10
1
mirror of https://github.com/pfloos/quack synced 2024-11-05 21:53:56 +01:00

create utils

This commit is contained in:
Pierre-Francois Loos 2019-03-20 13:38:42 +01:00
parent 0bb4e5f373
commit ca3a985d50
49 changed files with 1816 additions and 324 deletions

View File

@ -35,13 +35,13 @@ subroutine CalcBoysF(maxm,t,Fm)
do m=0,maxm
dm = dble(m)
Fm(m) = 1d0/(2d0*dm+1d0)
enddo
end do
else
do m=0,maxm
dm = dble(m)
! Fm(m) = gamma(dm+0.5d0)*gsl_sf_gamma_inc_P(dm+0.5d0,t)/(2d0*t**(dm+0.5d0))
Fm(m) = dgami(dm+0.5d0,t)/(2d0*t**(dm+0.5d0))
enddo
endif
end do
end if
end subroutine CalcBoysF

View File

@ -33,7 +33,7 @@ subroutine CalcOm(maxm,ExpPQi,NormPQSq,Om)
do m=0,maxm
dm =dble(m)
Om(m) = (2d0/sqrt(pi))*(-1d0)**dm*(1d0/ExpPQi)**(dm+0.5d0)*Fm(m)
enddo
end do
deallocate(Fm)

View File

@ -37,7 +37,7 @@ subroutine CalcOm3e(maxm,delta0,delta1,Y1,Y0,Om)
do m=0,maxm
Om(m) = (2d0/sqrt(pi))*OG*sqrt(delta0/(delta1-delta0))*(delta1/(delta1-delta0))**m
Om(m) = Om(m)*Fm(m)
enddo
end do
deallocate(Fm)

View File

@ -32,7 +32,7 @@ subroutine CalcOmERI(maxm,ExpY,NormYSq,Om)
do m=0,maxm
Om(m) = (2d0/sqrt(pi))*sqrt(ExpY)*Fm(m)
enddo
end do
deallocate(Fm)

View File

@ -32,7 +32,7 @@ subroutine CalcOmErf(maxm,ExpY,fG,NormYSq,Om)
do m=0,maxm
Om(m) = (2d0/sqrt(pi))*sqrt(fG)*(fG/ExpY)**m*Fm(m)
enddo
end do
deallocate(Fm)

View File

@ -33,7 +33,7 @@ subroutine CalcOmNuc(maxm,ExpPQi,NormPQSq,Om)
do m=0,maxm
dm =dble(m)
Om(m) = (2d0/sqrt(pi))*(1d0/ExpPQi)**(dm+0.5d0)*Fm(m)
enddo
end do
deallocate(Fm)

View File

@ -34,9 +34,9 @@ subroutine CalcOmYuk(maxm,ExpG,ExpY,fG,NormYSq,Om)
Om(m) = 0d0
do k=0,m
Om(m) = Om(m) + dbinom(m,k)*(ExpY/ExpG)**k*Fm(k)
enddo
end do
Om(m) = (2d0/sqrt(pi))*sqrt(ExpY)*(fG/ExpG)*exp(-fG*NormYSq)*Om(m)
enddo
end do
deallocate(Fm)

View File

@ -32,7 +32,7 @@ subroutine Compute2eInt(debug,chemist_notation,iType,nShell, &
integer,allocatable :: ShellFunctionB1(:,:),ShellFunctionB2(:,:)
double precision :: ExpBra(2),ExpKet(2)
double precision :: DBra(2),DKet(2)
double precision :: NormCoeff
double precision :: norm_coeff
integer :: iBasA1,iBasA2,iBasB1,iBasB2
integer :: iShA1,iShA2,iShB1,iShB2
@ -110,7 +110,7 @@ subroutine Compute2eInt(debug,chemist_notation,iType,nShell, &
iFile = 24
open(unit=iFile,file='int/Erf.dat')
endif
end if
!------------------------------------------------------------------------
! Loops over shell A1
@ -213,16 +213,16 @@ subroutine Compute2eInt(debug,chemist_notation,iType,nShell, &
do iKA1=1,KBra(1)
ExpBra(1) = ExpShell(iShA1,iKA1)
DBra(1) = DShell(iShA1,iKA1)*NormCoeff(ExpBra(1),AngMomBra(1,1:3))
DBra(1) = DShell(iShA1,iKA1)*norm_coeff(ExpBra(1),AngMomBra(1,1:3))
do iKA2=1,KBra(2)
ExpBra(2) = ExpShell(iShA2,iKA2)
DBra(2) = DShell(iShA2,iKA2)*NormCoeff(ExpBra(2),AngMomBra(2,1:3))
DBra(2) = DShell(iShA2,iKA2)*norm_coeff(ExpBra(2),AngMomBra(2,1:3))
do iKB1=1,KKet(1)
ExpKet(1) = ExpShell(iShB1,iKB1)
DKet(1) = DShell(iShB1,iKB1)*NormCoeff(ExpKet(1),AngMomKet(1,1:3))
DKet(1) = DShell(iShB1,iKB1)*norm_coeff(ExpKet(1),AngMomKet(1,1:3))
do iKB2=1,KKet(2)
ExpKet(2) = ExpShell(iShB2,iKB2)
DKet(2) = DShell(iShB2,iKB2)*NormCoeff(ExpKet(2),AngMomKet(2,1:3))
DKet(2) = DShell(iShB2,iKB2)*norm_coeff(ExpKet(2),AngMomKet(2,1:3))
call S2eInt(debug,iType,np2eInt,nSigp2eInt, &
ExpS,KG,DG,ExpG, &
@ -232,10 +232,10 @@ subroutine Compute2eInt(debug,chemist_notation,iType,nShell, &
c2eInt = c2eInt + DBra(1)*DBra(2)*DKet(1)*DKet(2)*p2eInt
enddo
enddo
enddo
enddo
end do
end do
end do
end do
call cpu_time(end_c2eInt)
nc2eInt = nc2eInt + 1
@ -252,7 +252,7 @@ subroutine Compute2eInt(debug,chemist_notation,iType,nShell, &
if(debug) then
write(*,'(A10,1X,F16.10,1X,I6,1X,I6,1X,I6,1X,I6)') &
'(a1b1|a2b2) = ',c2eInt,iBasA1,iBasB1,iBasA2,iBasB2
endif
end if
else
@ -261,38 +261,38 @@ subroutine Compute2eInt(debug,chemist_notation,iType,nShell, &
if(debug) then
write(*,'(A10,1X,F16.10,1X,I6,1X,I6,1X,I6,1X,I6)') &
'<a1a2|b1b2> = ',c2eInt,iBasA1,iBasA2,iBasB1,iBasB2
endif
end if
endif
endif
end if
end if
!------------------------------------------------------------------------
! End loops over contraction degrees
!------------------------------------------------------------------------
enddo
end do
deallocate(ShellFunctionB2)
enddo
end do
iBasB2 = 0
!------------------------------------------------------------------------
! End loops over shell B2
!------------------------------------------------------------------------
enddo
end do
deallocate(ShellFunctionA2)
enddo
end do
iBasA2 = 0
!------------------------------------------------------------------------
! End loops over shell A2
!------------------------------------------------------------------------
enddo
end do
deallocate(ShellFunctionB1)
enddo
end do
iBasB1 = 0
!------------------------------------------------------------------------
! End loops over shell B1
!------------------------------------------------------------------------
enddo
end do
deallocate(ShellFunctionA1)
enddo
end do
iBasA1 = 0
!------------------------------------------------------------------------
! End loops over shell A1

View File

@ -32,7 +32,7 @@ subroutine Compute3eInt(debug,iType,nShell, &
integer,allocatable :: ShellFunctionB1(:,:),ShellFunctionB2(:,:),ShellFunctionB3(:,:)
double precision :: ExpBra(3),ExpKet(3)
double precision :: DBra(3),DKet(3)
double precision :: NormCoeff
double precision :: norm_coeff
integer :: iBasA1,iBasA2,iBasA3,iBasB1,iBasB2,iBasB3
integer :: iShA1,iShA2,iShA3,iShB1,iShB2,iShB3
@ -80,7 +80,7 @@ subroutine Compute3eInt(debug,iType,nShell, &
elseif(iType == 3) then
iFile = 33
open(unit=iFile,file='int/3eInt_Type3.dat')
endif
end if
!------------------------------------------------------------------------
! Loops over shell A1
@ -229,22 +229,22 @@ subroutine Compute3eInt(debug,iType,nShell, &
do iKA1=1,KBra(1)
ExpBra(1) = ExpShell(iShA1,iKA1)
DBra(1) = DShell(iShA1,iKA1)*NormCoeff(ExpBra(1),AngMomBra(1,1:3))
DBra(1) = DShell(iShA1,iKA1)*norm_coeff(ExpBra(1),AngMomBra(1,1:3))
do iKA2=1,KBra(2)
ExpBra(2) = ExpShell(iShA2,iKA2)
DBra(2) = DShell(iShA2,iKA2)*NormCoeff(ExpBra(2),AngMomBra(2,1:3))
DBra(2) = DShell(iShA2,iKA2)*norm_coeff(ExpBra(2),AngMomBra(2,1:3))
do iKA3=1,KBra(3)
ExpBra(3) = ExpShell(iShA3,iKA3)
DBra(3) = DShell(iShA3,iKA3)*NormCoeff(ExpBra(3),AngMomBra(3,1:3))
DBra(3) = DShell(iShA3,iKA3)*norm_coeff(ExpBra(3),AngMomBra(3,1:3))
do iKB1=1,KKet(1)
ExpKet(1) = ExpShell(iShB1,iKB1)
DKet(1) = DShell(iShB1,iKB1)*NormCoeff(ExpKet(1),AngMomKet(1,1:3))
DKet(1) = DShell(iShB1,iKB1)*norm_coeff(ExpKet(1),AngMomKet(1,1:3))
do iKB2=1,KKet(2)
ExpKet(2) = ExpShell(iShB2,iKB2)
DKet(2) = DShell(iShB2,iKB2)*NormCoeff(ExpKet(2),AngMomKet(2,1:3))
DKet(2) = DShell(iShB2,iKB2)*norm_coeff(ExpKet(2),AngMomKet(2,1:3))
do iKB3=1,KKet(3)
ExpKet(3) = ExpShell(iShB3,iKB3)
DKet(3) = DShell(iShB3,iKB3)*NormCoeff(ExpKet(3),AngMomKet(3,1:3))
DKet(3) = DShell(iShB3,iKB3)*norm_coeff(ExpKet(3),AngMomKet(3,1:3))
call S3eInt(debug,iType,np3eInt,nSigp3eInt, &
ExpS,KG,DG,ExpG, &
@ -254,12 +254,12 @@ subroutine Compute3eInt(debug,iType,nShell, &
c3eInt = c3eInt + DBra(1)*DBra(2)*DBra(3)*DKet(1)*DKet(2)*DKet(3)*p3eInt
enddo
enddo
enddo
enddo
enddo
enddo
end do
end do
end do
end do
end do
end do
call cpu_time(end_c3eInt)
nc3eInt = nc3eInt + 1
@ -271,50 +271,50 @@ subroutine Compute3eInt(debug,iType,nShell, &
if(.true.) then
write(*,'(A15,1X,I6,1X,I6,1X,I6,1X,I6,1X,I6,1X,I6,1X,F16.10)') &
'(a1a2a3|b1b2b3) = ',iBasA1,iBasA2,iBasA3,iBasB1,iBasB2,iBasB3,c3eInt
endif
endif
end if
end if
!------------------------------------------------------------------------
! End loops over contraction degrees
!------------------------------------------------------------------------
enddo
end do
deallocate(ShellFunctionB3)
enddo
end do
iBasB3 = 0
!------------------------------------------------------------------------
! End loops over shell B3
!------------------------------------------------------------------------
enddo
end do
deallocate(ShellFunctionB2)
enddo
end do
iBasB2 = 0
!------------------------------------------------------------------------
! End loops over shell B2
!------------------------------------------------------------------------
enddo
end do
deallocate(ShellFunctionB1)
enddo
end do
iBasB1 = 0
!------------------------------------------------------------------------
! End loops over shell B1
!------------------------------------------------------------------------
enddo
end do
deallocate(ShellFunctionA3)
enddo
end do
iBasA3 = 0
!------------------------------------------------------------------------
! End loops over shell A3
!------------------------------------------------------------------------
enddo
end do
deallocate(ShellFunctionA2)
enddo
end do
iBasA2 = 0
!------------------------------------------------------------------------
! End loops over shell A2
!------------------------------------------------------------------------
enddo
end do
deallocate(ShellFunctionA1)
enddo
end do
iBasA1 = 0
!------------------------------------------------------------------------
! End loops over shell A1

View File

@ -29,7 +29,7 @@ subroutine Compute4eInt(debug,nEl,iType,nShell,ExpS, &
ShellFunctionC(:,:),ShellFunctionD(:,:)
double precision :: ExpA,ExpB,ExpC,ExpD
double precision,allocatable :: DA,DB,DC,DD
double precision :: NormCoeff
double precision :: norm_coeff
integer :: iBasA,iBasB,iBasC,iBasD
integer :: iShA,iShB,iShC,iShD
@ -166,16 +166,16 @@ subroutine Compute4eInt(debug,nEl,iType,nShell,ExpS, &
do iKA=1,KA
ExpA = ExpShell(iShA,iKA)
DA = DShell(iShA,iKA)*NormCoeff(ExpA,AngMomA)
DA = DShell(iShA,iKA)*norm_coeff(ExpA,AngMomA)
do iKB=1,KB
ExpB = ExpShell(iShB,iKB)
DB = DShell(iShB,iKB)*NormCoeff(ExpB,AngMomB)
DB = DShell(iShB,iKB)*norm_coeff(ExpB,AngMomB)
do iKC=1,KC
ExpC = ExpShell(iShC,iKC)
DC = DShell(iShC,iKC)*NormCoeff(ExpC,AngMomC)
DC = DShell(iShC,iKC)*norm_coeff(ExpC,AngMomC)
do iKD=1,KD
ExpD = ExpShell(iShD,iKD)
DD = DShell(iShD,iKD)*NormCoeff(ExpD,AngMomD)
DD = DShell(iShD,iKD)*norm_coeff(ExpD,AngMomD)
! Erf module
! call ErfInt(debug,npErf,nSigpErf, &
@ -188,10 +188,10 @@ subroutine Compute4eInt(debug,nEl,iType,nShell,ExpS, &
! cErf = cErf + DA*DB*DC*DD*pErf
enddo
enddo
enddo
enddo
end do
end do
end do
end do
call cpu_time(end_cErf)
ncErf = ncErf + 1
@ -203,36 +203,36 @@ subroutine Compute4eInt(debug,nEl,iType,nShell,ExpS, &
if(debug) then
write(*,'(A10,1X,F16.10,1X,I6,1X,I6,1X,I6,1X,I6)') &
'(ab|erf(r)/r|cd) = ',cErf,iBasA,iBasB,iBasC,iBasD
endif
endif
end if
end if
!------------------------------------------------------------------------
! End loops over contraction degrees
!------------------------------------------------------------------------
enddo
end do
deallocate(ShellFunctionD)
enddo
end do
iBasD = 0
!------------------------------------------------------------------------
! End loops over shell D
!------------------------------------------------------------------------
enddo
end do
deallocate(ShellFunctionC)
enddo
end do
iBasC = 0
!------------------------------------------------------------------------
! End loops over shell C
!------------------------------------------------------------------------
enddo
end do
deallocate(ShellFunctionB)
enddo
end do
iBasB = 0
!------------------------------------------------------------------------
! End loops over shell B
!------------------------------------------------------------------------
enddo
end do
deallocate(ShellFunctionA)
enddo
end do
iBasA = 0
!------------------------------------------------------------------------
! End loops over shell A

