4
1
mirror of https://github.com/pfloos/quack synced 2024-12-22 04:14:26 +01:00

saving worj in ccGW

This commit is contained in:
Pierre-Francois Loos 2024-09-30 23:06:00 +02:00
parent 80f393ded2
commit 172ca40f37
4 changed files with 337 additions and 358 deletions

View File

@ -162,12 +162,13 @@ subroutine RGW(dotest,doG0W0,doevGW,doqsGW,doufG0W0,doufGW,maxSCF,thresh,max_dii
! Perform CC-based G0W0 calculation
!------------------------------------------------------------------------
doccG0W0 = .false.
doccG0W0 = .true.
if(doccG0W0) then
call wall_time(start_GW)
call ccRG0W0(maxSCF,thresh,max_diis,nBas,nOrb,nC,nO,nV,nR,ERI_MO,ENuc,ERHF,eHF)
call ccRG0W0(maxSCF,thresh,max_diis,nBas,nOrb,nC,nO,nV,nR,nS,ERI_MO,ENuc,ERHF,eHF)
! call ccRG0W0_TDA(maxSCF,thresh,max_diis,nBas,nOrb,nC,nO,nV,nR,ERI_MO,ENuc,ERHF,eHF)
call wall_time(end_GW)
t_GW = end_GW - start_GW
@ -186,8 +187,7 @@ subroutine RGW(dotest,doG0W0,doevGW,doqsGW,doufG0W0,doufGW,maxSCF,thresh,max_dii
if(doccGW) then
call wall_time(start_GW)
! call ccRGW(maxSCF,thresh,nBas,nOrb,nC,nO,nV,nR,ERI_MO,ENuc,ERHF,eHF)
! call ccRGW_mat(maxSCF,thresh,nBas,nOrb,nC,nO,nV,nR,ERI_MO,ENuc,ERHF,eHF)
call ccRGW(maxSCF,thresh,nBas,nOrb,nC,nO,nV,nR,ERI_MO,ENuc,ERHF,eHF)
call wall_time(end_GW)
t_GW = end_GW - start_GW

View File

@ -1,4 +1,4 @@
subroutine ccRG0W0(maxSCF,thresh,max_diis,nBas,nOrb,nC,nO,nV,nR,ERI,ENuc,ERHF,eHF)
subroutine ccRG0W0(maxSCF,thresh,max_diis,nBas,nOrb,nC,nO,nV,nR,nS,ERI,ENuc,ERHF,eHF)
! CC-based GW module
@ -17,6 +17,7 @@ subroutine ccRG0W0(maxSCF,thresh,max_diis,nBas,nOrb,nC,nO,nV,nR,ERI,ENuc,ERHF,eH
integer,intent(in) :: nO
integer,intent(in) :: nV
integer,intent(in) :: nR
integer,intent(in) :: nS
double precision,intent(in) :: ENuc
double precision,intent(in) :: ERHF
double precision,intent(in) :: eHF(nOrb)
@ -27,17 +28,29 @@ subroutine ccRG0W0(maxSCF,thresh,max_diis,nBas,nOrb,nC,nO,nV,nR,ERI,ENuc,ERHF,eH
integer :: p,q,r,s
integer :: i,j,k,l
integer :: a,b,c,d
integer :: m
integer :: isp_W
logical :: TDA_W
logical :: dRPA
integer :: nSCF
double precision :: Conv
double precision :: EcRPA
double precision,allocatable :: Sig(:)
double precision,allocatable :: Z(:)
double precision,allocatable :: del(:,:,:)
double precision,allocatable :: vec(:,:,:)
double precision,allocatable :: res(:,:,:)
double precision,allocatable :: amp(:,:,:)
double precision,allocatable :: del(:,:)
double precision,allocatable :: vec(:,:)
double precision,allocatable :: res(:,:)
double precision,allocatable :: amp(:,:)
double precision,allocatable :: Aph(:,:)
double precision,allocatable :: Bph(:,:)
double precision,allocatable :: Om(:)
double precision,allocatable :: XpY(:,:)
double precision,allocatable :: XmY(:,:)
double precision,allocatable :: rho(:,:,:)
integer :: n_diis
double precision :: rcond
@ -55,16 +68,39 @@ subroutine ccRG0W0(maxSCF,thresh,max_diis,nBas,nOrb,nC,nO,nV,nR,ERI,ENuc,ERHF,eH
! Memory allocation
allocate(del(nO,nV,nOrb))
allocate(vec(nO,nV,nOrb))
allocate(res(nO,nV,nOrb))
allocate(amp(nO,nV,nOrb))
allocate(del(nS,nOrb))
allocate(vec(nS,nOrb))
allocate(res(nS,nOrb))
allocate(amp(nS,nOrb))
allocate(Sig(nOrb))
allocate(Z(nOrb))
allocate(r_diis(nO*nV*nOrb,max_diis))
allocate(t_diis(nO*nV*nOrb,max_diis))
allocate(r_diis(nS*nOrb,max_diis))
allocate(t_diis(nS*nOrb,max_diis))
!-------------------!
