10
1
mirror of https://github.com/pfloos/quack synced 2024-11-04 21:23:55 +01:00

CCD variants

This commit is contained in:
Pierre-Francois Loos 2020-03-21 22:50:43 +01:00
parent a3eef79eed
commit eee1a1c6b4
12 changed files with 492 additions and 206 deletions

View File

@ -1,74 +1,9 @@
1 10
S 8
1 19500.0000000 0.0005070
2 2923.0000000 0.0039230
3 664.5000000 0.0202000
4 187.5000000 0.0790100
5 60.6200000 0.2304390
6 21.4200000 0.4328720
7 7.9500000 0.3499640
8 0.8815000 -0.0078920
S 8
1 19500.0000000 -0.0001170
2 2923.0000000 -0.0009120
3 664.5000000 -0.0047170
4 187.5000000 -0.0190860
5 60.6200000 -0.0596550
6 21.4200000 -0.1400100
7 7.9500000 -0.1767820
8 0.8815000 0.6050430
1 3
S 3
1 38.3600000 0.0238090
2 5.7700000 0.1548910
3 1.2400000 0.4699870
S 1
1 2.2570000 1.0000000
S 1
1 0.3041000 1.0000000
P 3
1 43.8800000 0.0166650
2 9.9260000 0.1044720
3 2.9300000 0.3172600
1 0.2976000 1.0000000
P 1
1 0.9132000 1.0000000
P 1
1 0.2672000 1.0000000
D 1
1 3.1070000 1.0000000
D 1
1 0.8550000 1.0000000
F 1
1 1.9170000 1.0000000
2 10
S 8
1 19500.0000000 0.0005070
2 2923.0000000 0.0039230
3 664.5000000 0.0202000
4 187.5000000 0.0790100
5 60.6200000 0.2304390
6 21.4200000 0.4328720
7 7.9500000 0.3499640
8 0.8815000 -0.0078920
S 8
1 19500.0000000 -0.0001170
2 2923.0000000 -0.0009120
3 664.5000000 -0.0047170
4 187.5000000 -0.0190860
5 60.6200000 -0.0596550
6 21.4200000 -0.1400100
7 7.9500000 -0.1767820
8 0.8815000 0.6050430
S 1
1 2.2570000 1.0000000
S 1
1 0.3041000 1.0000000
P 3
1 43.8800000 0.0166650
2 9.9260000 0.1044720
3 2.9300000 0.3172600
P 1
1 0.9132000 1.0000000
P 1
1 0.2672000 1.0000000
D 1
1 3.1070000 1.0000000
D 1
1 0.8550000 1.0000000
F 1
1 1.9170000 1.0000000
1 1.2750000 1.0000000

View File

@ -2,14 +2,14 @@
T F F
# MP2 MP3 MP2-F12
T F F
# CCD CCSD CCSD(T) ringCCD ladderCCD
F F F F F
# CCD CCSD CCSD(T) drCCD rCCD lCCD pCCD
F F F T T T T
# CIS RPA RPAx ppRPA ADC
T T T F F
F T T T F
# G0F2 evGF2 G0F3 evGF3
F F F F
# G0W0 evGW qsGW
T F F
F F F
# G0T0 evGT qsGT
F F F
# MCMP2

View File

@ -1,5 +1,4 @@
# nAt nEla nElb nCore nRyd
2 9 9 0 0
1 1 1 0 0
# Znuc x y z
F 0. 0. 0.
F 0. 0. 2
He 0.0 0.0 0.0

View File

@ -1,4 +1,3 @@
2
1
F 0.0000000000 0.0000000000 0.0000000000
F 0.0000000000 0.0000000000 1.0583544980
He 0.0000000000 0.0000000000 0.0000000000

