current.F90

Go to the documentation of this file.

!! Copyright (C) 2008-2019 X. Andrade, F. Bonafe, R. Jestaedt, H. Appel
!!
!! This program is free software; you can redistribute it and/or modify
!! it under the terms of the GNU General Public License as published by
!! the Free Software Foundation; either version 2, or (at your option)
!! any later version.
!!
!! This program is distributed in the hope that it will be useful,
!! but WITHOUT ANY WARRANTY; without even the implied warranty of
!! MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
!! GNU General Public License for more details.
!!
!! You should have received a copy of the GNU General Public License
!! along with this program; if not, write to the Free Software
!! Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
!! 02110-1301, USA.
!!
 
#include "global.h"
 
module current_oct_m
  use accel_oct_m
  use batch_oct_m
  use batch_ops_oct_m
  use boundaries_oct_m
  use comm_oct_m
  use debug_oct_m
  use derivatives_oct_m
  use electron_space_oct_m
  use exchange_operator_oct_m
  use global_oct_m
  use grid_oct_m
  use hamiltonian_elec_oct_m
  use, intrinsic :: iso_fortran_env
  use ks_potential_oct_m
  use lalg_basic_oct_m
  use lda_u_oct_m
  use math_oct_m
  use magnetic_oct_m
  use mesh_function_oct_m
  use mesh_oct_m
  use messages_oct_m
  use namespace_oct_m
  use nonlocal_pseudopotential_oct_m
  use parser_oct_m
  use profiling_oct_m
  use projector_oct_m
  use scissor_oct_m
  use space_oct_m
  use states_elec_oct_m
  use states_elec_parallel_oct_m
  use types_oct_m
  use unit_oct_m
  use unit_system_oct_m
  use varinfo_oct_m
  use wfs_elec_oct_m
  use xc_oct_m
 
  implicit none
 
  private
 
  type current_t
    private
    integer :: method
    logical :: include_diamag
  end type current_t
 
 
  public ::                               &
    current_t,                            &
    current_init,                         &
    current_calculate,                    &
    current_calculate_mag,                &
    current_heat_calculate
 
  integer, parameter, public ::           &
    CURRENT_GRADIENT           = 1,       &
    current_gradient_corr      = 2,       &
    current_hamiltonian        = 3
 
contains
 
  subroutine current_init(this, namespace)
    type(current_t),   intent(out)   :: this
    type(namespace_t), intent(in)    :: namespace
 
    push_sub(current_init)
 
    !%Variable CurrentDensity
    !%Default gradient_corrected
    !%Type integer
    !%Section Hamiltonian
    !%Description
    !% This variable selects the method used to
    !% calculate the current density. For the moment this variable is
    !% for development purposes and users should not need to use
    !% it.
    !%Option gradient 1
    !% The calculation of current is done using the gradient operator. (Experimental)
    !%Option gradient_corrected 2
    !% The calculation of current is done using the gradient operator
    !% with additional corrections for the total current from non-local operators.
    !%Option hamiltonian 3
    !% The current density is obtained from the commutator of the
    !% Hamiltonian with the position operator. (Experimental)
    !%End
 
    call parse_variable(namespace, 'CurrentDensity', current_gradient_corr, this%method)
    if (.not. varinfo_valid_option('CurrentDensity', this%method)) call messages_input_error(namespace, 'CurrentDensity')
    if (this%method /= current_gradient_corr) then
      call messages_experimental("CurrentDensity /= gradient_corrected")
    end if
 