! Compute screening !
!-------------------!
! Spin manifold
isp_W = 1
TDA_W = .false.
dRPA = .true.
! Memory allocation
allocate(Om(nS),Aph(nS,nS),Bph(nS,nS),XpY(nS,nS),XmY(nS,nS),rho(nOrb,nOrb,nS))
call phLR_A(isp_W,dRPA,nOrb,nC,nO,nV,nR,nS,1d0,eHF,ERI,Aph)
call phLR_B(isp_W,dRPA,nOrb,nC,nO,nV,nR,nS,1d0,ERI,Bph)
call phLR(TDA_W,nS,Aph,Bph,EcRPA,Om,XpY,XmY)
call RGW_excitation_density(nOrb,nC,nO,nR,nS,ERI,XpY,rho)
deallocate(Aph,Bph,XpY,XmY)
! Initialization
@ -87,48 +123,26 @@ subroutine ccRG0W0(maxSCF,thresh,max_diis,nBas,nOrb,nC,nO,nV,nR,ERI,ENuc,ERHF,eH
r_diis(:,:) = 0d0
rcond = 0d0
amp(:,:,:) = 0d0
res(:,:,:) = 0d0
amp(:,:) = 0d0
res(:,:) = 0d0
! Compute energy differences
! Compute energy differences and coupling blocks
do i=nC+1,nO
do m=1,nS
do j=nC+1,nO
do a=1,nV-nR
del(i,a,j) = eHF(i) + eHF(j) - eHF(nO+a) - eHF(p)
del(m,j) = Om(m) + eHF(j) - eHF(p)
vec(m,j) = sqrt(2d0)*rho(p,j,m)
end do
end do
end do
do i=nC+1,nO
do a=1,nV-nR
do b=1,nV-nR
do m=1,nS
do b=1,nV-nR
del(i,a,nO+b) = eHF(nO+a) + eHF(nO+b) - eHF(i) - eHF(p)
del(m,nO+b) = Om(m) + eHF(nO+b) - eHF(p)
vec(m,nO+b) = sqrt(2d0)*rho(p,nO+b,m)
end do
end do
end do
do i=nC+1,nO
do a=1,nV-nR
do j=nC+1,nO
vec(i,a,j) = sqrt(2d0)*ERI(p,nO+a,i,j)
end do
end do
end do
do i=nC+1,nO
do a=1,nV-nR
do b=1,nV-nR
vec(i,a,nO+b) = sqrt(2d0)*ERI(p,i,nO+b,nO+a)
end do
end do
end do
@ -152,41 +166,23 @@ subroutine ccRG0W0(maxSCF,thresh,max_diis,nBas,nOrb,nC,nO,nV,nR,ERI,ENuc,ERHF,eH
! Compute residual for 2h1p sector
res(:,:,:) = vec(:,:,:) + (del(:,:,:) - Sig(p))*amp(:,:,:)
res(:,:) = vec(:,:) + (del(:,:) - Sig(p))*amp(:,:)
do i=nC+1,nO
do a=1,nV-nR
do j=nC+1,nO
do m=1,nS
do j=nC+1,nO
do k=nC+1,nO
do c=1,nV-nR
res(m,j) = res(m,j) - Om(m)*amp(m,j)
res(i,a,j) = res(i,a,j) - 2d0*ERI(j,nO+c,nO+a,k)*amp(i,c,k) &
- 2d0*ERI(i,nO+c,nO+a,k)*amp(k,c,j)
end do
end do
end do
end do
end do
! Compute residual for 2p1h sector
do i=nC+1,nO
do a=1,nV-nR
do b=1,nV-nR
do m=nC+1,nO
do b=1,nV-nR
do k=nC+1,nO
do c=1,nV-nR
res(i,a,nO+b) = res(i,a,nO+b) + 2d0*ERI(nO+a,k,i,nO+c)*amp(k,c,nO+b) &
+ 2d0*ERI(nO+b,k,i,nO+c)*amp(k,a,nO+c)
end do
end do
end do
res(m,nO+b) = res(m,nO+b) + Om(m)*amp(m,nO+b)
end do
end do
@ -196,14 +192,14 @@ subroutine ccRG0W0(maxSCF,thresh,max_diis,nBas,nOrb,nC,nO,nV,nR,ERI,ENuc,ERHF,eH
! Update amplitudes
