, lwork, &
|
||||
info )
|
||||
deallocate(work)
|
||||
|
||||
if (info < 0) then
|
||||
print *, irp_here, ': DSYEV: the ',-info,'-th argument had an illegal value'
|
||||
stop 2
|
||||
else if( info > 0 ) then
|
||||
write(*,*)'DSYEV Failed'
|
||||
stop 1
|
||||
end if
|
||||
|
||||
eigvectors = 0.d0
|
||||
eigvalues = 0.d0
|
||||
do j = 1, n
|
||||
eigvalues(j) = eigenvalues(j)
|
||||
do i = 1, n
|
||||
eigvectors(i,j) = A(i,j)
|
||||
enddo
|
||||
enddo
|
||||
deallocate(A,eigenvalues)
|
||||
end
|
||||
|
||||
|
||||
|
||||
|
||||
subroutine lapack_partial_diag(eigvalues,eigvectors,H,nmax,n,n_st)
|
||||
implicit none
|
||||
BEGIN_DOC
|
||||
|
@ -15,10 +15,10 @@ double precision function overlap_gaussian_x(A_center,B_center,alpha,beta,power_
|
||||
call give_explicit_poly_and_gaussian_x(P_new,P_center,p,fact_p,iorder_p,alpha,&
|
||||
beta,power_A,power_B,A_center,B_center,dim)
|
||||
|
||||
if(fact_p.lt.0.000001d0)then
|
||||
overlap_gaussian_x = 0.d0
|
||||
return
|
||||
endif
|
||||
! if(fact_p.lt.0.000001d0)then
|
||||
! overlap_gaussian_x = 0.d0
|
||||
! return
|
||||
! endif
|
||||
|
||||
overlap_gaussian_x = 0.d0
|
||||
integer :: i
|
||||
|
@ -49,17 +49,17 @@ double precision function binom_func(i,j)
|
||||
end
|
||||
|
||||
|
||||
BEGIN_PROVIDER [ double precision, binom, (0:20,0:20) ]
|
||||
&BEGIN_PROVIDER [ double precision, binom_transp, (0:20,0:20) ]
|
||||
BEGIN_PROVIDER [ double precision, binom, (0:40,0:40) ]
|
||||
&BEGIN_PROVIDER [ double precision, binom_transp, (0:40,0:40) ]
|
||||
implicit none
|
||||
BEGIN_DOC
|
||||
! Binomial coefficients
|
||||
END_DOC
|
||||
integer :: k,l
|
||||
double precision :: fact, f
|
||||
do k=0,20
|
||||
do k=0,40
|
||||
f = fact(k)
|
||||
do l=0,20
|
||||
do l=0,40
|
||||
binom(k,l) = f/(fact(l)*fact(k-l))
|
||||
binom_transp(l,k) = binom(k,l)
|
||||
enddo
|
||||
|
Loading…
Reference in New Issue
Block a user