10
1
mirror of https://github.com/pfloos/quack synced 2024-11-03 20:53:53 +01:00

Merge branch 'master' of github.com:pfloos/quack

This commit is contained in:
Clotilde Marut 2021-12-16 21:43:35 +01:00
commit 187c8a791c
18 changed files with 1005 additions and 39 deletions

View File

@ -1,5 +1,5 @@
# RHF UHF KS MOM
F F T F
T F F F
# MP2* MP3 MP2-F12
F F F
# CCD pCCD DCD CCSD CCSD(T)
@ -13,7 +13,7 @@
# G0F2* evGF2* qsGF2* G0F3 evGF3
F F F F F
# G0W0* evGW* qsGW* ufG0W0 ufGW
F F F F F
F F T F F
# G0T0 evGT qsGT
F F F
# MCMP2

View File

@ -1,18 +1,18 @@
# HF: maxSCF thresh DIIS n_diis guess_type ortho_type mix_guess stability
1024 0.0000001 T 5 1 1 F F
1024 0.00001 T 5 1 1 T F
# MP:
# CC: maxSCF thresh DIIS n_diis
64 0.0000001 T 5
64 0.00001 T 5
# spin: TDA singlet triplet spin_conserved spin_flip
F T T T T
# GF: maxSCF thresh DIIS n_diis lin eta renorm
256 0.00001 T 5 T 0.00367493 3
# GW/GT: maxSCF thresh DIIS n_diis lin eta COHSEX SOSEX TDA_W G0W GW0
256 0.00001 T 5 T 0.00 F F F F F
256 0.00001 T 5 T 0.0 F F F F F
# ACFDT: AC Kx XBS
F T F
F F F
# BSE: BSE dBSE dTDA evDyn
T T T F
F F F F
# MCMP2: nMC nEq nWalk dt nPrint iSeed doDrift
1000000 100000 10 0.3 10000 1234 T

View File

@ -1,4 +1,4 @@
2
H 0. 0. 0.
H 0. 0. 3.2
H 0. 0. 0.5

View File

@ -25,8 +25,8 @@ subroutine linear_response_C_pp(ispin,nBas,nC,nO,nV,nR,nOO,nVV,lambda,e,ERI,C_pp
! Define the chemical potential
eF = e(nO) + e(nO+1)
! eF = 0d0
! eF = e(nO) + e(nO+1)
eF = 0d0
! Build C matrix for the singlet manifold

View File

@ -25,8 +25,8 @@ subroutine linear_response_D_pp(ispin,nBas,nC,nO,nV,nR,nOO,nVV,lambda,e,ERI,D_pp
! Define the chemical potential
eF = e(nO) + e(nO+1)
! eF = 0d0
! eF = e(nO) + e(nO+1)
eF = 0d0
! Build the D matrix for the singlet manifold

View File

@ -47,7 +47,8 @@ subroutine linear_response_pp(ispin,TDA,nBas,nC,nO,nV,nR,nOO,nVV,lambda,e,ERI,Om
! Memory allocation
allocate(B(nVV,nOO),C(nVV,nVV),D(nOO,nOO),M(nOO+nVV,nOO+nVV),Z(nOO+nVV,nOO+nVV),Omega(nOO+nVV))
write(*,*) 'nOO', nOO
write(*,*) 'nVV', nVV
!-------------------------------------------------!
! Solve the p-p eigenproblem !
!-------------------------------------------------!
@ -62,7 +63,7 @@ subroutine linear_response_pp(ispin,TDA,nBas,nC,nO,nV,nR,nOO,nVV,lambda,e,ERI,Om
call linear_response_C_pp(ispin,nBas,nC,nO,nV,nR,nOO,nVV,lambda,e,ERI,C)
call linear_response_D_pp(ispin,nBas,nC,nO,nV,nR,nOO,nVV,lambda,e,ERI,D)
!call matout(nVV,nVV,C)
if(TDA) then
X1(:,:) = +C(:,:)
@ -76,7 +77,8 @@ subroutine linear_response_pp(ispin,TDA,nBas,nC,nO,nV,nR,nOO,nVV,lambda,e,ERI,Om
else
call linear_response_B_pp(ispin,nBas,nC,nO,nV,nR,nOO,nVV,lambda,ERI,B)
!call matout(nVV,nOO,B)
!call matout(nOO,nOO,D)
! Diagonal blocks
M( 1:nVV , 1:nVV) = + C(1:nVV,1:nVV)
@ -86,7 +88,7 @@ subroutine linear_response_pp(ispin,TDA,nBas,nC,nO,nV,nR,nOO,nVV,lambda,e,ERI,Om
M( 1:nVV ,nVV+1:nOO+nVV) = - B(1:nVV,1:nOO)
M(nVV+1:nOO+nVV, 1:nVV) = + transpose(B(1:nVV,1:nOO))
call matout(nOO+nVV,nOO+nVV,M)
! Diagonalize the p-h matrix
if(nOO+nVV > 0) call diagonalize_general_matrix(nOO+nVV,M,Omega,Z)