amp(:,:,:) = amp(:,:,:) - res(:,:,:)/del(:,:,:)
amp(:,:) = amp(:,:) - res(:,:)/del(:,:)
! DIIS extrapolation
if(max_diis > 1) then
n_diis = min(n_diis+1,max_diis)
call DIIS_extrapolation(rcond,nO*nV*nOrb,nO*nV*nOrb,n_diis,r_diis,t_diis,res,amp)
call DIIS_extrapolation(rcond,nS*nOrb,nS*nOrb,n_diis,r_diis,t_diis,res,amp)
end if
@ -211,13 +207,11 @@ subroutine ccRG0W0(maxSCF,thresh,max_diis,nBas,nOrb,nC,nO,nV,nR,ERI,ENuc,ERHF,eH
Sig(p) = 0d0
do i=nC+1,nO
do a=1,nV-nR
do q=nC+1,nOrb-nR
do m=1,nS
do q=nC+1,nOrb-nR
Sig(p) = Sig(p) + vec(i,a,q)*amp(i,a,q)
Sig(p) = Sig(p) + vec(m,q)*amp(m,q)
end do
end do
end do

263
src/GW/ccRG0W0_TDA.f90 Normal file
View File

@ -0,0 +1,263 @@
subroutine ccRG0W0_TDA(maxSCF,thresh,max_diis,nBas,nOrb,nC,nO,nV,nR,ERI,ENuc,ERHF,eHF)
! CC-based GW module (TDA screening)
implicit none
include 'parameters.h'
! Input variables
integer,intent(in) :: maxSCF
double precision,intent(in) :: thresh
integer,intent(in) :: max_diis
integer,intent(in) :: nBas
integer,intent(in) :: nOrb
integer,intent(in) :: nC
integer,intent(in) :: nO
integer,intent(in) :: nV
integer,intent(in) :: nR
double precision,intent(in) :: ENuc
double precision,intent(in) :: ERHF
double precision,intent(in) :: eHF(nOrb)
double precision,intent(in) :: ERI(nOrb,nOrb,nOrb,nOrb)
! Local variables
integer :: p,q,r,s
integer :: i,j,k,l
integer :: a,b,c,d
integer :: nSCF
double precision :: Conv
double precision,allocatable :: Sig(:)
double precision,allocatable :: Z(:)
double precision,allocatable :: del(:,:,:)
double precision,allocatable :: vec(:,:,:)
double precision,allocatable :: res(:,:,:)
double precision,allocatable :: amp(:,:,:)
integer :: n_diis
double precision :: rcond
double precision,allocatable :: r_diis(:,:)
double precision,allocatable :: t_diis(:,:)
! Hello world
write(*,*)
write(*,*)'*****************************'
write(*,*)'* CC-based G0W0 Calculation *'
write(*,*)'*****************************'
write(*,*)
! Memory allocation
allocate(del(nO,nV,nOrb))
allocate(vec(nO,nV,nOrb))
allocate(res(nO,nV,nOrb))
allocate(amp(nO,nV,nOrb))
allocate(Sig(nOrb))
allocate(Z(nOrb))
allocate(r_diis(nO*nV*nOrb,max_diis))
allocate(t_diis(nO*nV*nOrb,max_diis))
! Initialization
Sig(:) = 0d0
Z(:) = 1d0
!-------------------------!
! Main loop over orbitals !