    !%Variable CalculateDiamagneticCurrent
    !%Default no
    !%Type logical
    !%Section Hamiltonian
    !%Description
    !% This variable decides whether the current density arising from the non-uniform
    !% vector potential, defined as:
    !% <math> \vec{J}_{dmc}(\vec{r}, t)=-\frac{e^2}{m_e c_0} n(\vec{r}, t) \vec{A}(\vec{r},t)$ </math>
    !% is included in the total current density.
    !%End
    call parse_variable(namespace, 'CalculateDiamagneticCurrent', .false., this%include_diamag)
 
    pop_sub(current_init)
  end subroutine current_init
 
  ! ---------------------------------------------------------
 
  subroutine current_batch_accumulate(st, der, ik, ib, psib, gpsib)
    type(states_elec_t), intent(inout) :: st
    type(derivatives_t), intent(inout) :: der
    integer,             intent(in)    :: ik
    integer,             intent(in)    :: ib
    type(wfs_elec_t),    intent(in)    :: psib
    class(wfs_elec_t),   intent(in)    :: gpsib(:)
 
    integer :: ist, idir, ii, ip, idim, bsize, gsize
    complex(real64), allocatable :: psi(:, :), gpsi(:, :)
    real(real64), allocatable :: current_tmp(:, :)
    complex(real64) :: c_tmp
    real(real64) :: ww
    real(real64), allocatable :: weight(:)
    type(accel_mem_t) :: buff_weight, buff_current
    type(accel_kernel_t), save :: kernel
 
    safe_allocate(psi(1:der%mesh%np_part, 1:st%d%dim))
    safe_allocate(gpsi(1:der%mesh%np_part, 1:st%d%dim))
 
    if (st%d%ispin == spinors .or. (psib%status() == batch_device_packed .and. der%dim /= 3)) then
 
      do idir = 1, der%dim
        do ist = states_elec_block_min(st, ib), states_elec_block_max(st, ib)
 
          ww = st%kweights(ik)*st%occ(ist, ik)
          if (abs(ww) <= m_epsilon) cycle
 
          do idim = 1, st%d%dim
            ii = st%group%psib(ib, ik)%inv_index((/ist, idim/))
            call batch_get_state(psib, ii, der%mesh%np, psi(:, idim))
            call batch_get_state(gpsib(idir), ii, der%mesh%np, gpsi(:, idim))
          end do
 
          if (st%d%ispin /= spinors) then
            !$omp parallel do
            do ip = 1, der%mesh%np
              st%current_kpt(ip, idir, ik) = st%current_kpt(ip, idir, ik) + ww*aimag(conjg(psi(ip, 1))*gpsi(ip, 1))
            end do
            !$omp end parallel do
          else
            !$omp parallel do private(c_tmp)
            do ip = 1, der%mesh%np
              st%current_para(ip, idir, 1) = st%current_para(ip, idir, 1) + ww*aimag(conjg(psi(ip, 1))*gpsi(ip, 1))
              st%current_para(ip, idir, 2) = st%current_para(ip, idir, 2) + ww*aimag(conjg(psi(ip, 2))*gpsi(ip, 2))
              c_tmp = -m_half*(conjg(psi(ip, 2))*gpsi(ip, 1) - psi(ip, 1)*conjg(gpsi(ip, 2)))
              st%current_para(ip, idir, 3) = st%current_para(ip, idir, 3) - ww*aimag(c_tmp)
              st%current_para(ip, idir, 4) = st%current_para(ip, idir, 4) + ww*real(c_tmp, real64)
            end do
            !$omp end parallel do
          end if
 
        end do
      end do
 
    else if (psib%status() == batch_device_packed) then
 
      assert(der%dim == 3)
 
      safe_allocate(weight(1:psib%nst))
      do ist = 1, psib%nst
        weight(ist) = st%kweights(ik)*st%occ(psib%ist(ist), ik)
      end do
 
 
      call accel_create_buffer(buff_weight, accel_mem_read_only, type_float, psib%nst)
      call accel_write_buffer(buff_weight, psib%nst, weight, async=.true.)
 
