quantum-espresso/PW/setup.f90

898 lines
24 KiB
Fortran

!
! Copyright (C) 2001-2004 PWSCF 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 .
!
#include "f_defs.h"
!
!----------------------------------------------------------------------------
SUBROUTINE setup()
!----------------------------------------------------------------------------
!
! ... This routine
! ... 1) determines various parameters of the calculation
! ... 2) finds actual crystal symmetry, determine lattice
! ... 3) generates k-points corresponding to the crystal symmetry
!
! ... Calculated parameters:
! ... msh mesh point (atomic grid) for which R(msh) > Rcut = 10a.u.
! ... zv charge of each atomic type
! ... nelec total number of electrons
! ... nbnd total number of bands
! ... nbndx max number of bands used in iterative diagonalization
! ... tpiba 2 pi / a (a = lattice parameter)
! ... tpiba2 square of tpiba
! ... gcutm cut-off in g space
! ... gcutms cut-off in g space for smooth functions
! ... ethr convergence limit of iterative diagonalization
! ... at direct lattice vectors
! ... omega volume of the unit cell
! ... bg reciprocal lattice vectors
! ... s symmetry matrices in the direct lattice vectors basis
! ... nsym total number of symmetry operations
! ... ftau fractionary translations
! ... irt for each atom gives the corresponding symmetric
! ... invsym if true the system has inversion symmetry
! ... + non-collinear related quantities
! ... + spin-orbit related quantities
! ... + LDA+U-related quantities
!
USE kinds, ONLY : DP
USE constants, ONLY : eps8
USE parameters, ONLY : npsx, nchix, npk
USE io_global, ONLY : stdout
USE constants, ONLY : pi, degspin
USE cell_base, ONLY : at, bg, alat, tpiba, tpiba2, ibrav, symm_type
USE ions_base, ONLY : nat, tau, ntyp => nsp, ityp, zv
USE basis, ONLY : startingpot, natomwfc
USE gvect, ONLY : gcutm, ecutwfc, dual, nr1, nr2, nr3
USE gsmooth, ONLY : doublegrid, gcutms
USE klist, ONLY : xk, wk, xqq, nks, nelec, degauss, lgauss, &
lxkcry, nkstot, b_length, lcart
USE lsda_mod, ONLY : lsda, nspin, current_spin, isk, &
starting_magnetization
USE ktetra, ONLY : nk1, nk2, nk3, k1, k2, k3, tetra, ntetra, ltetra
USE symme, ONLY : s, irt, ftau, nsym, invsym
USE atom, ONLY : r, oc, chi, nchi, lchi, jchi, mesh, msh
USE pseud, ONLY : zp, nlc, nnl, alps, aps, lmax
USE wvfct, ONLY : nbnd, nbndx
USE control_flags, ONLY : tr2, ethr, alpha0, beta0, lscf, &
lmd, lpath, lphonon, david, isolve, imix, &
niter, noinv, nosym, modenum, lraman
USE relax, ONLY : dtau_ref, starting_diag_threshold
USE cellmd, ONLY : calc
USE uspp_param, ONLY : psd, betar, nbeta, dion, jjj, lll, tvanp
USE us, ONLY : okvan
USE ldaU, ONLY : d1, d2, d3, lda_plus_u, Hubbard_U, Hubbard_l, &
Hubbard_alpha, Hubbard_lmax
USE bp, ONLY : gdir, lberry, nppstr
USE fixed_occ, ONLY : f_inp, tfixed_occ
USE char, ONLY : sname
USE mp_global, ONLY : nimage, kunit
USE spin_orb, ONLY : lspinorb, domag
USE noncollin_module, ONLY : noncolin, npol, m_loc, i_cons, mcons, &
angle1, angle2
!
IMPLICIT NONE
!
! ... local variables
!
REAL(KIND=DP), PARAMETER :: &
rcut = 10.D0, &! cut-off radius for radial integrations
eps = 1.0D-12 ! small number
INTEGER :: &
na, &!
ir, &!
nt, &!
input_nks, &!
nrot, &!
iter, &!
ierr, &!
irot, &!
isym, &!
ipol, &!
jpol, &!
tipo, &!
is, &!
nb, &!
nbe, &
l, &!
ibnd !
LOGICAL :: &
so(npsx), &!
minus_q, &!
ltest !
REAL(KIND=DP) :: &
vionl, & !
iocc !