View File

@ -26,7 +26,7 @@ subroutine ComputeKin(debug,nShell, &
integer,allocatable :: ShellFunctionA(:,:),ShellFunctionB(:,:)
double precision :: ExpA,ExpB
double precision,allocatable :: DA,DB
double precision :: NormCoeff
double precision :: norm_coeff
integer :: iBasA,iBasB
integer :: iShA,iShB
@ -115,10 +115,10 @@ subroutine ComputeKin(debug,nShell, &
do iKA=1,KA
ExpA = ExpShell(iShA,iKA)
DA = DShell(iShA,iKA)*NormCoeff(ExpA,AngMomA)
DA = DShell(iShA,iKA)*norm_coeff(ExpA,AngMomA)
do iKB=1,KB
ExpB = ExpShell(iShB,iKB)
DB = DShell(iShB,iKB)*NormCoeff(ExpB,AngMomB)
DB = DShell(iShB,iKB)*norm_coeff(ExpB,AngMomB)
call KinInt(npKin,nSigpKin, &
ExpA,CenterA,AngMomA, &
@ -127,8 +127,8 @@ subroutine ComputeKin(debug,nShell, &
cKin = cKin + DA*DB*pKin
enddo
enddo
end do
end do
call cpu_time(end_cKin)
ncKin = ncKin + 1
@ -138,21 +138,21 @@ subroutine ComputeKin(debug,nShell, &
write(9,'(I6,I6,F20.15)') iBasA,iBasB,cKin
if(debug) then
write(*,'(A10,1X,F16.10,1X,I6,1X,I6)') '(a|T|b) = ',cKin,iBasA,iBasB
endif
endif
end if
end if
!------------------------------------------------------------------------
! End loops over contraction degrees
!------------------------------------------------------------------------
enddo
end do
deallocate(ShellFunctionB)
enddo
end do
iBasB = 0
!------------------------------------------------------------------------
! End loops over shell B
!------------------------------------------------------------------------
enddo
end do
deallocate(ShellFunctionA)
enddo
end do
iBasA = 0
!------------------------------------------------------------------------
! End loops over shell A

View File

@ -29,7 +29,7 @@ subroutine ComputeNuc(debug,nShell, &
integer,allocatable :: ShellFunctionA(:,:),ShellFunctionB(:,:)
double precision :: ExpA,ExpB,ZC
double precision,allocatable :: DA,DB
double precision :: NormCoeff
double precision :: norm_coeff
integer :: iBasA,iBasB
integer :: iShA,iShB,iNucC
@ -131,10 +131,10 @@ subroutine ComputeNuc(debug,nShell, &
do iKA=1,KA
ExpA = ExpShell(iShA,iKA)
DA = DShell(iShA,iKA)*NormCoeff(ExpA,AngMomA)
DA = DShell(iShA,iKA)*norm_coeff(ExpA,AngMomA)
do iKB=1,KB
ExpB = ExpShell(iShB,iKB)
DB = DShell(iShB,iKB)*NormCoeff(ExpB,AngMomB)
DB = DShell(iShB,iKB)*norm_coeff(ExpB,AngMomB)
call NucInt(debug,npNuc,nSigpNuc, &
ExpA,CenterA,AngMomA, &
@ -144,12 +144,12 @@ subroutine ComputeNuc(debug,nShell, &
cNuc = cNuc - DA*DB*ZC*pNuc
enddo
enddo
end do
end do
!------------------------------------------------------------------------
! End loops over contraction degrees
!------------------------------------------------------------------------
enddo
end do
call cpu_time(end_cNuc)
!------------------------------------------------------------------------
! End loops over nuclear centers C
@ -163,19 +163,19 @@ subroutine ComputeNuc(debug,nShell, &
if(debug) then
write(*,'(A10,1X,F16.10,1X,I6,1X,I6)') '(a|V|b) = ',cNuc,iBasA,iBasB
write(*,*)
endif
endif
end if
end if
enddo
end do
deallocate(ShellFunctionB)
enddo
end do
iBasB = 0
!------------------------------------------------------------------------
! End loops over shell B
!------------------------------------------------------------------------
enddo
end do
deallocate(ShellFunctionA)
enddo
end do
iBasA = 0
!------------------------------------------------------------------------
! End loops over shell A

View File

@ -26,7 +26,7 @@ subroutine ComputeOv(debug,nBas,nShell, &
integer,allocatable :: ShellFunctionA(:,:),ShellFunctionB(:,:)
double precision :: ExpA,ExpB
double precision,allocatable :: DA,DB
double precision :: NormCoeff
double precision :: norm_coeff
integer :: iBasA,iBasB
integer :: iShA,iShB
@ -117,10 +117,10 @@ subroutine ComputeOv(debug,nBas,nShell, &
do iKA=1,KA
ExpA = ExpShell(iShA,iKA)
DA = DShell(iShA,iKA)*NormCoeff(ExpA,AngMomA)
DA = DShell(iShA,iKA)*norm_coeff(ExpA,AngMomA)
do iKB=1,KB
ExpB = ExpShell(iShB,iKB)
DB = DShell(iShB,iKB)*NormCoeff(ExpB,AngMomB)
DB = DShell(iShB,iKB)*norm_coeff(ExpB,AngMomB)
call OvInt(npOv,nSigpOv, &
ExpA,CenterA,AngMomA, &
@ -129,8 +129,8 @@ subroutine ComputeOv(debug,nBas,nShell, &
cOv = cOv + DA*DB*pOv
enddo
enddo
end do
end do
call cpu_time(end_cOv)
ncOv = ncOv + 1
@ -141,22 +141,22 @@ subroutine ComputeOv(debug,nBas,nShell, &
write(8,'(I6,I6,F20.15)') iBasA,iBasB,cOv
if(debug) then
write(*,'(A10,1X,F16.10,1X,I6,1X,I6)') '(a|b) = ',cOv,iBasA,iBasB
endif
endif
end if
end if
!------------------------------------------------------------------------
! End loops over contraction degrees
!------------------------------------------------------------------------
enddo
end do
deallocate(ShellFunctionB)
enddo
end do
iBasB = 0
!------------------------------------------------------------------------
! End loops over shell B
!------------------------------------------------------------------------
enddo
end do
deallocate(ShellFunctionA)
enddo
end do
iBasA = 0
!------------------------------------------------------------------------
! End loops over shell A

View File

@ -44,7 +44,7 @@ subroutine FormVRR3e(ExpZ,ExpG,CenterZ,DY0,DY1,D2Y0,D2Y1,delta0,delta1,Y0,Y1)
do i=1,3
ZetaMat(i,i) = ExpZ(i)
enddo
end do
! print*,'Zeta'
! call matout(3,3,ZetaMat)
@ -64,20 +64,20 @@ subroutine FormVRR3e(ExpZ,ExpG,CenterZ,DY0,DY1,D2Y0,D2Y1,delta0,delta1,Y0,Y1)
do i=1,3
do j=1,i-1
GMat(i,j) = - ExpG(j,i)
enddo
end do
do j=i+1,3
GMat(i,j) = - ExpG(i,j)
enddo
enddo
end do
end do
do i=1,3
do j=1,i-1
GMat(i,i) = GMat(i,i) + ExpG(j,i)
enddo
end do
do j=i+1,3
GMat(i,i) = GMat(i,i) + ExpG(i,j)
enddo
enddo
end do
end do
! print*,'G'
! call matout(3,3,GMat)
@ -89,10 +89,10 @@ subroutine FormVRR3e(ExpZ,ExpG,CenterZ,DY0,DY1,D2Y0,D2Y1,delta0,delta1,Y0,Y1)
do k=1,3
CenterY(i,j,k) = CenterZ(i,k) - CenterZ(j,k)
Y2Mat(i,j) = Y2Mat(i,j) + CenterY(i,j,k)**2
enddo
end do
YMat(i,j) = sqrt(Y2Mat(i,j))
enddo
enddo
end do
end do
! print*,'Y'
! call matout(3,3,YMat)
@ -106,8 +106,8 @@ subroutine FormVRR3e(ExpZ,ExpG,CenterZ,DY0,DY1,D2Y0,D2Y1,delta0,delta1,Y0,Y1)
do j=1,3
Delta0Mat(i,j) = ZetaMat(i,j) + GMat(i,j)
Delta1Mat(i,j) = Delta0Mat(i,j) + CMat(i,j)
enddo
enddo
end do
end do
! Form the DY and D2Y matrices
@ -117,10 +117,10 @@ subroutine FormVRR3e(ExpZ,ExpG,CenterZ,DY0,DY1,D2Y0,D2Y1,delta0,delta1,Y0,Y1)
DYMat(i,j,k) = KappaCross(i,j,k)*YMat(j,k)/ExpZ(i)
do l=1,3
D2YMat(i,j,k,l) = 0.5d0*KappaCross(i,k,l)*KappaCross(j,k,l)/(ExpZ(i)*ExpZ(j))
enddo
enddo
enddo
enddo
end do
end do
end do
end do
! Compute the inverse of the Delta0 and Delta1 matrices
@ -130,8 +130,8 @@ subroutine FormVRR3e(ExpZ,ExpG,CenterZ,DY0,DY1,D2Y0,D2Y1,delta0,delta1,Y0,Y1)
do j=1,3
InvDelta0Mat(i,j) = Delta0Mat(i,j)
InvDelta1Mat(i,j) = Delta1Mat(i,j)
enddo
enddo
end do
end do
! call amove(3,3,Delta0Mat,InvDelta0Mat)
! call amove(3,3,Delta1Mat,InvDelta1Mat)
@ -150,10 +150,10 @@ subroutine FormVRR3e(ExpZ,ExpG,CenterZ,DY0,DY1,D2Y0,D2Y1,delta0,delta1,Y0,Y1)
do l=1,3
D0Mat(i,j) = D0Mat(i,k) + ZetaMat(i,k)*InvDelta0Mat(k,l)*ZetaMat(l,j)
D1Mat(i,j) = D1Mat(i,k) + ZetaMat(i,k)*InvDelta1Mat(k,l)*ZetaMat(l,j)
enddo
enddo
enddo
enddo
end do
end do
end do
end do
! Form the derivative matrices
@ -163,8 +163,8 @@ subroutine FormVRR3e(ExpZ,ExpG,CenterZ,DY0,DY1,D2Y0,D2Y1,delta0,delta1,Y0,Y1)
do j=1,3
call CalcTrAB(3,D0Mat,D2YMat,D2Y0(i,j))
call CalcTrAB(3,D1Mat,D2YMat,D2Y1(i,j))
enddo
enddo
end do
end do
! Compute Y0 and Y1

View File

@ -40,7 +40,7 @@ function G2eInt(debug,iType, &
do i=1,2
ExpZi(i) = 1d0/(ExpBra(i) + ExpKet(i))
enddo
end do
NormABSq = 0d0
do j=1,3
@ -49,20 +49,20 @@ function G2eInt(debug,iType, &
CenterAB(i,j) = CenterBra(i,j) - CenterKet(i,j)
CenterZA(i,j) = CenterZ(i,j) - CenterBra(i,j)
NormABSq(i) = NormABSq(i) + CenterAB(i,j)**2
enddo
enddo
end do
end do
do i=1,2
GAB(i) = (pi*ExpZi(i))**(1.5d0)*exp(-ExpBra(i)*ExpKet(i)*NormABSq(i)*ExpZi(i))
enddo
end do
! Pre-computed shell-quartet quantities
do i=1,2
do j=1,2
ExpY(i,j) = 1d0/(ExpZi(i) + ExpZi(j))
enddo
enddo
end do
end do
do i=1,2
do j=1,2
@ -70,9 +70,9 @@ function G2eInt(debug,iType, &
do k=1,3
CenterY(i,j,k) = CenterZ(i,k) - CenterZ(j,k)
NormYSq(i,j) = NormYSq(i,j) + CenterY(i,j,k)**2
enddo
enddo
enddo
end do
end do
end do
! fG = (ExpZ(1)*ExpZ(2)*ExpG)/(ExpZ(1)*ExpZ(2) + ExpZ(1)*ExpG + ExpZ(2)*ExpG)
fG = 1d0/(ExpZi(1) + 1d0/ExpG + ExpZi(2))
@ -84,7 +84,7 @@ function G2eInt(debug,iType, &
TotAngMomBra(i) = AngMomBra(i,1) + AngMomBra(i,2) + AngMomBra(i,3)
TotAngMomKet(i) = AngMomKet(i,1) + AngMomKet(i,2) + AngMomKet(i,3)
maxm = maxm + TotAngMomBra(i) + TotAngMomKet(i)
enddo
end do
! Pre-compute (00|00)^m
@ -97,7 +97,7 @@ function G2eInt(debug,iType, &
call CalcOmYuk(maxm,ExpG,ExpY(1,2),fG,NormYSq(1,2),Om)
elseif(iType == 4) then
call CalcOmErf(maxm,ExpY(1,2),fG,NormYSq(1,2),Om)
endif
end if
call cpu_time(finish_Om)
@ -107,9 +107,9 @@ function G2eInt(debug,iType, &
write(*,*) '(00|00)^m'
do i=0,maxm
write(*,*) i,Om(i)
enddo
end do
write(*,*)
endif
end if
!------------------------------------------------------------------------
! Launch reccurence relations!
@ -121,10 +121,10 @@ function G2eInt(debug,iType, &
a1a2b1b2 = Om(0)
else
a1a2b1b2 = VRR2e(0,AngMomBra,maxm,Om,ExpZi,ExpY,CenterZA,CenterY)
endif
end if
else
a1a2b1b2 = HRR2e(AngMomBra,AngMomKet,maxm,Om,ExpZi,ExpY,CenterAB,CenterZA,CenterY)
endif
end if
call cpu_time(finish_RR)