      call accel_create_buffer(buff_current, accel_mem_write_only, type_float, der%mesh%np*3)
 
      call accel_kernel_start_call(kernel, 'density.cu', 'current_accumulate')
 
      call accel_set_kernel_arg(kernel, 0, psib%nst)
      call accel_set_kernel_arg(kernel, 1, der%mesh%np)
      call accel_set_kernel_arg(kernel, 2, buff_weight)
      call accel_set_kernel_arg(kernel, 3, psib%ff_device)
      call accel_set_kernel_arg(kernel, 4, log2(int(psib%pack_size(1), int32)))
      call accel_set_kernel_arg(kernel, 5, gpsib(1)%ff_device)
      call accel_set_kernel_arg(kernel, 6, gpsib(2)%ff_device)
      call accel_set_kernel_arg(kernel, 7, gpsib(3)%ff_device)
      call accel_set_kernel_arg(kernel, 8, log2(int(gpsib(1)%pack_size(1), int32)))
      call accel_set_kernel_arg(kernel, 9, buff_current)
 
      ! Compute the grid size
      bsize = accel_kernel_block_size(kernel)
      call accel_grid_size(der%mesh%np, bsize, gsize)
 
      call accel_kernel_run(kernel, (/gsize/), (/bsize/))
 
      safe_allocate(current_tmp(1:der%mesh%np, 1:der%dim))
 
      call accel_read_buffer(buff_current, der%mesh%np, der%dim, current_tmp)
 
      call lalg_axpy(der%mesh%np, der%dim, m_one, current_tmp, st%current_kpt(:,:,ik))
 
      safe_deallocate_a(current_tmp)
 
      call accel_free_buffer(buff_weight)
      call accel_free_buffer(buff_current)
 
      safe_deallocate_a(weight)
 
    else
 
      assert(psib%is_packed() .eqv. gpsib(1)%is_packed())
 
      !$omp parallel private(ip, ist, ww, idir)
      do ii = 1, psib%nst
        ist = states_elec_block_min(st, ib) + ii - 1
        ww = st%kweights(ik)*st%occ(ist, ik)
        if (abs(ww) <= m_epsilon) cycle
 
        if (psib%is_packed()) then
          do idir = 1, der%dim
            !$omp do
            do ip = 1, der%mesh%np
              st%current_kpt(ip, idir, ik) = st%current_kpt(ip, idir, ik) &
                + ww*aimag(conjg(psib%zff_pack(ii, ip))*gpsib(idir)%zff_pack(ii, ip))
            end do
            !$omp end do nowait
          end do
        else
          do idir = 1, der%dim
            !$omp do
            do ip = 1, der%mesh%np
              st%current_kpt(ip, idir, ik) = st%current_kpt(ip, idir, ik) &
                + ww*aimag(conjg(psib%zff(ip, 1, ii))*gpsib(idir)%zff(ip, 1, ii))
            end do
            !$omp end do nowait
          end do
        end if
      end do
      !$omp end parallel
 
    end if
 
    safe_deallocate_a(psi)
    safe_deallocate_a(gpsi)
 
  end subroutine current_batch_accumulate
 
 
  ! ---------------------------------------------------------
  subroutine current_calculate(this, namespace, gr, hm, space, st)
    type(current_t),          intent(in)    :: this
    type(namespace_t),        intent(in)    :: namespace
    type(grid_t),             intent(inout) :: gr
    type(hamiltonian_elec_t), intent(inout) :: hm
    class(space_t),           intent(in)    :: space
    type(states_elec_t),      intent(inout) :: st
 
 
    call profiling_in("CURRENT_TOTAL")
    push_sub(current_calculate)
 
    call current_calculate_para(this, namespace, gr, hm, space, st)
    st%current = st%current_para
 
    if (this%include_diamag) then
      call current_calculate_dia_non_unif_vec_pot(gr%der, hm, st)
      st%current = st%current + st%current_dia
    end if
 
    call profiling_out("CURRENT_TOTAL")
    pop_sub(current_calculate)
 