INTEGER, EXTERNAL :: &
n_atom_wfc, &!
set_Hubbard_l
LOGICAL, EXTERNAL :: &
lchk_tauxk ! tests that atomic coordinates do not overlap
!
!
ALLOCATE( m_loc( 3, nat ) )
!
IF ( nimage > 1 .AND. .NOT. lpath ) &
CALL errore( 'setup', 'images parallelization not permitted', 1 )
!
DO nt = 1, ntyp
!
DO ir = 1, mesh(nt)
!
IF ( r(ir,nt) > rcut ) THEN
!
msh(nt) = ir
!
GO TO 5
!
END IF
!
END DO
!
msh(nt) = mesh(nt)
!
! ... force msh to be odd for simpson integration
!
5 msh(nt) = 2 * ( ( msh(nt) + 1 ) / 2 ) - 1
!
END DO
!
IF ( noncolin ) THEN
!
! ... wavefunctions are spinors with 2 components
!
npol = 2
!
! ... transform angles to radiants
!
DO nt = 1, ntyp
!
angle1(nt) = pi * angle1(nt) / 180.D0
angle2(nt) = pi * angle2(nt) / 180.D0
!
END DO
!
! ... Set the nomag variable to make a spin-orbit calculation with zero
! ... magnetization
!
IF ( lspinorb ) THEN
!
domag = .FALSE.
!
DO nt = 1, ntyp
!
domag = domag .OR. ( ABS( starting_magnetization(nt) ) > 1.D-6 )
!
END DO
!
ELSE
!
domag = .TRUE.
!
END IF
!
DO na = 1, nat
!
m_loc(1,na) = starting_magnetization(ityp(na)) * &
SIN( angle1(ityp(na)) ) * COS( angle2(ityp(na)) )
m_loc(2,na) = starting_magnetization(ityp(na)) * &
SIN( angle1(ityp(na)) ) * SIN( angle2(ityp(na)) )
m_loc(3,na) = starting_magnetization(ityp(na)) * &
COS( angle1(ityp(na)) )
END DO
!
IF ( i_cons == 2 ) THEN
!
! ... angle theta between the magnetic moments and the z-axis is
! ... constrained. Transform theta to radiants
!
DO na = 1, ntyp
!
mcons(1,na) = pi * mcons(1,na) / 180.D0
!
END DO
!
END IF
!
ELSE
!
! wavefunctions are scalars
!
npol = 1
!
END IF
!
! ... Compute the ionic charge for each atom type
!
zv(1:ntyp) = zp(1:ntyp)
!
! ... Set the number of electrons equal to the total ionic charge
!
IF ( nelec == 0.D0 ) THEN
!
#if defined (__PGI)
!
DO na = 1, nat
nelec = nelec + zv( ityp(na) )
END DO
!
#else
!
nelec = SUM( zv(ityp(1:nat)) )
!
#endif
!
END IF
!
! ... If the occupations are from input, check the consistency with the
! ... number of electrons
!
IF ( tfixed_occ ) THEN
!
iocc = 0
!
IF ( noncolin ) THEN
!
#if defined (__PGI)
!
DO ibnd = 1, nbnd
iocc = iocc + f_inp(ibnd,1)
END DO
!
#else
!
iocc = iocc + SUM( f_inp(1:nbnd,1) )
!
#endif
!
ELSE
!
DO is = 1, nspin
!
#if defined (__PGI)
!
DO ibnd = 1, nbnd
iocc = iocc + f_inp(ibnd,is)
END DO
!
#else
!
iocc = iocc + SUM( f_inp(1:nbnd,is) )
!
#endif
!
END DO
!
END IF
!
IF ( ABS( iocc - nelec ) > 1D-5 ) &
CALL errore( 'setup', 'strange occupations', 1 )
!
END IF
!
! ... For metals: check whether Gaussian broadening or Tetrahedron method
! ... is used
!
lgauss = ( ( degauss /= 0.D0 ) .AND. ( .NOT. tfixed_occ ) )
!
! ... Check: if there is an odd number of electrons, the crystal is a metal
!
IF ( lscf .AND. ABS( NINT( nelec / 2.D0 ) - nelec / 2.D0 ) > eps8 &
.AND. .NOT. lgauss .AND. .NOT. ltetra .AND. .NOT. tfixed_occ ) &
CALL errore( 'setup', 'the system is metallic, specify occupations', 1 )
!
! ... Check: spin-polarized calculations require tetrahedra or broadening
! or fixed occupation - the simple filling of levels is not
! implemented right now (it will yield an unpolarized system)
!
