quantum-espresso/atomic/dirsol.f90

387 lines
12 KiB
Fortran
Raw Normal View History

!
! Copyright (C) 2002 Vanderbilt group
! This file is distributed under the terms of the
! GNU General Public License. See the file `License'
! in the root directory of the present distribution,
! or http://www.gnu.org/copyleft/gpl.txt .
!
! This routine has been modified in order to be compatible with the
! ld1 code. The numerical algorithm is unchanged.
! ADC Nov 2003
!
!-------------------------------------------------------------------------
!
subroutine dirsol(idim1,mesh,ncur,lcur,jcur,it,e0,thresh,grid,snl,ruae,nstop)
!
! subroutine to compute solutions to the full dirac equation
!
! the dirac equation in rydberg units reads:
!
! df(r) k alpha
! ----- = + - f(r) - ( e - v(r) ) ----- g(r)
! dr r 2
!
! dg(r) k 4 alpha
! ----- = - - g(r) + ( e - v(r) + -------- ) ----- f(r)
! dr r alpha**2 2
!
! where
! alpha is the fine structure constant
! f(r) is r*minor component
! g(r) is r*major component
! k is quantum number
! k = - (l+1) if j = l+0.5
! k = + l if j = l-0.5
! IMPORTANT: on output, snl(:,1) contains the MAJOR component
! snl(:,2) contains the MINOR component
!
!----------------------------------------------------------------------------
!
use io_global, only : stdout
use kinds, only : DP
use radial_grids, only: radial_grid_type
use ld1inc, only : cau_fact, zed
implicit none
integer :: idim1
type(radial_grid_type),intent(in)::grid
real(DP) :: ruae(idim1), & ! the all electron potential
snl(idim1,2) ! the wavefunction
real(DP) :: e0, & ! the starting energy eigenvalue
jcur, & ! the j of the state
thresh
integer :: mesh, & ! the dimension of the mesh
it, & ! the iteration
ncur, & ! the n of the state
lcur ! the l of the state
real(DP) :: tbya, abyt, &
emin, emax, &
zz(idim1,2,2), &
tolinf,alpha2,alpha, &
yy(idim1,2), &
vzero, &
f0,f1,f2,g0,g1,g2, &
gout, gpout, &
gin, gpin, &
factor,gamma0, &
ecur, &
xw, decur, decurp
real(DP) :: f(idim1), int_0_inf_dr
integer :: itmax, & ! maximum number of iterations
iter, & ! current iteration
ir, & ! counter
ig, & ! auxiliary
kcur, & ! current k
nctp, & ! index of the classical turning point
nodes, & ! the number of nodes
ninf, & ! practical infinite
nstop ! 0 if all ok, 1 otherwise
!
! r o u t i n e i n i t i a l i s a t i o n
if (mesh.ne.grid%mesh) call errore('dirsol','mesh dimension is not as expected',1)
nstop=0
!
! set the maximum number of iterations for improving wavefunctions
!
itmax = 100
!
! set ( 2 / fine structure constant )
!tbya = 2.0_DP * 137.04_DP
tbya = 2.0_DP * cau_fact
! set ( fine structure constant / 2 )
abyt = 1.0_DP / tbya
if (jcur.eq.lcur+0.5_DP) then
kcur = - ( lcur + 1 )
else
kcur = lcur
endif
!
! set initial upper and lower bounds for the eigen value
emin = - 1.0e10_DP
emax = 1.0_DP
ecur=e0
!
do iter = 1,itmax
yy = 0.0_DP
!
! define the zz array
! ===================
!
if ( iter .eq. 1 ) then
do ir = 1,mesh
zz(ir,1,1) = grid%rab(ir) * DBLE(kcur) / grid%r(ir)
zz(ir,2,2) = - zz(ir,1,1)
enddo
endif
do ir = 1,mesh
zz(ir,1,2) = - grid%rab(ir) * ( ecur - ruae(ir) ) * abyt
zz(ir,2,1) = - zz(ir,1,2) + grid%rab(ir) * tbya
enddo
!
! ==============================================
! classical turning point and practical infinity
! ==============================================
!
do nctp = mesh,10,-1
if ( zz(nctp,1,2) .lt. 0.0_DP ) goto 240
enddo
call errore('dirsol', 'no classical turning point found', 1)
!