View File

@ -1,74 +1,9 @@
1 10
S 8
1 19500.0000000 0.0005070
2 2923.0000000 0.0039230
3 664.5000000 0.0202000
4 187.5000000 0.0790100
5 60.6200000 0.2304390
6 21.4200000 0.4328720
7 7.9500000 0.3499640
8 0.8815000 -0.0078920
S 8
1 19500.0000000 -0.0001170
2 2923.0000000 -0.0009120
3 664.5000000 -0.0047170
4 187.5000000 -0.0190860
5 60.6200000 -0.0596550
6 21.4200000 -0.1400100
7 7.9500000 -0.1767820
8 0.8815000 0.6050430
1 3
S 3
1 38.3600000 0.0238090
2 5.7700000 0.1548910
3 1.2400000 0.4699870
S 1
1 2.2570000 1.0000000
S 1
1 0.3041000 1.0000000
P 3
1 43.8800000 0.0166650
2 9.9260000 0.1044720
3 2.9300000 0.3172600
1 0.2976000 1.0000000
P 1
1 0.9132000 1.0000000
P 1
1 0.2672000 1.0000000
D 1
1 3.1070000 1.0000000
D 1
1 0.8550000 1.0000000
F 1
1 1.9170000 1.0000000
2 10
S 8
1 19500.0000000 0.0005070
2 2923.0000000 0.0039230
3 664.5000000 0.0202000
4 187.5000000 0.0790100
5 60.6200000 0.2304390
6 21.4200000 0.4328720
7 7.9500000 0.3499640
8 0.8815000 -0.0078920
S 8
1 19500.0000000 -0.0001170
2 2923.0000000 -0.0009120
3 664.5000000 -0.0047170
4 187.5000000 -0.0190860
5 60.6200000 -0.0596550
6 21.4200000 -0.1400100
7 7.9500000 -0.1767820
8 0.8815000 0.6050430
S 1
1 2.2570000 1.0000000
S 1
1 0.3041000 1.0000000
P 3
1 43.8800000 0.0166650
2 9.9260000 0.1044720
3 2.9300000 0.3172600
P 1
1 0.9132000 1.0000000
P 1
1 0.2672000 1.0000000
D 1
1 3.1070000 1.0000000
D 1
1 0.8550000 1.0000000
F 1
1 1.9170000 1.0000000
1 1.2750000 1.0000000

View File

@ -7,7 +7,7 @@ program QuAcK
logical :: doRHF,doUHF,doMOM
logical :: doMP2,doMP3,doMP2F12
logical :: doCCD,doCCSD,doCCSDT
logical :: do_ring_CCD,do_ladder_CCD
logical :: do_drCCD,do_rCCD,do_lCCD,do_pCCD
logical :: doCIS,doRPA,doRPAx
logical :: doppRPA,doADC
logical :: doG0F2,doevGF2,doG0F3,doevGF3
@ -114,15 +114,15 @@ program QuAcK
! Which calculations do you want to do?
call read_methods(doRHF,doUHF,doMOM, &
doMP2,doMP3,doMP2F12, &
doCCD,doCCSD,doCCSDT, &
do_ring_CCD,do_ladder_CCD, &
doCIS,doRPA,doRPAx, &
doppRPA,doADC, &
doG0F2,doevGF2,doG0F3,doevGF3, &
doG0W0,doevGW,doqsGW, &
doG0T0,doevGT,doqsGT, &
call read_methods(doRHF,doUHF,doMOM, &
doMP2,doMP3,doMP2F12, &
doCCD,doCCSD,doCCSDT, &
do_drCCD,do_rCCD,do_lCCD,do_pCCD, &
doCIS,doRPA,doRPAx, &
doppRPA,doADC, &
doG0F2,doevGF2,doG0F3,doevGF3, &
doG0W0,doevGW,doqsGW, &
doG0T0,doevGT,doqsGT, &
doMCMP2)
! Read options for methods
@ -392,14 +392,30 @@ program QuAcK
end if
!------------------------------------------------------------------------
! Perform direct ring CCD calculation
!------------------------------------------------------------------------
if(do_drCCD) then
call cpu_time(start_CCD)
call drCCD(maxSCF_CC,thresh_CC,n_diis_CC,nBas,nEl,ERI_MO_basis,ENuc,ERHF,eHF)
call cpu_time(end_CCD)
t_CCD = end_CCD - start_CCD
write(*,'(A65,1X,F9.3,A8)') 'Total CPU time for direct ring CCD = ',t_CCD,' seconds'
write(*,*)
end if
!------------------------------------------------------------------------
! Perform ring CCD calculation
!------------------------------------------------------------------------
if(do_ring_CCD) then
if(do_rCCD) then
call cpu_time(start_CCD)
call ring_CCD(maxSCF_CC,thresh_CC,n_diis_CC,nBas,nEl,ERI_MO_basis,ENuc,ERHF,eHF)
call rCCD(maxSCF_CC,thresh_CC,n_diis_CC,nBas,nEl,ERI_MO_basis,ENuc,ERHF,eHF)
call cpu_time(end_CCD)
t_CCD = end_CCD - start_CCD
@ -412,14 +428,30 @@ program QuAcK
! Perform ladder CCD calculation
!------------------------------------------------------------------------
if(do_ladder_CCD) then
if(do_lCCD) then
call cpu_time(start_CCD)
call ladder_CCD(maxSCF_CC,thresh_CC,n_diis_CC,nBas,nEl,ERI_MO_basis,ENuc,ERHF,eHF)
call lCCD(maxSCF_CC,thresh_CC,n_diis_CC,nBas,nEl,ERI_MO_basis,ENuc,ERHF,eHF)
call cpu_time(end_CCD)
t_CCD = end_CCD - start_CCD
write(*,'(A65,1X,F9.3,A8)') 'Total CPU time for CCD = ',t_CCD,' seconds'
write(*,'(A65,1X,F9.3,A8)') 'Total CPU time for ladder CCD = ',t_CCD,' seconds'
write(*,*)
end if
!------------------------------------------------------------------------
! Perform pair CCD calculation
!------------------------------------------------------------------------
if(do_pCCD) then
call cpu_time(start_CCD)
call pCCD(maxSCF_CC,thresh_CC,n_diis_CC,nBas,nEl,ERI_MO_basis,ENuc,ERHF,eHF)
call cpu_time(end_CCD)
t_CCD = end_CCD - start_CCD
write(*,'(A65,1X,F9.3,A8)') 'Total CPU time for pair CCD = ',t_CCD,' seconds'
write(*,*)
end if