View File

@ -49,7 +49,7 @@ function G3eInt(debug,iType, &
do i=1,3
ExpZ(i) = ExpBra(i) + ExpKet(i)
enddo
end do
NormABSq = 0d0
do i=1,3
@ -58,12 +58,12 @@ function G3eInt(debug,iType, &
CenterAB(i,j) = CenterBra(i,j) - CenterKet(i,j)
CenterZA(i,j) = CenterZ(i,j) - CenterBra(i,j)
NormABSq(i) = NormABSq(i) + CenterAB(i,j)**2
enddo
enddo
end do
end do
do i=1,3
GAB(i) = (pi/ExpZ(i))**(1.5d0)*exp(-ExpBra(i)*ExpKet(i)*NormABSq(i)/ExpZ(i))
enddo
end do
! Pre-computed shell-sextet quantities
@ -76,7 +76,7 @@ function G3eInt(debug,iType, &
TotAngMomBra(i) = AngMomBra(i,1) + AngMomBra(i,2) + AngMomBra(i,3)
TotAngMomKet(i) = AngMomKet(i,1) + AngMomKet(i,2) + AngMomKet(i,3)
maxm = maxm + TotAngMomBra(i) + TotAngMomKet(i)
enddo
end do
! Pre-compute (000|000)^m
@ -91,9 +91,9 @@ function G3eInt(debug,iType, &
write(*,*) '(000|000)^m'
do i=0,maxm
write(*,*) i,Om(i)
enddo
end do
write(*,*)
endif
end if
!------------------------------------------------------------------------
! Launch reccurence relations!
@ -104,10 +104,10 @@ function G3eInt(debug,iType, &
a1a2a3b1b2b3 = Om(0)
else
a1a2a3b1b2b3 = VRR3e(0,AngMomBra,maxm,Om,ExpZ,CenterZA,DY0,DY1,D2Y0,D2Y1)
endif
end if
else
a1a2a3b1b2b3 = HRR3e(AngMomBra,AngMomKet,maxm,Om,ExpZ,CenterAB,CenterZA,DY0,DY1,D2Y0,D2Y1)
endif
end if
call cpu_time(finish_RR)

View File

@ -87,7 +87,7 @@ function GF12Int(ExpG,ExpA,CenterA,AngMomA,ExpB,CenterB,AngMomB,ExpC,CenterC,Ang
do i=1,3
CenterRA(i) = CenterP(i) - CenterA(i) + fP*(CenterQ(i) - CenterP(i))
CenterRC(i) = CenterQ(i) - CenterC(i) + fQ*(CenterP(i) - CenterQ(i))
enddo
end do
!------------------------------------------------------------------------
! Launch reccurence relations!
!------------------------------------------------------------------------

View File

@ -23,8 +23,8 @@ subroutine GenerateShell(atot,nShellFunction,ShellFunction)
ShellFunction(ia,1) = ax
ShellFunction(ia,2) = ay
ShellFunction(ia,3) = az
endif
enddo
enddo
end if
end do
end do
end subroutine GenerateShell

View File

@ -31,7 +31,7 @@ recursive function HRR2e(AngMomBra,AngMomKet, &
NegAngMomKet(i) = AngMomKet(i,1) < 0 .or. AngMomKet(i,2) < 0 .or. AngMomKet(i,3) < 0
TotAngMomBra(i) = AngMomBra(i,1) + AngMomBra(i,2) + AngMomBra(i,3)
TotAngMomKet(i) = AngMomKet(i,1) + AngMomKet(i,2) + AngMomKet(i,3)
enddo
end do
!------------------------------------------------------------------------
! Termination condition
@ -52,8 +52,8 @@ recursive function HRR2e(AngMomBra,AngMomKet, &
do j=1,3
a1p(i,j) = AngMomBra(i,j)
b1m(i,j) = AngMomKet(i,j)
enddo
enddo
end do
end do
! Loop over cartesian directions
xyz = 0
if (AngMomKet(1,1) > 0) then
@ -64,7 +64,7 @@ recursive function HRR2e(AngMomBra,AngMomKet, &
xyz = 3
else
write(*,*) 'xyz = 0 in HRR2e!'
endif
end if
! End loop over cartesian directions
a1p(1,xyz) = a1p(1,xyz) + 1
b1m(1,xyz) = b1m(1,xyz) - 1
@ -78,8 +78,8 @@ recursive function HRR2e(AngMomBra,AngMomKet, &
do j=1,3
a2p(i,j) = AngMomBra(i,j)
b2m(i,j) = AngMomKet(i,j)
enddo
enddo
end do
end do
! Loop over cartesian directions
xyz = 0
if (AngMomKet(2,1) > 0) then
@ -90,12 +90,12 @@ recursive function HRR2e(AngMomBra,AngMomKet, &
xyz = 3
else
write(*,*) 'xyz = 0 in HRR2e!'
endif
end if
! End loop over cartesian directions
a2p(2,xyz) = a2p(2,xyz) + 1
b2m(2,xyz) = b2m(2,xyz) - 1
a1a2b1b2 = HRR2e(a2p,b2m,maxm,Om,ExpZi,ExpY,CenterAB,CenterZA,CenterY) &
+ CenterAB(2,xyz)*HRR2e(AngMomBra,b2m,maxm,Om,ExpZi,ExpY,CenterAB,CenterZA,CenterY)
endif
end if
end function HRR2e

View File

@ -30,7 +30,7 @@ recursive function HRR3e(AngMomBra,AngMomKet,maxm,Om,ExpZ,CenterAB,CenterZA,DY0,
NegAngMomKet(i) = AngMomKet(i,1) < 0 .or. AngMomKet(i,2) < 0 .or. AngMomKet(i,3) < 0
TotAngMomBra(i) = AngMomBra(i,1) + AngMomBra(i,2) + AngMomBra(i,3)
TotAngMomKet(i) = AngMomKet(i,1) + AngMomKet(i,2) + AngMomKet(i,3)
enddo
end do
!------------------------------------------------------------------------
! Termination condition
@ -50,8 +50,8 @@ recursive function HRR3e(AngMomBra,AngMomKet,maxm,Om,ExpZ,CenterAB,CenterZA,DY0,
do j=1,3
a1p(i,j) = AngMomBra(i,j)
b1m(i,j) = AngMomKet(i,j)
enddo
enddo
end do
end do
! Loop over cartesian directions
xyz = 0
if (AngMomKet(1,1) > 0) then
@ -62,7 +62,7 @@ recursive function HRR3e(AngMomBra,AngMomKet,maxm,Om,ExpZ,CenterAB,CenterZA,DY0,
xyz = 3
else
write(*,*) 'xyz = 0 in HRR3e!'
endif
end if
! End loop over cartesian directions
a1p(1,xyz) = a1p(1,xyz) + 1
b1m(1,xyz) = b1m(1,xyz) - 1
@ -77,8 +77,8 @@ recursive function HRR3e(AngMomBra,AngMomKet,maxm,Om,ExpZ,CenterAB,CenterZA,DY0,
do j=1,3
a2p(i,j) = AngMomBra(i,j)
b2m(i,j) = AngMomKet(i,j)
enddo
enddo
end do
end do
! Loop over cartesian directions
xyz = 0
if (AngMomKet(2,1) > 0) then
@ -89,7 +89,7 @@ recursive function HRR3e(AngMomBra,AngMomKet,maxm,Om,ExpZ,CenterAB,CenterZA,DY0,
xyz = 3
else
write(*,*) 'xyz = 0 in HRR3e!'
endif
end if
! End loop over cartesian directions
a2p(2,xyz) = a2p(2,xyz) + 1
b2m(2,xyz) = b2m(2,xyz) - 1
@ -104,8 +104,8 @@ recursive function HRR3e(AngMomBra,AngMomKet,maxm,Om,ExpZ,CenterAB,CenterZA,DY0,
do j=1,3
a3p(i,j) = AngMomBra(i,j)
b3m(i,j) = AngMomKet(i,j)
enddo
enddo
end do
end do
! Loop over cartesian directions
xyz = 0
if (AngMomKet(3,1) > 0) then
@ -116,13 +116,13 @@ recursive function HRR3e(AngMomBra,AngMomKet,maxm,Om,ExpZ,CenterAB,CenterZA,DY0,
xyz = 3
else
write(*,*) 'xyz = 0 in HRR3e!'
endif
end if
! End loop over cartesian directions
a3p(3,xyz) = a3p(3,xyz) + 1
b3m(3,xyz) = b3m(3,xyz) - 1
a1a2a3b1b2b3 = HRR3e(a3p,b3m,maxm,Om,ExpZ,CenterAB,CenterZA,DY0,DY1,D2Y0,D2Y1) &
+ CenterAB(3,xyz)* &
HRR3e(AngMomBra,b3m,maxm,Om,ExpZ,CenterAB,CenterZA,DY0,DY1,D2Y0,D2Y1)
endif
end if
end function HRR3e

View File

@ -48,7 +48,7 @@ recursive function HRRNuc(AngMomA,AngMomB,maxm,Om,ExpPi,CenterAB,CenterPA,Center
do i=1,3
ap(i) = AngMomA(i)
bm(i) = AngMomB(i)
enddo
end do
! Loop over cartesian directions
xyz = 0
if (AngMomB(1) > 0) then
@ -59,13 +59,13 @@ recursive function HRRNuc(AngMomA,AngMomB,maxm,Om,ExpPi,CenterAB,CenterPA,Center
xyz = 3
else
write(*,*) 'xyz = 0 in HRRNuc!'
endif
end if
! End loop over cartesian directions
ap(xyz) = ap(xyz) + 1
bm(xyz) = bm(xyz) - 1
Gab = HRRNuc(ap,bm,maxm,Om,ExpPi,CenterAB,CenterPA,CenterPC) &
+ CenterAB(xyz)*HRRNuc(AngMomA,bm,maxm,Om,ExpPi,CenterAB,CenterPA,CenterPC)
endif
endif
end if
end if
end function HRRNuc

View File

@ -22,7 +22,7 @@ recursive function HRROv(AngMomA,AngMomB,ExpPi,CenterAB,CenterPA) &
else
Gab = HRROv(AngMomA+1,AngMomB-1,ExpPi,CenterAB,CenterPA) &
+ CenterAB*HRROv(AngMomA,AngMomB-1,ExpPi,CenterAB,CenterPA)
endif
endif
end if
end if
end function HRROv

View File

@ -109,7 +109,7 @@ program IntPak
write(*,'(A65,1X,F9.3,A8)') 'Total CPU time = ',t_1eInt(iType),' seconds'
write(*,*)
endif
end if
!------------------------------------------------------------------------
! Compute one-electron kinetic integrals
@ -137,7 +137,7 @@ program IntPak
write(*,'(A65,1X,F9.3,A8)') 'Total CPU time = ',t_1eInt(iType),' seconds'
write(*,*)
endif
end if
!------------------------------------------------------------------------
! Compute one-electron nuclear attraction integrals
@ -166,7 +166,7 @@ program IntPak
write(*,'(A65,1X,F9.3,A8)') 'Total CPU time = ',t_1eInt(iType),' seconds'
write(*,*)
endif
end if
!------------------------------------------------------------------------
! Compute ERIs
@ -201,7 +201,7 @@ program IntPak
deallocate(DG,ExpG)
endif
end if
!------------------------------------------------------------------------
! Compute F12 two-electron integrals
@ -263,7 +263,7 @@ program IntPak
deallocate(DG,ExpG)
endif
end if
!------------------------------------------------------------------------
! Compute Yukawa two-electron integrals
@ -298,7 +298,7 @@ program IntPak
deallocate(DG,ExpG)
endif
end if
!------------------------------------------------------------------------
! Compute long-range Coulomb two-electron integrals
@ -334,7 +334,7 @@ program IntPak
deallocate(DG,ExpG)
endif
end if
!------------------------------------------------------------------------
! Compute three-electron integrals: Type 1 => chain C12 S23
@ -372,7 +372,7 @@ program IntPak
deallocate(DG,ExpG)
endif
end if
!------------------------------------------------------------------------
! Compute three-electron integrals: Type 2 => cyclic C12 S13 S23
@ -407,7 +407,7 @@ program IntPak
deallocate(DG,ExpG)
endif
end if
!------------------------------------------------------------------------
! Compute three-electron integrals: Type 3 => chain S13 S23
@ -442,7 +442,7 @@ program IntPak
deallocate(DG,ExpG)
endif
end if
!------------------------------------------------------------------------
! Compute four-electron integrals: Type 1 => chain C12 S14 S23
@ -477,7 +477,7 @@ program IntPak
deallocate(DG,ExpG)
endif
end if
!------------------------------------------------------------------------
! Compute four-electron integrals: Type 2 => trident C12 S13 S14
@ -511,7 +511,7 @@ program IntPak
deallocate(DG,ExpG)
endif
end if
!------------------------------------------------------------------------
! Compute four-electron integrals: Type 3 => chain C12 S13 S34
@ -546,7 +546,7 @@ program IntPak
deallocate(DG,ExpG)
endif
end if
!------------------------------------------------------------------------
! End of IntPak
!------------------------------------------------------------------------

View File

@ -71,6 +71,6 @@ subroutine KinInt(npKin,nSigpKin,ExpA,CenterA,AngMomA,ExpB,CenterB,AngMomB,pKin)
npKin = npKin + 1
if(abs(pKin) > 1d-15) then
nSigpKin = nSigpKin + 1
endif
end if
end subroutine KinInt

View File

@ -57,7 +57,7 @@ subroutine NucInt(debug,npNuc,nSigpNuc, &
CenterPC(i) = CenterP(i) - CenterC(i)
NormABSq = NormABSq + CenterAB(i)**2
NormPCSq = NormPCSq + CenterPC(i)**2
enddo
end do
G = (pi*ExpPi)**(1.5d0)*exp(-NormABSq/(ExpAi+ExpBi))
@ -81,9 +81,9 @@ subroutine NucInt(debug,npNuc,nSigpNuc, &
write(*,*) '(0|V|0)^m'
do i=0,maxm
write(*,*) i,Om(i)
enddo
end do
write(*,*)
endif
end if
!------------------------------------------------------------------------
! Launch reccurence relations!
@ -105,7 +105,7 @@ subroutine NucInt(debug,npNuc,nSigpNuc, &
if(abs(pNuc) > 1d-15) then
nSigpNuc = nSigpNuc + 1
! write(*,'(A10,1X,F16.10,1X,I6,1X,I6)') '[a|V|b] = ',pNuc
endif
end if
! Deallocate arrays