View File

@ -0,0 +1,119 @@
subroutine unrestricted_linear_response_B_pp(ispin,nBas,nC,nO,nV,nR,nPaa,nPab,nPbb,nPt,nHaa, &
nHab,nHbb,nHt,lambda,e,ERI_aaaa,ERI_aabb,ERI_bbbb,B_pp)
! Compute linear response
implicit none
include 'parameters.h'
! Input variables
integer,intent(in) :: ispin
integer,intent(in) :: nBas
integer,intent(in) :: nC(nspin)
integer,intent(in) :: nO(nspin)
integer,intent(in) :: nV(nspin)
integer,intent(in) :: nR(nspin)
integer,intent(in) :: nPaa
integer,intent(in) :: nPab
integer,intent(in) :: nPbb
integer,intent(in) :: nPt
integer,intent(in) :: nHaa
integer,intent(in) :: nHab
integer,intent(in) :: nHbb
integer,intent(in) :: nHt
double precision,intent(in) :: lambda
double precision,intent(in) :: e(nBas,nspin)
double precision,intent(in) :: ERI_aaaa(nBas,nBas,nBas,nBas)
double precision,intent(in) :: ERI_aabb(nBas,nBas,nBas,nBas)
double precision,intent(in) :: ERI_bbbb(nBas,nBas,nBas,nBas)
! Local variables
double precision :: eF
double precision,external :: Kronecker_delta
integer :: i,j,a,b,c,d,ij,ab
! Output variables
double precision,intent(out) :: B_pp(nPt,nHt)
eF = 0d0
!-----------------------------------------------
! Build B matrix for spin-conserving transitions
!-----------------------------------------------
if(ispin == 1) then
! aaaa block
ab = 0
do a=nO(1)+1,nBas-nR(1)
do b=a,nBas-nR(1)
ab = ab + 1
ij = 0
do i=nC(1)+1,nO(1)
do j=i+1,nO(1)
ij = ij + 1
B_pp(ab,ij) = lambda*(ERI_aaaa(a,b,i,j) - ERI_aaaa(a,b,j,i))
end do
end do
end do
end do
! bbbb block
ab = 0
do a=nO(2)+1,nBas-nR(2)
do b=a+1,nBas-nR(2)
ab = ab + 1
ij = 0
do i=nC(2)+1,nO(2)
do j=i+1,nO(2)
ij = ij + 1
B_pp(nPaa+ab,nHaa+ij) = lambda*(ERI_bbbb(a,b,i,j) - ERI_bbbb(a,b,j,i))
end do
end do
end do
end do
end if
!-----------------------------------------------
! Build B matrix for spin-flip transitions
!-----------------------------------------------
if(ispin == 2) then
B_pp(:,:) = 0d0
! abab block
ab = 0
do a=nO(1)+1,nBas-nR(1)
do b=nO(2)+1,nBas-nR(2)
ab = ab + 1
ij = 0
do i=nC(1)+1,nO(1)
do j=nC(2)+1,nO(2)
ij = ij + 1
B_pp(ab,ij) = lambda*ERI_aabb(a,b,i,j)
end do
end do
end do
end do
end if
end subroutine unrestricted_linear_response_B_pp

View File

@ -0,0 +1,115 @@
subroutine unrestricted_linear_response_C_pp(ispin,nBas,nC,nO,nV,nR,nPaa,nPab,nPbb,nPt,lambda,&
e,ERI_aaaa,ERI_aabb,ERI_bbbb,C_pp)
! Compute linear response
implicit none
include 'parameters.h'
! Input variables
integer,intent(in) :: ispin
integer,intent(in) :: nBas
integer,intent(in) :: nC(nspin)
integer,intent(in) :: nO(nspin)
integer,intent(in) :: nV(nspin)
integer,intent(in) :: nR(nspin)
integer,intent(in) :: nPaa
integer,intent(in) :: nPab
integer,intent(in) :: nPbb
integer,intent(in) :: nPt
double precision,intent(in) :: lambda
double precision,intent(in) :: e(nBas,nspin)
double precision,intent(in) :: ERI_aaaa(nBas,nBas,nBas,nBas)
double precision,intent(in) :: ERI_aabb(nBas,nBas,nBas,nBas)
double precision,intent(in) :: ERI_bbbb(nBas,nBas,nBas,nBas)
! Local variables
double precision :: eF
double precision,external :: Kronecker_delta
integer :: i,j,a,b,c,d,ab,cd
! Output variables
double precision,intent(out) :: C_pp(nPt,nPt)
eF = 0d0
!-----------------------------------------------
! Build C matrix for spin-conserving transitions
!-----------------------------------------------
if(ispin == 1) then
! aaaa block
ab = 0
do a=nO(1)+1,nBas-nR(1)
do b=a,nBas-nR(1)
ab = ab + 1
cd = 0
do c=nO(1)+1,nBas-nR(1)
do d=c,nBas-nR(1)
cd = cd + 1
C_pp(ab,cd) = (e(a,1) + e(b,1) - eF)*Kronecker_delta(a,c)*Kronecker_delta(b,d) &
+ lambda*(ERI_aaaa(a,b,c,d) - ERI_aaaa(a,b,d,c))
!write(*,*) C_pp(ab,cd)
end do
end do
end do
end do
! bbbb block
ab = 0
do a=nO(2)+1,nBas-nR(2)
do b=a,nBas-nR(2)
ab = ab + 1
cd = 0
do c=nO(2)+1,nBas-nR(2)
do d=c,nBas-nR(2)
cd = cd + 1
C_pp(nPaa+ab,nPaa+cd) = (e(a,2) + e(b,2) - eF)*Kronecker_delta(a,c) &
*Kronecker_delta(b,d) + lambda*(ERI_bbbb(a,b,c,d) - ERI_bbbb(a,b,d,c))
!write(*,*) 'nPaa+ab',nPaa+ab
end do
end do
end do
end do
end if
!
!-----------------------------------------------
! Build C matrix for spin-flip transitions
!-----------------------------------------------
if(ispin == 2) then
C_pp(:,:) = 0d0
! abab block
ab = 0
do a=nO(1)+1,nBas-nR(1)
do b=nO(2)+1,nBas-nR(2)
ab = ab + 1
cd = 0
do c=nO(1)+1,nBas-nR(1)
do d=nO(2)+1,nBas-nR(2)
cd = cd + 1
C_pp(ab,cd) = (e(a,1) + e(b,2))*Kronecker_delta(a,c) &
*Kronecker_delta(b,c) + lambda*ERI_aabb(a,b,c,d)
end do
end do
end do
end do
end if
end subroutine unrestricted_linear_response_C_pp