  end subroutine current_calculate
 
  ! ---------------------------------------------------------
  subroutine current_calculate_dia_non_unif_vec_pot(der, hm, st)
    type(derivatives_t),      intent(inout) :: der
    type(hamiltonian_elec_t), intent(in)    :: hm
    type(states_elec_t),      intent(inout) :: st
 
    integer :: ispin, idir, ip
 
    call profiling_in("CURRENT_DIA_NON_UNIF_A")
    push_sub(current_calculate_dia_non_unif_vec_pot)
 
    st%current_dia = m_zero
 
    if(allocated(hm%hm_base%vector_potential)) then
      do ispin = 1, st%d%nspin
        do idir = 1, der%dim
          !$omp parallel do
          do ip = 1, der%mesh%np
            ! the vector potential is assumed to be devided by c_0 already
            st%current_dia(ip, idir, ispin) = st%current_dia(ip, idir, ispin) + &
              st%rho(ip, ispin)*hm%hm_base%vector_potential(idir, ip)
          end do
          !$omp end parallel do
        end do
      end do
    end if
 
    call profiling_out("CURRENT_DIA_NON_UNIF_A")
    pop_sub(current_calculate_dia_non_unif_vec_pot)
 
  end subroutine current_calculate_dia_non_unif_vec_pot
 
  ! ---------------------------------------------------------
  subroutine current_calculate_mag(der, st)
    type(derivatives_t),      intent(inout) :: der
    type(states_elec_t),      intent(inout) :: st
 
    real(real64), allocatable :: magnetization_density(:, :), curl_mag(:, :)
 
    call profiling_in("CURRENT_MAG")
    push_sub(current_calculate_mag)
 
    st%current_mag = m_zero
 
    if (st%d%ispin /= unpolarized) then
      safe_allocate(magnetization_density(1:der%mesh%np_part, 1:der%dim))
      safe_allocate(curl_mag(1:der%mesh%np_part, 1:der%dim))
 
      call magnetic_density(der%mesh, st%d, st%rho, magnetization_density)
      call dderivatives_curl(der, magnetization_density, curl_mag)
 
      call lalg_axpy(der%mesh%np, der%dim, m_half, curl_mag, st%current_mag(:, :, 1))
 
      safe_deallocate_a(magnetization_density)
      safe_deallocate_a(curl_mag)
    end if
 
    call profiling_out("CURRENT_MAG")
    pop_sub(current_calculate_mag)
 
  end subroutine current_calculate_mag
 
  ! ---------------------------------------------------------
  subroutine current_calculate_para(this, namespace, gr, hm, space, st)
    type(current_t),             intent(in)    :: this
    type(namespace_t),           intent(in)    :: namespace
    type(grid_t),                intent(inout) :: gr
    type(hamiltonian_elec_t),    intent(inout) :: hm
    class(space_t),              intent(in)    :: space
    type(states_elec_t), target, intent(inout) :: st
 
    integer :: ik, ist, idir, idim, ip, ib, ii, ispin
    complex(real64), allocatable :: gpsi(:, :, :), psi(:, :), hpsi(:, :), rhpsi(:, :), rpsi(:, :), hrpsi(:, :)
    type(wfs_elec_t) :: hpsib, rhpsib, rpsib, hrpsib, epsib
    class(wfs_elec_t), allocatable :: commpsib(:)
    real(real64) :: ww
    complex(real64) :: c_tmp
 
    call profiling_in("CURRENT_PARA")
    push_sub(current_calculate_para)
 
    ! spin not implemented or tested
    assert(all(ubound(st%current_para) == (/gr%np_part, gr%der%dim, st%d%nspin/)))
    assert(all(ubound(st%current_kpt) == (/gr%np, gr%der%dim, st%d%kpt%end/)))
    assert(all(lbound(st%current_kpt) == (/1, 1, st%d%kpt%start/)))
 
    safe_allocate(psi(1:gr%np_part, 1:st%d%dim))
    safe_allocate(gpsi(1:gr%np, 1:gr%der%dim, 1:st%d%dim))
    safe_allocate(hpsi(1:gr%np_part, 1:st%d%dim))
    safe_allocate(rhpsi(1:gr%np_part, 1:st%d%dim))
    safe_allocate(rpsi(1:gr%np_part, 1:st%d%dim))
    safe_allocate(hrpsi(1:gr%np_part, 1:st%d%dim))
    safe_allocate_type_array(wfs_elec_t, commpsib, (1:gr%der%dim))
 