187
src/QuAcK/drCCD.f90 Normal file
View File

@ -0,0 +1,187 @@
subroutine drCCD(maxSCF,thresh,max_diis,nBas,nEl,ERI,ENuc,ERHF,eHF)
! Direct ring CCD module
implicit none
! Input variables
integer,intent(in) :: maxSCF
integer,intent(in) :: max_diis
double precision,intent(in) :: thresh
integer,intent(in) :: nBas,nEl
double precision,intent(in) :: ENuc,ERHF
double precision,intent(in) :: eHF(nBas)
double precision,intent(in) :: ERI(nBas,nBas,nBas,nBas)
! Local variables
integer :: nBas2
integer :: nO
integer :: nV
integer :: nSCF
double precision :: Conv
double precision :: EcMP2,EcMP3,EcMP4
double precision :: ECCD,EcCCD
double precision,allocatable :: seHF(:)
double precision,allocatable :: sERI(:,:,:,:)
double precision,allocatable :: eO(:)
double precision,allocatable :: eV(:)
double precision,allocatable :: delta_OOVV(:,:,:,:)
double precision,allocatable :: OOOO(:,:,:,:)
double precision,allocatable :: OOVV(:,:,:,:)
double precision,allocatable :: OVVO(:,:,:,:)
double precision,allocatable :: VVVV(:,:,:,:)
double precision,allocatable :: X1(:,:,:,:)
double precision,allocatable :: X2(:,:)
double precision,allocatable :: X3(:,:)
double precision,allocatable :: X4(:,:,:,:)
double precision,allocatable :: u(:,:,:,:)
double precision,allocatable :: v(:,:,:,:)
double precision,allocatable :: r2(:,:,:,:)
double precision,allocatable :: t2(:,:,:,:)
! Hello world
write(*,*)
write(*,*)'**************************************'
write(*,*)'| direct ring CCD calculation |'
write(*,*)'**************************************'
write(*,*)
! Spatial to spin orbitals
nBas2 = 2*nBas
allocate(seHF(nBas2),sERI(nBas2,nBas2,nBas2,nBas2))
call spatial_to_spin_MO_energy(nBas,eHF,nBas2,seHF)
call spatial_to_spin_ERI(nBas,ERI,nBas2,sERI)
! Antysymmetrize ERIs
! Define occupied and virtual spaces
nO = 2*nEl
nV = nBas2 - nO
! Form energy denominator
allocate(eO(nO),eV(nV))
allocate(delta_OOVV(nO,nO,nV,nV))
eO(:) = seHF(1:nO)
eV(:) = seHF(nO+1:nBas2)
call form_delta_OOVV(nO,nV,eO,eV,delta_OOVV)
deallocate(seHF)
! Create integral batches
allocate(OOOO(nO,nO,nO,nO),OOVV(nO,nO,nV,nV),OVVO(nO,nV,nV,nO),VVVV(nV,nV,nV,nV))
OOOO(:,:,:,:) = sERI( 1:nO , 1:nO , 1:nO , 1:nO )
OOVV(:,:,:,:) = sERI( 1:nO , 1:nO ,nO+1:nBas2,nO+1:nBas2)
OVVO(:,:,:,:) = sERI( 1:nO ,nO+1:nBas2,nO+1:nBas2, 1:nO )
VVVV(:,:,:,:) = sERI(nO+1:nBas2,nO+1:nBas2,nO+1:nBas2,nO+1:nBas2)
deallocate(sERI)
! MP2 guess amplitudes
allocate(t2(nO,nO,nV,nV))
t2(:,:,:,:) = -OOVV(:,:,:,:)/delta_OOVV(:,:,:,:)
EcMP2 = 0.25d0*dot_product(pack(OOVV,.true.),pack(t2,.true.))
EcMP4 = 0d0
! Initialization
allocate(r2(nO,nO,nV,nV),u(nO,nO,nV,nV),v(nO,nO,nV,nV))
allocate(X1(nO,nO,nO,nO),X2(nV,nV),X3(nO,nO),X4(nO,nO,nV,nV))
Conv = 1d0
nSCF = 0
!------------------------------------------------------------------------
! Main SCF loop
!------------------------------------------------------------------------
write(*,*)
write(*,*)'----------------------------------------------------'
write(*,*)'| direct ring CCD calculation |'
write(*,*)'----------------------------------------------------'
write(*,'(1X,A1,1X,A3,1X,A1,1X,A16,1X,A1,1X,A10,1X,A1,1X,A10,1X,A1,1X)') &
'|','#','|','E(CCD)','|','Ec(CCD)','|','Conv','|'
write(*,*)'----------------------------------------------------'
do while(Conv > thresh .and. nSCF < maxSCF)
! Increment
nSCF = nSCF + 1
! Compute residual
call form_ring_r(nO,nV,OVVO,OOVV,t2,r2)
r2(:,:,:,:) = OOVV(:,:,:,:) + delta_OOVV(:,:,:,:)*t2(:,:,:,:) + r2(:,:,:,:)
! Check convergence
Conv = maxval(abs(r2(:,:,:,:)))
! Update amplitudes
t2(:,:,:,:) = t2(:,:,:,:) - r2(:,:,:,:)/delta_OOVV(:,:,:,:)
! Compute correlation energy
EcCCD = 0.5d0*dot_product(pack(OOVV,.true.),pack(t2,.true.))
if(nSCF == 1) EcMP3 = 0.25d0*dot_product(pack(OOVV,.true.),pack(t2 + v/delta_OOVV,.true.))
! Dump results
ECCD = ERHF + EcCCD
write(*,'(1X,A1,1X,I3,1X,A1,1X,F16.10,1X,A1,1X,F10.6,1X,A1,1X,F10.6,1X,A1,1X)') &
'|',nSCF,'|',ECCD+ENuc,'|',EcCCD,'|',Conv,'|'
enddo
write(*,*)'----------------------------------------------------'
!------------------------------------------------------------------------
! End of SCF loop
!------------------------------------------------------------------------
! Did it actually converge?
if(nSCF == maxSCF) then
write(*,*)
write(*,*)'!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!'
write(*,*)' Convergence failed '
write(*,*)'!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!'
write(*,*)
stop
endif
write(*,*)
write(*,*)'----------------------------------------------------'
write(*,*)' direct ring CCD energy '
write(*,*)'----------------------------------------------------'
write(*,'(1X,A30,1X,F15.10)')' E(drCCD) = ',ECCD
write(*,'(1X,A30,1X,F15.10)')' Ec(drCCD) = ',EcCCD
write(*,*)'----------------------------------------------------'
write(*,*)
end subroutine drCCD