View File

@ -0,0 +1,115 @@
subroutine unrestricted_linear_response_D_pp(ispin,nBas,nC,nO,nV,nR,nHaa,nHab,nHbb,nHt,lambda,&
e,ERI_aaaa,ERI_aabb,ERI_bbbb,D_pp)
! Compute linear response
implicit none
include 'parameters.h'
! Input variables
integer,intent(in) :: ispin
integer,intent(in) :: nBas
integer,intent(in) :: nC(nspin)
integer,intent(in) :: nO(nspin)
integer,intent(in) :: nV(nspin)
integer,intent(in) :: nR(nspin)
integer,intent(in) :: nHaa
integer,intent(in) :: nHab
integer,intent(in) :: nHbb
integer,intent(in) :: nHt
double precision,intent(in) :: lambda
double precision,intent(in) :: e(nBas,nspin)
double precision,intent(in) :: ERI_aaaa(nBas,nBas,nBas,nBas)
double precision,intent(in) :: ERI_aabb(nBas,nBas,nBas,nBas)
double precision,intent(in) :: ERI_bbbb(nBas,nBas,nBas,nBas)
! Local variables
double precision :: eF
double precision,external :: Kronecker_delta
integer :: i,j,k,l,ij,kl
! Output variables
double precision,intent(out) :: D_pp(nHt,nHt)
eF = 0d0
!-----------------------------------------------
! Build D matrix for spin-conserving transitions
!-----------------------------------------------
if(ispin == 1) then
! aaaa block
ij = 0
do i=nC(1)+1,nO(1)
do j=i+1,nO(1)
ij = ij + 1
kl = 0
do k=nC(1)+1,nO(1)
do l=k+1,nO(1)
kl = kl + 1
D_pp(ij,kl) = -(e(i,1) + e(j,1) - eF)*Kronecker_delta(i,k)*Kronecker_delta(j,l) &
+ lambda*(ERI_aaaa(i,j,k,l) - ERI_aaaa(i,j,l,k))
end do
end do
end do
end do
! bbbb block
ij = 0
do i=nC(2)+1,nO(2)
do j=i+1,nO(2)
ij = ij + 1
kl = 0
do k=nC(2)+1,nO(2)
do l=k+1,nO(2)
kl = kl + 1
D_pp(nHaa+ij,nHaa+kl) = -(e(i,2) + e(j,2) - eF)*Kronecker_delta(i,k) &
*Kronecker_delta(j,l) + lambda*(ERI_bbbb(i,j,k,l) - ERI_bbbb(i,j,l,k))
end do
end do
end do
end do
end if
!-----------------------------------------------
! Build D matrix for spin-flip transitions
!-----------------------------------------------
if(ispin == 2) then
D_pp(:,:) = 0d0
! abab block
ij = 0
do i=nC(1)+1,nO(1)
do j=nC(2)+1,nO(2)
ij = ij + 1
kl = 0
do k=nC(1)+1,nO(1)
do l=nC(2)+1,nO(2)
kl = kl + 1
D_pp(ij,kl) = -(e(i,1) + e(j,2))*Kronecker_delta(i,k)&
*Kronecker_delta(j,l) +lambda*ERI_aabb(i,j,k,l)
end do
end do
end do
end do
end if
end subroutine unrestricted_linear_response_D_pp