IF ( lscf .AND. lsda &
.AND. .NOT. lgauss .AND. .NOT. ltetra .AND. .NOT. tfixed_occ ) &
CALL errore( 'setup', 'spin-polarized system, specify occupations', 1 )
!
! ... Set the number of occupied bands if not given in input
!
IF ( nbnd == 0 ) THEN
!
nbnd = NINT( nelec / degspin )
!
IF ( lgauss .OR. ltetra ) THEN
!
! ... metallic case: add 20% more bands, with a minimum of 4
!
nbnd = MAX( NINT( 1.2D0 * nelec / degspin ), ( nbnd + 4 ) )
!
END IF
!
! ... In the case of noncollinear magnetism, bands are NOT
! ... twofold degenerate :
!
IF ( noncolin ) nbnd = INT( degspin ) * nbnd
!
ELSE
!
IF ( nbnd < NINT( nelec / degspin ) .AND. lscf ) &
CALL errore( 'setup', 'too few bands', 1 )
!
IF ( nbnd < NINT( nelec ) .AND. lscf .AND. noncolin ) &
CALL errore( 'setup', 'too few bands', 1 )
!
END IF
!
! ... Here we set the precision of the diagonalization for the first scf
! ... iteration of for the first ionic step
! ... for subsequent steps ethr is automatically updated in electrons
!
ltest = ( ethr == 0.D0 )
!
IF ( lphonon .or. lraman ) THEN
!
! ... in the case of a phonon calculation ethr can not be specified
! ... in the input file
!
IF ( .NOT. ltest ) &
WRITE( UNIT = stdout, &
& FMT = '(5X,"diago_thr_init overwritten ", &
& "with conv_thr / nelec")' )
!
IF ( imix >= 0 ) ethr = 0.1D0 * MIN( 1.D-2, tr2 / nelec )
IF ( imix < 0 ) ethr = 0.1D0 * MIN( 1.D-6, SQRT( tr2 ) )
!
ELSE IF ( .NOT. lscf ) THEN
!
IF ( ltest ) THEN
!
IF ( imix >= 0 ) ethr = 0.1D0 * MIN( 1.D-2, tr2 / nelec )
IF ( imix < 0 ) ethr = 0.1D0 * MIN( 1.D-6, SQRT( tr2 ) )
!
END IF
!
ELSE
!
IF ( ltest ) THEN
!
IF ( startingpot == 'file' ) THEN
!
! ... if you think that the starting potential is good
! ... do not spoil it with a lousy first diagonalization :
! ... set a strict ethr in the input file (diago_thr_init)
!
IF ( imix >= 0 ) ethr = 1.D-5
IF ( imix < 0 ) ethr = 1.D-8
!
ELSE
!
! ... starting atomic potential is probably far from scf
! ... do not waste iterations in the first diagonalizations
!
IF ( imix >= 0 ) ethr = 1.0D-2
IF ( imix < 0 ) ethr = 1.0D-5
!
END IF
!
END IF
!
END IF
!
IF ( .NOT. lscf ) niter = 1
!
starting_diag_threshold = ethr
!
! check if spin-orbit is possible
!
lspinorb = lspinorb .AND. noncolin
!
! ... if this is not a spin-orbit calculation, all spin-orbit pseudopotentials
! ... are transformed into standard pseudopotentials
!
DO nt = 1, ntyp
!
so(nt) = .TRUE.
!
DO nb = 1, nbeta(nt)
!
so(nt) = so(nt) .AND. ( ABS( jjj(nb,nt) ) > 1.D-7 )
!
END DO
!
END DO
!
IF ( .NOT. lspinorb ) THEN
!
DO nt = 1, ntyp
!
IF ( so(nt) ) THEN
!
IF ( tvanp(nt) ) &
CALL errore( 'setup', 'US j-average not yet implemented', 1 )
!
nbe = 0
!
DO nb = 1, nbeta(nt)
!
nbe = nbe + 1
!
IF ( lll(nb,nt) /= 0 .AND. &
ABS( jjj(nb,nt) - lll(nb,nt) - 0.5D0 ) < 1.D-7 ) nbe = nbe - 1
END DO
!
nbeta(nt) = nbe
!
nbe = 0
!
DO nb = 1, nbeta(nt)
!
nbe = nbe + 1
!
l = lll(nbe,nt)
!