View File

@ -69,6 +69,6 @@ subroutine OvInt(npOv,nSigpOv,ExpA,CenterA,AngMomA,ExpB,CenterB,AngMomB,pOv)
npOv = npOv + 1
if(abs(pOv) > 1d-15) then
nSigpOv = nSigpOv + 1
endif
end if
end subroutine OvInt

View File

@ -46,7 +46,7 @@ subroutine S2eInt(debug,iType,np2eInt,nSigp2eInt, &
ExpKet(1),CenterKet(1,1:3),AngMomKet(1,1:3), &
ExpBra(2),CenterBra(2,1:3),AngMomBra(2,1:3), &
ExpKet(2),CenterKet(2,1:3),AngMomKet(2,1:3))
enddo
end do
else
do k=1,KG
ExpSG = ExpG(k)*ExpS**2
@ -55,8 +55,8 @@ subroutine S2eInt(debug,iType,np2eInt,nSigp2eInt, &
ExpSG, &
ExpBra,CenterBra,AngMomBra, &
ExpKet,CenterKet,AngMomKet)
enddo
endif
end do
end if
! Print result
@ -65,6 +65,6 @@ subroutine S2eInt(debug,iType,np2eInt,nSigp2eInt, &
if(abs(p2eInt) > 1d-15) then
nSigp2eInt = nSigp2eInt + 1
if(.false.) write(*,'(A15,1X,F16.10)') '[a1a2|b1b2] = ',p2eInt
endif
end if
end subroutine S2eInt

View File

@ -71,6 +71,6 @@ subroutine S3eInt(debug,iType,np3eInt,nSigp3eInt, &
if(abs(p3eInt) > 1d-15) then
nSigp3eInt = nSigp3eInt + 1
if(.false.) write(*,'(A15,1X,F16.10)') '[a1a2a3|b1b2b3] = ',p3eInt
endif
end if
end subroutine S3eInt

View File

@ -31,7 +31,7 @@ recursive function VRR2e(m,AngMomBra,maxm,Om,ExpZi,ExpY,CenterZA,CenterY) &
do i=1,2
NegAngMomBra(i) = AngMomBra(i,1) < 0 .or. AngMomBra(i,2) < 0 .or. AngMomBra(i,3) < 0
TotAngMomBra(i) = AngMomBra(i,1) + AngMomBra(i,2) + AngMomBra(i,3)
enddo
end do
fZ(1) = ExpY(1,2)*ExpZi(1)
fZ(2) = ExpY(1,2)*ExpZi(2)
@ -55,8 +55,8 @@ recursive function VRR2e(m,AngMomBra,maxm,Om,ExpZi,ExpY,CenterZA,CenterY) &
do j=1,3
a1m(i,j) = AngMomBra(i,j)
a1mm(i,j) = AngMomBra(i,j)
enddo
enddo
end do
end do
! Loop over cartesian directions
xyz = 0
if (AngMomBra(1,1) > 0) then
@ -67,7 +67,7 @@ recursive function VRR2e(m,AngMomBra,maxm,Om,ExpZi,ExpY,CenterZA,CenterY) &
xyz = 3
else
write(*,*) 'xyz = 0 in VRR2e!'
endif
end if
! End loop over cartesian directions
a1m(1,xyz) = a1m(1,xyz) - 1
a1mm(1,xyz) = a1mm(1,xyz) - 2
@ -82,7 +82,7 @@ recursive function VRR2e(m,AngMomBra,maxm,Om,ExpZi,ExpY,CenterZA,CenterY) &
+ 0.5d0*dble(AngMomBra(1,xyz)-1)*ExpZi(1)*( &
VRR2e(m,a1mm,maxm,Om,ExpZi,ExpY,CenterZA,CenterY) &
- fZ(1)*VRR2e(m+1,a1mm,maxm,Om,ExpZi,ExpY,CenterZA,CenterY))
endif
end if
!------------------------------------------------------------------------
! 2nd vertical recurrence relation (5 terms): (a0|c+0)^m
!------------------------------------------------------------------------
@ -92,8 +92,8 @@ recursive function VRR2e(m,AngMomBra,maxm,Om,ExpZi,ExpY,CenterZA,CenterY) &
a2m(i,j) = AngMomBra(i,j)
a2mm(i,j) = AngMomBra(i,j)
a1m2m(i,j) = AngMomBra(i,j)
enddo
enddo
end do
end do
! Loop over cartesian directions
xyz = 0
if (AngMomBra(2,1) > 0) then
@ -104,7 +104,7 @@ recursive function VRR2e(m,AngMomBra,maxm,Om,ExpZi,ExpY,CenterZA,CenterY) &
xyz = 3
else
write(*,*) 'xyz = 0 in VRR2e!'
endif
end if
! End loop over cartesian directions
a2m(2,xyz) = a2m(2,xyz) - 1
a2mm(2,xyz) = a2mm(2,xyz) - 2
@ -121,10 +121,10 @@ recursive function VRR2e(m,AngMomBra,maxm,Om,ExpZi,ExpY,CenterZA,CenterY) &
+ 0.5d0*dble(AngMomBra(2,xyz)-1)*ExpZi(2)*( &
VRR2e(m,a2mm,maxm,Om,ExpZi,ExpY,CenterZA,CenterY) &
- fZ(2)*VRR2e(m+1,a2mm,maxm,Om,ExpZi,ExpY,CenterZA,CenterY))
endif
end if
if(AngMomBra(1,xyz) > 0) &
a1a2 = a1a2 &
+ 0.5d0*dble(AngMomBra(1,xyz))*fZ(2)*ExpZi(1)*VRR2e(m+1,a1m2m,maxm,Om,ExpZi,ExpY,CenterZA,CenterY)
endif
end if
end function VRR2e

View File

@ -31,7 +31,7 @@ recursive function VRR3e(m,AngMomBra,maxm,Om,ExpZ,CenterZA,DY0,DY1,D2Y0,D2Y1) &
do i=1,3
NegAngMomBra(i) = AngMomBra(i,1) < 0 .or. AngMomBra(i,2) < 0 .or. AngMomBra(i,3) < 0
TotAngMomBra(i) = AngMomBra(i,1) + AngMomBra(i,2) + AngMomBra(i,3)
enddo
end do
!------------------------------------------------------------------------
! Termination condition
@ -51,8 +51,8 @@ recursive function VRR3e(m,AngMomBra,maxm,Om,ExpZ,CenterZA,DY0,DY1,D2Y0,D2Y1) &
do j=1,3
a1m(i,j) = AngMomBra(i,j)
a1mm(i,j) = AngMomBra(i,j)
enddo
enddo
end do
end do
! Loop over cartesian directions
xyz = 0
if (AngMomBra(1,1) > 0) then
@ -63,7 +63,7 @@ recursive function VRR3e(m,AngMomBra,maxm,Om,ExpZ,CenterZA,DY0,DY1,D2Y0,D2Y1) &
xyz = 3
else
write(*,*) 'xyz = 0 in VRR3e!'
endif
end if
! End loop over cartesian directions
a1m(1,xyz) = a1m(1,xyz) - 1
a1mm(1,xyz) = a1mm(1,xyz) - 2
@ -77,7 +77,7 @@ recursive function VRR3e(m,AngMomBra,maxm,Om,ExpZ,CenterZA,DY0,DY1,D2Y0,D2Y1) &
- (DY1(1) - DY0(1))*VRR3e(m+1,a1m, maxm,Om,ExpZ,CenterZA,DY0,DY1,D2Y0,D2Y1) &
+ dble(AngMomBra(1,xyz)-1)*(0.5d0/ExpZ(1) - D2Y0(1,1))*VRR3e(m, a1mm,maxm,Om,ExpZ,CenterZA,DY0,DY1,D2Y0,D2Y1) &
- dble(AngMomBra(1,xyz)-1)*(D2Y1(1,1) - D2Y0(1,1))*VRR3e(m+1,a1mm,maxm,Om,ExpZ,CenterZA,DY0,DY1,D2Y0,D2Y1)
endif
end if
!------------------------------------------------------------------------
! 2nd vertical recurrence relation (6 terms): (a1a2+0|000)^m
!------------------------------------------------------------------------
@ -87,8 +87,8 @@ recursive function VRR3e(m,AngMomBra,maxm,Om,ExpZ,CenterZA,DY0,DY1,D2Y0,D2Y1) &
a2m(i,j) = AngMomBra(i,j)
a2mm(i,j) = AngMomBra(i,j)
a1m2m(i,j) = AngMomBra(i,j)
enddo
enddo
end do
end do
! Loop over cartesian directions
xyz = 0
if (AngMomBra(2,1) > 0) then
@ -99,7 +99,7 @@ recursive function VRR3e(m,AngMomBra,maxm,Om,ExpZ,CenterZA,DY0,DY1,D2Y0,D2Y1) &
xyz = 3
else
write(*,*) 'xyz = 0 in VRR3e!'
endif
end if
! End loop over cartesian directions
a2m(2,xyz) = a2m(2,xyz) - 1
a2mm(2,xyz) = a2mm(2,xyz) - 2
@ -115,7 +115,7 @@ recursive function VRR3e(m,AngMomBra,maxm,Om,ExpZ,CenterZA,DY0,DY1,D2Y0,D2Y1) &
- (DY1(2) - DY0(2))*VRR3e(m+1,a2m, maxm,Om,ExpZ,CenterZA,DY0,DY1,D2Y0,D2Y1) &
+ dble(AngMomBra(2,xyz)-1)*(0.5d0/ExpZ(2) - D2Y0(2,2))*VRR3e(m, a2mm, maxm,Om,ExpZ,CenterZA,DY0,DY1,D2Y0,D2Y1) &
- dble(AngMomBra(2,xyz)-1)*(D2Y1(2,2) - D2Y0(2,2))*VRR3e(m+1,a2mm, maxm,Om,ExpZ,CenterZA,DY0,DY1,D2Y0,D2Y1)
endif
end if
if(AngMomBra(1,xyz) > 0) &
a1a2a3 = a1a2a3 &
+ dble(AngMomBra(1,xyz))*(-D2Y0(2,1))*VRR3e(m, a1m2m,maxm,Om,ExpZ,CenterZA,DY0,DY1,D2Y0,D2Y1) &
@ -130,8 +130,8 @@ recursive function VRR3e(m,AngMomBra,maxm,Om,ExpZ,CenterZA,DY0,DY1,D2Y0,D2Y1) &
a3mm(i,j) = AngMomBra(i,j)
a1m3m(i,j) = AngMomBra(i,j)
a2m3m(i,j) = AngMomBra(i,j)
enddo
enddo
end do
end do
! Loop over cartesian directions
xyz = 0
if (AngMomBra(3,1) > 0) then
@ -142,7 +142,7 @@ recursive function VRR3e(m,AngMomBra,maxm,Om,ExpZ,CenterZA,DY0,DY1,D2Y0,D2Y1) &
xyz = 3
else
write(*,*) 'xyz = 0 in VRR3e!'
endif
end if
! End loop over cartesian directions
a3m(3,xyz) = a3m(3,xyz) - 1
a3mm(3,xyz) = a3mm(3,xyz) - 2
@ -160,7 +160,7 @@ recursive function VRR3e(m,AngMomBra,maxm,Om,ExpZ,CenterZA,DY0,DY1,D2Y0,D2Y1) &
- (DY1(3) - DY0(3))*VRR3e(m+1,a3m, maxm,Om,ExpZ,CenterZA,DY0,DY1,D2Y0,D2Y1) &
+ dble(AngMomBra(3,xyz)-1)*(0.5d0/ExpZ(3) - D2Y0(3,3))*VRR3e(m, a3mm, maxm,Om,ExpZ,CenterZA,DY0,DY1,D2Y0,D2Y1) &
- dble(AngMomBra(3,xyz)-1)*(D2Y1(3,3) - D2Y0(3,3))*VRR3e(m+1,a3mm, maxm,Om,ExpZ,CenterZA,DY0,DY1,D2Y0,D2Y1)
endif
end if
if(dble(AngMomBra(1,xyz)) > 0) &
a1a2a3 = a1a2a3 &
+ dble(AngMomBra(1,xyz))*(-D2Y0(3,1))*VRR3e(m, a1m3m,maxm,Om,ExpZ,CenterZA,DY0,DY1,D2Y0,D2Y1) &
@ -169,6 +169,6 @@ recursive function VRR3e(m,AngMomBra,maxm,Om,ExpZ,CenterZA,DY0,DY1,D2Y0,D2Y1) &
a1a2a3 = a1a2a3 &
+ dble(AngMomBra(2,xyz))*(-D2Y0(3,2))*VRR3e(m, a2m3m,maxm,Om,ExpZ,CenterZA,DY0,DY1,D2Y0,D2Y1) &
- dble(AngMomBra(2,xyz))*(D2Y1(3,2) - D2Y0(3,2))*VRR3e(m+1,a2m3m,maxm,Om,ExpZ,CenterZA,DY0,DY1,D2Y0,D2Y1)
endif
end if
end function VRR3e

View File

@ -51,7 +51,7 @@ recursive function VRRNuc(m,AngMomA,maxm,Om,ExpPi,CenterAB,CenterPA,CenterPC) &
do i=1,3
am(i) = AngMomA(i)
amm(i) = AngMomA(i)
enddo
end do
! Loop over cartesian directions
xyz = 0
if (AngMomA(1) > 0) then
@ -62,7 +62,7 @@ recursive function VRRNuc(m,AngMomA,maxm,Om,ExpPi,CenterAB,CenterPA,CenterPC) &
xyz = 3
else
write(*,*) 'xyz = 0 in VRRNuc!'
endif
end if
! End loop over cartesian directions
am(xyz) = am(xyz) - 1
amm(xyz) = amm(xyz) - 2
@ -70,7 +70,7 @@ recursive function VRRNuc(m,AngMomA,maxm,Om,ExpPi,CenterAB,CenterPA,CenterPC) &
+ 0.5d0*dble(am(xyz))*ExpPi*VRRNuc(m,amm,maxm,Om,ExpPi,CenterAB,CenterPA,CenterPC) &
- CenterPC(xyz)*ExpPi*VRRNuc(m+1,am,maxm,Om,ExpPi,CenterAB,CenterPA,CenterPC) &
- 0.5d0*dble(am(xyz))*ExpPi**2*VRRNuc(m+1,amm,maxm,Om,ExpPi,CenterAB,CenterPA,CenterPC)
endif
endif
end if
end if
end function VRRNuc