View File

@ -0,0 +1,113 @@
subroutine unrestricted_linear_response_pp(ispin,TDA,nBas,nC,nO,nV,nR,nPaa,nPab,nPbb,nPt, &
nHaa,nHab,nHbb,nHt,lambda,e,ERI_aaaa,ERI_aabb,ERI_bbbb,Omega1,X1,Y1,Omega2,X2,Y2,&
EcRPA)
! Compute linear response for unrestricted formalism
implicit none
include 'parameters.h'
! Input variables
integer,intent(in) :: ispin
logical,intent(in) :: TDA
integer,intent(in) :: nBas
integer,intent(in) :: nC(nspin)
integer,intent(in) :: nO(nspin)
integer,intent(in) :: nV(nspin)
integer,intent(in) :: nR(nspin)
integer,intent(in) :: nPaa
integer,intent(in) :: nPab
integer,intent(in) :: nPbb
integer,intent(in) :: nPt
integer,intent(in) :: nHaa
integer,intent(in) :: nHab
integer,intent(in) :: nHbb
integer,intent(in) :: nHt
double precision,intent(in) :: lambda
double precision,intent(in) :: e(nBas,nspin)
double precision,intent(in) :: ERI_aaaa(nBas,nBas,nBas,nBas)
double precision,intent(in) :: ERI_aabb(nBas,nBas,nBas,nBas)
double precision,intent(in) :: ERI_bbbb(nBas,nBas,nBas,nBas)
! Local variables
double precision,external :: trace_matrix
double precision :: EcRPA1
double precision :: EcRPA2
double precision,allocatable :: C(:,:)
double precision,allocatable :: B(:,:)
double precision,allocatable :: D(:,:)
double precision,allocatable :: M(:,:)
double precision,allocatable :: Z(:,:)
double precision,allocatable :: Omega(:)
! Output variables
double precision,intent(out) :: Omega1(nPt)
double precision,intent(out) :: X1(nPt,nPt)
double precision,intent(out) :: Y1(nHt,nPt)
double precision,intent(out) :: Omega2(nHt)
double precision,intent(out) :: X2(nPt,nHt)
double precision,intent(out) :: Y2(nHt,nHt)
double precision,intent(out) :: EcRPA
! Memory allocation
allocate(C(nPt,nPt),B(nPt,nHt),D(nHt,nHt),M(nPt+nHt,nPt+nHt),Z(nPt+nHt,nPt+nHt),&
Omega(nPt+nHt))
!write(*,*) 'ispin', ispin
!write(*,*) 'nPt', nPt
!write(*,*) 'nHt', nHt
! Build C, B and D matrices for the pp channel
call unrestricted_linear_response_C_pp(ispin,nBas,nC,nO,nV,nR,nPaa,nPab,nPbb,nPt,lambda,&
e,ERI_aaaa,ERI_aabb,ERI_bbbb,C)
!call matout(nPt,nPt,C)
!write(*,*) 'Hello'
call unrestricted_linear_response_B_pp(ispin,nBas,nC,nO,nV,nR,nPaa,nPab,nPbb,nPt,nHaa,&
nHab,nHbb,nHt,lambda,ERI_aaaa,ERI_aabb,ERI_bbbb,B)
!call matout(nPt,nHt,B)
!write(*,*) 'Hello'
call unrestricted_linear_response_D_pp(ispin,nBas,nC,nO,nV,nR,nHaa,nHab,nHbb,nHt,lambda,&
ERI_aaaa,ERI_aabb,ERI_bbbb,D)
!call matout(nHt,nHt,D)
!write(*,*) 'Hello'
! Diagonal blocks
M( 1:nPt , 1:nPt) = + C(1:nPt,1:nPt)
M(nPt+1:nPt+nHt,nPt+1:nPt+nHt) = - D(1:nHt,1:nHt)
! Off-diagonal blocks
M( 1:nPt ,nPt+1:nHt+nPt) = - B(1:nPt,1:nHt)
M(nPt+1:nHt+nPt, 1:nPt) = + transpose(B(1:nPt,1:nHt))
!call matout(nPt+nHt,nPt+nHt,M)
! Diagonalize the p-h matrix
! if(nHt+nPt > 0) call diagonalize_general_matrix(nHt+nPt,M,Omega,Z)
! Split the various quantities in p-p and h-h parts
! call sort_ppRPA(nHt,nPt,Omega(:),Z(:,:),Omega1(:),X1(:,:),Y1(:,:),Omega2(:),X2(:,:),&
!Y2(:,:))
! end if Pourquoi ne faut-il pas de end if ici ?
! Compute the RPA correlation energy
! EcRPA = 0.5d0*( sum(Omega1(:)) - sum(Omega2(:)) - trace_matrix(nPt,C(:,:)) - trace_matrix(nHt,D(:,:)) )
! EcRPA1 = +sum(Omega1(:)) - trace_matrix(nPt,C(:,:))
! EcRPA2 = -sum(Omega2(:)) - trace_matrix(nHt,D(:,:))
! if(abs(EcRPA - EcRPA1) > 1d-6 .or. abs(EcRPA - EcRPA2) > 1d-6) &
! print*,'!!! Issue in pp-RPA linear reponse calculation RPA1 != RPA2 !!!'
end subroutine unrestricted_linear_response_pp

View File

@ -161,11 +161,13 @@ subroutine evGW(maxSCF,thresh,max_diis,doACFDT,exchange_kernel,doXBS,COHSEX,BSE,
if(G0W) then
! call regularized_self_energy_correlation_diag(COHSEX,eta,nBas,nC,nO,nV,nR,nS,eHF,OmRPA,rho_RPA,EcGM,SigC)
call self_energy_correlation_diag(COHSEX,eta,nBas,nC,nO,nV,nR,nS,eHF,OmRPA,rho_RPA,EcGM,SigC)
call renormalization_factor(COHSEX,eta,nBas,nC,nO,nV,nR,nS,eHF,OmRPA,rho_RPA,Z)
else
! call regularized_self_energy_correlation_diag(COHSEX,eta,nBas,nC,nO,nV,nR,nS,eGW,OmRPA,rho_RPA,EcGM,SigC)
call self_energy_correlation_diag(COHSEX,eta,nBas,nC,nO,nV,nR,nS,eGW,OmRPA,rho_RPA,EcGM,SigC)
call renormalization_factor(COHSEX,eta,nBas,nC,nO,nV,nR,nS,eGW,OmRPA,rho_RPA,Z)