IF ( l /= 0 ) THEN
!
vionl = ( ( l + 1.D0 ) * dion(nbe+1,nbe+1,nt) + &
l * dion(nbe,nbe,nt) ) / ( 2.D0 * l + 1.D0 )
!
betar(1:mesh(nt),nb,nt) = 1.D0 / ( 2.D0 * l + 1.D0 ) * &
( ( l + 1.D0 ) * SQRT( dion(nbe+1,nbe+1,nt) / vionl ) * &
betar(1:mesh(nt),nbe+1,nt) + &
l * SQRT( dion(nbe,nbe,nt) / vionl ) * &
betar(1:mesh(nt),nbe,nt) )
!
dion(nb,nb,nt) = vionl
!
nbe = nbe + 1
!
ELSE
!
betar(1:mesh(nt),nb,nt) = betar(1:mesh(nt),nbe,nt)
!
dion(nb,nb,nt) = dion(nbe,nbe,nt)
!
END IF
!
END DO
!
nbe = 0
!
DO nb = 1, nchi(nt)
!
nbe = nbe + 1
!
IF ( lchi(nb,nt) /= 0 .AND. &
ABS( jchi(nb,nt) - lchi(nb,nt) - 0.5D0 ) < 1.D-7 ) nbe = nbe - 1
!
END DO
!
nchi(nt) = nbe
!
nbe = 0
!
do nb = 1, nchi(nt)
!
nbe = nbe + 1
!
l = lchi(nbe,nt)
!
IF ( l /= 0 ) THEN
!
chi(1:mesh(nt),nb,nt)=( ( l + 1.D0 ) * chi(1:mesh(nt),nbe+1,nt)+ &
l * chi(1:mesh(nt),nbe,nt)) / ( 2.D0 * l + 1.D0 )
nbe = nbe + 1
!
ELSE
!
chi(1:mesh(nt),nb,nt) = chi(1:mesh(nt),nbe,nt)
!
END IF
!
END DO
!
END IF
!
END DO
!
END IF
!
! ... set number of atomic wavefunctions
!
natomwfc = n_atom_wfc( nat, npsx, ityp, nchix, nchi, oc, lchi, jchi )
!
! ... set the max number of bands used in iterative diagonalization
!
nbndx = nbnd
!
IF ( isolve == 0 ) nbndx = david * nbnd
!
! ... Set the units in real and reciprocal space
!
tpiba = 2.D0 * pi / alat
tpiba2 = tpiba**2
!
! ... Compute the cut-off of the G vectors
!
gcutm = dual * ecutwfc / tpiba2
!
doublegrid = ( dual > 4.D0 )
!
IF ( doublegrid ) THEN
!
gcutms = 4.D0 * ecutwfc / tpiba2
!
ELSE
!
gcutms = gcutm
!
END IF
!
! ... Generate the reciprocal lattice vectors
!
CALL recips( at(1,1), at(1,2), at(1,3), bg(1,1), bg(1,2), bg(1,3) )
!
! ... If lxkcry = .TRUE. , the input k-point components in crystal
! ... axis are transformed in cartesian coordinates
!
IF ( lxkcry ) CALL cryst_to_cart( nks, xk, bg, 1 )
!
! ... Test that atoms do not overlap
!
IF ( .NOT. ( lchk_tauxk( nat, tau, bg ) ) ) &
CALL errore( 'setup', 'Wrong atomic coordinates ', 1 )
!
! ... set dtau_ref for relaxation and dynamics
! ... this is done here because dtau_ref is updated in cg
!
dtau_ref = 0.2D0
!
! ... calculate dimensions of the FFT grid
!
CALL set_fft_dim()
!
! ... generate transformation matrices for the crystal point group
! ... First we generate all the symmetry matrices of the Bravais lattice
!
IF ( ibrav == 4 .OR. ibrav == 5 ) THEN
!
! ... here the hexagonal or trigonal bravais lattice
!
CALL hexsym( at, s, sname, nrot )
!
tipo = 2
!
ELSE IF ( ibrav >=1 .AND. ibrav <= 14 ) THEN
!
! ... here for the cubic bravais lattice
!
CALL cubicsym( at, s, sname, nrot )
!
tipo = 1
!
ELSE IF ( ibrav == 0 ) THEN
!
IF ( symm_type == 'cubic' ) THEN
!
tipo = 1
!
CALL cubicsym( at, s, sname, nrot )
!
ELSE IF ( symm_type == 'hexagonal' ) THEN
!
tipo = 2
!