View File

@ -1,4 +1,4 @@
subroutine ladder_CCD(maxSCF,thresh,max_diis,nBas,nEl,ERI,ENuc,ERHF,eHF)
subroutine lCCD(maxSCF,thresh,max_diis,nBas,nEl,ERI,ENuc,ERHF,eHF)
! Ladder CCD module
@ -52,7 +52,7 @@ subroutine ladder_CCD(maxSCF,thresh,max_diis,nBas,nEl,ERI,ENuc,ERHF,eHF)
write(*,*)
write(*,*)'**************************************'
write(*,*)'| ladder-CCD calculation |'
write(*,*)'| ladder CCD calculation |'
write(*,*)'**************************************'
write(*,*)
@ -123,7 +123,7 @@ subroutine ladder_CCD(maxSCF,thresh,max_diis,nBas,nEl,ERI,ENuc,ERHF,eHF)
!------------------------------------------------------------------------
write(*,*)
write(*,*)'----------------------------------------------------'
write(*,*)'| ladder-CCD calculation |'
write(*,*)'| ladder CCD calculation |'
write(*,*)'----------------------------------------------------'
write(*,'(1X,A1,1X,A3,1X,A1,1X,A16,1X,A1,1X,A10,1X,A1,1X,A10,1X,A1,1X)') &
'|','#','|','E(CCD)','|','Ec(CCD)','|','Conv','|'
@ -184,11 +184,11 @@ subroutine ladder_CCD(maxSCF,thresh,max_diis,nBas,nEl,ERI,ENuc,ERHF,eHF)
write(*,*)
write(*,*)'----------------------------------------------------'
write(*,*)' ladder-CCD energy '
write(*,*)' ladder CCD energy '
write(*,*)'----------------------------------------------------'
write(*,'(1X,A30,1X,F15.10)')' E(ladder-CCD) = ',ECCD
write(*,'(1X,A30,1X,F15.10)')' Ec(ladder-CCSD) = ',EcCCD
write(*,'(1X,A30,1X,F15.10)')' E(lCCD) = ',ECCD
write(*,'(1X,A30,1X,F15.10)')' Ec(lCCD) = ',EcCCD
write(*,*)'----------------------------------------------------'
write(*,*)
end subroutine ladder_CCD
end subroutine lCCD