View File

@ -192,11 +192,13 @@ subroutine qsGW(maxSCF,thresh,max_diis,doACFDT,exchange_kernel,doXBS,COHSEX,BSE,
if(G0W) then
! call regularized_self_energy_correlation(COHSEX,eta,nBas,nC,nO,nV,nR,nS,eHF,OmRPA,rho_RPA,EcGM,SigC)
call self_energy_correlation(COHSEX,eta,nBas,nC,nO,nV,nR,nS,eHF,OmRPA,rho_RPA,EcGM,SigC)
call renormalization_factor(COHSEX,eta,nBas,nC,nO,nV,nR,nS,eHF,OmRPA,rho_RPA,Z)
else
! call regularized_self_energy_correlation(COHSEX,eta,nBas,nC,nO,nV,nR,nS,eGW,OmRPA,rho_RPA,EcGM,SigC)
call self_energy_correlation(COHSEX,eta,nBas,nC,nO,nV,nR,nS,eGW,OmRPA,rho_RPA,EcGM,SigC)
call renormalization_factor(COHSEX,eta,nBas,nC,nO,nV,nR,nS,eGW,OmRPA,rho_RPA,Z)

View File

@ -0,0 +1,126 @@
subroutine regularized_self_energy_correlation(COHSEX,eta,nBas,nC,nO,nV,nR,nS,e,Omega,rho,EcGM,SigC)
! Compute correlation part of the regularized self-energy
implicit none
include 'parameters.h'
! Input variables
logical,intent(in) :: COHSEX
double precision,intent(in) :: eta
integer,intent(in) :: nBas,nC,nO,nV,nR,nS
double precision,intent(in) :: e(nBas)
double precision,intent(in) :: Omega(nS)
double precision,intent(in) :: rho(nBas,nBas,nS)
! Local variables
integer :: i,j,a,b
integer :: p,q,r
integer :: jb
double precision :: eps
double precision :: kappa
double precision :: fk
! Output variables
double precision,intent(out) :: SigC(nBas,nBas)
double precision,intent(out) :: EcGM
! Initialize
SigC(:,:) = 0d0
!---------------------------------------------!
! Parameters for regularized MP2 calculations !
!---------------------------------------------!
kappa = 1.1d0
!-----------------------------!
! COHSEX static approximation !
!-----------------------------!
if(COHSEX) then
! COHSEX: SEX of the COHSEX correlation self-energy
do p=nC+1,nBas-nR
do q=nC+1,nBas-nR
do i=nC+1,nO
do jb=1,nS
SigC(p,q) = SigC(p,q) + 4d0*rho(p,i,jb)*rho(q,i,jb)/Omega(jb)
end do
end do
end do
end do
! COHSEX: COH part of the COHSEX correlation self-energy
do p=nC+1,nBas-nR
do q=nC+1,nBas-nR
do r=nC+1,nBas-nR
do jb=1,nS
SigC(p,q) = SigC(p,q) - 2d0*rho(p,r,jb)*rho(q,r,jb)/Omega(jb)
end do
end do
end do
end do
EcGM = 0d0
do i=nC+1,nO
EcGM = EcGM + 0.5d0*SigC(i,i)
end do
else
!----------------!
! GW self-energy !
!----------------!
! Occupied part of the correlation self-energy
do p=nC+1,nBas-nR
do q=nC+1,nBas-nR
do i=nC+1,nO
do jb=1,nS
eps = e(p) - e(i) + Omega(jb)
fk = (1d0 - exp(-kappa*abs(eps)))**2/eps
SigC(p,q) = SigC(p,q) + 2d0*rho(p,i,jb)*rho(q,i,jb)*fk
end do
end do
end do
end do
! Virtual part of the correlation self-energy
do p=nC+1,nBas-nR
do q=nC+1,nBas-nR
do a=nO+1,nBas-nR
do jb=1,nS
eps = e(p) - e(a) - Omega(jb)
fk = (1d0 - exp(-kappa*abs(eps)))**2/eps
SigC(p,q) = SigC(p,q) + 2d0*rho(p,a,jb)*rho(q,a,jb)*fk
end do
end do
end do
end do
! GM correlation energy
EcGM = 0d0
do i=nC+1,nO
do a=nO+1,nBas-nR
do jb=1,nS
eps = e(a) - e(i) + Omega(jb)
fk = (1d0 - exp(-kappa*abs(eps)))**2/eps
EcGM = EcGM - 4d0*rho(a,i,jb)**2*fk
end do
end do
end do
end if
end subroutine regularized_self_energy_correlation