CALL hexsym( at, s, sname, nrot )
!
END IF
!
ELSE
!
CALL errore( 'setup', 'wrong ibrav', 1 )
!
END IF
!
! ... if noinv is .TRUE. eliminate all symmetries which exchange z with -z
!
IF ( noinv ) THEN
!
irot = 0
!
DO isym = 1, nrot
IF ( s(1,3,isym) == 0 .AND. s(3,1,isym) == 0 .AND. &
s(2,3,isym) == 0 .AND. s(3,2,isym) == 0 .AND. &
s(3,3,isym) == 1) THEN
!
irot = irot + 1
!
s(:,:,irot) = s(:,:,isym)
!
sname(irot) = sname(isym)
!
END IF
!
END DO
!
nrot = irot
!
END IF
!
! ... If nosym is true do not use any point-group symmetry
!
IF ( nosym ) nrot = 1
!
! ... Automatic generation of k-points (if required)
!
IF ( nks < 0 ) THEN
!
CALL setupkpoint( s, nrot, xk, wk, nks, npk, nk1, &
nk2, nk3, k1, k2, k3, at, bg, tipo )
!
ELSE IF ( nks == 0 ) THEN
!
IF ( lberry ) THEN
!
CALL kp_strings( nppstr, gdir, nrot, s, bg, npk, &
k1, k2, k3, nk1, nk2, nk3, nks, xk, wk )
!
nosym = .TRUE.
nrot = 1
nsym = 1
!
ELSE
!
CALL kpoint_grid( nrot, s, bg, npk, k1, k2, k3, &
nk1, nk2, nk3, nks, xk, wk )
!
END IF
!
END IF
!
! ... allocate space for irt
!
ALLOCATE( irt( 48, nat ) )
!
! ... "sgama" eliminates rotations that are not symmetry operations
! ... Input k-points are assumed to be given in the IBZ of the Bravais
! ... lattice, with the full point symmetry of the lattice.
! ... If some symmetries are missing in the crystal, "sgama" computes
! ... the missing k-points. If nosym is true (see above) we do not use
! ... any point-group symmetry and leave k-points unchanged.
!
input_nks = nks
!
CALL sgama( nrot, nat, s, sname, at, bg, tau, ityp, nsym, nr1, &
nr2, nr3, irt, ftau, npk, nks, xk, wk, invsym, minus_q, &
xqq, modenum, noncolin, m_loc )
!
CALL checkallsym( nsym, s, nat, tau, ityp, at, &
bg, nr1, nr2, nr3, irt, ftau )
!
! ... if dynamics is done the system should have no symmetries
! ... (inversion symmetry alone is allowed)
!
IF ( lmd .AND. ( nsym == 2 .AND. .NOT. invsym .OR. nsym > 2 ) &
.AND. .NOT. ( calc == 'mm' .OR. calc == 'nm' ) ) &
CALL errore( 'setup', 'Dynamics, you should have no symmetries', -1 )
!
! ... Calculate quantities used in tetrahedra method
!
IF ( ltetra ) THEN
!
ntetra = 6 * nk1 * nk2 * nk3
!
ALLOCATE( tetra( 4, ntetra ) )
!
CALL tetrahedra( nsym, s, minus_q, at, bg, npk, k1, k2, k3, &
nk1, nk2, nk3, nks, xk, wk, ntetra, tetra )
!
ELSE
!
ntetra = 0
!
END IF
!
! ... non scf calculation: do not change the number of k-points
!
ltest = ( nks /= input_nks ) .AND. &
( .NOT. lscf ) .AND. ( .NOT. ( lphonon .OR. lraman ) )
!
IF ( ltest ) THEN
!
WRITE( stdout, '(/,5X,"Only input k-points are used ", &
& "(inequivalent points not generated)",/)' )
!
nks = input_nks
!
END IF
!
! ... phonon calculation: add k+q to the list of k
!
IF ( lphonon ) CALL set_kplusq( xk, wk, xqq, nks, npk )
!
! ... raman calculation: add k+b to the list of k
!
IF ( lraman ) CALL set_kplusb(ibrav, xk, wk, b_length, nks, npk, lcart)
!
IF ( lsda ) THEN
!
! ... LSDA case: two different spin polarizations,
! ... each with its own kpoints
!
nspin = 2
!
CALL set_kup_and_kdw( xk, wk, isk, nks, npk )
!
ELSE IF ( noncolin ) THEN
!