View File

@ -22,7 +22,7 @@ recursive function VRROv(AngMomA,ExpPi,CenterPA) &
Ga = 1d0
else
Ga = CenterPA*VRROv(AngMomA-1,ExpPi,CenterPA) + 0.5d0*dble(AngMomA-1)*ExpPi*VRROv(AngMomA-2,ExpPi,CenterPA)
endif
endif
end if
end if
end function VRROv

1
src/IntPak/obj/.gitignore vendored Normal file
View File

@ -0,0 +1 @@
*.o

View File

@ -18,7 +18,7 @@ subroutine AO_values(doDrift,nBas,nShell,nWalk,CenterShell,TotAngMomShell,KShell
integer :: atot,nShellFunction,a(3)
integer,allocatable :: ShellFunction(:,:)
double precision :: rASq,xA,yA,zA,NormCoeff,prim
double precision :: rASq,xA,yA,zA,norm_coeff,prim
integer :: iSh,iShF,iK,iW,iEl,iBas,ixyz
@ -66,7 +66,7 @@ subroutine AO_values(doDrift,nBas,nShell,nWalk,CenterShell,TotAngMomShell,KShell
do iK=1,KShell(iSh)
! Calculate the exponential part
prim = DShell(iSh,iK)*NormCoeff(ExpShell(iSh,iK),a)*exp(-ExpShell(iSh,iK)*rASq)
prim = DShell(iSh,iK)*norm_coeff(ExpShell(iSh,iK),a)*exp(-ExpShell(iSh,iK)*rASq)
AO(iW,iEl,iBas) = AO(iW,iEl,iBas) + prim
if(doDrift) then

View File

@ -21,8 +21,8 @@ subroutine form_EOM_one_body(nO,nV,foo,fov,fvv,t1,t2,OOOV,OOVV,OVVV,cFoo,cFov,cF
! Local variables
integer :: i,j,k,l,m,n
integer :: a,b,c,d,e,f
integer :: i,j,k,l
integer :: a,b,c,d
double precision,allocatable :: tau(:,:,:,:)
! Output variables
@ -54,20 +54,20 @@ subroutine form_EOM_one_body(nO,nV,foo,fov,fvv,t1,t2,OOOV,OOVV,OVVV,cFoo,cFov,cF
do a=1,nV
do b=1,nV
do m=1,nO
cFvv(a,b) = cFvv(a,b) - fov(m,b)*t1(m,a)
do i=1,nO
cFvv(a,b) = cFvv(a,b) - fov(i,b)*t1(i,a)
end do
do m=1,nO
do f=1,nV
cFvv(a,b) = cFvv(a,b) + t1(m,f)*OVVV(m,a,f,b)
do i=1,nO
do c=1,nV
cFvv(a,b) = cFvv(a,b) + t1(i,c)*OVVV(i,a,c,b)
end do
end do
do m=1,nO
do n=1,nO
do e=1,nV
cFvv(a,b) = cFvv(a,b) - 0.5d0*tau(m,n,a,e)*OOVV(m,n,b,e)
do i=1,nO
do j=1,nO
do c=1,nV
cFvv(a,b) = cFvv(a,b) - 0.5d0*tau(i,j,a,c)*OOVV(i,j,b,c)
end do
end do
end do
@ -79,23 +79,23 @@ subroutine form_EOM_one_body(nO,nV,foo,fov,fvv,t1,t2,OOOV,OOVV,OVVV,cFoo,cFov,cF
cFoo(:,:) = foo(:,:)
do j=1,nO
do i=1,nO
do j=1,nO
do e=1,nV
cFoo(j,i) = cFoo(j,i) + t1(i,e)*fov(j,e)
do a=1,nV
cFoo(i,j) = cFoo(i,j) + t1(j,a)*fov(i,a)
end do
do m=1,nO
do e=1,nV
cFoo(j,i) = cFoo(j,i) + t1(m,e)*OVVV(j,m,i,e)
do k=1,nO
do a=1,nV
cFoo(i,j) = cFoo(i,j) + t1(k,a)*OVVV(i,k,j,a)
end do
end do
do m=1,nO
do e=1,nV
do f=1,nV
cFoo(j,i) = cFoo(j,i) + 0.5d0*tau(i,m,e,f)*OOVV(j,m,e,f)
do k=1,nO
do a=1,nV
do b=1,nV
cFoo(i,j) = cFoo(i,j) + 0.5d0*tau(j,k,a,b)*OOVV(i,k,a,b)
end do
end do
end do

View File

@ -72,19 +72,19 @@ subroutine form_EOM_two_body(nO,nV,foo,fov,fvv,t1,t2,cFoo,cFov,cFvv,
cWoooo(:,:,:,:) = OOOO(:,:,:,:)
do k=1,nO
do l=1,nO
do i=1,nO
do j=1,nO
do k=1,nO
do l=1,nO
do e=1,nV
cWoooo(k,l,i,j) = cWoooo(k,l,i,j) + t1(j,e)*OOOV(k,l,i,e)
cWoooo(k,l,i,j) = cWoooo(k,l,i,j) - t1(i,e)*OOOV(k,l,j,e)
cWoooo(i,j,k,l) = cWoooo(i,j,k,l) + t1(j,e)*OOOV(i,j,k,e)
cWoooo(i,j,k,l) = cWoooo(i,j,k,l) - t1(i,e)*OOOV(i,j,l,e)
end do
do e=1,nV
do f=1,nV
cWoooo(k,l,i,j) = cWoooo(k,l,i,j) + 0.5d0*tau(i,j,e,f)*OOVV(k,l,e,f)
do a=1,nV
do b=1,nV
cWoooo(i,j,k,l) = cWoooo(i,j,k,l) + 0.5d0*tau(k,l,a,b)*OOVV(i,j,a,b)
end do
end do
@ -102,14 +102,14 @@ subroutine form_EOM_two_body(nO,nV,foo,fov,fvv,t1,t2,cFoo,cFov,cFvv,
do c=1,nV
do d=1,nV
do m=1,nV
cWvvvv(a,b,c,d) = cWvvvv(a,b,c,d) - t1(m,b)*VOVV(a,m,c,d)
cWvvvv(a,b,c,d) = cWvvvv(a,b,c,d) + t1(m,a)*VOVV(b,m,c,d)
do i=1,nO
cWvvvv(a,b,c,d) = cWvvvv(a,b,c,d) - t1(i,b)*VOVV(a,i,c,d)
cWvvvv(a,b,c,d) = cWvvvv(a,b,c,d) + t1(i,a)*VOVV(b,i,c,d)
end do
do m=1,nV
do n=1,nV
cWvvvv(a,b,c,d) = cWvvvv(a,b,c,d) + tau(m,n,a,b)*OOVV(m,n,c,d)
do i=1,nO
do j=1,nO
cWvvvv(a,b,c,d) = cWvvvv(a,b,c,d) + tau(i,j,a,b)*OOVV(i,j,c,d)
end do
end do
@ -127,8 +127,8 @@ subroutine form_EOM_two_body(nO,nV,foo,fov,fvv,t1,t2,cFoo,cFov,cFvv,
do b=1,nV
do c=1,nV
do m=1,nV
cWvovv(a,i,b,c) = cWvovv(a,i,b,c) + t1(m,a)*OOVV(m,i,b,c)
do j=1,nO
cWvovv(a,i,b,c) = cWvovv(a,i,b,c) + t1(j,a)*OOVV(j,i,b,c)
end do
end do
@ -145,8 +145,8 @@ subroutine form_EOM_two_body(nO,nV,foo,fov,fvv,t1,t2,cFoo,cFov,cFvv,
do k=1,nO
do a=1,nO
do e=1,nV
cWooov(i,j,k,a) = cWooov(i,j,k,a) + t1(i,e)*OOVV(j,k,e,a)
do b=1,nV
cWooov(i,j,k,a) = cWooov(i,j,k,a) + t1(i,b)*OOVV(j,k,b,a)
end do
end do
@ -195,36 +195,38 @@ subroutine form_EOM_two_body(nO,nV,foo,fov,fvv,t1,t2,cFoo,cFov,cFvv,
do c=1,nV
do i=1,nO
do m=1,nO
do e=1,nV
cWvvvo(a,b,c,i) = cWvvvo(a,b,c,i) + t2(i,m,b,e)*VOVV(a,m,c,e)
cWvvvo(a,b,c,i) = cWvvvo(a,b,c,i) - t2(i,m,a,e)*VOVV(b,m,c,e)
do j=1,nO
do d=1,nV
cWvvvo(a,b,c,i) = cWvvvo(a,b,c,i) + t2(i,j,b,d)*VOVV(a,j,c,d)
cWvvvo(a,b,c,i) = cWvvvo(a,b,c,i) - t2(i,j,a,d)*VOVV(b,j,c,d)
end do
end do
do m=1,nO
do j=1,nO
do k=1,nO
cWvvvo(a,b,c,i) = cWvvvo(a,b,c,i) + 0.5d0*tau(j,k,a,b)*VOOO(c,i,j,k)
end do
end do
do j=1,nO
cWvvvo(a,b,c,i) = cWvvvo(a,b,c,i) + t2(i,j,a,b)*cFov(j,c)
end do
do d=1,nV
cWvvvo(a,b,c,i) = cWvvvo(a,b,c,i) + t1(i,d)*cWvvvv(a,b,c,d)
end do
do j=1,nO
cWvvvo(a,b,c,i) = cWvvvo(a,b,c,i) - t1(j,a)*OVVO(j,b,c,i)
cWvvvo(a,b,c,i) = cWvvvo(a,b,c,i) + t1(j,b)*OVVO(j,a,c,i)
do n=1,nO
cWvvvo(a,b,c,i) = cWvvvo(a,b,c,i) + 0.5d0*tau(m,n,a,b)*VOOO(c,i,m,n)
do d=1,nV
cWvvvo(a,b,c,i) = cWvvvo(a,b,c,i) + t1(j,a)*t2(k,i,b,d)*OOVV(j,k,c,d)
cWvvvo(a,b,c,i) = cWvvvo(a,b,c,i) - t1(j,b)*t2(k,i,a,d)*OOVV(j,k,c,d)
end do
end do
do m=1,nO
cWvvvo(a,b,c,i) = cWvvvo(a,b,c,i) + t2(i,m,a,b)*cFov(m,c)
end do
do e=1,nV
cWvvvo(a,b,c,i) = cWvvvo(a,b,c,i) + t1(i,e)*cWvvvv(a,b,c,e)
end do
do m=1,nO
cWvvvo(a,b,c,i) = cWvvvo(a,b,c,i) - t1(m,a)*OVVO(m,b,c,i)
cWvvvo(a,b,c,i) = cWvvvo(a,b,c,i) + t1(m,b)*OVVO(m,a,c,i)
do n=1,nO
do e=1,nV
cWvvvo(a,b,c,i) = cWvvvo(a,b,c,i) + t1(m,a)*t2(n,i,b,e)*OOVV(m,n,c,e)
cWvvvo(a,b,c,i) = cWvvvo(a,b,c,i) - t1(m,b)*t2(n,i,a,e)*OOVV(m,n,c,e)
end do
end do
end do
end do
@ -241,36 +243,38 @@ subroutine form_EOM_two_body(nO,nV,foo,fov,fvv,t1,t2,cFoo,cFov,cFvv,
do j=1,nO
do k=1,nO
do m=1,nO
do e=1,nV
cWovoo(i,a,j,k) = cWovoo(i,a,j,k) + t2(k,m,a,e)*OOOV(i,m,j,e)
cWovoo(i,a,j,k) = cWovoo(i,a,j,k) - t2(k,m,a,e)*OOOV(j,m,i,e)
do l=1,nO
do b=1,nV
cWovoo(i,a,j,k) = cWovoo(i,a,j,k) + t2(k,l,a,b)*OOOV(i,l,j,b)
cWovoo(i,a,j,k) = cWovoo(i,a,j,k) - t2(k,l,a,b)*OOOV(j,l,i,b)
end do
end do
do e=1,nV
do f=1,nV
cWovoo(i,a,j,k) = cWovoo(i,a,j,k) + 0.5d0*tau(j,k,e,f)*OVVV(i,a,e,f)
do b=1,nV
do c=1,nV
cWovoo(i,a,j,k) = cWovoo(i,a,j,k) + 0.5d0*tau(j,k,b,c)*OVVV(i,a,b,c)
end do
end do
do e=1,nV
cWovoo(i,a,j,k) = cWovoo(i,a,j,k) + t2(j,k,a,e)*cFov(i,e)
do b=1,nV
cWovoo(i,a,j,k) = cWovoo(i,a,j,k) + t2(j,k,a,b)*cFov(i,b)
end do
do m=1,nO
cWovoo(i,a,j,k) = cWovoo(i,a,j,k) + t1(m,a)*cWoooo(i,m,j,k)
do l=1,nO
cWovoo(i,a,j,k) = cWovoo(i,a,j,k) + t1(l,a)*cWoooo(i,l,j,k)
end do
do e=1,nV
cWovoo(i,a,j,k) = cWovoo(i,a,j,k) - t1(j,e)*OVVO(i,a,e,k)
cWovoo(i,a,j,k) = cWovoo(i,a,j,k) + t1(i,e)*OVVO(j,a,e,k)
do m=1,nO
do f=1,nV
cWovoo(i,a,j,k) = cWovoo(i,a,j,k) + t1(j,e)*t2(m,k,a,f)*OOVV(i,m,e,f)
cWovoo(i,a,j,k) = cWovoo(i,a,j,k) - t1(i,e)*t2(m,k,a,f)*OOVV(j,m,e,f)
do b=1,nV
cWovoo(i,a,j,k) = cWovoo(i,a,j,k) - t1(j,b)*OVVO(i,a,b,k)
cWovoo(i,a,j,k) = cWovoo(i,a,j,k) + t1(i,b)*OVVO(j,a,b,k)
do l=1,nO
do c=1,nV
cWovoo(i,a,j,k) = cWovoo(i,a,j,k) + t1(j,b)*t2(l,k,a,c)*OOVV(i,l,b,c)
cWovoo(i,a,j,k) = cWovoo(i,a,j,k) - t1(i,b)*t2(l,k,a,c)*OOVV(j,l,b,c)
end do
end do
end do
end do