View File

@ -0,0 +1,124 @@
subroutine regularized_self_energy_correlation_diag(COHSEX,eta,nBas,nC,nO,nV,nR,nS,e,Omega,rho,EcGM,SigC)
! Compute diagonal of the correlation part of the regularized self-energy
implicit none
include 'parameters.h'
! Input variables
logical,intent(in) :: COHSEX
double precision,intent(in) :: eta
integer,intent(in) :: nBas
integer,intent(in) :: nC
integer,intent(in) :: nO
integer,intent(in) :: nV
integer,intent(in) :: nR
integer,intent(in) :: nS
double precision,intent(in) :: e(nBas)
double precision,intent(in) :: Omega(nS)
double precision,intent(in) :: rho(nBas,nBas,nS)
! Local variables
integer :: i,a,p,q,jb
double precision :: eps
double precision,external :: SigC_dcgw
double precision :: kappa
double precision :: fk
! Output variables
double precision,intent(out) :: SigC(nBas)
double precision,intent(out) :: EcGM
! Initialize
SigC(:) = 0d0
!---------------------------------------------!
! Parameters for regularized MP2 calculations !
!---------------------------------------------!
kappa = 1.1d0
!-----------------------------
! COHSEX static self-energy
!-----------------------------
if(COHSEX) then
! COHSEX: SEX part of the COHSEX correlation self-energy
do p=nC+1,nBas-nR
do i=nC+1,nO
do jb=1,nS
SigC(p) = SigC(p) + 4d0*rho(p,i,jb)**2/Omega(jb)
end do
end do
end do
! COHSEX: COH part of the COHSEX correlation self-energy
do p=nC+1,nBas-nR
do q=nC+1,nBas-nR
do jb=1,nS
SigC(p) = SigC(p) - 2d0*rho(p,q,jb)**2/Omega(jb)
end do
end do
end do
! GM correlation energy
EcGM = 0d0
do i=nC+1,nO
EcGM = EcGM - SigC(i)
end do
!-----------------------------
! GW self-energy
!-----------------------------
else
! Occupied part of the correlation self-energy
do p=nC+1,nBas-nR
do i=nC+1,nO
do jb=1,nS
eps = e(p) - e(i) + Omega(jb)
fk = (1d0 - exp(-kappa*abs(eps)))**2/eps
SigC(p) = SigC(p) + 2d0*rho(p,i,jb)**2*fk
end do
end do
end do
! Virtual part of the correlation self-energy
do p=nC+1,nBas-nR
do a=nO+1,nBas-nR
do jb=1,nS
eps = e(p) - e(a) - Omega(jb)
fk = (1d0 - exp(-kappa*abs(eps)))**2/eps
SigC(p) = SigC(p) + 2d0*rho(p,a,jb)**2*fk
end do
end do
end do
! GM correlation energy
EcGM = 0d0
do i=nC+1,nO
do a=nO+1,nBas-nR
do jb=1,nS
eps = e(a) - e(i) + Omega(jb)
fk = (1d0 - exp(-kappa*abs(eps)))**2/eps
EcGM = EcGM - 4d0*rho(a,i,jb)**2*fk
end do
end do
end do
end if
end subroutine regularized_self_energy_correlation_diag