! jump point out of classical turning point loop
240 continue
if ( nctp .gt. mesh - 10 ) then
! write(stdout,*) 'State nlk=', ncur, lcur, kcur, nctp, mesh
! write(stdout,*) 'ecur, ecurmax=', ecur, ruae(mesh-10)
write(stdout,*) 'classical turning point too close to mesh',ncur,lcur,kcur
snl=0.0_DP
e0=e0-0.2_DP
nstop=1
goto 700
endif
!
tolinf = log(thresh) ** 2
do ninf = nctp+10,mesh
alpha2 = (ruae(ninf)-ecur) * (grid%r(ninf) - grid%r(nctp))**2
if ( alpha2 .gt. tolinf ) goto 260
enddo
!
! jump point out of practical infinity loop
260 continue
!
if (ninf.gt.mesh) ninf=mesh
!
! ===========================================================
! analytic start up of minor and major components from origin
! ===========================================================
!
! with finite nucleus so potential constant at origin we have
!
! f(r) = sum_n f_n r ** ( ig + n )
! g(r) = sum_n g_n r ** ( ig + n )
!
! with
!
! f_n+1 = - (ecur-v(0)) * abyt * g_n / ( ig - kcur + n + 1 )
! g_n+1 = (ecur-v(0)+tbya**2 ) * abyt * f_n / ( ig + kcur + n + 1)
!
! if kcur > 0 ig = + kcur , f_0 = 1 , g_0 = 0
! if kcur < 0 ig = - kcur , f_0 = 0 , g_1 = 1
!
! Uncomment the following instruction if you need this boundary conditions
!
! vzero = ruae(1)
!
! set f0 and g0
! if ( kcur .lt. 0 ) then
! ig = - kcur
! f0 = 0
! g0 = 1
! else
! ig = kcur
! f0 = 1
! g0 = 0
! endif
!
! f1 = - (ecur-vzero) * abyt * g0 / DBLE( ig - kcur + 1 )
! g1 = (ecur-vzero+tbya**2) * abyt * f0 / DBLE( ig + kcur + 1 )
! f2 = - (ecur-vzero) * abyt * g1 / DBLE( ig - kcur + 2 )
! g2 = (ecur-vzero+tbya**2) * abyt * f1 / DBLE( ig + kcur + 2 )
!
!
! do ir = 1,5
! yy(ir,1) = grid%r(ir)**ig * ( f0 + grid%r(ir) * ( f1 + grid%r(ir) * f2 ) )
! yy(ir,2) = grid%r(ir)**ig * ( g0 + grid%r(ir) * ( g1 + grid%r(ir) * g2 ) )
! enddo
!
! The following boundary conditions are for the Coulomb nuclear potential
! ADC 05/10/2007
!
vzero = ruae(1)+zed*2.0_DP/grid%r(1)
!
gamma0=sqrt(kcur**2-4.0_DP*(abyt*zed)**2)
if ( kcur .lt. 0 ) then
ig = - kcur
f0 = (kcur+gamma0)/(2.0_DP*abyt*zed)
g0 = 1.0_DP
else
ig = kcur
f0 = 1.0_DP
g0 = (kcur-gamma0)/(2.0_DP*abyt*zed)
endif
!
f1 = - (ecur-vzero) * abyt * g0 / ( gamma0 - kcur + 1.0_DP )
g1 = (ecur-vzero+tbya**2) * abyt * f0 / ( gamma0 + kcur + 1.0_DP )
f2 = - (ecur-vzero) * abyt * g1 / ( gamma0 - kcur + 2.0_DP )
g2 = (ecur-vzero+tbya**2) * abyt * f1 / ( gamma0 + kcur + 2.0_DP )
!
!
do ir = 1,5
yy(ir,1) = grid%r(ir)**gamma0*(f0+grid%r(ir)*(f1+grid%r(ir)*f2))
yy(ir,2) = grid%r(ir)**gamma0*(g0+grid%r(ir)*(g1+grid%r(ir)*g2))
enddo
! ===========================
! outward integration to nctp
! ===========================
!
! fifth order predictor corrector integration routine
call cfdsol(zz,yy,6,nctp,idim1)
!
! save major component and its gradient at nctp
gout = yy(nctp,2)
gpout = zz(nctp,2,1)*yy(nctp,1) + zz(nctp,2,2)*yy(nctp,2)
gpout = gpout / grid%rab(nctp)
!
! ==============================================
! start up of wavefunction at practical infinity
! ==============================================
!
do ir = ninf,ninf-4,-1
alpha = sqrt( ruae(ir) - ecur )
yy(ir,2) = exp ( - alpha * ( grid%r(ir) - grid%r(nctp) ) )
yy(ir,1) = ( DBLE(kcur)/grid%r(ir) - alpha ) * yy(ir,2)*tbya / &
& ( ecur - ruae(ir) + tbya ** 2 )
enddo
!
! ==========================
! inward integration to nctp
! ==========================
!
! fifth order predictor corrector integration routine
call cfdsol(zz,yy,ninf-5,nctp,idim1)
!
! save major component and its gradient at nctp
gin = yy(nctp,2)
gpin = zz(nctp,2,1)*yy(nctp,1) + zz(nctp,2,2)*yy(nctp,2)
gpin = gpin / grid%rab(nctp)
!
!
! ===============================================
! rescale tail to make major component continuous
! ===============================================
!
factor = gout / gin
do ir = nctp,ninf
yy(ir,1) = factor * yy(ir,1)
yy(ir,2) = factor * yy(ir,2)
enddo
!
gpin = gpin * factor
!
! =================================
! check that the number of nodes ok
! =================================
!
! count the number of nodes in major component
call nodeno(yy(1,2),1,ninf,nodes,idim1)
if ( nodes .lt. ncur - lcur - 1 ) then
! energy is too low
emin = ecur
! write(stdout,*) 'energy too low'
! write(stdout,'(i5,3f12.5,2i5)') &
! & iter,emin,ecur,emax,nodes,ncur-lcur-1
if ( ecur * 0.9_DP .gt. emax ) then
ecur = 0.5_DP * ecur + 0.5_DP * emax
else
ecur = 0.9_DP * ecur
endif
goto 370
endif
!
if ( nodes .gt. ncur - lcur - 1 ) then
! energy is too high
emax = ecur
!
! write(stdout,*) 'energy too high'
! write(stdout,'(i5,3f12.5,2i5)') &
! & iter,emin,ecur,emax,nodes,ncur-lcur-1
if ( ecur * 1.1_DP .lt. emin ) then
ecur = 0.5_DP * ecur + 0.5_DP * emin
else
ecur = 1.1_DP * ecur
endif
goto 370
endif
!
!
! =======================================================
! find normalisation of wavefunction
! =======================================================
!
do ir = 1,ninf
f(ir) = (yy(ir,1)**2 + yy(ir,2)**2)
enddo
factor=int_0_inf_dr(f,grid,ninf,2*ig)
!
!
! =========================================
! variational improvement of the eigenvalue
! =========================================
!
decur = gout * ( gpout - gpin ) / factor
!
! to prevent convergence problems:
! do not allow decur to exceed 20% of | ecur |
! do not allow decur to exceed 70% of distance to emin or emax
if (decur.gt.0.0_DP) then
emin=ecur
decurp=min(decur,-0.2_DP*ecur,0.7_DP*(emax-ecur))
else
emax=ecur
decurp=-min(-decur,-0.2_DP*ecur,0.7_DP*(ecur-emin))
endif
!
! write(stdout,'(i5,3f12.5,1p2e12.4)') &
! & iter,emin,ecur,emax,decur,decurp
!
! test to see whether eigenvalue converged
if ( abs(decur) .lt. thresh ) goto 400
ecur = ecur + decurp
!
! jump point from node check
370 continue
!
! =======================================================
! check that the iterative loop is not about to terminate
! =======================================================
!
if ( iter .eq. itmax ) then
! eigenfunction has not converged in allowed number of iterations
! write(stdout,999) it,ncur,lcur,jcur,e0,ecur
!999 format('iter',i4,' state',i4,i4,f4.1,' could not be converged.',/, &
! & ' starting energy for calculation was',f10.5, &
! & ' and end value =',f10.5)
write(stdout,*) 'state nlj',ncur,lcur,jcur, ' not converged'
snl=0.0_DP
nstop=1
goto 700
endif
!
! close iterative loop
enddo
!
! jump point on successful convergence of eigenvalue
400 continue
!
! normalize the wavefunction and exit
!
snl=0.0_DP
do ir=1,mesh
snl(ir,1)=yy(ir,2)/sqrt(factor)
snl(ir,2)=yy(ir,1)/sqrt(factor)
enddo
e0=ecur
700 continue
return
end subroutine dirsol