! ... noncolinear magnetism: potential and charge have dimension 4 (1+3)
!
nspin = 4
current_spin = 1
!
ELSE
!
! ... LDA case: the two spin polarizations are identical
!
wk(1:nks) = wk(1:nks) * degspin
nspin = 1
current_spin = 1
!
END IF
!
IF ( nks > npk ) CALL errore( 'setup', 'too many k points', nks )
!
#ifdef __PARA
!
! ... set the granularity for k-point distribution
!
IF ( ( ABS( xqq(1) ) < eps .AND. ABS( xqq(2) ) < eps .AND. &
ABS( xqq(3) ) < eps) .OR. ( .NOT. lphonon ) ) THEN
!
kunit = 1
!
ELSE
!
kunit = 2
!
ENDIF
!
IF ( lraman ) THEN
!
IF( lcart ) THEN
!
kunit = 7
!
ELSE
!
IF ( ibrav == 1 ) kunit = 7
IF ( ibrav == 2 ) kunit = 9
IF ( ibrav == 3 ) kunit = 13
!
END IF
!
END IF
!
! ... distribute the k-points (and their weights and spin indices)
!
CALL divide_et_impera( xk, wk, isk, lsda, nkstot, nks )
!
#else
!
! ... set nkstot which is used to write results for all k-points
!
nkstot = nks
!
#endif
!
! ... okvan = .TRUE. : at least one pseudopotential is US
!
okvan = ANY( tvanp(:) )
!
! ... initialize parameters for charge density extrapolation during dynamics
!
alpha0 = 1.D0
beta0 = 0.D0
!
! ... Needed for LDA+U
!
! ... initialize d1 and d2 to rotate the spherical harmonics
!
IF ( lda_plus_u ) THEN
!
Hubbard_lmax = -1
!
DO nt = 1, ntyp
!
IF ( Hubbard_U(nt) /= 0.D0 .OR. Hubbard_alpha(nt) /= 0.D0 ) THEN
!
Hubbard_l(nt) = set_Hubbard_l( psd(nt) )
!
Hubbard_lmax = MAX( Hubbard_lmax, Hubbard_l(nt) )
!
WRITE( UNIT = stdout, &
FMT = * ) ' HUBBARD L FOR TYPE ',psd(nt),' IS ', Hubbard_l(nt)
!
END IF
!
END DO
!
WRITE( UNIT = stdout, &
FMT = * ) ' MAXIMUM HUBBARD L IS ', Hubbard_lmax
!
IF ( Hubbard_lmax == -1 ) &
CALL errore( 'setup', &
& 'lda_plus_u calculation but Hubbard_l not set', 1 )
!
CALL d_matrix( d1, d2, d3 )
!
ELSE
!
Hubbard_lmax = 0
!
END IF
!
RETURN
!
END SUBROUTINE setup
!
!
!----------------------------------------------------------------------------
FUNCTION n_atom_wfc( nat, npsx, ityp, nchix, nchi, oc, lchi, jchi )
!----------------------------------------------------------------------------
!
! ... Find max number of bands needed
!
USE kinds, ONLY : DP
use noncollin_module, ONLY : noncolin
use spin_orb, ONLY : lspinorb
!
IMPLICIT NONE
!
INTEGER :: n_atom_wfc
INTEGER :: nat, npsx, ityp(nat), nchix, nchi(npsx), lchi(nchix,npsx)
REAL(KIND=DP) :: oc(nchix,npsx), jchi(nchix,npsx)
INTEGER :: na, nt, n
!
!
n_atom_wfc = 0
!
DO na = 1, nat
!
nt = ityp(na)
!
DO n = 1, nchi(nt)
!
IF ( oc(n,nt) >= 0.D0 ) THEN
!
IF ( noncolin ) THEN
!
IF ( lspinorb ) THEN
!
n_atom_wfc = n_atom_wfc + 2 * lchi(n,nt)
!
IF ( ABS( jchi(n,nt) - lchi(n,nt) - 0.5D0 ) < 1.D-6 ) &
n_atom_wfc = n_atom_wfc + 2
!
ELSE
!
n_atom_wfc = n_atom_wfc + 2 * ( 2 * lchi(n,nt) + 1 )
!
END IF
!
ELSE
!
n_atom_wfc = n_atom_wfc + 2 * lchi(n,nt) + 1
!
END IF
!
END IF
!
END DO
!
END DO
!
RETURN
!
END FUNCTION n_atom_wfc