View File

@ -1,6 +1,6 @@
subroutine MP2(nBas,nC,nO,nV,nR,ERI,ENuc,EHF,e,EcMP2)
! Perform second-order Moller-Plesset calculation
! Perform second-order Moller-Plesset calculation with and without regularizers
implicit none
@ -19,7 +19,15 @@ subroutine MP2(nBas,nC,nO,nV,nR,ERI,ENuc,EHF,e,EcMP2)
! Local variables
integer :: i,j,a,b
double precision :: eps,E2a,E2b
double precision :: kappa,sigm1,sigm2
double precision :: Dijab
double precision :: fs,fs2,fk
double precision :: E2d,E2ds,E2ds2,E2dk
double precision :: E2x,E2xs,E2xs2,E2xk
double precision :: EcsMP2,Ecs2MP2,EckMP2
! Output variables
@ -33,43 +41,129 @@ subroutine MP2(nBas,nC,nO,nV,nR,ERI,ENuc,EHF,e,EcMP2)
write(*,*)'************************************************'
write(*,*)
! Compute MP2 energy
!---------------------------------------------!
! Parameters for regularized MP2 calculations !
!---------------------------------------------!
kappa = 1.1d0
sigm1 = 0.7d0
sigm2 = 0.4d0
!--------------------------------------------------!
! Compute conventinal and regularized MP2 energies !
!--------------------------------------------------!
E2d = 0d0
E2ds = 0d0
E2ds2 = 0d0
E2dk = 0d0
E2x = 0d0
E2xs = 0d0
E2xs2 = 0d0
E2xk = 0d0
E2a = 0d0
E2b = 0d0
do i=nC+1,nO
do j=nC+1,nO
do a=nO+1,nBas-nR
do b=nO+1,nBas-nR
eps = e(i) + e(j) - e(a) - e(b)
Dijab = e(a) + e(b) - e(i) - e(j)
! Second-order ring diagram
! Second-order ring diagram
E2a = E2a + ERI(i,j,a,b)*ERI(i,j,a,b)/eps
fs = (1d0 - exp(-sigm1*Dijab))/Dijab
fs2 = (1d0 - exp(-sigm2*Dijab*Dijab))/Dijab
fk = (1d0 - exp(-kappa*Dijab))**2/Dijab
! Second-order exchange diagram
E2d = E2d - ERI(i,j,a,b)*ERI(i,j,a,b)/Dijab
E2ds = E2ds - ERI(i,j,a,b)*ERI(i,j,a,b)*fs
E2ds2 = E2ds2 - ERI(i,j,a,b)*ERI(i,j,a,b)*fs2
E2dk = E2dk - ERI(i,j,a,b)*ERI(i,j,a,b)*fk
E2b = E2b + ERI(i,j,a,b)*ERI(i,j,b,a)/eps
! Second-order exchange diagram
E2x = E2x - ERI(i,j,a,b)*ERI(i,j,b,a)/Dijab
E2xs = E2xs - ERI(i,j,a,b)*ERI(i,j,b,a)*fs
E2xs2 = E2xs2 - ERI(i,j,a,b)*ERI(i,j,b,a)*fs2
E2xk = E2xk - ERI(i,j,a,b)*ERI(i,j,b,a)*fk
enddo
enddo
enddo
enddo
EcMP2 = 2d0*E2a - E2b
EcMP2 = 2d0*E2d - E2x
EcsMP2 = 2d0*E2ds - E2xs
Ecs2MP2 = 2d0*E2ds2 - E2xs2
EckMP2 = 2d0*E2dk - E2xk
!------------!
! MP2 energy !
!------------!
write(*,*)
write(*,'(A32)') '--------------------------'
write(*,'(A32)') ' MP2 calculation '
write(*,'(A32)') '--------------------------'
write(*,'(A32,1X,F16.10)') ' MP2 correlation energy = ',EcMP2
write(*,'(A32,1X,F16.10)') ' Direct part = ',2d0*E2a
write(*,'(A32,1X,F16.10)') ' Exchange part = ',-E2b
write(*,'(A32,1X,F16.10)') ' Direct part = ',2d0*E2d
write(*,'(A32,1X,F16.10)') ' Exchange part = ',-E2x
write(*,'(A32)') '--------------------------'
write(*,'(A32,1X,F16.10)') ' MP2 electronic energy = ',EHF + EcMP2
write(*,'(A32,1X,F16.10)') ' MP2 total energy = ',ENuc + EHF + EcMP2
write(*,'(A32)') '--------------------------'
write(*,*)
!-------------------!
! sigma1-MP2 energy !
!-------------------!
write(*,*)
write(*,'(A32)') '--------------------------'
write(*,'(A32)') ' sigma-MP2 calculation '
write(*,'(A32)') '--------------------------'
write(*,'(A32,1X,F16.10)') ' MP2 correlation energy = ',EcsMP2
write(*,'(A32,1X,F16.10)') ' Direct part = ',2d0*E2ds
write(*,'(A32,1X,F16.10)') ' Exchange part = ',-E2xs
write(*,'(A32)') '--------------------------'
write(*,'(A32,1X,F16.10)') ' MP2 electronic energy = ',EHF + EcsMP2
write(*,'(A32,1X,F16.10)') ' MP2 total energy = ',ENuc + EHF + EcsMP2
write(*,'(A32)') '--------------------------'
write(*,*)
!--------------------!
! sigma^2-MP2 energy !
!--------------------!
write(*,*)
write(*,'(A32)') '--------------------------'
write(*,'(A32)') ' sigma^2-MP2 calculation '
write(*,'(A32)') '--------------------------'
write(*,'(A32,1X,F16.10)') ' MP2 correlation energy = ',Ecs2MP2
write(*,'(A32,1X,F16.10)') ' Direct part = ',2d0*E2ds2
write(*,'(A32,1X,F16.10)') ' Exchange part = ',-E2xs2
write(*,'(A32)') '--------------------------'
write(*,'(A32,1X,F16.10)') ' MP2 electronic energy = ',EHF + Ecs2MP2
write(*,'(A32,1X,F16.10)') ' MP2 total energy = ',ENuc + EHF + Ecs2MP2
write(*,'(A32)') '--------------------------'
write(*,*)
!------------------!
! kappa-MP2 energy !
!------------------!
write(*,*)
write(*,'(A32)') '--------------------------'
write(*,'(A32)') ' kappa-MP2 calculation '
write(*,'(A32)') '--------------------------'
write(*,'(A32,1X,F16.10)') ' MP2 correlation energy = ',EckMP2
write(*,'(A32,1X,F16.10)') ' Direct part = ',2d0*E2dk
write(*,'(A32,1X,F16.10)') ' Exchange part = ',-E2xk
write(*,'(A32)') '--------------------------'
write(*,'(A32,1X,F16.10)') ' MP2 electronic energy = ',EHF + EckMP2
write(*,'(A32,1X,F16.10)') ' MP2 total energy = ',ENuc + EHF + EckMP2
write(*,'(A32)') '--------------------------'
write(*,*)
end subroutine MP2

View File

@ -824,7 +824,15 @@ program QuAcK
if(doppRPA) then
call cpu_time(start_RPA)
if(unrestricted) then
call ppURPA(TDA,doACFDT,spin_conserved,spin_flip,nBas,nC,nO,nV,nR,ENuc,EUHF,ERI_MO_aaaa,ERI_MO_aabb,ERI_MO_bbbb,eHF)
else
call ppRPA(TDA,doACFDT,singlet,triplet,nBas,nC,nO,nV,nR,ENuc,ERHF,ERI_MO,eHF)
end if
call cpu_time(end_RPA)
t_RPA = end_RPA - start_RPA