195
src/QuAcK/pCCD.f90 Normal file
View File

@ -0,0 +1,195 @@
subroutine pCCD(maxSCF,thresh,max_diis,nBas,nEl,ERI,ENuc,ERHF,eHF)
! pair CCD module
implicit none
! Input variables
integer,intent(in) :: maxSCF
integer,intent(in) :: max_diis
double precision,intent(in) :: thresh
integer,intent(in) :: nBas,nEl
double precision,intent(in) :: ENuc,ERHF
double precision,intent(in) :: eHF(nBas)
double precision,intent(in) :: ERI(nBas,nBas,nBas,nBas)
! Local variables
integer :: nBas2
integer :: nO
integer :: nV
integer :: nSCF
double precision :: Conv
double precision :: EcMP2,EcMP3,EcMP4
double precision :: ECCD,EcCCD
double precision,allocatable :: seHF(:)
double precision,allocatable :: sERI(:,:,:,:)
double precision,allocatable :: dbERI(:,:,:,:)
double precision,allocatable :: eO(:)
double precision,allocatable :: eV(:)
double precision,allocatable :: delta_OOVV(:,:,:,:)
double precision,allocatable :: OOOO(:,:,:,:)
double precision,allocatable :: OOVV(:,:,:,:)
double precision,allocatable :: OVOV(:,:,:,:)
double precision,allocatable :: VVVV(:,:,:,:)
double precision,allocatable :: X1(:,:,:,:)
double precision,allocatable :: X2(:,:)
double precision,allocatable :: X3(:,:)
double precision,allocatable :: X4(:,:,:,:)
double precision,allocatable :: u(:,:,:,:)
double precision,allocatable :: v(:,:,:,:)
double precision,allocatable :: r2(:,:,:,:)
double precision,allocatable :: t2(:,:,:,:)
! Hello world
write(*,*)
write(*,*)'**************************************'
write(*,*)'| CCD calculation |'
write(*,*)'**************************************'
write(*,*)
! Spatial to spin orbitals
nBas2 = 2*nBas
allocate(seHF(nBas2),sERI(nBas2,nBas2,nBas2,nBas2))
call spatial_to_spin_MO_energy(nBas,eHF,nBas2,seHF)
call spatial_to_spin_ERI(nBas,ERI,nBas2,sERI)
! Antysymmetrize ERIs
allocate(dbERI(nBas2,nBas2,nBas2,nBas2))
call antisymmetrize_ERI(2,nBas2,sERI,dbERI)
deallocate(sERI)
! Define occupied and virtual spaces
nO = 2*nEl
nV = nBas2 - nO
! Form energy denominator
allocate(eO(nO),eV(nV))
allocate(delta_OOVV(nO,nO,nV,nV))
eO(:) = seHF(1:nO)
eV(:) = seHF(nO+1:nBas2)
call form_delta_OOVV(nO,nV,eO,eV,delta_OOVV)
deallocate(seHF)
! Create integral batches
allocate(OOOO(nO,nO,nO,nO),OOVV(nO,nO,nV,nV),OVOV(nO,nV,nO,nV),VVVV(nV,nV,nV,nV))
OOOO(:,:,:,:) = dbERI( 1:nO , 1:nO , 1:nO , 1:nO )
OOVV(:,:,:,:) = dbERI( 1:nO , 1:nO ,nO+1:nBas2,nO+1:nBas2)
