scemama/qp_plugins_scemama
1
0
Fork 0
mirror of https://gitlab.com/scemama/qp_plugins_scemama.git synced 2024-12-22 12:23:37 +01:00
Code Issues Releases Wiki Activity

Compare commits

base: scemama:d5192bcbc918fda3570dfe9b103b036c6fd7990e
Branches Tags
scemama:master
...
compare: scemama:501cc317d6f20685f2e9fb3363a5950e1bd91d15
Branches Tags
scemama:master

4 Commits
d5192bcbc9 ... 501cc317d6

Author SHA1 Message Date
Emmanuel Giner
501cc317d6 added the possibility to have a gradient based convergence in casscf 2021-07-02 18:04:05 +02:00
Emmanuel Giner
08b3f247f0 added the possibility to diagonalize the SXmatrix with Davidson 2021-07-02 17:46:24 +02:00
Emmanuel Giner
98c3948d6a changed hessian in order to better select the blocks 2021-07-02 16:02:33 +02:00
Emmanuel Giner
1edbd0b890 added the diagonal hessian 2021-07-01 18:16:23 +02:00
9 changed files with 688 additions and 347 deletions
Show Stats Expand all files Collapse all files

26
devel/casscf/EZFIO.cfg
View File

@ -10,12 +10,6 @@ doc: Calculated |FCI| energy + |PT2|
interface: ezfio
size: (determinants.n_states)
[cisd_guess]
type: logical
doc: If true, the CASSCF starts with a CISD wave function
interface: ezfio,provider,ocaml
default: True
[state_following_casscf]
type: logical
doc: If |true|, the CASSCF will try to follow the guess CI vector and orbitals
@ -23,6 +17,19 @@ interface: ezfio,provider,ocaml
default: False
[diag_hess_cas]
type: logical
doc: If |true|, only the DIAGONAL part of the hessian is retained for the CASSCF
interface: ezfio,provider,ocaml
default: False
[hess_cv_cv]
type: logical
doc: If |true|, the core-virtual - core-virtual part of the hessian is computed
interface: ezfio,provider,ocaml
default: True
[level_shift_casscf]
type: Positive_float
doc: Energy shift on the virtual MOs to improve SCF convergence
@ -35,3 +42,10 @@ type: logical
doc: If true, the two-rdm are computed with a fast algo
interface: ezfio,provider,ocaml
default: True
[criterion_casscf]
type: character*(32)
doc: choice of the criterion for the convergence of the casscf: can be energy or gradients
interface: ezfio, provider, ocaml
default: energy

1
devel/casscf/NEED
View File

@ -2,3 +2,4 @@ cipsi
selectors_full
generators_cas
two_body_rdm
dav_general_mat

48
devel/casscf/casscf.irp.f
View File

@ -4,10 +4,12 @@ program casscf
! TODO : Put the documentation of the program here
END_DOC
call reorder_orbitals_for_casscf
! no_vvvv_integrals = .True.
! touch no_vvvv_integrals
pt2_max = 0.02
no_vvvv_integrals = .True.
touch no_vvvv_integrals
pt2_max = 0.005
SOFT_TOUCH pt2_max
n_det_max_full = 500
touch n_det_max_full
call run_stochastic_cipsi
call run
end
@ -31,26 +33,38 @@ subroutine run
energy = eone+etwo+ecore
call write_time(6)
call write_int(6,iteration,'CAS-SCF iteration')
call write_double(6,energy,'CAS-SCF energy')
call write_double(6,energy_improvement, 'Predicted energy improvement')
call write_int(6,iteration,'CAS-SCF iteration = ')
call write_double(6,energy,'CAS-SCF energy = ')
call write_double(6,norm_grad_vec2,'Norm of gradients = ')
call write_double(6,norm_grad_vec2_tab(1), ' Core-active gradients = ')
call write_double(6,norm_grad_vec2_tab(2), ' Core-virtual gradients = ')
call write_double(6,norm_grad_vec2_tab(3), ' Active-virtual gradients = ')
call write_double(6,energy_improvement, 'Predicted energy improvement = ')
converged = dabs(energy_improvement) < thresh_scf
if(criterion_casscf == "energy")then
converged = dabs(energy_improvement) < thresh_scf
else if (criterion_casscf == "gradients")then
converged = norm_grad_vec2 < thresh_scf
else
converged = dabs(energy_improvement) < thresh_scf
endif
pt2_max = dabs(energy_improvement / pt2_relative_error)
mo_coef = NewOrbs
mo_occ = occnum
call save_mos
iteration += 1
N_det = max(N_det/2 ,N_states)
psi_det = psi_det_sorted
psi_coef = psi_coef_sorted
read_wf = .True.
call clear_mo_map
SOFT_TOUCH mo_coef N_det pt2_max psi_det psi_coef
if(iteration .gt. 3)then
state_following_casscf = state_following_casscf_save
touch state_following_casscf
if(.not.converged)then
iteration += 1
N_det = max(N_det/2 ,N_states)
psi_det = psi_det_sorted
psi_coef = psi_coef_sorted
read_wf = .True.
call clear_mo_map
SOFT_TOUCH mo_coef N_det pt2_max psi_det psi_coef
if(iteration .gt. 3)then
state_following_casscf = state_following_casscf_save
touch state_following_casscf
endif
endif
enddo