!-------------------------!
do p=nO,nO+1
! Initialization
Conv = 1d0
nSCF = 0
n_diis = 0
t_diis(:,:) = 0d0
r_diis(:,:) = 0d0
rcond = 0d0
amp(:,:,:) = 0d0
res(:,:,:) = 0d0
! Compute energy differences
do i=nC+1,nO
do j=nC+1,nO
do a=1,nV-nR
del(i,a,j) = eHF(i) + eHF(j) - eHF(nO+a) - eHF(p)
end do
end do
end do
do i=nC+1,nO
do a=1,nV-nR
do b=1,nV-nR
del(i,a,nO+b) = eHF(nO+a) + eHF(nO+b) - eHF(i) - eHF(p)
end do
end do
end do
do i=nC+1,nO
do a=1,nV-nR
do j=nC+1,nO
vec(i,a,j) = sqrt(2d0)*ERI(p,nO+a,i,j)
end do
end do
end do
do i=nC+1,nO
do a=1,nV-nR
do b=1,nV-nR
vec(i,a,nO+b) = sqrt(2d0)*ERI(p,i,nO+b,nO+a)
end do
end do
end do
!----------------------!
! Loop over amplitudes !
!----------------------!
write(*,*)
write(*,*)'-------------------------------------------------------------'
write(*,*)'| CC-based G0W0 calculation |'
write(*,*)'-------------------------------------------------------------'
write(*,'(1X,A1,1X,A3,1X,A1,1X,A15,1X,A1,1X,A15,1X,A1,1X,A15,1X,A1,1X)') &
'|','#','|','Sig_c (eV)','|','e_GW (eV)','|','Conv','|'
write(*,*)'-------------------------------------------------------------'
do while(Conv > thresh .and. nSCF < maxSCF)
! Increment
nSCF = nSCF + 1
! Compute residual for 2h1p sector
res(:,:,:) = vec(:,:,:) + (del(:,:,:) - Sig(p))*amp(:,:,:)
do i=nC+1,nO
do a=1,nV-nR
do j=nC+1,nO
do k=nC+1,nO
do c=1,nV-nR
res(i,a,j) = res(i,a,j) - 2d0*ERI(j,nO+c,nO+a,k)*amp(i,c,k)
! - 2d0*ERI(i,nO+c,nO+a,k)*amp(k,c,j)
end do
end do
end do
end do
end do
! Compute residual for 2p1h sector
do i=nC+1,nO
do a=1,nV-nR
do b=1,nV-nR
do k=nC+1,nO
do c=1,nV-nR
res(i,a,nO+b) = res(i,a,nO+b) + 2d0*ERI(nO+a,k,i,nO+c)*amp(k,c,nO+b)
! + 2d0*ERI(nO+b,k,i,nO+c)*amp(k,a,nO+c)
end do
end do
end do
end do
end do
! Check convergence
Conv = maxval(abs(res))
! Update amplitudes
amp(:,:,:) = amp(:,:,:) - res(:,:,:)/del(:,:,:)
! DIIS extrapolation
if(max_diis > 1) then
n_diis = min(n_diis+1,max_diis)
call DIIS_extrapolation(rcond,nO*nV*nOrb,nO*nV*nOrb,n_diis,r_diis,t_diis,res,amp)
end if
! Compute quasiparticle energy
Sig(p) = 0d0
do i=nC+1,nO
do a=1,nV-nR
do q=nC+1,nOrb-nR
Sig(p) = Sig(p) + vec(i,a,q)*amp(i,a,q)
end do
end do
end do
! Dump results
write(*,'(1X,A1,1X,I3,1X,A1,1X,F15.10,1X,A1,1X,F15.10,1X,A1,1X,F15.10,1X,A1,1X)') &
'|',nSCF,'|',Sig(p)*HaToeV,'|',(eHF(p)+Sig(p))*HaToeV,'|',Conv,'|'
end do
write(*,*)'-------------------------------------------------------------'
write(*,*)
!------------------------------------------------------------------------
! End of SCF loop
!------------------------------------------------------------------------
! Did it actually converge?
if(nSCF == maxSCF) then
write(*,*)
write(*,*)'!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!'
write(*,*)' Convergence failed '
write(*,*)'!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!'
write(*,*)
end if
write(*,*)'-------------------------------------------------------------------------------'
write(*,*)'| CC-based G0W0 calculation |'
write(*,*)'-------------------------------------------------------------------------------'
write(*,'(1X,A1,1X,A3,1X,A1,1X,A15,1X,A1,1X,A15,1X,A1,1X,A15,1X,A1,1X,A15,1X,A1,1X)') &
'|','Orb','|','e_HF (eV)','|','Sig_c (eV)','|','Z','|','e_QP (eV)','|'
write(*,*)'-------------------------------------------------------------------------------'
write(*,'(1X,A1,1X,I3,1X,A1,1X,F15.10,1X,A1,1X,F15.10,1X,A1,1X,F15.10,1X,A1,1X,F15.10,1X,A1,1X)') &
'|',p,'|',eHF(p)*HaToeV,'|',Sig(p)*HaToeV,'|',Z(p),'|',(eHF(p)+Sig(p))*HaToeV,'|'
write(*,*)'-------------------------------------------------------------------------------'
write(*,*)
end do
end subroutine

View File

@ -1,278 +0,0 @@
subroutine ccRG0W0_mat(maxSCF,thresh,nBas,nOrb,nC,nO,nV,nR,ERI,ENuc,ERHF,eHF)
! CC-based GW module
implicit none
include 'parameters.h'
! Input variables
integer,intent(in) :: maxSCF
double precision,intent(in) :: thresh
integer,intent(in) :: nBas
integer,intent(in) :: nOrb
integer,intent(in) :: nC
integer,intent(in) :: nO
integer,intent(in) :: nV
integer,intent(in) :: nR
double precision,intent(in) :: ENuc
double precision,intent(in) :: ERHF
double precision,intent(in) :: eHF(nOrb)
double precision,intent(in) :: ERI(nOrb,nOrb,nOrb,nOrb)
! Local variables
integer :: p,q
integer :: i,j,k,l
integer :: a,b,c,d
integer :: nSCF
double precision :: Conv
double precision,allocatable :: delta_2h1p(:)
double precision,allocatable :: delta_2p1h(:)
double precision,allocatable :: C_2h1p(:,:)
double precision,allocatable :: C_2p1h(:,:)
double precision,allocatable :: V_2h1p(:)
double precision,allocatable :: V_2p1h(:)
double precision,allocatable :: r_2h1p(:)
double precision,allocatable :: r_2p1h(:)
double precision,allocatable :: t_2h1p(:)
double precision,allocatable :: t_2p1h(:)
double precision,allocatable :: eGW(:)
double precision,allocatable :: Z(:)
double precision,external :: Kronecker_delta
integer :: n1h,n1p,n1h1p,n2h1p,n2p1h
integer :: ija,klc,iab,kcd
! Hello world
write(*,*)
write(*,*)'*****************************'
write(*,*)'* CC-based G0W0 Calculation *'
write(*,*)'*****************************'
write(*,*)
! Form energy denominator and guess amplitudes
n1h = nO
n1p = nV
n1h1p = n1h*n1p
n2h1p = n1h*n1h1p
n2p1h = n1p*n1h1p
allocate(delta_2h1p(n2h1p),delta_2p1h(n2p1h))
allocate(C_2h1p(n2h1p,n2h1p),C_2p1h(n2p1h,n2p1h))
allocate(V_2h1p(n2h1p),V_2p1h(n2p1h))
allocate(t_2h1p(n2h1p),t_2p1h(n2p1h))
allocate(r_2h1p(n2h1p),r_2p1h(n2p1h))
allocate(eGW(nOrb),Z(nOrb))
! Compute C2h1p and C2p1h
ija = 0
do i=nC+1,nO
do j=nC+1,nO
do a=nO+1,nOrb-nR
ija = ija + 1
klc = 0
do k=nC+1,nO
do l=nC+1,nO
do c=nO+1,nOrb-nR
klc = klc + 1
C_2h1p(ija,klc) = ((eHF(i) + eHF(j) - eHF(a))*Kronecker_delta(j,l)*Kronecker_delta(a,c) &
- 2d0*ERI(j,c,a,l))*Kronecker_delta(i,k)
end do
end do
end do
end do
end do
end do
iab = 0
do i=nC+1,nO
do a=nO+1,nOrb-nR
do b=nO+1,nOrb-nR
iab = iab + 1
kcd = 0
do k=nC+1,nO
do c=nO+1,nOrb-nR
do d=nO+1,nOrb-nR
kcd = kcd + 1
C_2p1h(iab,kcd) = ((eHF(a) + eHF(b) - eHF(i))*Kronecker_delta(i,k)*Kronecker_delta(a,c) &
+ 2d0*ERI(a,k,i,c))*Kronecker_delta(b,d)
end do
end do
end do
end do
end do
end do
!-------------------------!
! Main loop over orbitals !