OVOV(:,:,:,:) = dbERI( 1:nO ,nO+1:nBas2, 1:nO ,nO+1:nBas2)
VVVV(:,:,:,:) = dbERI(nO+1:nBas2,nO+1:nBas2,nO+1:nBas2,nO+1:nBas2)
deallocate(dbERI)
! MP2 guess amplitudes
allocate(t2(nO,nO,nV,nV))
t2(:,:,:,:) = -OOVV(:,:,:,:)/delta_OOVV(:,:,:,:)
EcMP2 = 0.25d0*dot_product(pack(OOVV,.true.),pack(t2,.true.))
EcMP4 = 0d0
! Initialization
allocate(r2(nO,nO,nV,nV),u(nO,nO,nV,nV),v(nO,nO,nV,nV))
allocate(X1(nO,nO,nO,nO),X2(nV,nV),X3(nO,nO),X4(nO,nO,nV,nV))
Conv = 1d0
nSCF = 0
!------------------------------------------------------------------------
! Main SCF loop
!------------------------------------------------------------------------
write(*,*)
write(*,*)'----------------------------------------------------'
write(*,*)'| pair CCD calculation |'
write(*,*)'----------------------------------------------------'
write(*,'(1X,A1,1X,A3,1X,A1,1X,A16,1X,A1,1X,A10,1X,A1,1X,A10,1X,A1,1X)') &
'|','#','|','E(pCCD)','|','Ec(pCCD)','|','Conv','|'
write(*,*)'----------------------------------------------------'
do while(Conv > thresh .and. nSCF < maxSCF)
! Increment
nSCF = nSCF + 1
! Form linear array
call form_u(nO,nV,OOOO,VVVV,OVOV,t2,u)
! Form interemediate arrays
call form_X(nO,nV,OOVV,t2,X1,X2,X3,X4)
! Form quadratic array
call form_v(nO,nV,X1,X2,X3,X4,t2,v)
! Compute residual
r2(:,:,:,:) = OOVV(:,:,:,:) + delta_OOVV(:,:,:,:)*t2(:,:,:,:) + u(:,:,:,:) + v(:,:,:,:)
! Check convergence
Conv = maxval(abs(r2(:,:,:,:)))
! Update amplitudes
t2(:,:,:,:) = t2(:,:,:,:) - r2(:,:,:,:)/delta_OOVV(:,:,:,:)
! Compute correlation energy
EcCCD = 0.25d0*dot_product(pack(OOVV,.true.),pack(t2,.true.))
if(nSCF == 1) EcMP3 = 0.25d0*dot_product(pack(OOVV,.true.),pack(t2 + v/delta_OOVV,.true.))
! Dump results
ECCD = ERHF + EcCCD
write(*,'(1X,A1,1X,I3,1X,A1,1X,F16.10,1X,A1,1X,F10.6,1X,A1,1X,F10.6,1X,A1,1X)') &
'|',nSCF,'|',ECCD+ENuc,'|',EcCCD,'|',Conv,'|'
enddo
write(*,*)'----------------------------------------------------'
!------------------------------------------------------------------------
! End of SCF loop
!------------------------------------------------------------------------
! Did it actually converge?
if(nSCF == maxSCF) then
write(*,*)
write(*,*)'!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!'
write(*,*)' Convergence failed '
write(*,*)'!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!'
write(*,*)
stop
endif
end subroutine pCCD

