quantum-espresso/GWW/simple_ip/diagonalization.f90

196 lines
6.8 KiB
Fortran

SUBROUTINE diagonalization(q,sh,input,eig,ik,kptns)
!------------------------------------------------------------------
!
! This routine calculates H(q)_ij in the optimal basis for a given q and then diagonalizes it.
! It reconstructs V_nonloc_ij from the interpolated projectors
! h_level = 0 only kinetic part
! h_level = 1 kinetic part + local
! h_level > 1 kinetic part + local + non-local
!
USE kinds, ONLY : DP
USE constants, ONLY : pi , rytoev
USE simple_ip_objects
USE input_simple_ip
USE io_global, ONLY : stdout
!
IMPLICIT NONE
!
REAL(kind=DP), INTENT(in) :: q(3) ! k-point (units 2*pi/alat)
INTEGER, INTENT(in) :: ik ! k-point in the local dense grid
TYPE(shirley) :: sh
TYPE(eigen) :: eig
TYPE(kpoints) :: kptns
TYPE(input_options_simple_ip) :: input
COMPLEX(kind=DP), DIMENSION(:,:), ALLOCATABLE :: h_mat
COMPLEX(kind=DP), DIMENSION(:), ALLOCATABLE :: work
REAL(kind=DP), DIMENSION(:), ALLOCATABLE :: rwork , energies_tmp
INTEGER, DIMENSION(:), ALLOCATABLE :: iwork , ifail
COMPLEX(kind=DP), DIMENSION(:,:,:), ALLOCATABLE :: b_nc , csca_nc
COMPLEX(kind=DP), DIMENSION(:,:), ALLOCATABLE :: b_c , csca_mat , csca_mat2 , deeaux , csca_nc2
REAL(kind=DP) :: k2 , abstol , vl , vu
INTEGER :: i, ndim, nt, na, j, ih, jh, ikb, jkb, lwork, info, low_eig, high_eig ,ntot_e , num_bands
!
!call start_clock('diagonalization')
!
if(associated(eig%energy)) deallocate(eig%energy)
allocate(eig%energy(sh%num_bands))
if(associated(eig%wave_func)) deallocate(eig%wave_func)
allocate(eig%wave_func(sh%ntot_e,sh%num_bands))
!
allocate(h_mat(sh%ntot_e,sh%ntot_e),energies_tmp(sh%ntot_e))
h_mat(1:sh%ntot_e,1:sh%ntot_e) = 0.d0
energies_tmp(1:sh%ntot_e) = 0.d0
!
! Construction of the matrix H(q) = (1/2)*[q**2 + q*K1 + K0] + Vloc + Vnonloc(q) in the Shirley basis
! The factor 1/2 in front of the kinetic term is not needed in the QE units
!
! Diagonal part: q^2
k2 = q(1)*q(1) + q(2)*q(2) + q(3)*q(3)
k2 = k2 * (2.0*pi/sh%alat)**2
do i=1,sh%ntot_e
h_mat(i,i) = k2
enddo
!
! Kinetic term: q*K1
do i=1,3
h_mat(1:sh%ntot_e,1:sh%ntot_e) = h_mat(1:sh%ntot_e,1:sh%ntot_e) + &
& q(i)*sh%h1(1:sh%ntot_e,1:sh%ntot_e,i) * (2.0*pi/sh%alat) !*0.5
enddo
!
! Kinetic term: K0
h_mat(1:sh%ntot_e,1:sh%ntot_e) = h_mat(1:sh%ntot_e,1:sh%ntot_e) + sh%h0(1:sh%ntot_e,1:sh%ntot_e)
!
! Vloc
if (input%h_level>0) then
h_mat(1:sh%ntot_e,1:sh%ntot_e) = h_mat(1:sh%ntot_e,1:sh%ntot_e) + sh%Vloc(1:sh%ntot_e,1:sh%ntot_e)
endif
!
! Vnonloc
!call start_clock('diago_vnloc')
if (input%h_level>1) then
!
if (sh%noncolin) then
ndim = sh%nkb * sh%npol * sh%ntot_e
allocate(b_nc(sh%nkb,sh%npol,sh%ntot_e))
if(input%nonlocal_interpolation) then
call trilinear_parallel_nc(kptns,ik,sh,b_nc)
else
b_nc(1:sh%nkb,1:sh%npol,1:sh%ntot_e) = sh%beck_nc(1:sh%nkb, 1:sh%npol, 1:sh%ntot_e,ik)
endif
else
ndim = sh%nkb * sh%ntot_e
allocate(b_c(sh%nkb,sh%ntot_e))
if(input%nonlocal_interpolation) then
call trilinear_parallelc(kptns,ik,sh,b_c)
else
b_c(1:sh%nkb,1:sh%ntot_e) = sh%beckc(1:sh%nkb, 1:sh%ntot_e,ik)
endif
endif
!
! Build Vnonloc
if (sh%noncolin) then
!
allocate(csca_nc(sh%nkb,sh%npol,sh%ntot_e),csca_nc2(sh%ntot_e,sh%ntot_e))
csca_nc(:,:,:) = 0.d0
do nt=1,sh%ntyp
if (sh%nh(nt)==0) CYCLE
do na=1,sh%nat
if ( sh%ityp(na) == nt ) then
do j=1,sh%ntot_e
do jh=1,sh%nh(nt)
jkb = sh%ofsbeta(na)+jh
do ih=1,sh%nh(nt)
ikb = sh%ofsbeta(na)+ih
!
csca_nc(ikb,1,j) = csca_nc(ikb,1,j) + &
sh%deeq_nc(ih,jh,na,1)*b_nc(jkb,1,j)+ &
sh%deeq_nc(ih,jh,na,2)*b_nc(jkb,2,j)
csca_nc(ikb,2,j) = csca_nc(ikb,2,j) + &
sh%deeq_nc(ih,jh,na,3)*b_nc(jkb,1,j)+&
sh%deeq_nc(ih,jh,na,4)*b_nc(jkb,2,j)
!
enddo
enddo
enddo
endif
enddo
enddo
!
call ZGEMM ('C', 'N', sh%ntot_e, sh%ntot_e, sh%npol*sh%nkb, ( 1.d0, 0.d0 ) , b_nc, &
sh%npol*sh%nkb, csca_nc, sh%npol*sh%nkb, ( 0.d0, 0.d0 ) , csca_nc2, sh%ntot_e)
h_mat(1:sh%ntot_e,1:sh%ntot_e) = h_mat(1:sh%ntot_e,1:sh%ntot_e) + csca_nc2(1:sh%ntot_e,1:sh%ntot_e)
!
deallocate(csca_nc,csca_nc2)
!
else
!
allocate(csca_mat(sh%nkb,sh%ntot_e), csca_mat2(sh%ntot_e,sh%ntot_e))
!
do nt=1,sh%ntyp
if (sh%nh(nt)==0) cycle
allocate ( deeaux(sh%nh(nt),sh%nh(nt)) )
do na=1,sh%nat
if ( sh%ityp(na) == nt ) then
!
deeaux(1:sh%nh(nt),1:sh%nh(nt)) = sh%deeqc(1:sh%nh(nt),1:sh%nh(nt),na)
call ZGEMM('N','N', sh%nh(nt), sh%ntot_e, sh%nh(nt), (1.d0,0.d0), &
deeaux, sh%nh(nt), b_c(sh%ofsbeta(na)+1,1), sh%nkb, &
(0.d0, 0.d0), csca_mat(sh%ofsbeta(na)+1,1), sh%nkb )
!
endif
enddo
deallocate(deeaux)
enddo
!
CALL ZGEMM( 'C', 'N', sh%ntot_e, sh%ntot_e, sh%nkb, ( 1.d0, 0.d0 ) , b_c, &
sh%nkb, csca_mat, sh%nkb, ( 0.d0, 0.d0 ) , csca_mat2, sh%ntot_e )
h_mat(1:sh%ntot_e,1:sh%ntot_e) = h_mat(1:sh%ntot_e,1:sh%ntot_e) + csca_mat2(1:sh%ntot_e,1:sh%ntot_e)
!
deallocate(csca_mat, csca_mat2)
!
endif
!
if (sh%noncolin) then
deallocate(b_nc)
else
deallocate(b_c)
endif
!
endif
!call stop_clock('diago_vnloc')
!
! Hamiltonian diagonalization
allocate(rwork(7*sh%ntot_e),iwork(5*sh%ntot_e),ifail(sh%ntot_e))
allocate(work(1))
lwork = -1
abstol = -1.d0
vl = 0.d0
vu = 0.d0
low_eig = sh%num_nbndv(1)-sh%num_val+1
high_eig = sh%num_nbndv(1)+sh%num_cond
num_bands = sh%num_bands
!
CALL ZHEEVX( 'V', 'I', 'U', sh%ntot_e, h_mat, sh%ntot_e, vl, vu, low_eig, high_eig, abstol, &
& num_bands, energies_tmp, eig%wave_func, sh%ntot_e, work, lwork, rwork, iwork, ifail, info)
!
lwork = int(work(1))
deallocate(work)
allocate(work(1:lwork))
!
!Diagonalization of H(k) using ZHEEVX (only the first sh%num_bands energies/wavefunctions are explicitly calculated)
!call start_clock('diago_zheevx')
call ZHEEVX( 'V', 'I', 'U', sh%ntot_e, h_mat, sh%ntot_e, vl, vu, low_eig, high_eig, abstol, &
& num_bands, energies_tmp, eig%wave_func, sh%ntot_e, work, lwork, rwork, iwork, ifail, info)
!call stop_clock('diago_zheevx')
!
if (info /= 0) write(stdout,*) 'ERROR in the diagonalization of the k-dependent Hamiltonian', info
eig%energy(1:sh%num_bands) = energies_tmp(1:sh%num_bands)
!
deallocate(h_mat, rwork, work, iwork, ifail, energies_tmp)
!
!call stop_clock('diagonalization')
!
END SUBROUTINE diagonalization