    !$omp parallel private(idir, ip, ispin)
    do ik = st%d%kpt%start, st%d%kpt%end
      do idir = 1, gr%der%dim
        !$omp do
        do ip = 1, gr%np
          st%current_kpt(ip, idir, ik) = m_zero
        end do
        !$omp end do nowait
      end do
    end do
    do ispin = 1, st%d%nspin
      do idir = 1, gr%der%dim
        !$omp do
        do ip = 1, gr%np
          st%current_para(ip, idir, ispin) = m_zero
        end do
        !$omp end do nowait
      end do
    end do
    !$omp end parallel
 
    select case (this%method)
 
    case (current_hamiltonian)
      if (family_is_hybrid(hm%xc)) then
        if (.not. hm%exxop%useACE) then
          hm%exxop%st => st
          call states_elec_parallel_remote_access_start(hm%exxop%st)
        end if
      end if
 
      do ik = st%d%kpt%start, st%d%kpt%end
        ispin = st%d%get_spin_index(ik)
        do ib = st%group%block_start, st%group%block_end
 
          call st%group%psib(ib, ik)%do_pack(copy = .true.)
 
          call st%group%psib(ib, ik)%copy_to(hpsib)
          call st%group%psib(ib, ik)%copy_to(rhpsib)
          call st%group%psib(ib, ik)%copy_to(rpsib)
          call st%group%psib(ib, ik)%copy_to(hrpsib)
 
          call boundaries_set(gr%der%boundaries, gr, st%group%psib(ib, ik))
          call zhamiltonian_elec_apply_batch(hm, namespace, gr, st%group%psib(ib, ik), hpsib, set_bc = .false.)
 
          do idir = 1, gr%der%dim
 
            call batch_mul(gr%np, gr%x_t(:, idir), hpsib, rhpsib)
            call batch_mul(gr%np_part, gr%x_t(:, idir), st%group%psib(ib, ik), rpsib)
 
            call zhamiltonian_elec_apply_batch(hm, namespace, gr, rpsib, hrpsib, set_bc = .false.)
 
            do ist = states_elec_block_min(st, ib), states_elec_block_max(st, ib)
              ww = st%kweights(ik)*st%occ(ist, ik)
              if (ww <= m_epsilon) cycle
 
              do idim = 1, st%d%dim
                ii = st%group%psib(ib, ik)%inv_index((/ist, idim/))
                call batch_get_state(st%group%psib(ib, ik), ii, gr%np, psi(:, idim))
                call batch_get_state(hrpsib, ii, gr%np, hrpsi(:, idim))
                call batch_get_state(rhpsib, ii, gr%np, rhpsi(:, idim))
              end do
 
              if (st%d%ispin /= spinors) then
                !$omp parallel do
                do ip = 1, gr%np
                  st%current_kpt(ip, idir, ik) = st%current_kpt(ip, idir, ik) &
                    - ww*aimag(conjg(psi(ip, 1))*hrpsi(ip, 1) - conjg(psi(ip, 1))*rhpsi(ip, 1))
                end do
                !$omp end parallel do
              else
                !$omp parallel do  private(c_tmp)
                do ip = 1, gr%np
                  st%current_para(ip, idir, 1) = st%current_para(ip, idir, 1) + &
                    ww*aimag(conjg(psi(ip, 1))*hrpsi(ip, 1) - conjg(psi(ip, 1))*rhpsi(ip, 1))
                  st%current_para(ip, idir, 2) = st%current_para(ip, idir, 2) + &
                    ww*aimag(conjg(psi(ip, 2))*hrpsi(ip, 2) - conjg(psi(ip, 2))*rhpsi(ip, 2))
                  c_tmp = -m_half*m_zi*(conjg(psi(ip, 2))*hrpsi(ip, 1) - conjg(psi(ip, 2))*rhpsi(ip, 1) &
                    -psi(ip, 1)*conjg(hrpsi(ip, 2)) - psi(ip, 1)*conjg(rhpsi(ip, 2)))
                  st%current_para(ip, idir, 3) = st%current_para(ip, idir, 3) + ww*real(c_tmp, real64)
                  st%current_para(ip, idir, 4) = st%current_para(ip, idir, 4) + ww*aimag(c_tmp)
                end do
                !$omp end parallel do
              end if
 