1
src/QuAcK/obj/.gitingore Normal file
View File

@ -0,0 +1 @@
*.o

View File

@ -21,7 +21,7 @@ subroutine AO_values_grid(nBas,nShell,CenterShell,TotAngMomShell,KShell,DShell,E
integer :: atot,nShellFunction,a(3)
integer,allocatable :: ShellFunction(:,:)
double precision :: rASq,xA,yA,zA,NormCoeff,prim
double precision :: rASq,xA,yA,zA,norm_coeff,prim
integer :: iSh,iShF,iK,iG,iBas
@ -68,7 +68,7 @@ subroutine AO_values_grid(nBas,nShell,CenterShell,TotAngMomShell,KShell,DShell,E
! Calculate the exponential part
prim = DShell(iSh,iK)*NormCoeff(ExpShell(iSh,iK),a)*exp(-ExpShell(iSh,iK)*rASq)
prim = DShell(iSh,iK)*norm_coeff(ExpShell(iSh,iK),a)*exp(-ExpShell(iSh,iK)*rASq)
AO(iBas,iG) = AO(iBas,iG) + prim
prim = -2d0*ExpShell(iSh,iK)*prim

1
src/eDFT/obj/.gitingore Normal file
View File

@ -0,0 +1 @@
*.o

170
src/utils/elements.f90 Normal file
View File

@ -0,0 +1,170 @@
function element_number(element_name)
implicit none
integer,parameter :: nelement_max = 103
character(len=2),intent(in) :: element_name
integer :: element_number
character(len=2),parameter :: element_list(nelement_max) = &
(/' H', 'He', & ! 2
'Li','Be', ' B',' C',' N',' O',' F','Ne', & ! 10
'Na','Mg', 'Al','Si',' P',' S','Cl','Ar', & ! 18
' K','Ca','Sc','Ti',' V','Cr','Mn','Fe','Co','Ni','Cu','Zn','Ga','Ge','As','Se','Br','Kr', & ! 36
'Rb','Sr',' Y','Zr','Nb','Mo','Tc','Ru','Rh','Pd','Ag','Cd','In','Sn','Sb','Te',' I','Xe', & ! 54
'Cs','Ba', & ! 56
'La','Ce','Pr','Nd','Pm','Sm','Eu','Gd','Tb','Dy','Ho','Er','Tm','Yb', & ! 70
'Lu','Hf','Ta',' W','Re','Os','Ir','Pt','Au','Hg','Tl','Pb','Bi','Po','At','Rn', & ! 86
'Fr','Ra', & ! 88
'Ac','Th','Pa',' U','Np','Pu','Am','Cm','Bk','Cf','Es','Fm','Md','No', & ! 102
'Lr' & ! 103
/)
!=====
integer :: ielement
!=====
ielement=1
do while( ADJUSTL(element_name) /= ADJUSTL(element_list(ielement)) )
if( ielement == nelement_max ) then
write(*,'(a,a)') ' Input symbol ',element_name
write(*,'(a,i3,a)') ' Element symbol is not one of first ',nelement_max,' elements'
write(*,*) '!!! element symbol not understood !!!'
stop
endif
ielement = ielement + 1
enddo
element_number = ielement
end function element_number
function element_core(zval,zatom)
implicit none
double precision,intent(in) :: zval
double precision,intent(in) :: zatom
integer :: element_core
!=====
!
! If zval /= zatom, this is certainly an effective core potential
! and no core states should be frozen.
if( ABS(zval - zatom) > 1d0-3 ) then
element_core = 0
else
if( zval <= 4.00001d0 ) then ! up to Be
element_core = 0
else if( zval <= 12.00001d0 ) then ! up to Mg
element_core = 1
else if( zval <= 30.00001d0 ) then ! up to Ca
element_core = 5
else if( zval <= 48.00001d0 ) then ! up to Sr
element_core = 9
else
write(*,*) '!!! not imlemented in element_core !!!'
stop
endif
endif
end function element_core
function element_covalent_radius(zatom)
! Return covalent radius of an atom
implicit none
include 'parameters.h'
integer,intent(in) :: zatom
double precision :: element_covalent_radius
!
! Data from Cambridge Structural Database
! http://en.wikipedia.org/wiki/Covalent_radius
!
! Values are first given in picometer
! They will be converted in bohr later on
select case(zatom)
case( 1)
element_covalent_radius = 31.
case( 2)
element_covalent_radius = 28.
case( 3)
element_covalent_radius = 128.
case( 4)
element_covalent_radius = 96.
case( 5)
element_covalent_radius = 84.
case( 6)
element_covalent_radius = 73.
case( 7)
element_covalent_radius = 71.
case( 8)
element_covalent_radius = 66.
case( 9)
element_covalent_radius = 57.
case(10) ! Ne.
element_covalent_radius = 58.
case(11)
element_covalent_radius = 166.
case(12)
element_covalent_radius = 141.
case(13)
element_covalent_radius = 121.
case(14)
element_covalent_radius = 111.
case(15)
element_covalent_radius = 107.
case(16)
element_covalent_radius = 105.
case(17)
element_covalent_radius = 102.
case(18) ! Ar.
element_covalent_radius = 106.
case(19)
element_covalent_radius = 203.
case(20)
element_covalent_radius = 176.
case(21)
element_covalent_radius = 170.
case(22)
element_covalent_radius = 160.
case(23)
element_covalent_radius = 153.
case(24)
element_covalent_radius = 139.
case(25)
element_covalent_radius = 145.
case(26)
element_covalent_radius = 145.
case(27)
element_covalent_radius = 140.
case(28)
element_covalent_radius = 124.
case(29)
element_covalent_radius = 132.
case(30)
element_covalent_radius = 122.
case(31)
element_covalent_radius = 120.
case(32)
element_covalent_radius = 119.
case(34)
element_covalent_radius = 120.
case(35)
element_covalent_radius = 120.
case(36) ! Kr.
element_covalent_radius = 116.
case default
write(*,*) '!!! covalent radius not available !!!'
stop
end select
! pm to bohr conversion
element_covalent_radius = element_covalent_radius*pmtoau
end function element_covalent_radius

29
src/utils/norm_coeff.f90 Normal file
View File

@ -0,0 +1,29 @@
function norm_coeff(alpha,a)
implicit none
! Input variables
double precision,intent(in) :: alpha
integer,intent(in) :: a(3)
! local variable
double precision :: pi,dfa(3),dfac
integer :: atot
! Output variable
double precision norm_coeff
pi = 4d0*atan(1d0)
atot = a(1) + a(2) + a(3)
dfa(1) = dfac(2*a(1))/(2d0**a(1)*dfac(a(1)))
dfa(2) = dfac(2*a(2))/(2d0**a(2)*dfac(a(2)))
dfa(3) = dfac(2*a(3))/(2d0**a(3)*dfac(a(3)))
norm_coeff = (2d0*alpha/pi)**(3d0/2d0)*(4d0*alpha)**atot
norm_coeff = norm_coeff/(dfa(1)*dfa(2)*dfa(3))
norm_coeff = sqrt(norm_coeff)
end function norm_coeff

1
src/utils/obj/.gitignore vendored Normal file
View File

@ -0,0 +1 @@
*.o

View File

@ -0,0 +1,176 @@
subroutine read_auxiliary_basis(NAtoms,XYZAtoms,nShell,CenterShell, &
TotAngMomShell,KShell,DShell,ExpShell)
! Read auxiliary basis set information
implicit none
include 'parameters.h'
! Input variables
integer,intent(in) :: NAtoms
double precision,intent(in) :: XYZAtoms(NAtoms,3)
! Local variables
integer :: nShAt,iAt
integer :: i,j,k
character :: shelltype
! Output variables
integer,intent(out) :: nShell
double precision,intent(out) :: CenterShell(maxShell,3)
integer,intent(out) :: TotAngMomShell(maxShell),KShell(maxShell)
double precision,intent(out) :: DShell(maxShell,maxK),ExpShell(maxShell,maxK)
!------------------------------------------------------------------------
! Primary basis set information
!------------------------------------------------------------------------
! Open file with basis set specification
open(unit=2,file='input/basis')
! Read basis information
write(*,'(A28)') 'Gaussian basis set'
write(*,'(A28)') '------------------'
nShell = 0
do i=1,NAtoms
read(2,*) iAt,nShAt
write(*,'(A28,1X,I16)') 'Atom n. ',iAt
write(*,'(A28,1X,I16)') 'number of shells ',nShAt
write(*,'(A28)') '------------------'
! Basis function centers
do j=1,nShAt
nShell = nShell + 1
do k=1,3
CenterShell(nShell,k) = XYZAtoms(iAt,k)
enddo
! Shell type and contraction degree
read(2,*) shelltype,KShell(nShell)
if(shelltype == "S") then
TotAngMomShell(nShell) = 0
write(*,'(A28,1X,I16)') 's-type shell with K = ',KShell(nShell)
elseif(shelltype == "P") then
TotAngMomShell(nShell) = 1
write(*,'(A28,1X,I16)') 'p-type shell with K = ',KShell(nShell)
elseif(shelltype == "D") then
TotAngMomShell(nShell) = 2
write(*,'(A28,1X,I16)') 'd-type shell with K = ',KShell(nShell)
elseif(shelltype == "F") then
TotAngMomShell(nShell) = 3
write(*,'(A28,1X,I16)') 'f-type shell with K = ',KShell(nShell)
elseif(shelltype == "G") then
TotAngMomShell(nShell) = 4
write(*,'(A28,1X,I16)') 'g-type shell with K = ',KShell(nShell)
elseif(shelltype == "H") then
TotAngMomShell(nShell) = 5
write(*,'(A28,1X,I16)') 'h-type shell with K = ',KShell(nShell)
elseif(shelltype == "I") then
TotAngMomShell(nShell) = 6
write(*,'(A28,1X,I16)') 'i-type shell with K = ',KShell(nShell)
endif
! Read exponents and contraction coefficients
write(*,'(A28,1X,A16,A16)') '','Exponents','Contraction'
do k=1,Kshell(nShell)
read(2,*) ExpShell(nShell,k),DShell(nShell,k)
write(*,'(A28,1X,F16.10,F16.10)') '',ExpShell(nShell,k),DShell(nShell,k)
enddo
enddo
write(*,'(A28)') '------------------'
enddo
! Total number of shells
write(*,'(A28,1X,I16)') 'Number of shells in OBS',nShell
write(*,'(A28)') '------------------'
write(*,*)
! Close file with basis set specification
close(unit=2)
!------------------------------------------------------------------------
! Auxiliary basis set information
!------------------------------------------------------------------------
! Open file with auxilairy basis specification
open(unit=3,file='input/auxbasis')
! Read basis information
write(*,'(A28)') 'Auxiliary basis set'
write(*,'(A28)') '-------------------'
do i=1,NAtoms
read(3,*) iAt,nShAt
write(*,'(A28,1X,I16)') 'Atom n. ',iAt
write(*,'(A28,1X,I16)') 'number of shells ',nShAt
write(*,'(A28)') '------------------'
! Basis function centers
do j=1,nShAt
nShell = nShell + 1
do k=1,3
CenterShell(nShell,k) = XYZAtoms(iAt,k)
enddo
! Shell type and contraction degree
read(3,*) shelltype,KShell(nShell)
if(shelltype == "S") then
TotAngMomShell(nShell) = 0
write(*,'(A28,1X,I16)') 's-type shell with K = ',KShell(nShell)
elseif(shelltype == "P") then
TotAngMomShell(nShell) = 1
write(*,'(A28,1X,I16)') 'p-type shell with K = ',KShell(nShell)
elseif(shelltype == "D") then
TotAngMomShell(nShell) = 2
write(*,'(A28,1X,I16)') 'd-type shell with K = ',KShell(nShell)
elseif(shelltype == "F") then
TotAngMomShell(nShell) = 3
write(*,'(A28,1X,I16)') 'f-type shell with K = ',KShell(nShell)
elseif(shelltype == "G") then
TotAngMomShell(nShell) = 4
write(*,'(A28,1X,I16)') 'g-type shell with K = ',KShell(nShell)
elseif(shelltype == "H") then
TotAngMomShell(nShell) = 5
write(*,'(A28,1X,I16)') 'h-type shell with K = ',KShell(nShell)
elseif(shelltype == "I") then
TotAngMomShell(nShell) = 6
write(*,'(A28,1X,I16)') 'i-type shell with K = ',KShell(nShell)
endif
! Read exponents and contraction coefficients
write(*,'(A28,1X,A16,A16)') '','Exponents','Contraction'
do k=1,Kshell(nShell)
read(3,*) ExpShell(nShell,k),DShell(nShell,k)
write(*,'(A28,1X,F16.10,F16.10)') '',ExpShell(nShell,k),DShell(nShell,k)
enddo
enddo
write(*,'(A28)') '------------------'
enddo
! Total number of shells
write(*,'(A28,1X,I16)') 'Number of shells in ABS',nShell
write(*,'(A28)') '------------------'
write(*,*)
! Close file with basis set specification
close(unit=3)
end subroutine read_auxiliary_basis