44
devel/casscf/dav_sx_mat.irp.f Normal file
View File

@ -0,0 +1,44 @@
subroutine davidson_diag_sx_mat(N_st, u_in, energies)
implicit none
integer, intent(in) :: N_st
double precision, intent(out) :: u_in(nMonoEx+1,n_states_diag), energies(N_st)
integer :: i,j,N_st_tmp, dim_in, sze, N_st_diag_in
integer, allocatable :: list_guess(:)
double precision, allocatable :: H_jj(:)
logical :: converged
N_st_diag_in = n_states_diag
provide SXmatrix
sze = nMonoEx+1
dim_in = sze
allocate(H_jj(sze), list_guess(sze))
H_jj(1) = 0.d0
N_st_tmp = 1
list_guess(1) = 1
do j = 2, nMonoEx+1
H_jj(j) = SXmatrix(j,j)
if(H_jj(j).lt.0.d0)then
list_guess(N_st_tmp) = j
N_st_tmp += 1
endif
enddo
if(N_st_tmp .ne. N_st)then
print*,'Pb in davidson_diag_sx_mat'
print*,'N_st_tmp .ne. N_st'
print*,N_st_tmp, N_st
stop
endif
print*,'Number of possibly interesting states = ',N_st
print*,'Corresponding diagonal elements of the SX matrix '
u_in = 0.d0
do i = 1, N_st
j = list_guess(i)
print*,'i,j',i,j
print*,'SX(i,i) = ',H_jj(j)
u_in(j,i) = 1.d0
enddo
call davidson_general(u_in,H_jj,energies,dim_in,sze,N_st,N_st_diag_in,converged,SXmatrix)
print*,'energies = ',energies
end

61
devel/casscf/gradient.irp.f
View File

@ -8,15 +8,31 @@ BEGIN_PROVIDER [ integer, nMonoEx ]
nMonoEx=n_core_inact_orb*n_act_orb+n_core_inact_orb*n_virt_orb+n_act_orb*n_virt_orb
END_PROVIDER
BEGIN_PROVIDER [integer, n_c_a_prov]
&BEGIN_PROVIDER [integer, n_c_v_prov]
&BEGIN_PROVIDER [integer, n_a_v_prov]
implicit none
n_c_a_prov = n_core_inact_orb * n_act_orb
n_c_v_prov = n_core_inact_orb * n_virt_orb
n_a_v_prov = n_act_orb * n_virt_orb
END_PROVIDER
BEGIN_PROVIDER [integer, excit, (2,nMonoEx)]
&BEGIN_PROVIDER [character*3, excit_class, (nMonoEx)]
&BEGIN_PROVIDER [integer, list_idx_c_a, (3,n_c_a_prov) ]
&BEGIN_PROVIDER [integer, list_idx_c_v, (3,n_c_v_prov) ]
&BEGIN_PROVIDER [integer, list_idx_a_v, (3,n_a_v_prov) ]
&BEGIN_PROVIDER [integer, mat_idx_c_a, (n_core_inact_orb,n_act_orb)
&BEGIN_PROVIDER [integer, mat_idx_c_v, (n_core_inact_orb,n_virt_orb)
&BEGIN_PROVIDER [integer, mat_idx_a_v, (n_act_orb,n_virt_orb)
BEGIN_DOC
! a list of the orbitals involved in the excitation
END_DOC
implicit none
integer :: i,t,a,ii,tt,aa,indx
integer :: i,t,a,ii,tt,aa,indx,indx_tmp
indx=0
indx_tmp = 0
do ii=1,n_core_inact_orb
i=list_core_inact(ii)
do tt=1,n_act_orb
@ -25,9 +41,15 @@ END_PROVIDER
excit(1,indx)=i
excit(2,indx)=t
excit_class(indx)='c-a'
indx_tmp += 1
list_idx_c_a(1,indx_tmp) = indx
list_idx_c_a(2,indx_tmp) = ii
list_idx_c_a(3,indx_tmp) = tt
mat_idx_c_a(ii,tt) = indx
end do
end do
indx_tmp = 0
do ii=1,n_core_inact_orb
i=list_core_inact(ii)
do aa=1,n_virt_orb
@ -36,9 +58,15 @@ END_PROVIDER
excit(1,indx)=i
excit(2,indx)=a
excit_class(indx)='c-v'
indx_tmp += 1
list_idx_c_v(1,indx_tmp) = indx
list_idx_c_v(2,indx_tmp) = ii
list_idx_c_v(3,indx_tmp) = aa
mat_idx_c_v(ii,aa) = indx
end do
end do
indx_tmp = 0
do tt=1,n_act_orb
t=list_act(tt)
do aa=1,n_virt_orb
@ -47,6 +75,11 @@ END_PROVIDER
excit(1,indx)=t
excit(2,indx)=a
excit_class(indx)='a-v'
indx_tmp += 1
list_idx_a_v(1,indx_tmp) = indx
list_idx_a_v(2,indx_tmp) = tt
list_idx_a_v(3,indx_tmp) = aa
mat_idx_a_v(tt,aa) = indx
end do
end do
@ -60,7 +93,9 @@ END_PROVIDER
END_PROVIDER
BEGIN_PROVIDER [real*8, gradvec2, (nMonoEx)]
BEGIN_PROVIDER [real*8, gradvec2, (nMonoEx)]
&BEGIN_PROVIDER [real*8, norm_grad_vec2]
&BEGIN_PROVIDER [real*8, norm_grad_vec2_tab, (3)]
BEGIN_DOC
! calculate the orbital gradient <Psi| H E_pq |Psi> from density
! matrices and integrals; Siegbahn et al, Phys Scr 1980
@ -69,13 +104,14 @@ BEGIN_PROVIDER [real*8, gradvec2, (nMonoEx)]
implicit none
integer :: i,t,a,indx
real*8 :: gradvec_it,gradvec_ia,gradvec_ta
real*8 :: norm_grad
indx=0
norm_grad_vec2_tab = 0.d0
do i=1,n_core_inact_orb
do t=1,n_act_orb
indx+=1
gradvec2(indx)=gradvec_it(i,t)
norm_grad_vec2_tab(1) += gradvec2(indx)*gradvec2(indx)
end do
end do
@ -83,6 +119,7 @@ BEGIN_PROVIDER [real*8, gradvec2, (nMonoEx)]
do a=1,n_virt_orb
indx+=1
gradvec2(indx)=gradvec_ia(i,a)
norm_grad_vec2_tab(2) += gradvec2(indx)*gradvec2(indx)
end do
end do
@ -90,17 +127,23 @@ BEGIN_PROVIDER [real*8, gradvec2, (nMonoEx)]
do a=1,n_virt_orb
indx+=1
gradvec2(indx)=gradvec_ta(t,a)
norm_grad_vec2_tab(3) += gradvec2(indx)*gradvec2(indx)
end do
end do
norm_grad=0.d0
norm_grad_vec2=0.d0
do indx=1,nMonoEx
norm_grad+=gradvec2(indx)*gradvec2(indx)
norm_grad_vec2+=gradvec2(indx)*gradvec2(indx)
end do
norm_grad=sqrt(norm_grad)
write(6,*)
write(6,*) ' Norm of the orbital gradient (via D, P and integrals): ', norm_grad
write(6,*)
do i = 1, 3
norm_grad_vec2_tab(i) = dsqrt(norm_grad_vec2_tab(i))
enddo
norm_grad_vec2=sqrt(norm_grad_vec2)
if(bavard)then
write(6,*)
write(6,*) ' Norm of the orbital gradient (via D, P and integrals): ', norm_grad_vec2
write(6,*)
endif
END_PROVIDER