!-------------------------!
eGW(:) = eHF(:)
do p=nO,nO
! Initialization
Conv = 1d0
nSCF = 0
t_2h1p(:) = 0d0
t_2p1h(:) = 0d0
! Compute energy differences
ija = 0
do i=nC+1,nO
do j=nC+1,nO
do a=nO+1,nOrb-nR
ija = ija + 1
delta_2h1p(ija) = eHF(i) + eHF(j) - eHF(a) - eHF(p)
end do
end do
end do
iab = 0
do i=nC+1,nO
do a=nO+1,nOrb-nR
do b=nO+1,nOrb-nR
iab = iab + 1
delta_2p1h(iab) = eHF(a) + eHF(b) - eHF(i) - eHF(p)
end do
end do
end do
klc = 0
do k=nC+1,nO
do l=nC+1,nO
do c=nO+1,nOrb-nR
klc = klc + 1
V_2h1p(klc) = sqrt(2d0)*ERI(p,c,k,l)
end do
end do
end do
kcd = 0
do k=nC+1,nO
do c=nO+1,nOrb-nR
do d=nO+1,nOrb-nR
kcd = kcd + 1
V_2p1h(kcd) = sqrt(2d0)*ERI(p,k,d,c)
end do
end do
end do
!----------------------!
! Loop over amplitudes !
!----------------------!
write(*,*)
write(*,*)'----------------------------------------------'
write(*,*)'| CC-based G0W0 calculation |'
write(*,*)'----------------------------------------------'
write(*,'(1X,A1,1X,A3,1X,A1,1X,A10,1X,A1,1X,A10,1X,A1,1X,A10,1X,A1,1X)') &
'|','#','|','HF','|','G0W0','|','Conv','|'
write(*,*)'----------------------------------------------'
do while(Conv > thresh .and. nSCF < maxSCF)
! Increment
nSCF = nSCF + 1
! Compute residual for 2h1p sector
r_2h1p = V_2h1p + matmul(C_2h1p,t_2h1p) - t_2h1p*eHF(p) &
- dot_product(t_2h1p,V_2h1p)*t_2h1p - t_2h1p*dot_product(V_2p1h,t_2p1h)
! Compute residual for 2p1h sector
r_2p1h = V_2p1h + matmul(C_2p1h,t_2p1h) - t_2p1h*eHF(p) &
- t_2p1h*dot_product(V_2h1p,t_2h1p) - dot_product(t_2p1h,V_2p1h)*t_2p1h
! Check convergence
Conv = max(maxval(abs(r_2h1p)),maxval(abs(r_2p1h)))
! Update amplitudes
t_2h1p(:) = t_2h1p(:) - r_2h1p(:)/delta_2h1p(:)
t_2p1h(:) = t_2p1h(:) - r_2p1h(:)/delta_2p1h(:)
! Compute quasiparticle energies
eGW(p) = eHF(p) + dot_product(V_2h1p,t_2h1p) + dot_product(V_2p1h,t_2p1h)
! Renormalization factor
Z(:) = 1d0
! Dump results
write(*,'(1X,A1,1X,I3,1X,A1,1X,F10.6,1X,A1,1X,F10.6,1X,A1,1X,F10.6,1X,A1,1X)') &
'|',nSCF,'|',eHF(p)*HaToeV,'|',eGW(p)*HaToeV,'|',Conv,'|'
end do
write(*,*)'----------------------------------------------'
!------------------------------------------------------------------------
! End of SCF loop
!------------------------------------------------------------------------
! Did it actually converge?
if(nSCF == maxSCF) then
write(*,*)
write(*,*)'!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!'
write(*,*)' Convergence failed '
write(*,*)'!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!'
write(*,*)
stop
end if
write(*,*)'-------------------------------------------------------------------------------'
write(*,*)' CC-G0W0 calculation '
write(*,*)'-------------------------------------------------------------------------------'
write(*,'(1X,A1,1X,A3,1X,A1,1X,A15,1X,A1,1X,A15,1X,A1,1X,A15,1X,A1,1X,A15,1X,A1,1X)') &
'|','#','|','e_HF (eV)','|','Sig_c (eV)','|','Z','|','e_QP (eV)','|'
write(*,*)'-------------------------------------------------------------------------------'
write(*,'(1X,A1,1X,I3,1X,A1,1X,F15.6,1X,A1,1X,F15.6,1X,A1,1X,F15.6,1X,A1,1X,F15.6,1X,A1,1X)') &
'|',p,'|',eHF(p)*HaToeV,'|',(eGW(p)-eHF(p))*HaToeV,'|',Z(p),'|',eGW(p)*HaToeV,'|'
write(*,*)'-------------------------------------------------------------------------------'
end do
end subroutine