118
src/utils/read_basis.f90 Normal file
View File

@ -0,0 +1,118 @@
subroutine read_basis(nNuc,rNuc,nBas,nO,nV,nShell,TotAngMomShell,CenterShell,KShell,DShell,ExpShell)
! Read basis set information
implicit none
include 'parameters.h'
! Input variables
integer,intent(in) :: nNuc,nO
double precision,intent(in) :: rNuc(nNuc,ncart)
! Local variables
integer :: nShAt,iNuc,iShell
integer :: i,j,k
character :: shelltype
! Output variables
integer,intent(out) :: nShell,nBas
double precision,intent(out) :: CenterShell(maxShell,ncart)
integer,intent(out) :: TotAngMomShell(maxShell),KShell(maxShell)
double precision,intent(out) :: DShell(maxShell,maxK),ExpShell(maxShell,maxK)
integer,intent(out) :: nV
!------------------------------------------------------------------------
! Primary basis set information
!------------------------------------------------------------------------
! Open file with basis set specification
open(unit=2,file='input/basis')
! Read basis information
write(*,'(A28)') 'Gaussian basis set'
write(*,'(A28)') '------------------'
nShell = 0
do i=1,nNuc
read(2,*) iNuc,nShAt
write(*,'(A28,1X,I16)') 'Atom n. ',iNuc
write(*,'(A28,1X,I16)') 'number of shells ',nShAt
write(*,'(A28)') '------------------'
! Basis function centers
do j=1,nShAt
nShell = nShell + 1
do k=1,ncart
CenterShell(nShell,k) = rNuc(iNuc,k)
enddo
! Shell type and contraction degree
read(2,*) shelltype,KShell(nShell)
if(shelltype == "S") then
TotAngMomShell(nShell) = 0
write(*,'(A28,1X,I16)') 's-type shell with K = ',KShell(nShell)
elseif(shelltype == "P") then
TotAngMomShell(nShell) = 1
write(*,'(A28,1X,I16)') 'p-type shell with K = ',KShell(nShell)
elseif(shelltype == "D") then
TotAngMomShell(nShell) = 2
write(*,'(A28,1X,I16)') 'd-type shell with K = ',KShell(nShell)
elseif(shelltype == "F") then
TotAngMomShell(nShell) = 3
write(*,'(A28,1X,I16)') 'f-type shell with K = ',KShell(nShell)
elseif(shelltype == "G") then
TotAngMomShell(nShell) = 4
write(*,'(A28,1X,I16)') 'g-type shell with K = ',KShell(nShell)
elseif(shelltype == "H") then
TotAngMomShell(nShell) = 5
write(*,'(A28,1X,I16)') 'h-type shell with K = ',KShell(nShell)
elseif(shelltype == "I") then
TotAngMomShell(nShell) = 6
write(*,'(A28,1X,I16)') 'i-type shell with K = ',KShell(nShell)
endif
! Read exponents and contraction coefficients
write(*,'(A28,1X,A16,A16)') '','Exponents','Contraction'
do k=1,Kshell(nShell)
read(2,*) ExpShell(nShell,k),DShell(nShell,k)
write(*,'(A28,1X,F16.10,F16.10)') '',ExpShell(nShell,k),DShell(nShell,k)
enddo
enddo
write(*,'(A28)') '------------------'
enddo
! Total number of shells
write(*,'(A28,1X,I16)') 'Number of shells',nShell
write(*,'(A28)') '------------------'
write(*,*)
! Close file with basis set specification
close(unit=2)
! Calculate number of basis functions
nBas = 0
do iShell=1,nShell
nBas = nBas + (TotAngMomShell(iShell)*TotAngMomShell(iShell) + 3*TotAngMomShell(iShell) + 2)/2
enddo
write(*,'(A28)') '------------------'
write(*,'(A28,1X,I16)') 'Number of basis functions',NBas
write(*,'(A28)') '------------------'
write(*,*)
! Number of virtual orbitals
nV = nBas - nO
end subroutine read_basis

View File

@ -0,0 +1,68 @@
subroutine read_geometry(nNuc,ZNuc,rNuc,ENuc)
! Read molecular geometry
implicit none
include 'parameters.h'
! Ouput variables
integer,intent(in) :: nNuc
! Local variables
integer :: i,j
double precision :: RAB
character(len=2) :: El
integer,external :: element_number
! Ouput variables
double precision,intent(out) :: ZNuc(nNuc),rNuc(nNuc,ncart),ENuc
! Open file with geometry specification
open(unit=1,file='input/molecule')
! Read geometry
read(1,*)
read(1,*)
read(1,*)
do i=1,nNuc
read(1,*) El,rNuc(i,1),rNuc(i,2),rNuc(i,3)
ZNuc(i) = element_number(El)
enddo
! Compute nuclear repulsion energy
ENuc = 0
do i=1,nNuc-1
do j=i+1,nNuc
RAB = (rNuc(i,1)-rNuc(j,1))**2 + (rNuc(i,2)-rNuc(j,2))**2 + (rNuc(i,3)-rNuc(j,3))**2
ENuc = ENuc + ZNuc(i)*ZNuc(j)/sqrt(RAB)
enddo
enddo
! Close file with geometry specification
close(unit=1)
! Print geometry
write(*,'(A28)') '------------------'
write(*,'(A28)') 'Molecular geometry'
write(*,'(A28)') '------------------'
do i=1,nNuc
write(*,'(A28,1X,I16)') 'Atom n. ',i
write(*,'(A28,1X,F16.10)') 'Z = ',ZNuc(i)
write(*,'(A28,1X,F16.10,F16.10,F16.10)') 'Atom coordinates:',(rNuc(i,j),j=1,ncart)
enddo
write(*,*)
write(*,'(A28)') '------------------'
write(*,'(A28,1X,F16.10)') 'Nuclear repulsion energy = ',ENuc
write(*,'(A28)') '------------------'
write(*,*)
end subroutine read_geometry

View File

@ -0,0 +1,119 @@
subroutine read_integrals(nBas,S,T,V,Hc,G)
! Read one- and two-electron integrals from files
implicit none
! Input variables
integer,intent(in) :: nBas
! Local variables
logical :: debug
integer :: mu,nu,la,si
double precision :: Ov,Kin,Nuc,ERI
double precision :: scale
! Output variables
double precision,intent(out) :: S(nBas,nBas),T(nBas,nBas),V(nBas,nBas),Hc(nBas,nBas),G(nBas,nBas,nBas,nBas)
! Open file with integrals
debug = .false.
scale = 1d0
open(unit=8 ,file='int/Ov.dat')
open(unit=9 ,file='int/Kin.dat')
open(unit=10,file='int/Nuc.dat')
open(unit=11,file='int/ERI.dat')
! Read overlap integrals
S = 0d0
do
read(8,*,end=8) mu,nu,Ov
S(mu,nu) = Ov
enddo
8 close(unit=8)
! Read kinetic integrals
T = 0d0
do
read(9,*,end=9) mu,nu,Kin
T(mu,nu) = Kin/scale**2
enddo
9 close(unit=9)
! Read nuclear integrals
V = 0d0
do
read(10,*,end=10) mu,nu,Nuc
V(mu,nu) = Nuc
enddo
10 close(unit=10)
! Define core Hamiltonian
Hc = T + V
! Read nuclear integrals
G = 0d0
do
read(11,*,end=11) mu,nu,la,si,ERI
ERI = ERI/scale
! <12|34>
G(mu,nu,la,si) = ERI
! <32|14>
G(la,nu,mu,si) = ERI
! <14|32>
G(mu,si,la,nu) = ERI
! <34|12>
G(la,si,mu,nu) = ERI
! <41|23>
G(si,mu,nu,la) = ERI
! <23|41>
G(nu,la,si,mu) = ERI
! <21|43>
G(nu,mu,si,la) = ERI
! <43|21>
G(si,la,nu,mu) = ERI
enddo
11 close(unit=11)
! Print results
if(debug) then
write(*,'(A28)') '----------------------'
write(*,'(A28)') 'Overlap integrals'
write(*,'(A28)') '----------------------'
call matout(nBas,nBas,S)
write(*,*)
write(*,'(A28)') '----------------------'
write(*,'(A28)') 'Kinetic integrals'
write(*,'(A28)') '----------------------'
call matout(nBas,nBas,T)
write(*,*)
write(*,'(A28)') '----------------------'
write(*,'(A28)') 'Nuclear integrals'
write(*,'(A28)') '----------------------'
call matout(nBas,nBas,V)
write(*,*)
write(*,'(A28)') '----------------------'
write(*,'(A28)') 'Electron repulsion integrals'
write(*,'(A28)') '----------------------'
do la=1,nBas
do si=1,nBas
call matout(nBas,nBas,G(1,1,la,si))
enddo
enddo
write(*,*)
endif
end subroutine read_integrals

View File

@ -0,0 +1,64 @@
subroutine read_molecule(nNuc,nEl,nO,nC,nR)
! Read number of atoms and number of electrons
implicit none
include 'parameters.h'
! Input variables
integer,intent(out) :: nNuc
integer,intent(out) :: nEl(nspin)
integer,intent(out) :: nO(nspin)
integer,intent(out) :: nC(nspin)
integer,intent(out) :: nR(nspin)
! Local variables
integer :: nCore
integer :: nRyd
! Open file with geometry specification
open(unit=1,file='input/molecule')
! Read number of atoms and number of electrons
read(1,*)
read(1,*) nNuc,nEl(1),nEl(2),nCore,nRyd
if(mod(nCore,2) /= 0 .or. mod(nRyd,2) /= 0) then
print*, 'The number of core and Rydberg electrons must be even!'
stop
end if
nO(:) = nEl(:)
nC(:) = nCore/2
nR(:) = nRyd/2
! Print results
write(*,'(A28)') '----------------------'
write(*,'(A28,1X,I16)') 'Number of atoms',nNuc
write(*,'(A28)') '----------------------'
write(*,*)
write(*,'(A28)') '----------------------'
write(*,'(A28,1X,I16)') 'Number of spin-up electrons',nEl(1)
write(*,'(A28,1X,I16)') 'Number of spin-down electrons',nEl(2)
write(*,'(A28,1X,I16)') ' Total number of electrons',sum(nEl(:))
write(*,'(A28)') '----------------------'
write(*,*)
write(*,'(A28)') '----------------------'
write(*,'(A28,1X,I16)') 'Number of core electrons',sum(nC(:))
write(*,'(A28,1X,I16)') 'Number of Rydberg electrons',sum(nR(:))
write(*,'(A28)') '----------------------'
write(*,*)
! Close file with geometry specification
close(unit=1)
end subroutine read_molecule