479
devel/casscf/hessian.irp.f
View File

@ -1,303 +1,5 @@
use bitmasks
BEGIN_PROVIDER [real*8, hessmat, (nMonoEx,nMonoEx)]
BEGIN_DOC
! calculate the orbital hessian 2 <Psi| E_pq H E_rs |Psi>
! + <Psi| E_pq E_rs H |Psi> + <Psi| E_rs E_pq H |Psi> by hand,
! determinant per determinant, as for the gradient
!
! we assume that we have natural active orbitals
END_DOC
implicit none
integer :: indx,ihole,ipart
integer :: jndx,jhole,jpart
character*3 :: iexc,jexc
real*8 :: res
if (bavard) then
write(6,*) ' providing Hessian matrix hessmat '
write(6,*) ' nMonoEx = ',nMonoEx
endif
do indx=1,nMonoEx
do jndx=1,nMonoEx
hessmat(indx,jndx)=0.D0
end do
end do
do indx=1,nMonoEx
ihole=excit(1,indx)
ipart=excit(2,indx)
iexc=excit_class(indx)
do jndx=indx,nMonoEx
jhole=excit(1,jndx)
jpart=excit(2,jndx)
jexc=excit_class(jndx)
call calc_hess_elem(ihole,ipart,jhole,jpart,res)
hessmat(indx,jndx)=res
hessmat(jndx,indx)=res
end do
end do
END_PROVIDER
subroutine calc_hess_elem(ihole,ipart,jhole,jpart,res)
BEGIN_DOC
! eq 19 of Siegbahn et al, Physica Scripta 1980
! we calculate 2 <Psi| E_pq H E_rs |Psi>
! + <Psi| E_pq E_rs H |Psi> + <Psi| E_rs E_pq H |Psi>
! average over all states is performed.
! no transition between states.
END_DOC
implicit none
integer :: ihole,ipart,ispin,mu,istate
integer :: jhole,jpart,jspin
integer :: mu_pq, mu_pqrs, mu_rs, mu_rspq, nu_rs,nu
real*8 :: res
integer(bit_kind), allocatable :: det_mu(:,:)
integer(bit_kind), allocatable :: det_nu(:,:)
integer(bit_kind), allocatable :: det_mu_pq(:,:)
integer(bit_kind), allocatable :: det_mu_rs(:,:)
integer(bit_kind), allocatable :: det_nu_rs(:,:)
integer(bit_kind), allocatable :: det_mu_pqrs(:,:)
integer(bit_kind), allocatable :: det_mu_rspq(:,:)
real*8 :: i_H_psi_array(N_states),phase,phase2,phase3
real*8 :: i_H_j_element
allocate(det_mu(N_int,2))
allocate(det_nu(N_int,2))
allocate(det_mu_pq(N_int,2))
allocate(det_mu_rs(N_int,2))
allocate(det_nu_rs(N_int,2))
allocate(det_mu_pqrs(N_int,2))
allocate(det_mu_rspq(N_int,2))
integer :: mu_pq_possible
integer :: mu_rs_possible
integer :: nu_rs_possible
integer :: mu_pqrs_possible
integer :: mu_rspq_possible
res=0.D0
! the terms <0|E E H |0>
do mu=1,n_det
! get the string of the determinant
call det_extract(det_mu,mu,N_int)
do ispin=1,2
! do the monoexcitation pq on it
call det_copy(det_mu,det_mu_pq,N_int)
call do_signed_mono_excitation(det_mu,det_mu_pq,mu_pq &
,ihole,ipart,ispin,phase,mu_pq_possible)
if (mu_pq_possible.eq.1) then
! possible, but not necessarily in the list
! do the second excitation
do jspin=1,2
call det_copy(det_mu_pq,det_mu_pqrs,N_int)
call do_signed_mono_excitation(det_mu_pq,det_mu_pqrs,mu_pqrs&
,jhole,jpart,jspin,phase2,mu_pqrs_possible)
! excitation possible
if (mu_pqrs_possible.eq.1) then
call i_H_psi(det_mu_pqrs,psi_det,psi_coef,N_int &
,N_det,N_det,N_states,i_H_psi_array)
do istate=1,N_states
res+=i_H_psi_array(istate)*psi_coef(mu,istate)*phase*phase2
end do
end if
! try the de-excitation with opposite sign
call det_copy(det_mu_pq,det_mu_pqrs,N_int)
call do_signed_mono_excitation(det_mu_pq,det_mu_pqrs,mu_pqrs&
,jpart,jhole,jspin,phase2,mu_pqrs_possible)
phase2=-phase2
! excitation possible
if (mu_pqrs_possible.eq.1) then
call i_H_psi(det_mu_pqrs,psi_det,psi_coef,N_int &
,N_det,N_det,N_states,i_H_psi_array)
do istate=1,N_states
res+=i_H_psi_array(istate)*psi_coef(mu,istate)*phase*phase2
end do
end if
end do
end if
! exchange the notion of pq and rs
! do the monoexcitation rs on the initial determinant
call det_copy(det_mu,det_mu_rs,N_int)
call do_signed_mono_excitation(det_mu,det_mu_rs,mu_rs &
,jhole,jpart,ispin,phase2,mu_rs_possible)
if (mu_rs_possible.eq.1) then
! do the second excitation
do jspin=1,2
call det_copy(det_mu_rs,det_mu_rspq,N_int)
call do_signed_mono_excitation(det_mu_rs,det_mu_rspq,mu_rspq&
,ihole,ipart,jspin,phase3,mu_rspq_possible)
! excitation possible (of course, the result is outside the CAS)
if (mu_rspq_possible.eq.1) then
call i_H_psi(det_mu_rspq,psi_det,psi_coef,N_int &