View File

@ -104,10 +104,10 @@ subroutine ppRPA(singlet_manifold,triplet_manifold,nBas,nC,nO,nV,nR,ENuc,ERHF,ER
write(*,*)
write(*,*)'-------------------------------------------------------------------------------'
write(*,'(2X,A40,F15.6)') 'Tr@ppRPA correlation energy (singlet) =',Ec_ppRPA(1)
write(*,'(2X,A40,F15.6)') 'Tr@ppRPA correlation energy (triplet) =',3d0*Ec_ppRPA(2)
write(*,'(2X,A40,F15.6)') 'Tr@ppRPA correlation energy =',Ec_ppRPA(1) + 3d0*Ec_ppRPA(2)
write(*,'(2X,A40,F15.6)') 'Tr@ppRPA total energy =',ENuc + ERHF + Ec_ppRPA(1) + 3d0*Ec_ppRPA(2)
write(*,'(2X,A50,F20.10)') 'Tr@ppRPA correlation energy (singlet) =',Ec_ppRPA(1)
write(*,'(2X,A50,F20.10)') 'Tr@ppRPA correlation energy (triplet) =',3d0*Ec_ppRPA(2)
write(*,'(2X,A50,F20.10)') 'Tr@ppRPA correlation energy =',Ec_ppRPA(1) + 3d0*Ec_ppRPA(2)
write(*,'(2X,A50,F20.10)') 'Tr@ppRPA total energy =',ENuc + ERHF + Ec_ppRPA(1) + 3d0*Ec_ppRPA(2)
write(*,*)'-------------------------------------------------------------------------------'
write(*,*)

View File

@ -1,4 +1,4 @@
subroutine ring_CCD(maxSCF,thresh,max_diis,nBas,nEl,ERI,ENuc,ERHF,eHF)
subroutine rCCD(maxSCF,thresh,max_diis,nBas,nEl,ERI,ENuc,ERHF,eHF)
! Ring CCD module
@ -52,7 +52,7 @@ subroutine ring_CCD(maxSCF,thresh,max_diis,nBas,nEl,ERI,ENuc,ERHF,eHF)
write(*,*)
write(*,*)'**************************************'
write(*,*)'| ring-CCD calculation |'
write(*,*)'| ring CCD calculation |'
write(*,*)'**************************************'
write(*,*)
@ -123,7 +123,7 @@ subroutine ring_CCD(maxSCF,thresh,max_diis,nBas,nEl,ERI,ENuc,ERHF,eHF)
!------------------------------------------------------------------------
write(*,*)
write(*,*)'----------------------------------------------------'
write(*,*)'| ring-CCD calculation |'
write(*,*)'| ring CCD calculation |'
write(*,*)'----------------------------------------------------'
write(*,'(1X,A1,1X,A3,1X,A1,1X,A16,1X,A1,1X,A10,1X,A1,1X,A10,1X,A1,1X)') &
'|','#','|','E(CCD)','|','Ec(CCD)','|','Conv','|'
@ -184,11 +184,11 @@ subroutine ring_CCD(maxSCF,thresh,max_diis,nBas,nEl,ERI,ENuc,ERHF,eHF)
write(*,*)
write(*,*)'----------------------------------------------------'
write(*,*)' ring-CCD energy '
write(*,*)' ring CCD energy '
write(*,*)'----------------------------------------------------'
write(*,'(1X,A30,1X,F15.10)')' E(ring-CCD) = ',ECCD
write(*,'(1X,A30,1X,F15.10)')' Ec(ring-CCD) = ',EcCCD
write(*,'(1X,A30,1X,F15.10)')' E(rCCD) = ',ECCD
write(*,'(1X,A30,1X,F15.10)')' Ec(rCCD) = ',EcCCD
write(*,*)'----------------------------------------------------'
write(*,*)
end subroutine ring_CCD
end subroutine rCCD