146
src/RPA/ppURPA.f90 Normal file
View File

@ -0,0 +1,146 @@
subroutine ppURPA(TDA,doACFDT,spin_conserved,spin_flip,nBas,nC,nO,nV,nR,ENuc,EUHF,ERI_aaaa,ERI_aabb,ERI_bbbb,e)
! Perform unrestricted pp-RPA calculations
implicit none
include 'parameters.h'
! Input variables
logical,intent(in) :: TDA
logical,intent(in) :: doACFDT
logical,intent(in) :: spin_conserved
logical,intent(in) :: spin_flip
integer,intent(in) :: nBas
integer,intent(in) :: nC(nspin)
integer,intent(in) :: nO(nspin)
integer,intent(in) :: nV(nspin)
integer,intent(in) :: nR(nspin)
double precision,intent(in) :: ENuc
double precision,intent(in) :: EUHF
double precision,intent(in) :: e(nBas,nspin)
double precision,intent(in) :: ERI_aaaa(nBas,nBas,nBas,nBas)
double precision,intent(in) :: ERI_aabb(nBas,nBas,nBas,nBas)
double precision,intent(in) :: ERI_bbbb(nBas,nBas,nBas,nBas)
! Local variables
integer :: ispin
integer :: nPaa,nPbb,nPab,nP_sc,nP_sf
integer :: nHaa,nHbb,nHab,nH_sc,nH_sf
double precision,allocatable :: Omega1sc(:),Omega1sf(:)
double precision,allocatable :: X1sc(:,:),X1sf(:,:)
double precision,allocatable :: Y1sc(:,:),Y1sf(:,:)
double precision,allocatable :: Omega2sc(:),Omega2sf(:)
double precision,allocatable :: X2sc(:,:),X2sf(:,:)
double precision,allocatable :: Y2sc(:,:),Y2sf(:,:)
double precision :: Ec_ppURPA(nspin)
double precision :: EcAC(nspin)
! Hello world
write(*,*)
write(*,*)'****************************************'
write(*,*)'| particle-particle URPA calculation |'
write(*,*)'****************************************'
write(*,*)
! Initialization
Ec_ppURPA(:) = 0d0
EcAC(:) = 0d0
! Useful quantities
!spin-conserved quantities
nPaa = nV(1)*(nV(1)-1)/2
nPbb = nV(2)*(nV(2)-1)/2
nP_sc = nPaa + nPbb
nHaa = nO(1)*(nO(1)-1)/2
nHbb = nO(2)*(nO(2)-1)/2
nH_sc = nHaa + nHbb
!spin-flip quantities
nPab = nV(1)*nV(2)
nHab = nO(1)*nO(2)
nP_sf = nPab
nH_sf = nPab
! Memory allocation
allocate(Omega1sc(nP_sc),X1sc(nP_sc,nP_sc),Y1sc(nH_sc,nP_sc), &
Omega2sc(nH_sc),X2sc(nP_sc,nH_sc),Y2sc(nH_sc,nH_sc))
allocate(Omega1sf(nP_sf),X1sf(nP_sf,nP_sf),Y1sf(nH_sf,nP_sf), &
Omega2sf(nH_sf),X2sf(nP_sf,nH_sf),Y2sf(nH_sf,nH_sf))
! Spin-conserved manifold
if(spin_conserved) then
ispin = 1
call unrestricted_linear_response_pp(ispin,TDA,nBas,nC,nO,nV,nR,nPaa,nPab,nPbb,nP_sc, &
nHaa,nHab,nHbb,nH_sc,1d0,e,ERI_aaaa,ERI_aabb,ERI_bbbb,Omega1sc,X1sc,Y1sc,Omega2sc,X2sc,Y2sc,&
Ec_ppURPA(ispin))
write(*,*) 'Hello!'
call print_excitation('pp-RPA (N+2)',5,nP_sc,Omega1sc)
call print_excitation('pp-RPA (N-2)',5,nH_sc,Omega2sc)
endif
! Spin-flip manifold
if(spin_flip) then
ispin = 2
call unrestricted_linear_response_pp(ispin,TDA,nBas,nC,nO,nV,nR,nPaa,nPab,nPbb,nP_sf, &
nHaa,nHab,nHbb,nH_sf,1d0,e,ERI_aaaa,ERI_aabb,ERI_bbbb,Omega1sf,X1sf,Y1sf,Omega2sf,X2sf,Y2sf,&
Ec_ppURPA(ispin))
call print_excitation('pp-RPA (N+2)',6,nP_sf,Omega1sf)
call print_excitation('pp-RPA (N-2)',6,nH_sf,Omega2sf)
endif
write(*,*)
write(*,*)'-------------------------------------------------------------------------------'
write(*,'(2X,A50,F20.10)') 'Tr@ppRPA correlation energy (spin-conserved) =',Ec_ppURPA(1)
write(*,'(2X,A50,F20.10)') 'Tr@ppRPA correlation energy (spin-flip) =',3d0*Ec_ppURPA(2)
write(*,'(2X,A50,F20.10)') 'Tr@ppRPA correlation energy =',Ec_ppURPA(1) + 3d0*Ec_ppURPA(2)
write(*,'(2X,A50,F20.10)') 'Tr@ppRPA total energy =',ENuc + EUHF + Ec_ppURPA(1) + 3d0*Ec_ppURPA(2)
write(*,*)'-------------------------------------------------------------------------------'
write(*,*)
! Compute the correlation energy via the adiabatic connection
! if(doACFDT) then
! write(*,*) '---------------------------------------------------------'
! write(*,*) 'Adiabatic connection version of pp-RPA correlation energy'
! write(*,*) '---------------------------------------------------------'
! write(*,*)
! call ACFDT_pp(TDA,singlet,triplet,nBas,nC,nO,nV,nR,nS,ERI,e,EcAC)
! write(*,*)
! write(*,*)'-------------------------------------------------------------------------------'
! write(*,'(2X,A50,F20.10,A3)') 'AC@ppRPA correlation energy (singlet) =',EcAC(1),' au'
! write(*,'(2X,A50,F20.10,A3)') 'AC@ppRPA correlation energy (triplet) =',EcAC(2),' au'
! write(*,'(2X,A50,F20.10,A3)') 'AC@ppRPA correlation energy =',EcAC(1) + EcAC(2),' au'
! write(*,'(2X,A50,F20.10,A3)') 'AC@ppRPA total energy =',ENuc + ERHF + EcAC(1) + EcAC(2),' au'
! write(*,*)'-------------------------------------------------------------------------------'
! write(*,*)
! end if
end subroutine ppURPA