533
src/utils/utils.f90 Normal file
View File

@ -0,0 +1,533 @@
!------------------------------------------------------------------------
function Kronecker_delta(i,j) result(delta)
! Kronecker Delta
implicit none
! Input variables
integer,intent(in) :: i,j
! Output variables
double precision :: delta
if(i == j) then
delta = 1d0
else
delta = 0d0
endif
end function Kronecker_delta
function KroneckerDelta(i,j) result(delta)
! Kronecker Delta
implicit none
! Input variables
integer,intent(in) :: i,j
! Output variables
integer :: delta
if(i == j) then
delta = 1
else
delta = 0
endif
end function KroneckerDelta
!------------------------------------------------------------------------
subroutine matout(m,n,A)
! Print the MxN array A
implicit none
integer,parameter :: ncol = 5
double precision,parameter :: small = 1d-10
integer,intent(in) :: m,n
double precision,intent(in) :: A(m,n)
double precision :: B(ncol)
integer :: ilower,iupper,num,i,j
do ilower=1,n,ncol
iupper = min(ilower + ncol - 1,n)
num = iupper - ilower + 1
write(*,'(3X,10(9X,I6))') (j,j=ilower,iupper)
do i=1,m
do j=ilower,iupper
B(j-ilower+1) = A(i,j)
enddo
do j=1,num
if(abs(B(j)) < small) B(j) = 0d0
enddo
write(*,'(I7,10F15.8)') i,(B(j),j=1,num)
enddo
enddo
end subroutine matout
!------------------------------------------------------------------------
subroutine trace_vector(n,v,Tr)
! Calculate the trace of the vector v of length n
!!! Please use the intrinsic fortran sum() !!!
implicit none
! Input variables
integer,intent(in) :: n
double precision,intent(in) :: v(n)
! Local variables
integer :: i
! Output variables
double precision,intent(out) :: Tr
Tr = 0d0
do i=1,n
Tr = Tr + v(i)
enddo
end subroutine trace_vector
!------------------------------------------------------------------------
function trace_matrix(n,A) result(Tr)
! Calculate the trace of the square matrix A
implicit none
! Input variables
integer,intent(in) :: n
double precision,intent(in) :: A(n,n)
! Local variables
integer :: i
! Output variables
double precision :: Tr
Tr = 0d0
do i=1,n
Tr = Tr + A(i,i)
enddo
end function trace_matrix
!------------------------------------------------------------------------
subroutine compute_error(nData,Mean,Var,Error)
! Calculate the statistical error
implicit none
! Input variables
double precision,intent(in) :: nData,Mean(3)
! Output variables
double precision,intent(out) :: Error(3)
double precision,intent(inout):: Var(3)
Error = sqrt((Var-Mean**2/nData)/nData/(nData-1d0))
end subroutine compute_error
!------------------------------------------------------------------------
subroutine identity_matrix(N,A)
! Set the matrix A to the identity matrix
implicit none
! Input variables
integer,intent(in) :: N
! Local viaruabkes
integer :: i
! Output variables
double precision,intent(out) :: A(N,N)
A = 0d0
do i=1,N
A(i,i) = 1d0
enddo
end subroutine identity_matrix
!------------------------------------------------------------------------
subroutine prepend(N,M,A,b)
! Prepend the vector b of size N into the matrix A of size NxM
implicit none
! Input variables
integer,intent(in) :: N,M
double precision,intent(in) :: b(N)
! Local viaruabkes
integer :: i,j
! Output variables
double precision,intent(out) :: A(N,M)
! print*,'b in append'
! call matout(N,1,b)
do i=1,N
do j=M-1,1,-1
A(i,j+1) = A(i,j)
enddo
A(i,1) = b(i)
enddo
end subroutine prepend
!------------------------------------------------------------------------
subroutine append(N,M,A,b)
! Append the vector b of size N into the matrix A of size NxM
implicit none
! Input variables
integer,intent(in) :: N,M
double precision,intent(in) :: b(N)
! Local viaruabkes
integer :: i,j
! Output variables
double precision,intent(out) :: A(N,M)
do i=1,N
do j=2,M
A(i,j-1) = A(i,j)
enddo
A(i,M) = b(i)
enddo
end subroutine append
!------------------------------------------------------------------------
subroutine AtDA(N,A,D,B)
! Perform B = At.D.A where A is a NxN matrix and D is a diagonal matrix given
! as a vector of length N
implicit none
! Input variables
integer,intent(in) :: N
double precision,intent(in) :: A(N,N),D(N)
! Local viaruabkes
integer :: i,j,k
! Output variables
double precision,intent(out) :: B(N,N)
B = 0d0
do i=1,N
do j=1,N
do k=1,N
B(i,k) = B(i,k) + A(j,i)*D(j)*A(j,k)
enddo
enddo
enddo
end subroutine AtDA
!------------------------------------------------------------------------
subroutine ADAt(N,A,D,B)
! Perform B = A.D.At where A is a NxN matrix and D is a diagonal matrix given
! as a vector of length N
implicit none
! Input variables
integer,intent(in) :: N
double precision,intent(in) :: A(N,N),D(N)
! Local viaruabkes
integer :: i,j,k
! Output variables
double precision,intent(out) :: B(N,N)
B = 0d0
do i=1,N
do j=1,N
do k=1,N
B(i,k) = B(i,k) + A(i,j)*D(j)*A(k,j)
enddo
enddo
enddo
end subroutine ADAt
!------------------------------------------------------------------------
subroutine DA(N,D,A)
! Perform A <- D.A where A is a NxN matrix and D is a diagonal matrix given
! as a vector of length N
implicit none
integer,intent(in) :: N
integer :: i,j,k
double precision,intent(in) :: D(N)
double precision,intent(inout):: A(N,N)
do i=1,N
do j=1,N
A(i,j) = D(i)*A(i,j)
enddo
enddo
end subroutine DA
!------------------------------------------------------------------------
subroutine AD(N,A,D)
! Perform A <- A.D where A is a NxN matrix and D is a diagonal matrix given
! as a vector of length N
implicit none
integer,intent(in) :: N
integer :: i,j,k
double precision,intent(in) :: D(N)
double precision,intent(inout):: A(N,N)
do i=1,N
do j=1,N
A(i,j) = A(i,j)*D(j)
enddo
enddo
end subroutine AD
!------------------------------------------------------------------------
subroutine print_warning(message)
! Print warning
implicit none
character(len=*),intent(in) :: message
write(*,*) message
end subroutine print_warning
!------------------------------------------------------------------------
subroutine CalcTrAB(n,A,B,Tr)
! Calculate the trace of the square matrix A.B
implicit none
! Input variables
integer,intent(in) :: n
double precision,intent(in) :: A(n,n),B(n,n)
! Local variables
integer :: i,j
! Output variables
double precision,intent(out) :: Tr
Tr = 0d0
do i=1,n
do j=1,n
Tr = Tr + A(i,j)*B(j,i)
enddo
enddo
end subroutine CalcTrAB
!------------------------------------------------------------------------
function EpsilonSwitch(i,j) result(delta)
! Epsilon function
implicit none
! Input variables
integer,intent(in) :: i,j
integer :: delta
if(i <= j) then
delta = 1
else
delta = -1
endif
end function EpsilonSwitch
!------------------------------------------------------------------------
function KappaCross(i,j,k) result(kappa)
! kappa(i,j,k) = eps(i,j) delta(i,k) - eps(k,i) delta(i,j)
implicit none
! Input variables
integer,intent(in) :: i,j,k
! Local variables
integer :: EpsilonSwitch,KroneckerDelta
double precision :: kappa
kappa = dble(EpsilonSwitch(i,j)*KroneckerDelta(i,k) - EpsilonSwitch(k,i)*KroneckerDelta(i,j))
end function KappaCross
!------------------------------------------------------------------------
subroutine CalcInv3(a,det)
! Calculate the inverse and the determinant of a 3x3 matrix
implicit none
double precision,intent(inout) :: a(3,3)
double precision, intent(inout) :: det
double precision :: b(3,3)
integer :: i,j
det = a(1,1)*(a(2,2)*a(3,3)-a(2,3)*a(3,2)) &
- a(1,2)*(a(2,1)*a(3,3)-a(2,3)*a(3,1)) &
+ a(1,3)*(a(2,1)*a(3,2)-a(2,2)*a(3,1))
do i=1,3
b(i,1) = a(i,1)
b(i,2) = a(i,2)
b(i,3) = a(i,3)
enddo
a(1,1) = b(2,2)*b(3,3) - b(2,3)*b(3,2)
a(2,1) = b(2,3)*b(3,1) - b(2,1)*b(3,3)
a(3,1) = b(2,1)*b(3,2) - b(2,2)*b(3,1)
a(1,2) = b(1,3)*b(3,2) - b(1,2)*b(3,3)
a(2,2) = b(1,1)*b(3,3) - b(1,3)*b(3,1)
a(3,2) = b(1,2)*b(3,1) - b(1,1)*b(3,2)
a(1,3) = b(1,2)*b(2,3) - b(1,3)*b(2,2)
a(2,3) = b(1,3)*b(2,1) - b(1,1)*b(2,3)
a(3,3) = b(1,1)*b(2,2) - b(1,2)*b(2,1)
do i=1,3
do j=1,3
a(i,j) = a(i,j)/det
enddo
enddo
end subroutine CalcInv3
!------------------------------------------------------------------------
subroutine CalcInv4(a,det)
implicit none
double precision,intent(inout) :: a(4,4)
double precision,intent(inout) :: det
double precision :: b(4,4)
integer :: i,j
det = a(1,1)*(a(2,2)*(a(3,3)*a(4,4)-a(3,4)*a(4,3)) &
-a(2,3)*(a(3,2)*a(4,4)-a(3,4)*a(4,2)) &
+a(2,4)*(a(3,2)*a(4,3)-a(3,3)*a(4,2))) &
- a(1,2)*(a(2,1)*(a(3,3)*a(4,4)-a(3,4)*a(4,3)) &
-a(2,3)*(a(3,1)*a(4,4)-a(3,4)*a(4,1)) &
+a(2,4)*(a(3,1)*a(4,3)-a(3,3)*a(4,1))) &
+ a(1,3)*(a(2,1)*(a(3,2)*a(4,4)-a(3,4)*a(4,2)) &
-a(2,2)*(a(3,1)*a(4,4)-a(3,4)*a(4,1)) &
+a(2,4)*(a(3,1)*a(4,2)-a(3,2)*a(4,1))) &
- a(1,4)*(a(2,1)*(a(3,2)*a(4,3)-a(3,3)*a(4,2)) &
-a(2,2)*(a(3,1)*a(4,3)-a(3,3)*a(4,1)) &
+a(2,3)*(a(3,1)*a(4,2)-a(3,2)*a(4,1)))
do i=1,4
b(1,i) = a(1,i)
b(2,i) = a(2,i)
b(3,i) = a(3,i)
b(4,i) = a(4,i)
enddo
a(1,1) = b(2,2)*(b(3,3)*b(4,4)-b(3,4)*b(4,3))-b(2,3)*(b(3,2)*b(4,4)-b(3,4)*b(4,2))+b(2,4)*(b(3,2)*b(4,3)-b(3,3)*b(4,2))
a(2,1) = -b(2,1)*(b(3,3)*b(4,4)-b(3,4)*b(4,3))+b(2,3)*(b(3,1)*b(4,4)-b(3,4)*b(4,1))-b(2,4)*(b(3,1)*b(4,3)-b(3,3)*b(4,1))
a(3,1) = b(2,1)*(b(3,2)*b(4,4)-b(3,4)*b(4,2))-b(2,2)*(b(3,1)*b(4,4)-b(3,4)*b(4,1))+b(2,4)*(b(3,1)*b(4,2)-b(3,2)*b(4,1))
a(4,1) = -b(2,1)*(b(3,2)*b(4,3)-b(3,3)*b(4,2))+b(2,2)*(b(3,1)*b(4,3)-b(3,3)*b(4,1))-b(2,3)*(b(3,1)*b(4,2)-b(3,2)*b(4,1))
a(1,2) = -b(1,2)*(b(3,3)*b(4,4)-b(3,4)*b(4,3))+b(1,3)*(b(3,2)*b(4,4)-b(3,4)*b(4,2))-b(1,4)*(b(3,2)*b(4,3)-b(3,3)*b(4,2))
a(2,2) = b(1,1)*(b(3,3)*b(4,4)-b(3,4)*b(4,3))-b(1,3)*(b(3,1)*b(4,4)-b(3,4)*b(4,1))+b(1,4)*(b(3,1)*b(4,3)-b(3,3)*b(4,1))
a(3,2) = -b(1,1)*(b(3,2)*b(4,4)-b(3,4)*b(4,2))+b(1,2)*(b(3,1)*b(4,4)-b(3,4)*b(4,1))-b(1,4)*(b(3,1)*b(4,2)-b(3,2)*b(4,1))
a(4,2) = b(1,1)*(b(3,2)*b(4,3)-b(3,3)*b(4,2))-b(1,2)*(b(3,1)*b(4,3)-b(3,3)*b(4,1))+b(1,3)*(b(3,1)*b(4,2)-b(3,2)*b(4,1))
a(1,3) = b(1,2)*(b(2,3)*b(4,4)-b(2,4)*b(4,3))-b(1,3)*(b(2,2)*b(4,4)-b(2,4)*b(4,2))+b(1,4)*(b(2,2)*b(4,3)-b(2,3)*b(4,2))
a(2,3) = -b(1,1)*(b(2,3)*b(4,4)-b(2,4)*b(4,3))+b(1,3)*(b(2,1)*b(4,4)-b(2,4)*b(4,1))-b(1,4)*(b(2,1)*b(4,3)-b(2,3)*b(4,1))
a(3,3) = b(1,1)*(b(2,2)*b(4,4)-b(2,4)*b(4,2))-b(1,2)*(b(2,1)*b(4,4)-b(2,4)*b(4,1))+b(1,4)*(b(2,1)*b(4,2)-b(2,2)*b(4,1))
a(4,3) = -b(1,1)*(b(2,2)*b(4,3)-b(2,3)*b(4,2))+b(1,2)*(b(2,1)*b(4,3)-b(2,3)*b(4,1))-b(1,3)*(b(2,1)*b(4,2)-b(2,2)*b(4,1))
a(1,4) = -b(1,2)*(b(2,3)*b(3,4)-b(2,4)*b(3,3))+b(1,3)*(b(2,2)*b(3,4)-b(2,4)*b(3,2))-b(1,4)*(b(2,2)*b(3,3)-b(2,3)*b(3,2))
a(2,4) = b(1,1)*(b(2,3)*b(3,4)-b(2,4)*b(3,3))-b(1,3)*(b(2,1)*b(3,4)-b(2,4)*b(3,1))+b(1,4)*(b(2,1)*b(3,3)-b(2,3)*b(3,1))
a(3,4) = -b(1,1)*(b(2,2)*b(3,4)-b(2,4)*b(3,2))+b(1,2)*(b(2,1)*b(3,4)-b(2,4)*b(3,1))-b(1,4)*(b(2,1)*b(3,2)-b(2,2)*b(3,1))
a(4,4) = b(1,1)*(b(2,2)*b(3,3)-b(2,3)*b(3,2))-b(1,2)*(b(2,1)*b(3,3)-b(2,3)*b(3,1))+b(1,3)*(b(2,1)*b(3,2)-b(2,2)*b(3,1))
do i=1,4
do j=1,4
a(i,j) = a(i,j)/det
enddo
enddo
end subroutine CalcInv4

207
src/utils/wrap_lapack.f90 Normal file
View File

@ -0,0 +1,207 @@
!subroutine eigenvalues_non_symmetric_matrix(N,A,e)
!
!! Diagonalize a square matrix
!
! implicit none
!
!! Input variables
!
! integer,intent(in) :: N
! double precision,intent(inout):: A(N,N)
! double precision,intent(out) :: e(N)
!
!! Local variables
!
! integer :: lwork,info
! double precision,allocatable :: work(:)
!
!! Memory allocation
!
! allocate(eRe(N),eIm(N),work(3*N))
! lwork = size(work)
!
! call DGEEV('N','N',N,A,N, eRe, eIm, 0d0,1, VR,LDVR, WORK, LWORK, INFO )
!
! if(info /= 0) then
! print*,'Problem in diagonalize_matrix (dseev)!!'
! stop
! endif
!
!end subroutine eigenvalues_non_symmetric_matrix
subroutine diagonalize_matrix(N,A,e)
! Diagonalize a square matrix
implicit none
! Input variables
integer,intent(in) :: N
double precision,intent(inout):: A(N,N)
double precision,intent(out) :: e(N)
! Local variables
integer :: lwork,info
double precision,allocatable :: work(:)
! Memory allocation
allocate(work(3*N))
lwork = size(work)
call dsyev('V','U',N,A,N,e,work,lwork,info)
if(info /= 0) then
print*,'Problem in diagonalize_matrix (dsyev)!!'
endif
end subroutine diagonalize_matrix
subroutine svd(N,A,U,D,Vt)
! Compute A = U.D.Vt
! Dimension of A is NxN
implicit none
integer, intent(in) :: N
double precision,intent(in) :: A(N,N)
double precision,intent(out) :: U(N,N)
double precision,intent(out) :: Vt(N,N)
double precision,intent(out) :: D(N)
double precision,allocatable :: work(:)
integer :: info,lwork
double precision,allocatable :: scr(:,:)
allocate (scr(N,N))
scr(:,:) = A(:,:)
! Find optimal size for temporary arrays
allocate(work(1))
lwork = -1
call dgesvd('A','A',N,N,scr,N,D,U,N,Vt,N,work,lwork,info)
lwork = int(work(1))
deallocate(work)
allocate(work(lwork))
call dgesvd('A','A',N,N,scr,N,D,U,N,Vt,N,work,lwork,info)
deallocate(work,scr)
if (info /= 0) then
print *, info, ': SVD failed'
stop
endif
end
subroutine inverse_matrix(N,A,B)
! Returns the inverse of the square matrix A in B
implicit none
integer,intent(in) :: N
double precision, intent(in) :: A(N,N)
double precision, intent(out) :: B(N,N)
integer :: info,lwork
integer, allocatable :: ipiv(:)
double precision,allocatable :: work(:)
allocate (ipiv(N),work(N*N))
lwork = size(work)
B(1:N,1:N) = A(1:N,1:N)
call dgetrf(N,N,B,N,ipiv,info)
if (info /= 0) then
print*,info
stop 'error in inverse (dgetrf)!!'
endif
call dgetri(N,B,N,ipiv,work,lwork,info)
if (info /= 0) then
print *, info
stop 'error in inverse (dgetri)!!'
endif
deallocate(ipiv,work)
end subroutine inverse_matrix
subroutine linear_solve(N,A,b,x,rcond)
! Solve the linear system A.x = b where A is a NxN matrix
! and x and x are vectors of size N
implicit none
integer,intent(in) :: N
double precision,intent(in) :: A(N,N),b(N),rcond
double precision,intent(out) :: x(N)
integer :: info,lwork
double precision :: ferr,berr
integer,allocatable :: ipiv(:),iwork(:)
double precision,allocatable :: AF(:,:),work(:)
lwork = 3*N
allocate(AF(N,N),ipiv(N),work(lwork),iwork(N))
call dsysvx('N','U',N,1,A,N,AF,N,ipiv,b,N,x,N,rcond,ferr,berr,work,lwork,iwork,info)
! if (info /= 0) then
! print *, info
! stop 'error in linear_solve (dsysvx)!!'
! endif
end subroutine linear_solve
subroutine easy_linear_solve(N,A,b,x)
! Solve the linear system A.x = b where A is a NxN matrix
! and x and x are vectors of size N
implicit none
integer,intent(in) :: N
double precision,intent(in) :: A(N,N),b(N)
double precision,intent(out) :: x(N)
integer :: info,lwork
integer,allocatable :: ipiv(:)
double precision,allocatable :: work(:)
allocate(ipiv(N),work(N*N))
lwork = size(work)
x = b
call dsysv('U',N,1,A,N,ipiv,x,N,work,lwork,info)
if (info /= 0) then
print *, info
stop 'error in linear_solve (dsysv)!!'
endif
end subroutine easy_linear_solve