,N_det,N_det,N_states,i_H_psi_array)
do istate=1,N_states
res+=i_H_psi_array(istate)*psi_coef(mu,istate)*phase2*phase3
end do
end if
! we may try the de-excitation, with opposite sign
call det_copy(det_mu_rs,det_mu_rspq,N_int)
call do_signed_mono_excitation(det_mu_rs,det_mu_rspq,mu_rspq&
,ipart,ihole,jspin,phase3,mu_rspq_possible)
phase3=-phase3
! excitation possible (of course, the result is outside the CAS)
if (mu_rspq_possible.eq.1) then
call i_H_psi(det_mu_rspq,psi_det,psi_coef,N_int &
,N_det,N_det,N_states,i_H_psi_array)
do istate=1,N_states
res+=i_H_psi_array(istate)*psi_coef(mu,istate)*phase2*phase3
end do
end if
end do
end if
!
! the operator E H E, we have to do a double loop over the determinants
! we still have the determinant mu_pq and the phase in memory
if (mu_pq_possible.eq.1) then
do nu=1,N_det
call det_extract(det_nu,nu,N_int)
do jspin=1,2
call det_copy(det_nu,det_nu_rs,N_int)
call do_signed_mono_excitation(det_nu,det_nu_rs,nu_rs &
,jhole,jpart,jspin,phase2,nu_rs_possible)
! excitation possible ?
if (nu_rs_possible.eq.1) then
call i_H_j(det_mu_pq,det_nu_rs,N_int,i_H_j_element)
do istate=1,N_states
res+=2.D0*i_H_j_element*psi_coef(mu,istate) &
*psi_coef(nu,istate)*phase*phase2
end do
end if
end do
end do
end if
end do
end do
! state-averaged Hessian
res*=1.D0/dble(N_states)
end subroutine calc_hess_elem
BEGIN_PROVIDER [real*8, hessmat2, (nMonoEx,nMonoEx)]
BEGIN_DOC
! explicit hessian matrix from density matrices and integrals
! of course, this will be used for a direct Davidson procedure later
! we will not store the matrix in real life
! formulas are broken down as functions for the 6 classes of matrix elements
!
END_DOC
implicit none
integer :: i,j,t,u,a,b,indx,jndx,bstart,ustart,indx_shift
real*8 :: hessmat_itju
real*8 :: hessmat_itja
real*8 :: hessmat_itua
real*8 :: hessmat_iajb
real*8 :: hessmat_iatb
real*8 :: hessmat_taub
if (bavard) then
write(6,*) ' providing Hessian matrix hessmat2 '
write(6,*) ' nMonoEx = ',nMonoEx
endif
!$OMP PARALLEL DEFAULT(NONE) &
!$OMP SHARED(hessmat2,n_core_inact_orb,n_act_orb,n_virt_orb,nMonoEx) &
!$OMP PRIVATE(i,indx,jndx,j,ustart,t,u,a,bstart,indx_shift)
!$OMP DO
do i=1,n_core_inact_orb
do t=1,n_act_orb
indx = t + (i-1)*n_act_orb
jndx=indx
do j=i,n_core_inact_orb
if (i.eq.j) then
ustart=t
else
ustart=1
end if
do u=ustart,n_act_orb
hessmat2(jndx,indx)=hessmat_itju(i,t,j,u)
jndx+=1
end do
end do
do j=1,n_core_inact_orb
do a=1,n_virt_orb
hessmat2(jndx,indx)=hessmat_itja(i,t,j,a)
jndx+=1
end do
end do
do u=1,n_act_orb
do a=1,n_virt_orb
hessmat2(jndx,indx)=hessmat_itua(i,t,u,a)
jndx+=1
end do
end do
end do
end do
!$OMP END DO NOWAIT
indx_shift = n_core_inact_orb*n_act_orb
!$OMP DO
do a=1,n_virt_orb
do i=1,n_core_inact_orb
indx = a + (i-1)*n_virt_orb + indx_shift
jndx=indx
do j=i,n_core_inact_orb
if (i.eq.j) then
bstart=a
else
bstart=1
end if
do b=bstart,n_virt_orb
hessmat2(jndx,indx)=hessmat_iajb(i,a,j,b)
jndx+=1
end do
end do
do t=1,n_act_orb
do b=1,n_virt_orb
hessmat2(jndx,indx)=hessmat_iatb(i,a,t,b)
jndx+=1
end do
end do
end do
end do
!$OMP END DO NOWAIT
indx_shift += n_core_inact_orb*n_virt_orb
!$OMP DO
do a=1,n_virt_orb
do t=1,n_act_orb
indx = a + (t-1)*n_virt_orb + indx_shift
jndx=indx
do u=t,n_act_orb
if (t.eq.u) then
bstart=a
else
bstart=1
end if
do b=bstart,n_virt_orb
hessmat2(jndx,indx)=hessmat_taub(t,a,u,b)
jndx+=1
end do
end do
end do
end do
!$OMP END DO
!$OMP END PARALLEL
do jndx=1,nMonoEx
do indx=1,jndx-1
hessmat2(indx,jndx) = hessmat2(jndx,indx)
enddo
enddo
END_PROVIDER
real*8 function hessmat_itju(i,t,j,u)
BEGIN_DOC
! the orbital hessian for core/inactive -> active, core/inactive -> active
@ -654,3 +356,184 @@ BEGIN_PROVIDER [real*8, hessdiag, (nMonoEx)]
!$OMP END PARALLEL
END_PROVIDER
BEGIN_PROVIDER [double precision, hessmat, (nMonoEx,nMonoEx)]
implicit none
integer :: i,j,t,u,a,b
integer :: indx,indx_tmp, jndx, jndx_tmp
integer :: ustart,bstart
real*8 :: hessmat_itju
real*8 :: hessmat_itja
real*8 :: hessmat_itua
real*8 :: hessmat_iajb
real*8 :: hessmat_iatb
real*8 :: hessmat_taub
! c-a c-v a-v
! c-a | X X X
! c-v | X X
! a-v | X
provide mo_two_e_integrals_in_map
hessmat = 0.d0
!$OMP PARALLEL DEFAULT(NONE) &
!$OMP SHARED(hessmat,n_c_a_prov,list_idx_c_a,n_core_inact_orb,n_act_orb,mat_idx_c_a) &
!$OMP PRIVATE(indx_tmp,indx,i,t,j,u,ustart,jndx)
!$OMP DO
!!!! < Core-active| H |Core-active >
! Core-active excitations
do indx_tmp = 1, n_c_a_prov
indx = list_idx_c_a(1,indx_tmp)
i = list_idx_c_a(2,indx_tmp)
t = list_idx_c_a(3,indx_tmp)
! Core-active excitations
do j = 1, n_core_inact_orb
if (i.eq.j) then
ustart=t
else
ustart=1
end if
do u=ustart,n_act_orb
jndx = mat_idx_c_a(j,u)
hessmat(jndx,indx) = hessmat_itju(i,t,j,u)
hessmat(indx,jndx) = hessmat(jndx,indx)
enddo
enddo
enddo
!$OMP END DO NOWAIT
!$OMP END PARALLEL
!$OMP PARALLEL DEFAULT(NONE) &
!$OMP SHARED(hessmat,n_c_a_prov,n_c_v_prov,list_idx_c_a,list_idx_c_v) &
!$OMP PRIVATE(indx_tmp,jndx_tmp,indx,i,t,j,a,jndx)
!$OMP DO
!!!! < Core-active| H |Core-VIRTUAL >
! Core-active excitations
do indx_tmp = 1, n_c_a_prov
indx = list_idx_c_a(1,indx_tmp)
i = list_idx_c_a(2,indx_tmp)
t = list_idx_c_a(3,indx_tmp)
! Core-VIRTUAL excitations
do jndx_tmp = 1, n_c_v_prov
jndx = list_idx_c_v(1,jndx_tmp)
j = list_idx_c_v(2,jndx_tmp)
a = list_idx_c_v(3,jndx_tmp)
hessmat(jndx,indx) = hessmat_itja(i,t,j,a)
hessmat(indx,jndx) = hessmat(jndx,indx)
enddo
enddo
!$OMP END DO NOWAIT
!$OMP END PARALLEL
!$OMP PARALLEL DEFAULT(NONE) &
!$OMP SHARED(hessmat,n_c_a_prov,n_a_v_prov,list_idx_c_a,list_idx_a_v) &
!$OMP PRIVATE(indx_tmp,jndx_tmp,indx,i,t,u,a,jndx)
!$OMP DO
!!!! < Core-active| H |ACTIVE-VIRTUAL >
! Core-active excitations
do indx_tmp = 1, n_c_a_prov
indx = list_idx_c_a(1,indx_tmp)
i = list_idx_c_a(2,indx_tmp)
t = list_idx_c_a(3,indx_tmp)
! ACTIVE-VIRTUAL excitations
do jndx_tmp = 1, n_a_v_prov
jndx = list_idx_a_v(1,jndx_tmp)
u = list_idx_a_v(2,jndx_tmp)
a = list_idx_a_v(3,jndx_tmp)
hessmat(jndx,indx) = hessmat_itua(i,t,u,a)
hessmat(indx,jndx) = hessmat(jndx,indx)
enddo
enddo
!$OMP END DO NOWAIT
!$OMP END PARALLEL
if(hess_cv_cv)then
!$OMP PARALLEL DEFAULT(NONE) &
!$OMP SHARED(hessmat,n_c_v_prov,list_idx_c_v,n_core_inact_orb,n_virt_orb,mat_idx_c_v) &
!$OMP PRIVATE(indx_tmp,indx,i,a,j,b,bstart,jndx)
!$OMP DO
!!!!! < Core-VIRTUAL | H |Core-VIRTUAL >
! Core-VIRTUAL excitations
do indx_tmp = 1, n_c_v_prov
indx = list_idx_c_v(1,indx_tmp)
i = list_idx_c_v(2,indx_tmp)
a = list_idx_c_v(3,indx_tmp)
! Core-VIRTUAL excitations
do j = 1, n_core_inact_orb
if (i.eq.j) then
bstart=a
else
bstart=1
end if
do b=bstart,n_virt_orb
jndx = mat_idx_c_v(j,b)
hessmat(jndx,indx) = hessmat_iajb(i,a,j,b)
hessmat(indx,jndx) = hessmat(jndx,indx)
enddo
enddo
enddo
!$OMP END DO NOWAIT
!$OMP END PARALLEL
endif
!$OMP PARALLEL DEFAULT(NONE) &
!$OMP SHARED(hessmat,n_c_v_prov,n_a_v_prov,list_idx_c_v,list_idx_a_v) &
!$OMP PRIVATE(indx_tmp,jndx_tmp,indx,i,a,t,b,jndx)
!$OMP DO
!!!! < Core-VIRTUAL | H |Active-VIRTUAL >
! Core-VIRTUAL excitations
do indx_tmp = 1, n_c_v_prov
indx = list_idx_c_v(1,indx_tmp)
i = list_idx_c_v(2,indx_tmp)
a = list_idx_c_v(3,indx_tmp)
! Active-VIRTUAL excitations
do jndx_tmp = 1, n_a_v_prov
jndx = list_idx_a_v(1,jndx_tmp)
t = list_idx_a_v(2,jndx_tmp)
b = list_idx_a_v(3,jndx_tmp)
hessmat(jndx,indx) = hessmat_iatb(i,a,t,b)
hessmat(indx,jndx) = hessmat(jndx,indx)
enddo
enddo
!$OMP END DO NOWAIT
!$OMP END PARALLEL
!$OMP PARALLEL DEFAULT(NONE) &
!$OMP SHARED(hessmat,n_a_v_prov,list_idx_a_v,n_act_orb,n_virt_orb,mat_idx_a_v) &
!$OMP PRIVATE(indx_tmp,indx,t,a,u,b,bstart,jndx)
!$OMP DO
!!!! < Active-VIRTUAL | H |Active-VIRTUAL >
! Active-VIRTUAL excitations
do indx_tmp = 1, n_a_v_prov
indx = list_idx_a_v(1,indx_tmp)
t = list_idx_a_v(2,indx_tmp)
a = list_idx_a_v(3,indx_tmp)
! Active-VIRTUAL excitations
do u=t,n_act_orb
if (t.eq.u) then
bstart=a
else
bstart=1
end if
do b=bstart,n_virt_orb
jndx = mat_idx_a_v(u,b)
hessmat(jndx,indx) = hessmat_taub(t,a,u,b)
hessmat(indx,jndx) = hessmat(jndx,indx)
enddo
enddo
enddo
!$OMP END DO NOWAIT
!$OMP END PARALLEL
END_PROVIDER

310
devel/casscf/hessian_old.irp.f Normal file
View File

@ -0,0 +1,310 @@
use bitmasks
BEGIN_PROVIDER [real*8, hessmat_old, (nMonoEx,nMonoEx)]
BEGIN_DOC
! calculate the orbital hessian 2 <Psi| E_pq H E_rs |Psi>
! + <Psi| E_pq E_rs H |Psi> + <Psi| E_rs E_pq H |Psi> by hand,
! determinant per determinant, as for the gradient
!
! we assume that we have natural active orbitals
END_DOC
implicit none
integer :: indx,ihole,ipart
integer :: jndx,jhole,jpart
character*3 :: iexc,jexc
real*8 :: res
if (bavard) then
write(6,*) ' providing Hessian matrix hessmat_old '
write(6,*) ' nMonoEx = ',nMonoEx
endif
do indx=1,nMonoEx
do jndx=1,nMonoEx
hessmat_old(indx,jndx)=0.D0
end do
end do
do indx=1,nMonoEx
ihole=excit(1,indx)
ipart=excit(2,indx)
iexc=excit_class(indx)
do jndx=indx,nMonoEx
jhole=excit(1,jndx)
jpart=excit(2,jndx)
jexc=excit_class(jndx)
call calc_hess_elem(ihole,ipart,jhole,jpart,res)
hessmat_old(indx,jndx)=res
hessmat_old(jndx,indx)=res
end do
end do
END_PROVIDER
subroutine calc_hess_elem(ihole,ipart,jhole,jpart,res)
BEGIN_DOC
! eq 19 of Siegbahn et al, Physica Scripta 1980
! we calculate 2 <Psi| E_pq H E_rs |Psi>
! + <Psi| E_pq E_rs H |Psi> + <Psi| E_rs E_pq H |Psi>
! average over all states is performed.
! no transition between states.
END_DOC
implicit none
integer :: ihole,ipart,ispin,mu,istate
integer :: jhole,jpart,jspin
integer :: mu_pq, mu_pqrs, mu_rs, mu_rspq, nu_rs,nu
real*8 :: res
integer(bit_kind), allocatable :: det_mu(:,:)
integer(bit_kind), allocatable :: det_nu(:,:)
integer(bit_kind), allocatable :: det_mu_pq(:,:)
integer(bit_kind), allocatable :: det_mu_rs(:,:)
integer(bit_kind), allocatable :: det_nu_rs(:,:)
integer(bit_kind), allocatable :: det_mu_pqrs(:,:)
integer(bit_kind), allocatable :: det_mu_rspq(:,:)
real*8 :: i_H_psi_array(N_states),phase,phase2,phase3
real*8 :: i_H_j_element
allocate(det_mu(N_int,2))
allocate(det_nu(N_int,2))
allocate(det_mu_pq(N_int,2))
allocate(det_mu_rs(N_int,2))
allocate(det_nu_rs(N_int,2))
allocate(det_mu_pqrs(N_int,2))
allocate(det_mu_rspq(N_int,2))
integer :: mu_pq_possible
integer :: mu_rs_possible
integer :: nu_rs_possible
integer :: mu_pqrs_possible
integer :: mu_rspq_possible
res=0.D0
! the terms <0|E E H |0>
do mu=1,n_det
! get the string of the determinant
call det_extract(det_mu,mu,N_int)
do ispin=1,2
! do the monoexcitation pq on it
call det_copy(det_mu,det_mu_pq,N_int)
call do_signed_mono_excitation(det_mu,det_mu_pq,mu_pq &
,ihole,ipart,ispin,phase,mu_pq_possible)
if (mu_pq_possible.eq.1) then
! possible, but not necessarily in the list
! do the second excitation
do jspin=1,2
call det_copy(det_mu_pq,det_mu_pqrs,N_int)
call do_signed_mono_excitation(det_mu_pq,det_mu_pqrs,mu_pqrs&
,jhole,jpart,jspin,phase2,mu_pqrs_possible)
! excitation possible
if (mu_pqrs_possible.eq.1) then
call i_H_psi(det_mu_pqrs,psi_det,psi_coef,N_int &
,N_det,N_det,N_states,i_H_psi_array)
do istate=1,N_states
res+=i_H_psi_array(istate)*psi_coef(mu,istate)*phase*phase2
end do
end if
! try the de-excitation with opposite sign
call det_copy(det_mu_pq,det_mu_pqrs,N_int)
call do_signed_mono_excitation(det_mu_pq,det_mu_pqrs,mu_pqrs&
,jpart,jhole,jspin,phase2,mu_pqrs_possible)
phase2=-phase2
! excitation possible
if (mu_pqrs_possible.eq.1) then
call i_H_psi(det_mu_pqrs,psi_det,psi_coef,N_int &
,N_det,N_det,N_states,i_H_psi_array)
do istate=1,N_states
res+=i_H_psi_array(istate)*psi_coef(mu,istate)*phase*phase2
end do
end if
end do
end if
! exchange the notion of pq and rs
! do the monoexcitation rs on the initial determinant
call det_copy(det_mu,det_mu_rs,N_int)
call do_signed_mono_excitation(det_mu,det_mu_rs,mu_rs &
,jhole,jpart,ispin,phase2,mu_rs_possible)
if (mu_rs_possible.eq.1) then
! do the second excitation
do jspin=1,2
call det_copy(det_mu_rs,det_mu_rspq,N_int)
call do_signed_mono_excitation(det_mu_rs,det_mu_rspq,mu_rspq&
,ihole,ipart,jspin,phase3,mu_rspq_possible)
! excitation possible (of course, the result is outside the CAS)
if (mu_rspq_possible.eq.1) then
call i_H_psi(det_mu_rspq,psi_det,psi_coef,N_int &
,N_det,N_det,N_states,i_H_psi_array)
do istate=1,N_states
res+=i_H_psi_array(istate)*psi_coef(mu,istate)*phase2*phase3
end do
end if
! we may try the de-excitation, with opposite sign
call det_copy(det_mu_rs,det_mu_rspq,N_int)
call do_signed_mono_excitation(det_mu_rs,det_mu_rspq,mu_rspq&
,ipart,ihole,jspin,phase3,mu_rspq_possible)
phase3=-phase3
! excitation possible (of course, the result is outside the CAS)
if (mu_rspq_possible.eq.1) then
call i_H_psi(det_mu_rspq,psi_det,psi_coef,N_int &
,N_det,N_det,N_states,i_H_psi_array)
do istate=1,N_states
res+=i_H_psi_array(istate)*psi_coef(mu,istate)*phase2*phase3
end do
end if
end do
end if
!
! the operator E H E, we have to do a double loop over the determinants
! we still have the determinant mu_pq and the phase in memory
if (mu_pq_possible.eq.1) then
do nu=1,N_det
call det_extract(det_nu,nu,N_int)
do jspin=1,2
call det_copy(det_nu,det_nu_rs,N_int)
call do_signed_mono_excitation(det_nu,det_nu_rs,nu_rs &
,jhole,jpart,jspin,phase2,nu_rs_possible)
! excitation possible ?
if (nu_rs_possible.eq.1) then
call i_H_j(det_mu_pq,det_nu_rs,N_int,i_H_j_element)
do istate=1,N_states
res+=2.D0*i_H_j_element*psi_coef(mu,istate) &
*psi_coef(nu,istate)*phase*phase2
end do
end if
end do
end do
end if
end do
end do
! state-averaged Hessian
res*=1.D0/dble(N_states)
end subroutine calc_hess_elem
BEGIN_PROVIDER [real*8, hessmat_peter, (nMonoEx,nMonoEx)]
BEGIN_DOC
! explicit hessian matrix from density matrices and integrals
! of course, this will be used for a direct Davidson procedure later
! we will not store the matrix in real life
! formulas are broken down as functions for the 6 classes of matrix elements
!
END_DOC
implicit none
integer :: i,j,t,u,a,b,indx,jndx,bstart,ustart,indx_shift
real*8 :: hessmat_itju
real*8 :: hessmat_itja
real*8 :: hessmat_itua
real*8 :: hessmat_iajb
real*8 :: hessmat_iatb
real*8 :: hessmat_taub
if (bavard) then
write(6,*) ' providing Hessian matrix hessmat_peter '
write(6,*) ' nMonoEx = ',nMonoEx
endif
provide mo_two_e_integrals_in_map
!$OMP PARALLEL DEFAULT(NONE) &
!$OMP SHARED(hessmat_peter,n_core_inact_orb,n_act_orb,n_virt_orb,nMonoEx) &
!$OMP PRIVATE(i,indx,jndx,j,ustart,t,u,a,bstart,indx_shift)
!$OMP DO
! (DOUBLY OCCUPIED ---> ACT )
do i=1,n_core_inact_orb
do t=1,n_act_orb
indx = t + (i-1)*n_act_orb
jndx=indx
! (DOUBLY OCCUPIED ---> ACT )
do j=i,n_core_inact_orb
if (i.eq.j) then
ustart=t
else
ustart=1
end if
do u=ustart,n_act_orb
hessmat_peter(jndx,indx)=hessmat_itju(i,t,j,u)
jndx+=1
end do
end do
! (DOUBLY OCCUPIED ---> VIRTUAL)
do j=1,n_core_inact_orb
do a=1,n_virt_orb
hessmat_peter(jndx,indx)=hessmat_itja(i,t,j,a)
jndx+=1
end do
end do
! (ACTIVE ---> VIRTUAL)
do u=1,n_act_orb
do a=1,n_virt_orb
hessmat_peter(jndx,indx)=hessmat_itua(i,t,u,a)
jndx+=1
end do
end do
end do
end do
!$OMP END DO NOWAIT
indx_shift = n_core_inact_orb*n_act_orb
!$OMP DO
! (DOUBLY OCCUPIED ---> VIRTUAL)
do a=1,n_virt_orb
do i=1,n_core_inact_orb
indx = a + (i-1)*n_virt_orb + indx_shift
jndx=indx
! (DOUBLY OCCUPIED ---> VIRTUAL)
do j=i,n_core_inact_orb
if (i.eq.j) then
bstart=a
else
bstart=1
end if
do b=bstart,n_virt_orb
hessmat_peter(jndx,indx)=hessmat_iajb(i,a,j,b)
jndx+=1
end do
end do
! (ACT ---> VIRTUAL)
do t=1,n_act_orb
do b=1,n_virt_orb
hessmat_peter(jndx,indx)=hessmat_iatb(i,a,t,b)
jndx+=1
end do
end do
end do
end do
!$OMP END DO NOWAIT
indx_shift += n_core_inact_orb*n_virt_orb
!$OMP DO
! (ACT ---> VIRTUAL)
do a=1,n_virt_orb
do t=1,n_act_orb
indx = a + (t-1)*n_virt_orb + indx_shift
jndx=indx
! (ACT ---> VIRTUAL)
do u=t,n_act_orb
if (t.eq.u) then
bstart=a
else
bstart=1
end if
do b=bstart,n_virt_orb
hessmat_peter(jndx,indx)=hessmat_taub(t,a,u,b)
jndx+=1
end do
end do
end do
end do
!$OMP END DO
!$OMP END PARALLEL
do jndx=1,nMonoEx
do indx=1,jndx-1
hessmat_peter(indx,jndx) = hessmat_peter(jndx,indx)
enddo
enddo
END_PROVIDER

62
devel/casscf/neworbs.irp.f
View File

@ -1,4 +1,5 @@
BEGIN_PROVIDER [real*8, SXmatrix, (nMonoEx+1,nMonoEx+1)]
BEGIN_PROVIDER [real*8, SXmatrix, (nMonoEx+1,nMonoEx+1)]
&BEGIN_PROVIDER [integer, n_guess_sx_mat ]
implicit none
BEGIN_DOC
! Single-excitation matrix
@ -16,24 +17,34 @@ BEGIN_PROVIDER [real*8, SXmatrix, (nMonoEx+1,nMonoEx+1)]
SXmatrix(1,i+1)=gradvec2(i)
SXmatrix(1+i,1)=gradvec2(i)
end do
do i=1,nMonoEx
do j=1,nMonoEx
SXmatrix(i+1,j+1)=hessmat2(i,j)
SXmatrix(j+1,i+1)=hessmat2(i,j)
end do
end do
if(diag_hess_cas)then
do i = 1, nMonoEx
SXmatrix(i+1,i+1) = hessdiag(i)
enddo
else
do i=1,nMonoEx
do j=1,nMonoEx
SXmatrix(i+1,j+1)=hessmat(i,j)
SXmatrix(j+1,i+1)=hessmat(i,j)
end do
end do
endif
do i = 1, nMonoEx
SXmatrix(i+1,i+1) += level_shift_casscf
enddo
n_guess_sx_mat = 1
do i = 1, nMonoEx
if(SXmatrix(i+1,i+1).lt.0.d0 )then
n_guess_sx_mat += 1
endif
enddo
if (bavard) then
do i=2,nMonoEx
write(6,*) ' diagonal of the Hessian : ',i,hessmat2(i,i)
write(6,*) ' diagonal of the Hessian : ',i,hessmat(i,i)
end do
end if
END_PROVIDER
BEGIN_PROVIDER [real*8, SXeigenvec, (nMonoEx+1,nMonoEx+1)]
@ -42,11 +53,32 @@ END_PROVIDER
BEGIN_DOC
! Eigenvectors/eigenvalues of the single-excitation matrix
END_DOC
call lapack_diag(SXeigenval,SXeigenvec,SXmatrix,nMonoEx+1,nMonoEx+1)
if(nMonoEx+1.gt.n_det_max_full)then
if(bavard)then
print*,'Using the Davidson algorithm to diagonalize the SXmatrix'
endif
double precision, allocatable :: u_in(:,:),energies(:)
allocate(u_in(nMonoEx+1,n_states_diag),energies(n_guess_sx_mat))
call davidson_diag_sx_mat(n_guess_sx_mat, u_in, energies)
integer :: i,j
SXeigenvec = 0.d0
SXeigenval = 0.d0
do i = 1, n_guess_sx_mat
SXeigenval(i) = energies(i)
do j = 1, nMonoEx+1
SXeigenvec(j,i) = u_in(j,i)
enddo
enddo
else
if(bavard)then
print*,'Diagonalize the SXmatrix with Jacobi'
endif
call lapack_diag(SXeigenval,SXeigenvec,SXmatrix,nMonoEx+1,nMonoEx+1)
endif
if (bavard) then
write(6,*) ' SXdiag : lowest 5 eigenvalues '
write(6,*) ' SXdiag : lowest eigenvalues '
write(6,*) ' 1 - ',SXeigenval(1),SXeigenvec(1,1)
if(nmonoex.gt.0)then
if(n_guess_sx_mat.gt.0)then
write(6,*) ' 2 - ',SXeigenval(2),SXeigenvec(1,2)
write(6,*) ' 3 - ',SXeigenval(3),SXeigenvec(1,3)
write(6,*) ' 4 - ',SXeigenval(4),SXeigenvec(1,4)
@ -77,8 +109,8 @@ END_PROVIDER
best_vector_ovrlp_casscf = -1000
do i=1,nMonoEx+1
if (SXeigenval(i).lt.0.D0) then
if (abs(SXeigenvec(1,i)).gt.best_overlap_casscf) then
best_overlap_casscf=abs(SXeigenvec(1,i))
if (dabs(SXeigenvec(1,i)).gt.best_overlap_casscf) then
best_overlap_casscf=dabs(SXeigenvec(1,i))
best_vector_ovrlp_casscf = i
end if
end if

4
devel/casscf/swap_orb.irp.f
View File

@ -45,12 +45,12 @@
imono = max_overlap(i)
iorb = excit(1,imono)
jorb = excit(2,imono)
if (excit_class(imono) == "c-a" .and.hessmat2(imono,imono).gt.0.d0)then ! core --> active rotation
if (excit_class(imono) == "c-a" .and.hessmat(imono,imono).gt.0.d0)then ! core --> active rotation
n_orb_swap += 1
orb_swap(1,n_orb_swap) = iorb ! core
orb_swap(2,n_orb_swap) = jorb ! active
index_orb_swap(n_orb_swap) = imono
else if (excit_class(imono) == "a-v" .and.hessmat2(imono,imono).gt.0.d0)then ! active --> virtual rotation
else if (excit_class(imono) == "a-v" .and.hessmat(imono,imono).gt.0.d0)then ! active --> virtual rotation
n_orb_swap += 1
orb_swap(1,n_orb_swap) = jorb ! virtual
orb_swap(2,n_orb_swap) = iorb ! active