            end do
 
          end do
 
          call st%group%psib(ib, ik)%do_unpack(copy = .false.)
 
          call hpsib%end()
          call rhpsib%end()
          call rpsib%end()
          call hrpsib%end()
 
        end do
      end do
 
      if (family_is_hybrid(hm%xc)) then
        if (.not. hm%exxop%useACE) then
          call states_elec_parallel_remote_access_stop(hm%exxop%st)
          nullify(hm%exxop%st)
        end if
      end if
 
    case (current_gradient, current_gradient_corr)
 
      if (this%method == current_gradient_corr .and. .not. family_is_mgga_with_exc(hm%xc) &
        .and. hm%theory_level /= hartree_fock &
        .and. hm%theory_level /= generalized_kohn_sham_dft &
        .and. hm%theory_level /= rdmft &
        .and. hm%vnl%apply_projector_matrices) then
 
        ! we can use the packed version
 
        do ik = st%d%kpt%start, st%d%kpt%end
          ispin = st%d%get_spin_index(ik)
          do ib = st%group%block_start, st%group%block_end
 
            ! copy st%group%psib(ib, ik) to epsib and set the phase
            call hm%phase%copy_and_set_phase(gr, st%d%kpt, st%group%psib(ib, ik), epsib)
 
            ! this now takes non-orthogonal axis into account
            call zderivatives_batch_grad(gr%der, epsib, commpsib, set_bc=.false.)
 
            call hm%vnl%zposition_commutator(gr, st%d, gr%der%boundaries%spiral, epsib, commpsib, async=.true.)
 
            call zlda_u_commute_r(hm%lda_u, gr, space, st%d, namespace, epsib, commpsib)
 
            call current_batch_accumulate(st, gr%der, ik, ib, epsib, commpsib)
 
            do idir = 1, gr%der%dim
              call commpsib(idir)%end()
            end do
 
            call epsib%end()
 
            call accel_finish()
          end do
        end do
 
      else
 
        ! use the slow non-packed version
 
        do ik = st%d%kpt%start, st%d%kpt%end
          ispin = st%d%get_spin_index(ik)
          do ist = st%st_start, st%st_end
 
            ww = st%kweights(ik)*st%occ(ist, ik)
            if (abs(ww) <= m_epsilon) cycle
 
            call states_elec_get_state(st, gr, ist, ik, psi)
 
            if (hm%phase%is_allocated()) then
              call hm%phase%apply_to_single(psi, gr%np, st%d%dim, ik, .false.)
              ! apply phase correction while setting boundary -> memory needs to be
              ! accessed only once
              do idim = 1, st%d%dim
                call boundaries_set(gr%der%boundaries, gr, psi(:, idim), phase_correction = hm%phase%phase_corr(:, ik), &
                  buff_phase_corr = hm%phase%buff_phase_corr, offset=int((ik-st%d%kpt%start)*(gr%np_part-gr%np)))
              end do
            else
              do idim = 1, st%d%dim
                call boundaries_set(gr%der%boundaries, gr, psi(:, idim))
              end do
            end if
 
            do idim = 1, st%d%dim
              call zderivatives_grad(gr%der, psi(:, idim), gpsi(:, :, idim), set_bc = .false.)
            end do
 
            if (this%method == current_gradient_corr) then
              !A nonlocal contribution from the MGGA potential must be included
              !This must be done first, as this is like a position-dependent mass
              call hm%ks_pot%zcurrent_mass_renormalization(gpsi, gr%der%dim, st%d%dim, ispin)
 