View File

@ -1,12 +1,12 @@
subroutine read_methods(doRHF,doUHF,doMOM, &
doMP2,doMP3,doMP2F12, &
doCCD,doCCSD,doCCSDT, &
do_ring_CCD,do_ladder_CCD, &
doCIS,doRPA,doRPAx, &
doppRPA,doADC, &
doG0F2,doevGF2,doG0F3,doevGF3, &
doG0W0,doevGW,doqsGW, &
doG0T0,doevGT,doqsGT, &
subroutine read_methods(doRHF,doUHF,doMOM, &
doMP2,doMP3,doMP2F12, &
doCCD,doCCSD,doCCSDT, &
do_drCCD,do_rCCD,do_lCCD,do_pCCD, &
doCIS,doRPA,doRPAx, &
doppRPA,doADC, &
doG0F2,doevGF2,doG0F3,doevGF3, &
doG0W0,doevGW,doqsGW, &
doG0T0,doevGT,doqsGT, &
doMCMP2)
! Read desired methods
@ -18,7 +18,7 @@ subroutine read_methods(doRHF,doUHF,doMOM, &
logical,intent(out) :: doRHF,doUHF,doMOM
logical,intent(out) :: doMP2,doMP3,doMP2F12
logical,intent(out) :: doCCD,doCCSD,doCCSDT
logical,intent(out) :: do_ring_CCD,do_ladder_CCD
logical,intent(out) :: do_drCCD,do_rCCD,do_lCCD,do_pCCD
logical,intent(out) :: doCIS,doRPA,doRPAx,doppRPA,doADC
logical,intent(out) :: doG0F2,doevGF2,doG0F3,doevGF3
logical,intent(out) :: doG0W0,doevGW,doqsGW
@ -27,7 +27,7 @@ subroutine read_methods(doRHF,doUHF,doMOM, &
! Local variables
character(len=1) :: answer1,answer2,answer3,answer4,answer5
character(len=1) :: answer1,answer2,answer3,answer4,answer5,answer6,answer7
! Open file with method specification
@ -47,8 +47,10 @@ subroutine read_methods(doRHF,doUHF,doMOM, &
doCCSD = .false.
doCCSDT = .false.
do_ring_CCD = .false.
do_ladder_CCD = .false.
do_drCCD = .false.
do_rCCD = .false.
do_lCCD = .false.
do_pCCD = .false.
doCIS = .false.
doRPA = .false.
@ -90,12 +92,14 @@ subroutine read_methods(doRHF,doUHF,doMOM, &
! Read CC methods
read(1,*)
read(1,*) answer1,answer2,answer3,answer4,answer5
if(answer1 == 'T') doCCD = .true.
if(answer2 == 'T') doCCSD = .true.
if(answer3 == 'T') doCCSDT = .true.
if(answer4 == 'T') do_ring_CCD = .true.
if(answer5 == 'T') do_ladder_CCD = .true.
read(1,*) answer1,answer2,answer3,answer4,answer5,answer6,answer7
if(answer1 == 'T') doCCD = .true.
if(answer2 == 'T') doCCSD = .true.
if(answer3 == 'T') doCCSDT = .true.
if(answer4 == 'T') do_drCCD = .true.
if(answer5 == 'T') do_rCCD = .true.
if(answer6 == 'T') do_lCCD = .true.
if(answer7 == 'T') do_pCCD = .true.
! Read excited state methods