10
0
mirror of https://github.com/LCPQ/quantum_package synced 2024-06-26 15:12:14 +02:00

Finalized DIIS

This commit is contained in:
Anthony Scemama 2017-06-02 23:53:44 +02:00
parent 54b5e86608
commit de4ba1961f
4 changed files with 126 additions and 145 deletions

View File

@ -1,68 +1,74 @@
begin_template
BEGIN_PROVIDER [ double precision, threshold_DIIS_nonzero ]
implicit none
BEGIN_DOC
! If threshold_DIIS is zero, choose sqrt(thresh_scf)
END_DOC
if (threshold_DIIS == 0.d0) then
threshold_DIIS_nonzero = dsqrt(thresh_scf)
else
threshold_DIIS_nonzero = threshold_DIIS
endif
ASSERT (threshold_DIIS_nonzero >= 0.d0)
begin_provider [double precision, FPS_SPF_Matrix_AO_$alpha, (AO_num, AO_num)]
END_PROVIDER
BEGIN_PROVIDER [double precision, FPS_SPF_Matrix_AO, (AO_num, AO_num)]
implicit none
begin_doc
! Commutator FPS - SPF
end_doc
double precision, allocatable:: scratch(:,:)
allocate( &
scratch(AO_num_align, AO_num) &
)
BEGIN_DOC
! Commutator FPS - SPF
END_DOC
double precision, allocatable :: scratch(:,:)
allocate( &
scratch(AO_num_align, AO_num) &
)
! Compute FP
call dgemm('N','N',AO_num,AO_num,AO_num, &
1.d0, &
Fock_Matrix_AO,Size(Fock_Matrix_AO,1), &
HF_Density_Matrix_AO,Size(HF_Density_Matrix_AO,1), &
0.d0, &
scratch,Size(scratch,1))
! Compute FPS
call dgemm('N','N',AO_num,AO_num,AO_num, &
1.d0, &
scratch,Size(scratch,1), &
AO_Overlap,Size(AO_Overlap,1), &
0.d0, &
FPS_SPF_Matrix_AO,Size(FPS_SPF_Matrix_AO,1))
! Compute SP
call dgemm('N','N',AO_num,AO_num,AO_num, &
1.d0, &
AO_Overlap,Size(AO_Overlap,1), &
HF_Density_Matrix_AO,Size(HF_Density_Matrix_AO,1), &
0.d0, &
scratch,Size(scratch,1))
! Compute FPS - SPF
call dgemm('N','N',AO_num,AO_num,AO_num, &
-1.d0, &
scratch,Size(scratch,1), &
Fock_Matrix_AO,Size(Fock_Matrix_AO,1), &
1.d0, &
FPS_SPF_Matrix_AO,Size(FPS_SPF_Matrix_AO,1))
! Compute FP
END_PROVIDER
call dgemm('N','N',AO_num,AO_num,AO_num, &
1.d0, &
Fock_Matrix_AO_$alpha,Size(Fock_Matrix_AO_$alpha,1), &
HF_Density_Matrix_AO_$alpha,Size(HF_Density_Matrix_AO_$alpha,1), &
0.d0, &
scratch,Size(scratch,1))
! Compute FPS
call dgemm('N','N',AO_num,AO_num,AO_num, &
1.d0, &
scratch,Size(scratch,1), &
AO_Overlap,Size(AO_Overlap,1), &
0.d0, &
FPS_SPF_Matrix_AO_$alpha,Size(FPS_SPF_Matrix_AO_$alpha,1))
! Compute SP
call dgemm('N','N',AO_num,AO_num,AO_num, &
1.d0, &
AO_Overlap,Size(AO_Overlap,1), &
HF_Density_Matrix_AO_$alpha,Size(HF_Density_Matrix_AO_$alpha,1), &
0.d0, &
scratch,Size(scratch,1))
! Compute FPS - SPF
call dgemm('N','N',AO_num,AO_num,AO_num, &
-1.d0, &
scratch,Size(scratch,1), &
Fock_Matrix_AO_$alpha,Size(Fock_Matrix_AO_$alpha,1), &
1.d0, &
FPS_SPF_Matrix_AO_$alpha,Size(FPS_SPF_Matrix_AO_$alpha,1))
end_provider
begin_provider [double precision, FPS_SPF_Matrix_MO_$alpha, (mo_tot_num, mo_tot_num)]
bEGIN_PROVIDER [double precision, FPS_SPF_Matrix_MO, (mo_tot_num, mo_tot_num)]
implicit none
begin_doc
! Commutator FPS - SPF in MO basis
end_doc
call ao_to_mo(FPS_SPF_Matrix_AO_$alpha, size(FPS_SPF_Matrix_AO_$alpha,1), &
FPS_SPF_Matrix_MO_$alpha, size(FPS_SPF_Matrix_MO_$alpha,1))
end_provider
call ao_to_mo(FPS_SPF_Matrix_AO, size(FPS_SPF_Matrix_AO,1), &
FPS_SPF_Matrix_MO, size(FPS_SPF_Matrix_MO,1))
END_PROVIDER
subst [alpha]
alpha ;;
beta ;;
end_template
BEGIN_PROVIDER [ double precision, eigenvalues_Fock_matrix_AO, (AO_num) ]
&BEGIN_PROVIDER [ double precision, eigenvectors_Fock_matrix_AO, (AO_num_align,AO_num) ]
@ -78,11 +84,11 @@ end_template
integer :: i,j
lwork = 3*AO_num - 1
allocate( &
scratch(AO_num_align,AO_num), &
work(lwork), &
Xt(AO_num,AO_num) &
)
allocate( &
scratch(AO_num_align,AO_num), &
work(lwork), &
Xt(AO_num,AO_num) &
)
! Calculate Xt
@ -162,10 +168,11 @@ BEGIN_PROVIDER [ double precision, X_matrix_AO, (AO_num_align,AO_num) ]
num_linear_dependencies = 0
do i=1,AO_num
print*,D(i)
if(abs(D(i)) < threshold_overlap_AO_eigenvalues) then
if(abs(D(i)) <= threshold_overlap_AO_eigenvalues) then
D(i) = 0.d0
num_linear_dependencies += 1
else
ASSERT (D(i) > 0.d0)
D(i) = 1.d0/sqrt(D(i))
endif
do j=1,AO_num

View File

@ -12,13 +12,13 @@ default: 15
[threshold_diis]
type: Threshold
doc: Threshold on the convergence of the DIIS error vector during a Hartree-Fock calculation
doc: Threshold on the convergence of the DIIS error vector during a Hartree-Fock calculation. If 0. is chosen, the square root of thresh_scf will be used.
interface: ezfio,provider,ocaml
default: 1.e-6
default: 0.
[thresh_scf]
type: Threshold
doc: Threshold on the convergence of the Hartree Fock energy
doc: Threshold on the convergence of the Hartree Fock energy.
interface: ezfio,provider,ocaml
default: 1.e-12
@ -32,11 +32,11 @@ default: 128
type: Positive_float
doc: Energy shift on the virtual MOs to improve SCF convergence
interface: ezfio,provider,ocaml
default: 0.4
default: 0.0
[scf_algorithm]
type: character*(32)
doc: Type of SCF algorithm used. Possible choices are [ Damp | DIIS]
doc: Type of SCF algorithm used. Possible choices are [ Simple | DIIS]
interface: ezfio,provider,ocaml
default: DIIS

View File

@ -8,20 +8,16 @@ END_DOC
double precision :: energy_SCF,energy_SCF_previous,Delta_energy_SCF
double precision :: max_error_DIIS,max_error_DIIS_alpha,max_error_DIIS_beta
double precision, allocatable :: Fock_matrix_DIIS_alpha(:,:,:),error_matrix_DIIS_alpha(:,:,:)
double precision, allocatable :: Fock_matrix_DIIS_beta (:,:,:),error_matrix_DIIS_beta (:,:,:)
double precision, allocatable :: Fock_matrix_DIIS(:,:,:),error_matrix_DIIS(:,:,:)
integer :: iteration_SCF,dim_DIIS,index_dim_DIIS
integer :: dim_DIIS_alpha, dim_DIIS_beta
integer :: i,j
allocate( &
Fock_matrix_DIIS_alpha(ao_num,ao_num,max_dim_DIIS), &
Fock_matrix_DIIS_beta (ao_num,ao_num,max_dim_DIIS), &
error_matrix_DIIS_alpha(ao_num,ao_num,max_dim_DIIS), &
error_matrix_DIIS_beta (ao_num,ao_num,max_dim_DIIS) &
)
allocate( &
Fock_matrix_DIIS (ao_num,ao_num,max_dim_DIIS), &
error_matrix_DIIS(ao_num,ao_num,max_dim_DIIS) &
)
call write_time(output_hartree_fock)
@ -35,17 +31,16 @@ END_DOC
! Initialize energies and density matrices
energy_SCF_previous = HF_energy
Delta_energy_SCF = 0.d0
iteration_SCF = 0
dim_DIIS_alpha = 0
dim_DIIS_beta = 0
dim_DIIS = 0
max_error_DIIS = 1.d0
Delta_energy_SCF = 1.d0
iteration_SCF = 0
dim_DIIS = 0
max_error_DIIS = 1.d0
!
! Start of main SCF loop
!
do while((max_error_DIIS > threshold_DIIS) .and. (iteration_SCF < n_it_SCF_max))
do while((max_error_DIIS > threshold_DIIS_nonzero) .and. (iteration_SCF < n_it_SCF_max) .and. dabs(Delta_energy_SCF) > thresh_SCF)
! Increment cycle number
@ -55,46 +50,38 @@ END_DOC
dim_DIIS = min(dim_DIIS+1,max_dim_DIIS)
! Store Fock and error matrices at each iteration
do j=1,ao_num
do i=1,ao_num
index_dim_DIIS = mod(dim_DIIS-1,max_dim_DIIS)+1
Fock_matrix_DIIS_alpha (i,j,index_dim_DIIS) = Fock_matrix_AO_alpha(i,j)
Fock_matrix_DIIS_beta (i,j,index_dim_DIIS) = Fock_matrix_AO_beta (i,j)
error_matrix_DIIS_alpha(i,j,index_dim_DIIS) = FPS_SPF_matrix_AO_alpha(i,j)
error_matrix_DIIS_beta (i,j,index_dim_DIIS) = FPS_SPF_matrix_AO_beta (i,j)
if (scf_algorithm == 'DIIS') then
! Store Fock and error matrices at each iteration
do j=1,ao_num
do i=1,ao_num
index_dim_DIIS = mod(dim_DIIS-1,max_dim_DIIS)+1
Fock_matrix_DIIS (i,j,index_dim_DIIS) = Fock_matrix_AO(i,j)
error_matrix_DIIS(i,j,index_dim_DIIS) = FPS_SPF_matrix_AO(i,j)
enddo
enddo
enddo
! Compute the extrapolated Fock matrix
! Compute the extrapolated Fock matrix
dim_DIIS_alpha = dim_DIIS
call extrapolate_Fock_matrix( &
error_matrix_DIIS_alpha,Fock_matrix_DIIS_alpha, &
Fock_matrix_AO_alpha,size(Fock_matrix_AO_alpha,1), &
iteration_SCF,dim_DIIS_alpha &
)
call extrapolate_Fock_matrix( &
error_matrix_DIIS,Fock_matrix_DIIS, &
Fock_matrix_AO,size(Fock_matrix_AO,1), &
iteration_SCF,dim_DIIS &
)
dim_DIIS_beta = dim_DIIS
call extrapolate_Fock_matrix( &
error_matrix_DIIS_beta,Fock_matrix_DIIS_beta, &
Fock_matrix_AO_beta,size(Fock_matrix_AO_beta,1), &
iteration_SCF,dim_DIIS_beta &
)
Fock_matrix_AO_alpha = Fock_matrix_AO*0.5d0
Fock_matrix_AO_Beta = Fock_matrix_AO*0.5d0
touch Fock_matrix_AO_alpha Fock_matrix_AO_beta
dim_DIIS = min(dim_DIIS_alpha,dim_DIIS_beta)
touch Fock_matrix_AO_alpha Fock_matrix_AO_beta
endif
MO_coef = eigenvectors_Fock_matrix_MO
touch MO_coef
! Calculate error vectors
! Calculate error vectors
max_error_DIIS_alpha = maxval(Abs(FPS_SPF_Matrix_MO_alpha))
max_error_DIIS_beta = maxval(Abs(FPS_SPF_Matrix_MO_beta ))
max_error_DIIS = max(max_error_DIIS_alpha,max_error_DIIS_beta)
max_error_DIIS = maxval(Abs(FPS_SPF_Matrix_MO))
! SCF energy
@ -102,11 +89,15 @@ END_DOC
Delta_Energy_SCF = energy_SCF - energy_SCF_previous
energy_SCF_previous = energy_SCF
! Print results at the end of each iteration
! Print results at the end of each iteration
write(output_hartree_fock,'(I4, 1X, F16.10, 1X, F16.10, 1X, F16.10, 1X, I3)') &
iteration_SCF, energy_SCF, Delta_energy_SCF, max_error_DIIS, dim_DIIS
if (Delta_energy_SCF < 0.d0) then
call save_mos
endif
enddo
!
@ -178,7 +169,7 @@ END_DOC
B_matrix_DIIS(i,j) = 0.d0
do k=1,ao_num
B_matrix_DIIS(i,j) += scratch(k,k)
B_matrix_DIIS(i,j) = B_matrix_DIIS(i,j) + scratch(k,k)
enddo
enddo
enddo
@ -225,29 +216,27 @@ END_DOC
stop 'bug in DIIS'
endif
if (rcond > 1.d-8) then
if (rcond > 1.d-14) then
! Compute extrapolated Fock matrix
! Compute extrapolated Fock matrix
Fock_matrix_AO_(:,:) = 0.d0
do k=1,dim_DIIS
!$OMP PARALLEL DO PRIVATE(i,j,k) DEFAULT(SHARED)
do j=1,ao_num
do i=1,ao_num
Fock_matrix_AO_(i,j) += X_vector_DIIS(k)*Fock_matrix_DIIS(i,j,dim_DIIS-k+1)
Fock_matrix_AO_(i,j) = 0.d0
enddo
do k=1,dim_DIIS
do i=1,ao_num
Fock_matrix_AO_(i,j) = Fock_matrix_AO_(i,j) + &
X_vector_DIIS(k)*Fock_matrix_DIIS(i,j,dim_DIIS-k+1)
enddo
enddo
enddo
enddo
!$OMP END PARALLEL DO
else
write(*,*) 'Re-initialize DIIS!!'
dim_DIIS = 0
endif
! do i=1,ao_num
! do j=1,ao_num
! write(*,*) Fock_matrix_AO_(i,j)
! enddo
! enddo
end

View File

@ -56,24 +56,9 @@ subroutine run
! Choose SCF algorithm
if(scf_algorithm == 'Damp') then
call damping_SCF
else if(scf_algorithm == 'DIIS') then
call Roothaan_Hall_SCF
else
write(*,*) 'Unrecognized SCF algorithm: '//scf_algorithm
stop 1
endif
! call damping_SCF
call Roothaan_Hall_SCF
end
subroutine debug
implicit none
integer :: i,j
do i=1,mo_tot_num
do j=1,mo_tot_num
print *, i, j, fps_spf_matrix_mo_alpha(i,j)
enddo
enddo
end