              !A nonlocal contribution from the pseudopotential must be included
              call zprojector_commute_r_allatoms_alldir(hm%ep%proj, hm%ions, gr, st%d%dim, &
                gr%der%boundaries, ik, psi, gpsi)
              !A nonlocal contribution from the scissor must be included
              if (hm%scissor%apply) then
                call scissor_commute_r(hm%scissor, gr, ik, psi, gpsi)
              end if
 
              call zlda_u_commute_r_single(hm%lda_u, gr, space, st%d, namespace, ist, ik, &
                psi, gpsi, hm%phase%is_allocated())
 
              call zexchange_operator_commute_r(hm%exxop, namespace, gr, st%d, ik, psi, gpsi)
 
            end if
 
            if (st%d%ispin /= spinors) then
              do idir = 1, gr%der%dim
                !$omp parallel do
                do ip = 1, gr%np
                  st%current_kpt(ip, idir, ik) = st%current_kpt(ip, idir, ik) + &
                    ww*aimag(conjg(psi(ip, 1))*gpsi(ip, idir, 1))
                end do
                !$omp end parallel do
              end do
            else
              do idir = 1, gr%der%dim
                !$omp parallel do  private(c_tmp)
                do ip = 1, gr%np
                  st%current_para(ip, idir, 1) = st%current_para(ip, idir, 1) + &
                    ww*aimag(conjg(psi(ip, 1))*gpsi(ip, idir, 1))
                  st%current_para(ip, idir, 2) = st%current_para(ip, idir, 2) + &
                    ww*aimag(conjg(psi(ip, 2))*gpsi(ip, idir, 2))
                  c_tmp = -m_half*m_zi*(conjg(psi(ip, 2))*gpsi(ip, idir, 1) - psi(ip, 1)*conjg(gpsi(ip, idir, 2)))
                  st%current_para(ip, idir, 3) = st%current_para(ip, idir, 3) + ww*real(c_tmp, real64)
                  st%current_para(ip, idir, 4) = st%current_para(ip, idir, 4) + ww*aimag(c_tmp)
                end do
                !$omp end parallel do
              end do
            end if
 
          end do
        end do
 
      end if
 
    case default
 
      assert(.false.)
 
    end select
 
    if (st%d%ispin /= spinors) then
      !We sum the current over k-points
      do ik = st%d%kpt%start, st%d%kpt%end
        ispin = st%d%get_spin_index(ik)
        call lalg_axpy(gr%np, gr%der%dim, m_one, st%current_kpt(:, :, ik), st%current_para(:, :, ispin))
      end do
    end if
 
    if (st%parallel_in_states .or. st%d%kpt%parallel) then
      call comm_allreduce(st%st_kpt_mpi_grp, st%current_para, dim = (/gr%np, gr%der%dim, st%d%nspin/))
    end if
 
    if (st%symmetrize_density) then
      do ispin = 1, st%d%nspin
        call dgrid_symmetrize_vector_field(gr, st%current_para(:, :, ispin), suppress_warning = .true.)
      end do
    end if
 
    safe_deallocate_a(gpsi)
    safe_deallocate_a(psi)
    safe_deallocate_a(hpsi)
    safe_deallocate_a(rhpsi)
    safe_deallocate_a(rpsi)
    safe_deallocate_a(hrpsi)
    safe_deallocate_a(commpsib)
 
    call profiling_out("CURRENT_PARA")
    pop_sub(current_calculate_para)
 
  end subroutine current_calculate_para
 
  ! ---------------------------------------------------------
  subroutine current_heat_calculate(space, der, hm, st, current)
    class(space_t),           intent(in)    :: space
    type(derivatives_t),      intent(in)    :: der
    type(hamiltonian_elec_t), intent(in)    :: hm
    type(states_elec_t),      intent(in)    :: st
    real(real64),             intent(out)   :: current(:, :, :)
 
    integer :: ik, ist, idir, idim, ip, ispin, ndim
    complex(real64), allocatable :: gpsi(:, :, :), psi(:, :), g2psi(:, :, :, :)
    complex(real64) :: tmp
 
    push_sub(current_heat_calculate)
 
    assert(space%is_periodic())
    assert(st%d%dim == 1)
 
    ndim = space%dim
 
    safe_allocate(psi(1:der%mesh%np_part, 1:st%d%dim))
    safe_allocate(gpsi(1:der%mesh%np_part, 1:ndim, 1:st%d%dim))
    safe_allocate(g2psi(1:der%mesh%np, 1:ndim, 1:ndim, 1:st%d%dim))
 
    do ip = 1, der%mesh%np
      current(ip, 1:ndim, 1:st%d%nspin) = st%current(ip, 1:ndim, 1:st%d%nspin)*hm%ep%vpsl(ip)
    end do
 
 
    do ik = st%d%kpt%start, st%d%kpt%end
      ispin = st%d%get_spin_index(ik)
      do ist = st%st_start, st%st_end
 
        if (abs(st%kweights(ik)*st%occ(ist, ik)) <= m_epsilon) cycle
 
        call states_elec_get_state(st, der%mesh, ist, ik, psi)
        do idim = 1, st%d%dim
          call boundaries_set(der%boundaries, der%mesh, psi(:, idim))
        end do
 
        if (hm%phase%is_allocated()) then
          call hm%phase%apply_to_single(psi, der%mesh%np_part, st%d%dim, ik, conjugate = .false.)
        end if
 
        do idim = 1, st%d%dim
          call zderivatives_grad(der, psi(:, idim), gpsi(:, :, idim), set_bc = .false.)
        end do
        do idir = 1, ndim
          if (hm%phase%is_allocated()) then
            call hm%phase%apply_to_single(gpsi(:, idir, :), der%mesh%np, st%d%dim, ik, conjugate = .true.)
          end if
 
          do idim = 1, st%d%dim
            call boundaries_set(der%boundaries, der%mesh, gpsi(:,idir, idim))
          end do
 
          if (hm%phase%is_allocated()) then
            call hm%phase%apply_to_single(gpsi(:, idir, :), der%mesh%np_part, st%d%dim, ik, conjugate = .false.)
          end if
 
          do idim = 1, st%d%dim
            call zderivatives_grad(der, gpsi(:, idir, idim), g2psi(:, :, idir, idim), set_bc = .false.)
          end do
        end do
        idim = 1
        do ip = 1, der%mesh%np
          do idir = 1, ndim
            tmp = sum(conjg(g2psi(ip, 1:ndim, idir, idim))*gpsi(ip, 1:ndim, idim)) - &
              sum(conjg(gpsi(ip, 1:ndim, idim))*g2psi(ip, 1:ndim, idir, idim))
            tmp = tmp - conjg(gpsi(ip, idir, idim))*sum(g2psi(ip, 1:ndim, 1:ndim, idim)) + &
              sum(conjg(g2psi(ip, 1:ndim, 1:ndim, idim)))*gpsi(ip, idir, idim)
            current(ip, idir, ispin) = current(ip, idir, ispin) + st%kweights(ik)*st%occ(ist, ik)*aimag(tmp)/8.0_real64
          end do
        end do
      end do
    end do
 
 
    pop_sub(current_heat_calculate)
 
  end subroutine current_heat_calculate
 
end module current_oct_m
 
!! Local Variables:
!! mode: f90
!! coding: utf-8
!! End: