optproperties.cc

Go to the documentation of this file.

/* Copyright (C) 2003-2012 Claudia Emde <claudia.emde@dlr.de>
 
   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., 59 Temple Place - Suite 330, Boston, MA 02111-1307,
   USA. */
 
/*===========================================================================
  === File description
  ===========================================================================*/
 
/*===========================================================================
  === External declarations
  ===========================================================================*/
 
#include "optproperties.h"
#include <cfloat>
#include <cmath>
#include <stdexcept>
#include "array.h"
#include "arts.h"
#include "check_input.h"
#include "interpolation.h"
#include "logic.h"
#include "math_funcs.h"
#include "matpackVII.h"
#include "messages.h"
#include "xml_io.h"
 
extern const Numeric DEG2RAD;
extern const Numeric RAD2DEG;
extern const Numeric PI;
 
#define F11 pha_mat_int[0]
#define F12 pha_mat_int[1]
#define F22 pha_mat_int[2]
#define F33 pha_mat_int[3]
#define F34 pha_mat_int[4]
#define F44 pha_mat_int[5]
 
 
/*
void methodname(//Output
                type& name,
                //Input
                const type& name)
{
}
*/
 
 
void opt_prop_Bulk(    //Output
    Tensor5& ext_mat,  // (nf,nT,ndir,nst,nst)
    Tensor4& abs_vec,  // (nf,nT,ndir,nst)
    Index& ptype,
    //Input
    const ArrayOfTensor5& ext_mat_ss,  // [nss](nf,nT,ndir,nst,nst)
    const ArrayOfTensor4& abs_vec_ss,  // [nss](nf,nT,ndir,nst)
    const ArrayOfIndex& ptypes_ss) {
  assert(ext_mat_ss.nelem() == abs_vec_ss.nelem());
 
  ext_mat = ext_mat_ss[0];
  abs_vec = abs_vec_ss[0];
 
  for (Index i_ss = 1; i_ss < ext_mat_ss.nelem(); i_ss++) {
    ext_mat += ext_mat_ss[i_ss];
    abs_vec += abs_vec_ss[i_ss];
  }
  ptype = max(ptypes_ss);
}
 
 
void opt_prop_ScatSpecBulk(   //Output
    ArrayOfTensor5& ext_mat,  // [nss](nf,nT,ndir,nst,nst)
    ArrayOfTensor4& abs_vec,  // [nss](nf,nT,ndir,nst)
    ArrayOfIndex& ptype,
    //Input
    const ArrayOfArrayOfTensor5& ext_mat_se,  // [nss][nse](nf,nT,ndir,nst,nst)
    const ArrayOfArrayOfTensor4& abs_vec_se,  // [nss][nse](nf,nT,ndir,nst)
    const ArrayOfArrayOfIndex& ptypes_se,
    ConstMatrixView pnds,
    ConstMatrixView t_ok) {
  assert(t_ok.nrows() == pnds.nrows());
  assert(t_ok.ncols() == pnds.ncols());
  assert(TotalNumberOfElements(ext_mat_se) == pnds.nrows());
  assert(TotalNumberOfElements(abs_vec_se) == pnds.nrows());
  assert(ext_mat_se.nelem() == abs_vec_se.nelem());
 
  const Index nT = pnds.ncols();
  const Index nf = abs_vec_se[0][0].nbooks();
  const Index nDir = abs_vec_se[0][0].nrows();
  const Index stokes_dim = abs_vec_se[0][0].ncols();
 
  const Index nss = ext_mat_se.nelem();
  ext_mat.resize(nss);
  abs_vec.resize(nss);
  ptype.resize(nss);
  Tensor4 ext_tmp;
  Tensor3 abs_tmp;
 
  Index i_se_flat = 0;
 
  for (Index i_ss = 0; i_ss < nss; i_ss++) {
    assert(ext_mat_se[i_ss].nelem() == abs_vec_se[i_ss].nelem());
    assert(nT == ext_mat_se[i_ss][0].nbooks());
    assert(nT == abs_vec_se[i_ss][0].npages());
 
    ext_mat[i_ss].resize(nf, nT, nDir, stokes_dim, stokes_dim);
    ext_mat[i_ss] = 0.;
    abs_vec[i_ss].resize(nf, nT, nDir, stokes_dim);
    abs_vec[i_ss] = 0.;
 
    for (Index i_se = 0; i_se < ext_mat_se[i_ss].nelem(); i_se++) {
      assert(nT == ext_mat_se[i_ss][i_se].nbooks());
      assert(nT == abs_vec_se[i_ss][i_se].npages());
 
      for (Index Tind = 0; Tind < nT; Tind++) {
        if (pnds(i_se_flat, Tind) != 0.) {
          if (t_ok(i_se_flat, Tind) > 0.) {
            ext_tmp = ext_mat_se[i_ss][i_se](joker, Tind, joker, joker, joker);
            ext_tmp *= pnds(i_se_flat, Tind);
            ext_mat[i_ss](joker, Tind, joker, joker, joker) += ext_tmp;
 
            abs_tmp = abs_vec_se[i_ss][i_se](joker, Tind, joker, joker);
            abs_tmp *= pnds(i_se_flat, Tind);
            abs_vec[i_ss](joker, Tind, joker, joker) += abs_tmp;
          } else {
            ostringstream os;
            os << "Interpolation error for (flat-array) scattering element #"
               << i_se_flat << "\n"
               << "at location/temperature point #" << Tind << "\n";
            throw runtime_error(os.str());
          }
        }
      }
      i_se_flat++;
    }
    ptype[i_ss] = max(ptypes_se[i_ss]);
  }
}
 
 
void opt_prop_NScatElems(            //Output
    ArrayOfArrayOfTensor5& ext_mat,  // [nss][nse](nf,nT,ndir,nst,nst)
    ArrayOfArrayOfTensor4& abs_vec,  // [nss][nse](nf,nT,ndir,nst)
    ArrayOfArrayOfIndex& ptypes,
    Matrix& t_ok,
    //Input
    const ArrayOfArrayOfSingleScatteringData& scat_data,
    const Index& stokes_dim,
    const Vector& T_array,
    const Matrix& dir_array,
    const Index& f_index,
    const Index& t_interp_order) {
  Index f_start, nf;
  if (f_index < 0) {
    nf = scat_data[0][0].ext_mat_data.nshelves();
    f_start = 0;
    //f_end = f_start+nf;
  } else {
    nf = 1;
    if (scat_data[0][0].ext_mat_data.nshelves() == 1)
      f_start = 0;
    else
      f_start = f_index;
    //f_end = f_start+nf;
  }
 
  const Index nT = T_array.nelem();
  const Index nDir = dir_array.nrows();
 
  const Index nss = scat_data.nelem();
  ext_mat.resize(nss);
  abs_vec.resize(nss);
  ptypes.resize(nss);
 
  const Index Nse_all = TotalNumberOfElements(scat_data);
  t_ok.resize(Nse_all, nT);
  Index i_se_flat = 0;
 
  for (Index i_ss = 0; i_ss < nss; i_ss++) {
    Index nse = scat_data[i_ss].nelem();
    ext_mat[i_ss].resize(nse);
    abs_vec[i_ss].resize(nse);
    ptypes[i_ss].resize(nse);
 
    for (Index i_se = 0; i_se < nse; i_se++) {
      ext_mat[i_ss][i_se].resize(nf, nT, nDir, stokes_dim, stokes_dim);
      abs_vec[i_ss][i_se].resize(nf, nT, nDir, stokes_dim);
 
      opt_prop_1ScatElem(ext_mat[i_ss][i_se],
                         abs_vec[i_ss][i_se],
                         ptypes[i_ss][i_se],
                         t_ok(i_se_flat, joker),
                         scat_data[i_ss][i_se],
                         T_array,
                         dir_array,
                         f_start,
                         t_interp_order);
      i_se_flat++;
    }
  }
  assert(i_se_flat == Nse_all);
}
 
 
void opt_prop_1ScatElem(  //Output
    Tensor5View ext_mat,  // nf, nT, ndir, nst, nst
    Tensor4View abs_vec,  // nf, nT, ndir, nst
    Index& ptype,
    VectorView t_ok,
    //Input
    const SingleScatteringData& ssd,
    const Vector& T_array,
    const Matrix& dir_array,
    const Index& f_start,
    const Index& t_interp_order) {
  // FIXME: this is prob best done in scat_data_checkedCalc (or
  // cloudbox_checkedCalc) to have it done once and for all. Here at max assert.
 
  // At very first check validity of the scatt elements ptype (so far we only
  // handle PTYPE_TOTAL_RND and PTYPE_AZIMUTH_RND).
  /*
  if( ssd.ptype != PTYPE_TOTAL_RND and ssd.ptype != PTYPE_AZIMUTH_RND )
  {
    ostringstream os;
    os << "Only ptypes " << PTYPE_TOTAL_RND << " and " << PTYPE_AZIMUTH_RND
       << " can be handled.\n"
       << "Encountered scattering element with ptype " << ssd.ptype
       << ", though.";
    throw runtime_error( os.str() );
  }
  */
 
  assert(ssd.ptype == PTYPE_TOTAL_RND or ssd.ptype == PTYPE_AZIMUTH_RND);
 
  const Index nf = ext_mat.nshelves();
  assert(abs_vec.nbooks() == nf);
  if (nf > 1) {
    assert(nf == ssd.f_grid.nelem());
  }
 
  const Index nTout = T_array.nelem();
  assert(ext_mat.nbooks() == nTout);
  assert(abs_vec.npages() == nTout);
  assert(t_ok.nelem() == nTout);
 
  const Index nDir = dir_array.nrows();
  assert(ext_mat.npages() == nDir);
  assert(abs_vec.nrows() == nDir);
 
  const Index stokes_dim = abs_vec.ncols();
  assert(ext_mat.nrows() == stokes_dim);
  assert(ext_mat.ncols() == stokes_dim);
 
  ptype = ssd.ptype;
 
  // Determine T-interpol order as well as interpol positions and weights (they
  // are the same for all directions (and freqs), ie it is sufficient to
  // calculate them once).
  const Index nTin = ssd.T_grid.nelem();
  Index this_T_interp_order;
  Matrix T_itw;
  ArrayOfGridPosPoly T_gp(nTout);
  ssd_tinterp_parameters(t_ok,
                         this_T_interp_order,
                         T_gp,
                         T_itw,
                         ssd.T_grid,
                         T_array,
                         t_interp_order);
 
  // Now loop over requested directions (and apply simultaneously for all freqs):
  // 1) extract/interpolate direction (but not for tot.random)
  // 2) apply T-interpol
  if (ptype == PTYPE_TOTAL_RND)
    if (this_T_interp_order < 0)  // just extract (and unpack) ext and abs data
                                  // for the fs, the Tin, and the dirin and sort
                                  // (copy) into the output fs, Ts, and dirs.
    {
      Tensor3 ext_mat_tmp(nf, stokes_dim, stokes_dim);
      Matrix abs_vec_tmp(nf, stokes_dim);
      for (Index find = 0; find < nf; find++) {
        ext_mat_SSD2Stokes(ext_mat_tmp(find, joker, joker),
                           ssd.ext_mat_data(find + f_start, 0, 0, 0, joker),
                           stokes_dim,
                           ptype);
        abs_vec_SSD2Stokes(abs_vec_tmp(find, joker),
                           ssd.abs_vec_data(find + f_start, 0, 0, 0, joker),
                           stokes_dim,
                           ptype);
      }
      for (Index Tind = 0; Tind < nTout; Tind++)
        for (Index dind = 0; dind < nDir; dind++) {
          ext_mat(joker, Tind, dind, joker, joker) = ext_mat_tmp;
          abs_vec(joker, Tind, dind, joker) = abs_vec_tmp;
        }
    } else  // T-interpolation required (but not dir). To be done on the compact
            // ssd format.
    {
      Tensor4 ext_mat_tmp(nf, nTout, stokes_dim, stokes_dim);
      Tensor3 abs_vec_tmp(nf, nTout, stokes_dim);
      Matrix ext_mat_tmp_ssd(nTout, ssd.ext_mat_data.ncols());
      Matrix abs_vec_tmp_ssd(nTout, ssd.abs_vec_data.ncols());
      for (Index find = 0; find < nf; find++) {
        for (Index nst = 0; nst < ext_mat_tmp_ssd.ncols(); nst++)
          interp(ext_mat_tmp_ssd(joker, nst),
                 T_itw,
                 ssd.ext_mat_data(find + f_start, joker, 0, 0, nst),
                 T_gp);
        for (Index Tind = 0; Tind < nTout; Tind++)
          if (t_ok[Tind] > 0.)
            ext_mat_SSD2Stokes(ext_mat_tmp(find, Tind, joker, joker),
                               ext_mat_tmp_ssd(Tind, joker),
                               stokes_dim,
                               ptype);
          else
            ext_mat_tmp(find, Tind, joker, joker) = 0.;
 
        for (Index nst = 0; nst < abs_vec_tmp_ssd.ncols(); nst++)
          interp(abs_vec_tmp_ssd(joker, nst),
                 T_itw,
                 ssd.abs_vec_data(find + f_start, joker, 0, 0, nst),
                 T_gp);
        for (Index Tind = 0; Tind < nTout; Tind++)
          if (t_ok[Tind] > 0.)
            abs_vec_SSD2Stokes(abs_vec_tmp(find, Tind, joker),
                               abs_vec_tmp_ssd(Tind, joker),
                               stokes_dim,
                               ptype);
          else
            abs_vec_tmp(find, Tind, joker) = 0.;
      }
 
      for (Index dind = 0; dind < nDir; dind++) {
        ext_mat(joker, joker, dind, joker, joker) = ext_mat_tmp;
        abs_vec(joker, joker, dind, joker) = abs_vec_tmp;
      }
    }
 
  else  // dir-interpolation for non-tot-random particles
  {
    // as we don't allow other than az.random here, ext and abs will only vary
    // with za, not with aa. Hence, we could be smart here and calc data only
    // for unique za (and copy the rest). however, this smartness might cost as
    // well. so for now, we leave that and blindly loop over the given direction
    // array, and leave smartness in setting up directional array to the calling
    // methods.
 
    // derive the direction interpolation weights.
    ArrayOfGridPos dir_gp(nDir);
    gridpos(dir_gp, ssd.za_grid, dir_array(joker, 0));
    Matrix dir_itw(nDir, 2);  // only interpolating in za, ie 1D linear interpol
    interpweights(dir_itw, dir_gp);
 
    const Index next = ssd.ext_mat_data.ncols();
    const Index nabs = ssd.abs_vec_data.ncols();
 
    if (this_T_interp_order < 0)  // T only needs to be extracted.
    {
      Matrix ext_mat_tmp_ssd(nDir, next);
      Matrix abs_vec_tmp_ssd(nDir, nabs);
      Tensor4 ext_mat_tmp(nf, nDir, stokes_dim, stokes_dim);
      Tensor3 abs_vec_tmp(nf, nDir, stokes_dim);
      for (Index find = 0; find < nf; find++) {
        for (Index nst = 0; nst < next; nst++)
          interp(ext_mat_tmp_ssd(joker, nst),
                 dir_itw,
                 ssd.ext_mat_data(find + f_start, 0, joker, 0, nst),
                 dir_gp);
        for (Index Dind = 0; Dind < nDir; Dind++)
          ext_mat_SSD2Stokes(ext_mat_tmp(find, Dind, joker, joker),
                             ext_mat_tmp_ssd(Dind, joker),
                             stokes_dim,
                             ptype);
 
        for (Index nst = 0; nst < nabs; nst++)
          interp(abs_vec_tmp_ssd(joker, nst),
                 dir_itw,
                 ssd.abs_vec_data(find + f_start, 0, joker, 0, nst),
                 dir_gp);
        for (Index Dind = 0; Dind < nDir; Dind++)
          abs_vec_SSD2Stokes(abs_vec_tmp(find, Dind, joker),
                             abs_vec_tmp_ssd(Dind, joker),
                             stokes_dim,
                             ptype);
      }
 
      for (Index Tind = 0; Tind < nTout; Tind++) {
        ext_mat(joker, Tind, joker, joker, joker) = ext_mat_tmp;
        abs_vec(joker, Tind, joker, joker) = abs_vec_tmp;
      }
    } else  // T- and dir-interpolation required. To be done on the compact ssd
            // format.
    {
      Tensor3 ext_mat_tmp_ssd(nTin, nDir, next);
      Tensor3 abs_vec_tmp_ssd(nTin, nDir, nabs);
      Matrix ext_mat_tmp(nTout, next);
      Matrix abs_vec_tmp(nTout, nabs);
 
      for (Index find = 0; find < nf; find++) {
        for (Index Tind = 0; Tind < nTin; Tind++) {
          for (Index nst = 0; nst < next; nst++)
            interp(ext_mat_tmp_ssd(Tind, joker, nst),
                   dir_itw,
                   ssd.ext_mat_data(find + f_start, Tind, joker, 0, nst),
                   dir_gp);
          for (Index nst = 0; nst < nabs; nst++)
            interp(abs_vec_tmp_ssd(Tind, joker, nst),
                   dir_itw,
                   ssd.abs_vec_data(find + f_start, Tind, joker, 0, nst),
                   dir_gp);
        }
 
        for (Index Dind = 0; Dind < nDir; Dind++) {
          for (Index nst = 0; nst < next; nst++)
            interp(ext_mat_tmp(joker, nst),
                   T_itw,
                   ext_mat_tmp_ssd(joker, Dind, nst),
                   T_gp);
          for (Index Tind = 0; Tind < nTout; Tind++)
            ext_mat_SSD2Stokes(ext_mat(find, Tind, Dind, joker, joker),
                               ext_mat_tmp(Tind, joker),
                               stokes_dim,
                               ptype);
 
          for (Index nst = 0; nst < nabs; nst++)
            interp(abs_vec_tmp(joker, nst),
                   T_itw,
                   abs_vec_tmp_ssd(joker, Dind, nst),
                   T_gp);
          for (Index Tind = 0; Tind < nTout; Tind++)
            abs_vec_SSD2Stokes(abs_vec(find, Tind, Dind, joker),
                               abs_vec_tmp(Tind, joker),
                               stokes_dim,
                               ptype);
        }
      }
    }
  }
}
 
 
void ext_mat_SSD2Stokes(  //Output
    MatrixView ext_mat_stokes,
    //Input
    ConstVectorView ext_mat_ssd,
    const Index& stokes_dim,
    const Index& ptype) {
  // for now, no handling of PTYPE_GENERAL. should be ensured somewhere in the
  // calling methods, though.
  assert(ptype <= PTYPE_AZIMUTH_RND);
 
  ext_mat_stokes = 0.;
 
  for (Index ist = 0; ist < stokes_dim; ist++) {
    ext_mat_stokes(ist, ist) = ext_mat_ssd[0];
  }
 
  if (ptype > PTYPE_TOTAL_RND) {
    switch (stokes_dim) {
      case 4: {
        ext_mat_stokes(2, 3) = ext_mat_ssd[2];
        ext_mat_stokes(3, 2) = -ext_mat_ssd[2];
      } /* FALLTHROUGH */
      case 3: {
        // nothing to be done here. but we need this for executing the below
        // also in case of stokes_dim==3.
      }
      case 2: {
        ext_mat_stokes(0, 1) = ext_mat_ssd[1];
        ext_mat_stokes(1, 0) = ext_mat_ssd[1];
      }
    }
  }
}
 
 
void abs_vec_SSD2Stokes(  //Output
    VectorView abs_vec_stokes,
    //Input
    ConstVectorView abs_vec_ssd,
    const Index& stokes_dim,
    const Index& ptype) {
  // for now, no handling of PTYPE_GENERAL. should be ensured somewhere in the
  // calling methods, though.
  assert(ptype <= PTYPE_AZIMUTH_RND);
 
  abs_vec_stokes = 0.;
 
  abs_vec_stokes[0] = abs_vec_ssd[0];
 
  if (ptype > PTYPE_TOTAL_RND and stokes_dim > 1) {
    abs_vec_stokes[1] = abs_vec_ssd[1];
  }
}
 
 
void pha_mat_Bulk(     //Output
    Tensor6& pha_mat,  // (nf,nT,npdir,nidir,nst,nst)
    Index& ptype,
    //Input
    const ArrayOfTensor6& pha_mat_ss,  // [nss](nf,nT,npdir,nidir,nst,nst)
    const ArrayOfIndex& ptypes_ss) {
  pha_mat = pha_mat_ss[0];
 
  for (Index i_ss = 1; i_ss < pha_mat_ss.nelem(); i_ss++)
    pha_mat += pha_mat_ss[i_ss];
 
  ptype = max(ptypes_ss);
}
 
 
void pha_mat_ScatSpecBulk(    //Output
    ArrayOfTensor6& pha_mat,  // [nss](nf,nT,npdir,nidir,nst,nst)
    ArrayOfIndex& ptype,
    //Input
    const ArrayOfArrayOfTensor6&
        pha_mat_se,  // [nss][nse](nf,nT,npdir,nidir,nst,nst)
    const ArrayOfArrayOfIndex& ptypes_se,
    ConstMatrixView pnds,
    ConstMatrixView t_ok) {
  assert(t_ok.nrows() == pnds.nrows());
  assert(t_ok.ncols() == pnds.ncols());
  assert(TotalNumberOfElements(pha_mat_se) == pnds.nrows());
 
  const Index nT = pnds.ncols();
  const Index nf = pha_mat_se[0][0].nvitrines();
  const Index npDir = pha_mat_se[0][0].nbooks();
  const Index niDir = pha_mat_se[0][0].npages();
  const Index stokes_dim = pha_mat_se[0][0].ncols();
 
  const Index nss = pha_mat_se.nelem();
  pha_mat.resize(nss);
  ptype.resize(nss);
  Tensor5 pha_tmp;
 
  Index i_se_flat = 0;
 
  for (Index i_ss = 0; i_ss < nss; i_ss++) {
    assert(nT == pha_mat_se[i_ss][0].nshelves());
 
    pha_mat[i_ss].resize(nf, nT, npDir, niDir, stokes_dim, stokes_dim);
    pha_mat[i_ss] = 0.;
 
    for (Index i_se = 0; i_se < pha_mat_se[i_ss].nelem(); i_se++) {
      assert(nT == pha_mat_se[i_ss][i_se].nshelves());
 
      for (Index Tind = 0; Tind < nT; Tind++) {
        if (pnds(i_se_flat, Tind) != 0.) {
          if (t_ok(i_se_flat, Tind) > 0.) {
            pha_tmp =
                pha_mat_se[i_ss][i_se](joker, Tind, joker, joker, joker, joker);
            pha_tmp *= pnds(i_se_flat, Tind);
            pha_mat[i_ss](joker, Tind, joker, joker, joker, joker) += pha_tmp;
          } else {
            ostringstream os;
            os << "Interpolation error for (flat-array) scattering element #"
               << i_se_flat << "\n"
               << "at location/temperature point #" << Tind << "\n";
            throw runtime_error(os.str());
          }
        }
      }
      i_se_flat++;
    }
    ptype[i_ss] = max(ptypes_se[i_ss]);
  }
}
 
 
void pha_mat_NScatElems(             //Output
    ArrayOfArrayOfTensor6& pha_mat,  // [nss][nse](nf,nT,npdir,nidir,nst,nst)
    ArrayOfArrayOfIndex& ptypes,
    Matrix& t_ok,
    //Input
    const ArrayOfArrayOfSingleScatteringData& scat_data,
    const Index& stokes_dim,
    const Vector& T_array,
    const Matrix& pdir_array,
    const Matrix& idir_array,
    const Index& f_index,
    const Index& t_interp_order) {
  Index f_start, nf;
  if (f_index < 0) {
    nf = scat_data[0][0].pha_mat_data.nlibraries();
    f_start = 0;
    //f_end = f_start+nf;
  } else {
    nf = 1;
    if (scat_data[0][0].pha_mat_data.nlibraries() == 1)
      f_start = 0;
    else
      f_start = f_index;
    //f_end = f_start+nf;
  }
 
  const Index nT = T_array.nelem();
  const Index npDir = pdir_array.nrows();
  const Index niDir = idir_array.nrows();
 
  const Index nss = scat_data.nelem();
  pha_mat.resize(nss);
  ptypes.resize(nss);
 
  const Index Nse_all = TotalNumberOfElements(scat_data);
  t_ok.resize(Nse_all, nT);
  Index i_se_flat = 0;
 
  for (Index i_ss = 0; i_ss < nss; i_ss++) {
    Index nse = scat_data[i_ss].nelem();
    pha_mat[i_ss].resize(nse);
    ptypes[i_ss].resize(nse);
 
    for (Index i_se = 0; i_se < nse; i_se++) {
      pha_mat[i_ss][i_se].resize(nf, nT, npDir, niDir, stokes_dim, stokes_dim);
 
      pha_mat_1ScatElem(pha_mat[i_ss][i_se],
                        ptypes[i_ss][i_se],
                        t_ok(i_se_flat, joker),
                        scat_data[i_ss][i_se],
                        T_array,
                        pdir_array,
                        idir_array,
                        f_start,
                        t_interp_order);
      i_se_flat++;
    }
  }
  assert(i_se_flat == Nse_all);
}
 
 
void pha_mat_1ScatElem(   //Output
    Tensor6View pha_mat,  // nf, nT, npdir, nidir, nst, nst
    Index& ptype,
    VectorView t_ok,
    //Input
    const SingleScatteringData& ssd,
    const Vector& T_array,
    const Matrix& pdir_array,
    const Matrix& idir_array,
    const Index& f_start,
    const Index& t_interp_order) {
  assert(ssd.ptype == PTYPE_TOTAL_RND or ssd.ptype == PTYPE_AZIMUTH_RND);
 
  const Index nf = pha_mat.nvitrines();
  if (nf > 1) {
    assert(nf == ssd.f_grid.nelem());
  }
 
  const Index nTout = T_array.nelem();
  assert(pha_mat.nshelves() == nTout);
  assert(t_ok.nelem() == nTout);
 
  const Index npDir = pdir_array.nrows();
  assert(pha_mat.nbooks() == npDir);
 
  const Index niDir = idir_array.nrows();
  assert(pha_mat.npages() == niDir);
 
  const Index stokes_dim = pha_mat.ncols();
  assert(pha_mat.nrows() == stokes_dim);
 
  ptype = ssd.ptype;
 
  // Determine T-interpol order as well as interpol positions and weights (they
  // are the same for all directions (and freqs), ie it is sufficient to
  // calculate them once).
  const Index nTin = ssd.T_grid.nelem();
  Index this_T_interp_order;
  Matrix T_itw;
  ArrayOfGridPosPoly T_gp(nTout);
  ssd_tinterp_parameters(t_ok,
                         this_T_interp_order,
                         T_gp,
                         T_itw,
                         ssd.T_grid,
                         T_array,
                         t_interp_order);
 
  // Now loop over requested directions (and apply simultaneously for all freqs):
  // 1) interpolate direction
  // 2) apply T-interpol
  if (ptype == PTYPE_TOTAL_RND) {
    // determine how many of the compact stokes elements we will need.
    // restrict interpolations to those.
    Index npha;
    if (stokes_dim == 1)
      npha = 1;
    else if (stokes_dim < 4)  // stokes_dim==2 || stokes_dim==3
      npha = 4;
    else
      npha = 6;
    if (this_T_interp_order < 0)  // just extract (and unpack) pha data for the
    // fs and Tin, and interpolate in sca-angs, and
    // sort (copy) into the output fs, Ts, and dirs.
    {
      for (Index pdir = 0; pdir < npDir; pdir++)
        for (Index idir = 0; idir < niDir; idir++) {
          // calc scat ang theta from incident and prop dirs
          Numeric theta = scat_angle(pdir_array(pdir, 0),
                                     pdir_array(pdir, 1),
                                     idir_array(idir, 0),
                                     idir_array(idir, 1));
 
          // get scat angle interpolation weights
          GridPos dir_gp;
          gridpos(dir_gp, ssd.za_grid, theta * RAD2DEG);
          Vector dir_itw(2);
          interpweights(dir_itw, dir_gp);
 
          Vector pha_mat_int(npha, 0.);
          Matrix pha_mat_tmp(stokes_dim, stokes_dim);
          for (Index find = 0; find < nf; find++) {
            // perform the scat angle interpolation
            for (Index nst = 0; nst < npha; nst++)
              pha_mat_int[nst] = interp(
                  dir_itw,
                  ssd.pha_mat_data(find + f_start, 0, joker, 0, 0, 0, nst),
                  dir_gp);
 
            // convert from scat to lab frame
            pha_mat_labCalc(pha_mat_tmp(joker, joker),
                            pha_mat_int,
                            pdir_array(pdir, 0),
                            pdir_array(pdir, 1),
                            idir_array(idir, 0),
                            idir_array(idir, 1),
                            theta);
 
            for (Index Tind = 0; Tind < nTout; Tind++)
              pha_mat(find, Tind, pdir, idir, joker, joker) = pha_mat_tmp;
          }
        }
    } else  // T-interpolation required. To be done on the compact ssd format.
    {
      for (Index pdir = 0; pdir < npDir; pdir++)
        for (Index idir = 0; idir < niDir; idir++) {
          // calc scat ang theta from incident and prop dirs
          Numeric theta = scat_angle(pdir_array(pdir, 0),
                                     pdir_array(pdir, 1),
                                     idir_array(idir, 0),
                                     idir_array(idir, 1));
 
          // get scat angle interpolation weights
          GridPos dir_gp;
          gridpos(dir_gp, ssd.za_grid, theta * RAD2DEG);
          Vector dir_itw(2);
          interpweights(dir_itw, dir_gp);
 
          Matrix pha_mat_int(nTin, npha, 0.);
          Matrix pha_mat_tmp(nTout, npha, 0.);
          for (Index find = 0; find < nf; find++) {
            for (Index Tind = 0; Tind < nTin; Tind++)
              // perform the scat angle interpolation
              for (Index nst = 0; nst < npha; nst++) {
                pha_mat_int(Tind, nst) = interp(
                    dir_itw,
                    ssd.pha_mat_data(find + f_start, Tind, joker, 0, 0, 0, nst),
                    dir_gp);
              }
            // perform the T-interpolation
            for (Index nst = 0; nst < npha; nst++) {
              interp(pha_mat_tmp(joker, nst),
                     T_itw,
                     pha_mat_int(joker, nst),
                     T_gp);
            }
            // FIXME: it's probably better to do the frame conversion first,
            // then the T-interpolation (which is better depends on how many T-
            // aka vertical grid points we have. for single point, T-interpol
            // first is better, for a full vertical grid, frame conversion first
            // should be faster.)
            for (Index Tind = 0; Tind < nTout; Tind++) {
              // convert from scat to lab frame
              pha_mat_labCalc(pha_mat(find, Tind, pdir, idir, joker, joker),
                              pha_mat_tmp(Tind, joker),
                              pdir_array(pdir, 0),
                              pdir_array(pdir, 1),
                              idir_array(idir, 0),
                              idir_array(idir, 1),
                              theta);
            }
          }
        }
    }
  } else  // dir-interpolation for non-tot-random particles
          // Data is already stored in the laboratory frame,
          // but it is compressed a little. Details elsewhere.
  {
    Index nDir = npDir * niDir;
    Vector adelta_aa(nDir);
    Matrix delta_aa(npDir, niDir);
    ArrayOfGridPos daa_gp(nDir), pza_gp(nDir), iza_gp(nDir);
    ArrayOfGridPos pza_gp_tmp(npDir), iza_gp_tmp(niDir);
 
    gridpos(pza_gp_tmp, ssd.za_grid, pdir_array(joker, 0));
    gridpos(iza_gp_tmp, ssd.za_grid, idir_array(joker, 0));
 
    Index j = 0;
    for (Index pdir = 0; pdir < npDir; pdir++) {
      for (Index idir = 0; idir < niDir; idir++) {
        delta_aa(pdir, idir) =
            pdir_array(pdir, 1) - idir_array(idir, 1) +
            (pdir_array(pdir, 1) - idir_array(idir, 1) < -180) * 360 -
            (pdir_array(pdir, 1) - idir_array(idir, 1) > 180) * 360;
        adelta_aa[j] = abs(delta_aa(pdir, idir));
        pza_gp[j] = pza_gp_tmp[pdir];
        iza_gp[j] = iza_gp_tmp[idir];
        j++;
      }
    }
 
    gridpos(daa_gp, ssd.aa_grid, adelta_aa);
 
    Matrix dir_itw(nDir, 8);
    interpweights(dir_itw, pza_gp, daa_gp, iza_gp);
 
    if (this_T_interp_order < 0)  // T only needs to be extracted.
    {
      Tensor3 pha_mat_int(nDir, stokes_dim, stokes_dim, 0.);
      Tensor4 pha_mat_tmp(npDir, niDir, stokes_dim, stokes_dim);
 
      for (Index find = 0; find < nf; find++) {
        // perform the (tri-linear) angle interpolation. but only for the
        // pha_mat elements that we actually need.
 
        for (Index ist1 = 0; ist1 < stokes_dim; ist1++)
          for (Index ist2 = 0; ist2 < stokes_dim; ist2++)
            interp(
                pha_mat_int(joker, ist1, ist2),
                dir_itw,
                ssd.pha_mat_data(
                    find + f_start, 0, joker, joker, joker, 0, ist1 * 4 + ist2),
                pza_gp,
                daa_gp,
                iza_gp);
 
        // sort direction-combined 1D-array back into prop and incident
        // direction matrix
        Index i = 0;
        for (Index pdir = 0; pdir < npDir; pdir++)
          for (Index idir = 0; idir < niDir; idir++) {
            pha_mat_tmp(pdir, idir, joker, joker) =
                pha_mat_int(i, joker, joker);
            i++;
          }
 
        if (stokes_dim > 2) {
          for (Index pdir = 0; pdir < npDir; pdir++)
            for (Index idir = 0; idir < niDir; idir++)
              if (delta_aa(pdir, idir) < 0.) {
                pha_mat_tmp(pdir, idir, 0, 2) *= -1;
                pha_mat_tmp(pdir, idir, 1, 2) *= -1;
                pha_mat_tmp(pdir, idir, 2, 0) *= -1;
                pha_mat_tmp(pdir, idir, 2, 1) *= -1;
              }
        }
 
        if (stokes_dim > 3) {
          for (Index pdir = 0; pdir < npDir; pdir++)
            for (Index idir = 0; idir < niDir; idir++)
              if (delta_aa(pdir, idir) < 0.) {
                pha_mat_tmp(pdir, idir, 0, 3) *= -1;
                pha_mat_tmp(pdir, idir, 1, 3) *= -1;
                pha_mat_tmp(pdir, idir, 3, 0) *= -1;
                pha_mat_tmp(pdir, idir, 3, 1) *= -1;
              }
        }
 
        for (Index Tind = 0; Tind < nTout; Tind++)
          pha_mat(find, Tind, joker, joker, joker, joker) = pha_mat_tmp;
      }
    }
 
    else  // T- and dir-interpolation required. To be done on the compact ssd
          // format.
    {
      Tensor4 pha_mat_int(nTin, nDir, stokes_dim, stokes_dim, 0.);
 
      for (Index find = 0; find < nf; find++) {
        // perform the (tri-linear) angle interpolation. but only for the
        // pha_mat elements that we actually need.
        for (Index Tind = 0; Tind < nTin; Tind++) {
          for (Index ist1 = 0; ist1 < stokes_dim; ist1++)
            for (Index ist2 = 0; ist2 < stokes_dim; ist2++)
              interp(pha_mat_int(Tind, joker, ist1, ist2),
                     dir_itw,
                     ssd.pha_mat_data(find + f_start,
                                      Tind,
                                      joker,
                                      joker,
                                      joker,
                                      0,
                                      ist1 * 4 + ist2),
                     pza_gp,
                     daa_gp,
                     iza_gp);
        }
 
        // perform the T-interpolation and simultaneously sort
        // direction-combined 1D-array back into prop and incident direction
        // matrix.
        Index i = 0;
        for (Index pdir = 0; pdir < npDir; pdir++)
          for (Index idir = 0; idir < niDir; idir++) {
            for (Index ist1 = 0; ist1 < stokes_dim; ist1++)
              for (Index ist2 = 0; ist2 < stokes_dim; ist2++)
                interp(pha_mat(find, joker, pdir, idir, ist1, ist2),
                       T_itw,
                       pha_mat_int(joker, i, ist1, ist2),
                       T_gp);
            i++;
          }
 
        if (stokes_dim > 2) {
          for (Index pdir = 0; pdir < npDir; pdir++)
            for (Index idir = 0; idir < niDir; idir++)
              if (delta_aa(pdir, idir) < 0.) {
                pha_mat(find, joker, pdir, idir, 0, 2) *= -1;
                pha_mat(find, joker, pdir, idir, 1, 2) *= -1;
                pha_mat(find, joker, pdir, idir, 2, 0) *= -1;
                pha_mat(find, joker, pdir, idir, 2, 1) *= -1;
              }
        }
 
        if (stokes_dim > 3) {
          for (Index pdir = 0; pdir < npDir; pdir++)
            for (Index idir = 0; idir < niDir; idir++)
              if (delta_aa(pdir, idir) < 0.) {
                pha_mat(find, joker, pdir, idir, 0, 3) *= -1;
                pha_mat(find, joker, pdir, idir, 1, 3) *= -1;
                pha_mat(find, joker, pdir, idir, 3, 0) *= -1;
                pha_mat(find, joker, pdir, idir, 3, 1) *= -1;
              }
        }
      }
    }
  }
}
 
 
void FouComp_1ScatElem(       //Output
    Tensor7View pha_mat_fou,  // nf, nT, npdir, nidir, nst, nst, m
    Index& ptype,
    VectorView t_ok,
    //Input
    const SingleScatteringData& ssd,
    const Vector& T_array,
    const Vector& pdir_array,
    const Vector& idir_array,
    const Index& f_start,
    const Index& t_interp_order,
    const Index& naa_totran) {
  assert(ssd.ptype == PTYPE_TOTAL_RND or ssd.ptype == PTYPE_AZIMUTH_RND);
 
  const Index nf = pha_mat_fou.nlibraries();
  if (nf > 1) {
    assert(nf == ssd.f_grid.nelem());
  }
 
  const Index nTout = T_array.nelem();
  assert(pha_mat_fou.nvitrines() == nTout);
  assert(t_ok.nelem() == nTout);
 
  const Index npDir = pdir_array.nelem();
  assert(pha_mat_fou.nshelves() == npDir);
  const Index niDir = idir_array.nelem();
  assert(pha_mat_fou.nbooks() == niDir);
 
  const Index stokes_dim = pha_mat_fou.nrows();
  assert(pha_mat_fou.npages() == stokes_dim);
  // currently code is only prepared for stokes_dim up to 2 (nothing else needed in
  // RT4 and generally in azimuth-symmetrical system)
  assert(stokes_dim < 3);
 
  const Index nmodes = pha_mat_fou.ncols();
  // currently code is only prepared for fourier mode 0 (nothing else needed in
  // RT4 and generally in azimuth-symmetrical system)
  assert(nmodes == 0);
 
  ptype = ssd.ptype;
 
  // Determine T-interpol order as well as interpol positions and weights (they
  // are the same for all directions (and freqs), ie it is sufficient to
  // calculate them once).
  const Index nTin = ssd.T_grid.nelem();
  Index this_T_interp_order;
  Matrix T_itw;
  ArrayOfGridPosPoly T_gp(nTout);
  ssd_tinterp_parameters(t_ok,
                         this_T_interp_order,
                         T_gp,
                         T_itw,
                         ssd.T_grid,
                         T_array,
                         t_interp_order);
 
  // 1) derive Fourier component(s)
  // 2) apply T-interpol
  if (ptype == PTYPE_TOTAL_RND) {
    // DCalculate azimuth angles and their integration weights for Fourier
    // component derivation (they are only determined by naa_totran).
    if (naa_totran < 3) {
      ostringstream os;
      os << "Azimuth grid size for scatt matrix extraction"
         << " (*naa_totran*) must be >3.\n"
         << "Yours is " << naa_totran << ".\n";
      throw runtime_error(os.str());
    }
    Vector aa_grid;
    nlinspace(aa_grid, 0, 180, naa_totran);
    Numeric daa_totran =
        1. / float(naa_totran - 1);  // 2*180./360./(naa_totran-1)
    Vector theta(naa_totran);
    ArrayOfGridPos theta_gp;
    Matrix theta_itw(naa_totran, 2);
 
    Index npha;
    if (stokes_dim == 1)
      npha = 1;
    else if (stokes_dim < 4)  // stokes_dim==2 || stokes_dim==3
      npha = 4;
    else
      npha = 6;
 
    Matrix pha_mat(stokes_dim, stokes_dim);
    Matrix pha_mat_angint(naa_totran, npha);
    Tensor3 Fou_int(nTin, stokes_dim, stokes_dim);
 
    for (Index idir = 0; idir < niDir; idir++)
      for (Index pdir = 0; pdir < npDir; pdir++) {
        for (Index iaa = 0; iaa < naa_totran; iaa++)
          // calc scat ang theta from incident and prop dirs
          theta[iaa] =
              scat_angle(pdir_array[pdir], aa_grid[iaa], idir_array[idir], 0.);
 
        // get scat angle interpolation weights
        gridpos(theta_gp, ssd.za_grid, theta * RAD2DEG);
        interpweights(theta_itw, theta_gp);
 
        for (Index find = 0; find < nf; find++) {
          Fou_int = 0.;
          for (Index Tind = 0; Tind < nTin; Tind++) {
            // perform the scat angle interpolation
            for (Index nst = 0; nst < npha; nst++)
              interp(
                  pha_mat_angint(joker, nst),
                  theta_itw,
                  ssd.pha_mat_data(find + f_start, Tind, joker, 0, 0, 0, nst),
                  theta_gp);
            for (Index iaa = 0; iaa < naa_totran; iaa++) {
              // convert from scat to lab frame
              pha_mat_labCalc(pha_mat,
                              pha_mat_angint(iaa, joker),
                              pdir_array[pdir],
                              aa_grid[iaa],
                              idir_array[idir],
                              0.,
                              theta[iaa]);
              // and sum up/integrate
              // FIXME: can the integration probably be done in lab frame?
              // test!
              if (iaa == 0 || iaa == naa_totran - 1)
                pha_mat *= (daa_totran / 2.);
              else
                pha_mat *= daa_totran;
              Fou_int(Tind, joker, joker) += pha_mat;
            }
          }
 
          if (this_T_interp_order <
              0)  // T only needs to be sorted into pha_mat_fou.
          {
            for (Index Tind = 0; Tind < nTout; Tind++)
              pha_mat_fou(find, Tind, pdir, idir, joker, joker, 0) =
                  Fou_int(0, joker, joker);
          } else {
            for (Index ist1 = 0; ist1 < stokes_dim; ist1++)
              for (Index ist2 = 0; ist2 < stokes_dim; ist2++)
                for (Index im = 0; im < nmodes; im++)
                  interp(pha_mat_fou(find, joker, pdir, idir, ist1, ist2, im),
                         T_itw,
                         Fou_int(joker, ist1, ist2),
                         T_gp);
          }
        }
      }
  } else  // derive Fourier component(s) on scattering elements own za_grid (using
          // given aa_grid), then 2D-interpolate inc and sca polar directions.
  // Do calcs only for required pha_mat components depending on stokes_dim.
  {
    Index nza = ssd.za_grid.nelem();
    Index naa = ssd.aa_grid.nelem();
    ConstVectorView za_datagrid = ssd.za_grid;
    ConstVectorView aa_datagrid = ssd.aa_grid;
    assert(aa_datagrid[0] == 0.);
    assert(aa_datagrid[naa - 1] == 180.);
    Vector daa(naa);
 
    // Precalculate azimuth integration weights for this azimuthally randomly
    // oriented scat element (need to do this per scat element as ssd.aa_grid is
    // scat element specific (might change between elements) and need to do this
    // on actual grid instead of grid number since the grid can, at least
    // theoretically be non-equidistant).
    daa[0] = (aa_datagrid[1] - aa_datagrid[0]) / 360.;
    for (Index iaa = 1; iaa < naa - 1; iaa++)
      daa[iaa] = (aa_datagrid[iaa + 1] - aa_datagrid[iaa - 1]) / 360.;
    daa[naa - 1] = (aa_datagrid[naa - 1] - aa_datagrid[naa - 2]) / 360.;
 
    // Precalculate polar angle interpolation grid positions and weights
    ArrayOfGridPos pdir_za_gp, idir_za_gp;
    gridpos(pdir_za_gp, za_datagrid, pdir_array);
    gridpos(idir_za_gp, za_datagrid, idir_array);
    Tensor3 dir_itw(npDir, niDir, 4);
    interpweights(dir_itw, pdir_za_gp, idir_za_gp);
 
    Tensor4 Fou_ssd(nza, nza, stokes_dim, stokes_dim);
    Tensor5 Fou_angint(nTin, npDir, niDir, stokes_dim, stokes_dim);
 
    for (Index find = 0; find < nf; find++) {
      for (Index Tind = 0; Tind < nTin; Tind++) {
        // first, extract the phase matrix at the scatt elements own polar angle
        // grid and integrate over azimuth deriving their respective azimuthal
        // (Fourier series) 0-mode
        Fou_ssd = 0.;
        for (Index iza = 0; iza < nza; iza++)
          for (Index sza = 0; sza < nza; sza++) {
            for (Index iaa = 0; iaa < naa; iaa++) {
              // FIXME: are there any possible shortcuts for specific stokes
              // components (specifically, some where F0(ist1,ist2)==0?)
              for (Index ist1 = 0; ist1 < stokes_dim; ist1++)
                for (Index ist2 = 0; ist2 < stokes_dim; ist2++)
                  Fou_ssd(sza, iza, ist1, ist2) +=
                      daa[iaa] * ssd.pha_mat_data(find + f_start,
                                                  Tind,
                                                  sza,
                                                  iaa,
                                                  iza,
                                                  0,
                                                  ist1 * 4 + ist2);
            }
          }
 
        // second, interpolate the extracted azimuthal mode to the stream directions
        for (Index ist1 = 0; ist1 < stokes_dim; ist1++)
          for (Index ist2 = 0; ist2 < stokes_dim; ist2++)
            // FIXME: do we need to apply any sign changes?
            interp(Fou_angint(Tind, joker, joker, ist1, ist2),
                   dir_itw,
                   Fou_ssd(joker, joker, ist1, ist2),
                   pdir_za_gp,
                   idir_za_gp);
      }
 
      if (this_T_interp_order <
          0)  // T only needs to be sorted into pha_mat_fou.
      {
        for (Index Tind = 0; Tind < nTout; Tind++)
          pha_mat_fou(find, Tind, joker, joker, joker, joker, 0) =
              Fou_angint(0, joker, joker, joker, joker);
      } else {
        for (Index pdir = 0; pdir < npDir; pdir++)
          for (Index idir = 0; idir < niDir; idir++)
            for (Index ist1 = 0; ist1 < stokes_dim; ist1++)
              for (Index ist2 = 0; ist2 < stokes_dim; ist2++)
                for (Index im = 0; im < nmodes; im++)
                  interp(pha_mat_fou(find, joker, pdir, idir, ist1, ist2, im),
                         T_itw,
                         Fou_angint(joker, pdir, idir, ist1, ist2),
                         T_gp);
      }
    }
  }
}
 
 
void abs_vecTransform(  //Output and Input
    StokesVector& abs_vec_lab,
    //Input
    ConstTensor3View abs_vec_data,
    ConstVectorView za_datagrid,
    ConstVectorView aa_datagrid _U_,
    const PType& ptype,
    const Numeric& za_sca _U_,
    const Numeric& aa_sca _U_,
    const Verbosity& verbosity) {
  const Index stokes_dim = abs_vec_lab.StokesDimensions();
  assert(abs_vec_lab.NumberOfFrequencies() == 1);
 
  if (stokes_dim > 4 || stokes_dim < 1) {
    throw runtime_error(
        "The dimension of the stokes vector \n"
        "must be 1,2,3 or 4");
  }
 
  switch (ptype) {
    case PTYPE_GENERAL: {
      /*
         TO ANY DEVELOPER:
         current usage of coordinate systems in scattering solvers (RT and SSD
         extraction) and general radiative transfer is not consistent. Not an
         issue as long as only PTYPE_TOTAL_RND and PTYPE_AZIMUTH_RND are used,
         but will be a problem for PTYPE_GENERAL, ie needs to be fixed BEFORE
         adding PTYPE_GENERAL support (see AUG appendix for more info).
      */
 
      CREATE_OUT0;
      out0 << "Case PTYPE_GENERAL not yet implemented. \n";
      break;
    }
    case PTYPE_TOTAL_RND: {
      // The first element of the vector corresponds to the absorption
      // coefficient which is stored in the database, the others are 0.
 
      abs_vec_lab.SetZero();
 
      abs_vec_lab.Kjj()[0] = abs_vec_data(0, 0, 0);
      break;
    }
 
    case PTYPE_AZIMUTH_RND:  //Added by Cory Davis 9/12/03
    {
      assert(abs_vec_data.ncols() == 2);
 
      // In the case of azimuthally randomly oriented particles, only the first
      // two elements of the absorption coefficient vector are non-zero.
      // These values are dependent on the zenith angle of propagation.
 
      // 1st interpolate data by za_sca
      GridPos gp;
      Vector itw(2);
 
      gridpos(gp, za_datagrid, za_sca);
      interpweights(itw, gp);
 
      abs_vec_lab.SetZero();
 
      abs_vec_lab.Kjj()[0] = interp(itw, abs_vec_data(Range(joker), 0, 0), gp);
 
      if (stokes_dim == 1) {
        break;
      }
      abs_vec_lab.K12()[0] = interp(itw, abs_vec_data(Range(joker), 0, 1), gp);
      break;
    }
    default: {
      CREATE_OUT0;
      out0 << "Not all ptype cases are implemented\n";
    }
  }
}
 
 
void ext_matTransform(  //Output and Input
    PropagationMatrix& ext_mat_lab,
    //Input
    ConstTensor3View ext_mat_data,
    ConstVectorView za_datagrid,
    ConstVectorView aa_datagrid _U_,
    const PType& ptype,
    const Numeric& za_sca,
    const Numeric& aa_sca _U_,
    const Verbosity& verbosity) {
  const Index stokes_dim = ext_mat_lab.StokesDimensions();
  assert(ext_mat_lab.NumberOfFrequencies() == 1);
 
  if (stokes_dim > 4 || stokes_dim < 1) {
    throw runtime_error(
        "The dimension of the stokes vector \n"
        "must be 1,2,3 or 4");
  }
 
  switch (ptype) {
    case PTYPE_GENERAL: {
      /*
         TO ANY DEVELOPER:
         current usage of coordinate systems in scattering solvers (RT and SSD
         extraction) and general radiative transfer is not consistent. Not an
         issue as long as only PTYPE_TOTAL_RND and PTYPE_AZIMUTH_RND are used,
         but will be a problem for PTYPE_GENERAL, ie needs to be fixed BEFORE
         adding PTYPE_GENERAL support (see AUG appendix for more info).
      */
 
      CREATE_OUT0;
      out0 << "Case PTYPE_GENERAL not yet implemented. \n";
      break;
    }
    case PTYPE_TOTAL_RND: {
      assert(ext_mat_data.ncols() == 1);
 
      // In the case of randomly oriented particles the extinction matrix is
      // diagonal. The value of each element of the diagonal is the
      // extinction cross section, which is stored in the database.
 
      ext_mat_lab.SetZero();
 
      ext_mat_lab.Kjj() = ext_mat_data(0, 0, 0);
      break;
    }
 
    case PTYPE_AZIMUTH_RND:  //Added by Cory Davis 9/12/03
    {
      assert(ext_mat_data.ncols() == 3);
 
      // In the case of azimuthally randomly oriented particles, the extinction
      // matrix has only 3 independent non-zero elements Kjj, K12=K21, and K34=-K43.
      // These values are dependent on the zenith angle of propagation.
 
      // 1st interpolate data by za_sca
      GridPos gp;
      Vector itw(2);
      Numeric Kjj;
      Numeric K12;
      Numeric K34;
 
      gridpos(gp, za_datagrid, za_sca);
      interpweights(itw, gp);
 
      ext_mat_lab.SetZero();
 
      Kjj = interp(itw, ext_mat_data(Range(joker), 0, 0), gp);
      ext_mat_lab.Kjj() = Kjj;
 
      if (stokes_dim < 2) {
        break;
      }
 
      K12 = interp(itw, ext_mat_data(Range(joker), 0, 1), gp);
      ext_mat_lab.K12()[0] = K12;
 
      if (stokes_dim < 4) {
        break;
      }
 
      K34 = interp(itw, ext_mat_data(Range(joker), 0, 2), gp);
      ext_mat_lab.K34()[0] = K34;
      break;
    }
    default: {
      CREATE_OUT0;
      out0 << "Not all ptype cases are implemented\n";
    }
  }
}
 
 
void pha_matTransform(  //Output
    MatrixView pha_mat_lab,
    //Input
    ConstTensor5View pha_mat_data,
    ConstVectorView za_datagrid,
    ConstVectorView aa_datagrid,
    const PType& ptype,
    const Index& za_sca_idx,
    const Index& aa_sca_idx,
    const Index& za_inc_idx,
    const Index& aa_inc_idx,
    ConstVectorView za_grid,
    ConstVectorView aa_grid,
    const Verbosity& verbosity) {
  const Index stokes_dim = pha_mat_lab.ncols();
 
  Numeric za_sca = za_grid[za_sca_idx];
  Numeric aa_sca = aa_grid[aa_sca_idx];
  Numeric za_inc = za_grid[za_inc_idx];
  Numeric aa_inc = aa_grid[aa_inc_idx];
 
  if (stokes_dim > 4 || stokes_dim < 1) {
    throw runtime_error(
        "The dimension of the stokes vector \n"
        "must be 1,2,3 or 4");
  }
 
  switch (ptype) {
    case PTYPE_GENERAL: {
      /*
         TO ANY DEVELOPER:
         current usage of coordinate systems in scattering solvers (RT and SSD
         extraction) and general radiative transfer is not consistent. Not an
         issue as long as only PTYPE_TOTAL_RND and PTYPE_AZIMUTH_RND are used,
         but will be a problem for PTYPE_GENERAL, ie needs to be fixed BEFORE
         adding PTYPE_GENERAL support (see AUG appendix for more info).
      */
 
      CREATE_OUT0;
      out0 << "Case PTYPE_GENERAL not yet implemented. \n";
      break;
    }
    case PTYPE_TOTAL_RND: {
      // Calculate the scattering angle and interpolate the data in it:
      Numeric theta_rad = scat_angle(za_sca, aa_sca, za_inc, aa_inc);
      const Numeric theta = RAD2DEG * theta_rad;
 
      // Interpolation of the data on the scattering angle:
      Vector pha_mat_int(6);
      interpolate_scat_angle(pha_mat_int, pha_mat_data, za_datagrid, theta);
 
      // Calculate the phase matrix in the laboratory frame:
      pha_mat_labCalc(
          pha_mat_lab, pha_mat_int, za_sca, aa_sca, za_inc, aa_inc, theta_rad);
 
      break;
    }
 
    case PTYPE_AZIMUTH_RND:  //Added by Cory Davis
                             //Data is already stored in the laboratory frame,
      //but it is compressed a little.  Details elsewhere.
      {
        assert(pha_mat_data.ncols() == 16);
        assert(pha_mat_data.npages() == za_datagrid.nelem());
        Numeric delta_aa = aa_sca - aa_inc + (aa_sca - aa_inc < -180) * 360 -
                           (aa_sca - aa_inc > 180) *
                               360;  //delta_aa corresponds to the "books"
                                     //dimension of pha_mat_data
        GridPos za_sca_gp;
        GridPos delta_aa_gp;
        GridPos za_inc_gp;
        Vector itw(8);
 
        gridpos(delta_aa_gp, aa_datagrid, abs(delta_aa));
        gridpos(za_inc_gp, za_datagrid, za_inc);
        gridpos(za_sca_gp, za_datagrid, za_sca);
 
        interpweights(itw, za_sca_gp, delta_aa_gp, za_inc_gp);
 
        pha_mat_lab(0, 0) =
            interp(itw,
                   pha_mat_data(Range(joker), Range(joker), Range(joker), 0, 0),
                   za_sca_gp,
                   delta_aa_gp,
                   za_inc_gp);
        if (stokes_dim == 1) {
          break;
        }
        pha_mat_lab(0, 1) =
            interp(itw,
                   pha_mat_data(Range(joker), Range(joker), Range(joker), 0, 1),
                   za_sca_gp,
                   delta_aa_gp,
                   za_inc_gp);
        pha_mat_lab(1, 0) =
            interp(itw,
                   pha_mat_data(Range(joker), Range(joker), Range(joker), 0, 4),
                   za_sca_gp,
                   delta_aa_gp,
                   za_inc_gp);
        pha_mat_lab(1, 1) =
            interp(itw,
                   pha_mat_data(Range(joker), Range(joker), Range(joker), 0, 5),
                   za_sca_gp,
                   delta_aa_gp,
                   za_inc_gp);
        if (stokes_dim == 2) {
          break;
        }
        if (delta_aa >= 0) {
          pha_mat_lab(0, 2) = interp(
              itw,
              pha_mat_data(Range(joker), Range(joker), Range(joker), 0, 2),
              za_sca_gp,
              delta_aa_gp,
              za_inc_gp);
          pha_mat_lab(1, 2) = interp(
              itw,
              pha_mat_data(Range(joker), Range(joker), Range(joker), 0, 6),
              za_sca_gp,
              delta_aa_gp,
              za_inc_gp);
          pha_mat_lab(2, 0) = interp(
              itw,
              pha_mat_data(Range(joker), Range(joker), Range(joker), 0, 8),
              za_sca_gp,
              delta_aa_gp,
              za_inc_gp);
          pha_mat_lab(2, 1) = interp(
              itw,
              pha_mat_data(Range(joker), Range(joker), Range(joker), 0, 9),
              za_sca_gp,
              delta_aa_gp,
              za_inc_gp);
        } else {
          pha_mat_lab(0, 2) = -interp(
              itw,
              pha_mat_data(Range(joker), Range(joker), Range(joker), 0, 2),
              za_sca_gp,
              delta_aa_gp,
              za_inc_gp);
          pha_mat_lab(1, 2) = -interp(
              itw,
              pha_mat_data(Range(joker), Range(joker), Range(joker), 0, 6),
              za_sca_gp,
              delta_aa_gp,
              za_inc_gp);
          pha_mat_lab(2, 0) = -interp(
              itw,
              pha_mat_data(Range(joker), Range(joker), Range(joker), 0, 8),
              za_sca_gp,
              delta_aa_gp,
              za_inc_gp);
          pha_mat_lab(2, 1) = -interp(
              itw,
              pha_mat_data(Range(joker), Range(joker), Range(joker), 0, 9),
              za_sca_gp,
              delta_aa_gp,
              za_inc_gp);
        }
        pha_mat_lab(2, 2) = interp(
            itw,
            pha_mat_data(Range(joker), Range(joker), Range(joker), 0, 10),
            za_sca_gp,
            delta_aa_gp,
            za_inc_gp);
        if (stokes_dim == 3) {
          break;
        }
        if (delta_aa >= 0) {
          pha_mat_lab(0, 3) = interp(
              itw,
              pha_mat_data(Range(joker), Range(joker), Range(joker), 0, 3),
              za_sca_gp,
              delta_aa_gp,
              za_inc_gp);
          pha_mat_lab(1, 3) = interp(
              itw,
              pha_mat_data(Range(joker), Range(joker), Range(joker), 0, 7),
              za_sca_gp,
              delta_aa_gp,
              za_inc_gp);
          pha_mat_lab(3, 0) = interp(
              itw,
              pha_mat_data(Range(joker), Range(joker), Range(joker), 0, 12),
              za_sca_gp,
              delta_aa_gp,
              za_inc_gp);
          pha_mat_lab(3, 1) = interp(
              itw,
              pha_mat_data(Range(joker), Range(joker), Range(joker), 0, 13),
              za_sca_gp,
              delta_aa_gp,
              za_inc_gp);
        } else {
          pha_mat_lab(0, 3) = -interp(
              itw,
              pha_mat_data(Range(joker), Range(joker), Range(joker), 0, 3),
              za_sca_gp,
              delta_aa_gp,
              za_inc_gp);
          pha_mat_lab(1, 3) = -interp(
              itw,
              pha_mat_data(Range(joker), Range(joker), Range(joker), 0, 7),
              za_sca_gp,
              delta_aa_gp,
              za_inc_gp);
          pha_mat_lab(3, 0) = -interp(
              itw,
              pha_mat_data(Range(joker), Range(joker), Range(joker), 0, 12),
              za_sca_gp,
              delta_aa_gp,
              za_inc_gp);
          pha_mat_lab(3, 1) = -interp(
              itw,
              pha_mat_data(Range(joker), Range(joker), Range(joker), 0, 13),
              za_sca_gp,
              delta_aa_gp,
              za_inc_gp);
        }
        pha_mat_lab(2, 3) = interp(
            itw,
            pha_mat_data(Range(joker), Range(joker), Range(joker), 0, 11),
            za_sca_gp,
            delta_aa_gp,
            za_inc_gp);
        pha_mat_lab(3, 2) = interp(
            itw,
            pha_mat_data(Range(joker), Range(joker), Range(joker), 0, 14),
            za_sca_gp,
            delta_aa_gp,
            za_inc_gp);
        pha_mat_lab(3, 3) = interp(
            itw,
            pha_mat_data(Range(joker), Range(joker), Range(joker), 0, 15),
            za_sca_gp,
            delta_aa_gp,
            za_inc_gp);
        break;
      }
 
    default: {
      CREATE_OUT0;
      out0 << "Not all ptype cases are implemented\n";
    }
  }
}
 
 
void ext_matFromabs_vec(  //Output
    MatrixView ext_mat,
    //Input
    ConstVectorView abs_vec,
    const Index& stokes_dim) {
  assert(stokes_dim >= 1 && stokes_dim <= 4);
  assert(ext_mat.nrows() == stokes_dim);
  assert(ext_mat.ncols() == stokes_dim);
  assert(abs_vec.nelem() == stokes_dim);
 
  // first: diagonal elements
  for (Index is = 0; is < stokes_dim; is++) {
    ext_mat(is, is) += abs_vec[0];
  }
  // second: off-diagonal elements, namely first row and column
  for (Index is = 1; is < stokes_dim; is++) {
    ext_mat(0, is) += abs_vec[is];
    ext_mat(is, 0) += abs_vec[is];
  }
}
 
 
void ssd_tinterp_parameters(  //Output
    VectorView t_ok,
    Index& this_T_interp_order,
    ArrayOfGridPosPoly& T_gp,
    Matrix& T_itw,
    //Input
    ConstVectorView T_grid,
    const Vector& T_array,
    const Index& t_interp_order) {
  const Index nTin = T_grid.nelem();
  const Index nTout = T_array.nelem();
 
  this_T_interp_order = -1;
 
  if (nTin > 1) {
    this_T_interp_order = min(t_interp_order, nTin - 1);
    T_itw.resize(nTout, this_T_interp_order + 1);
 
    // we need to check T-grid exceedance. and catch these cases (because T
    // is assumed to correspond to a location and T-exceedance is ok when pnd=0
    // there. however, here we do not know pnd.) and report them to have the
    // calling method dealing with it.
 
    // we set the extrapolfax explicitly and use it here as well as in
    // gridpos_poly below.
    const Numeric extrapolfac = 0.5;
    const Numeric lowlim = T_grid[0] - extrapolfac * (T_grid[1] - T_grid[0]);
    const Numeric uplim =
        T_grid[nTin - 1] + extrapolfac * (T_grid[nTin - 1] - T_grid[nTin - 2]);
 
    bool any_T_exceed = false;
    for (Index Tind = 0; Tind < nTout; Tind++) {
      if (T_array[Tind] < lowlim || T_array[Tind] > uplim) {
        t_ok[Tind] = -1.;
        any_T_exceed = true;
      } else
        t_ok[Tind] = 1.;
    }
 
    if (any_T_exceed) {
      GridPosPoly dummy_gp;
      dummy_gp.idx.resize(this_T_interp_order + 1);
      dummy_gp.w.resize(this_T_interp_order + 1);
      for (Index i = 0; i <= this_T_interp_order; ++i) dummy_gp.idx[i] = i;
      dummy_gp.w = 0.;
      dummy_gp.w[0] = 1.;
      bool grid_unchecked = true;
 
      for (Index iT = 0; iT < nTout; iT++) {
        if (t_ok[iT] < 0)
          // setup T-exceed points with dummy grid positions
          T_gp[iT] = dummy_gp;
        else {
          if (grid_unchecked) {
            chk_interpolation_grids(
                "Temperature interpolation in pha_mat_1ScatElem",
                T_grid,
                T_array[Range(iT, 1)],
                this_T_interp_order);
            grid_unchecked = false;
          }
          gridpos_poly(
              T_gp[iT], T_grid, T_array[iT], this_T_interp_order, extrapolfac);
        }
      }
    } else
      gridpos_poly(T_gp, T_grid, T_array, this_T_interp_order, extrapolfac);
    interpweights(T_itw, T_gp);
  } else {
    t_ok = 1.;
  }
}
 
 
Numeric scat_angle(const Numeric& za_sca,
                   const Numeric& aa_sca,
                   const Numeric& za_inc,
                   const Numeric& aa_inc) {
  Numeric theta_rad;
  Numeric ANG_TOL = 1e-7;
 
  // CPD 5/10/03.
  // Two special cases are implemented here to avoid NaNs that can sometimes
  // occur in in the acos() formula in forward and backscattering cases.
  //
  // GH 2011-05-31
  // Consider not only aa_sca-aa_inc ~= 0, but also aa_sca-aa_inc ~= 360.
 
  if ((abs(aa_sca - aa_inc) < ANG_TOL) ||
      (abs(abs(aa_sca - aa_inc) - 360) < ANG_TOL)) {
    theta_rad = DEG2RAD * abs(za_sca - za_inc);
  } else if (abs(abs(aa_sca - aa_inc) - 180) < ANG_TOL) {
    theta_rad = DEG2RAD * (za_sca + za_inc);
    if (theta_rad > PI) {
      theta_rad = 2 * PI - theta_rad;
    }
  } else {
    const Numeric za_sca_rad = za_sca * DEG2RAD;
    const Numeric za_inc_rad = za_inc * DEG2RAD;
    const Numeric aa_sca_rad = aa_sca * DEG2RAD;
    const Numeric aa_inc_rad = aa_inc * DEG2RAD;
 
    theta_rad =
        acos(cos(za_sca_rad) * cos(za_inc_rad) +
             sin(za_sca_rad) * sin(za_inc_rad) * cos(aa_sca_rad - aa_inc_rad));
  }
  return theta_rad;
}
 
 
void interpolate_scat_angle(  //Output:
    VectorView pha_mat_int,
    //Input:
    ConstTensor5View pha_mat_data,
    ConstVectorView za_datagrid,
    const Numeric theta) {
  GridPos thet_gp;
  gridpos(thet_gp, za_datagrid, theta);
 
  Vector itw(2);
  interpweights(itw, thet_gp);
 
  assert(pha_mat_data.ncols() == 6);
  for (Index i = 0; i < 6; i++) {
    pha_mat_int[i] = interp(itw, pha_mat_data(joker, 0, 0, 0, i), thet_gp);
  }
}
 
 
void pha_mat_labCalc(  //Output:
    MatrixView pha_mat_lab,
    //Input:
    ConstVectorView pha_mat_int,
    const Numeric& za_sca,
    const Numeric& aa_sca,
    const Numeric& za_inc,
    const Numeric& aa_inc,
    const Numeric& theta_rad) {
  const Index stokes_dim = pha_mat_lab.ncols();
 
  if (std::isnan(F11)) {
    throw runtime_error(
        "NaN value(s) detected in *pha_mat_labCalc* (0,0). Could the "
        "input data contain NaNs? Please check with *scat_dataCheck*. If "
        "input data are OK and you critically need the ongoing calculations, "
        "try to change the observation LOS slightly. If you can reproduce "
        "this error, please contact Patrick in order to help tracking down "
        "the reason to this problem. If you see this message occasionally "
        "when doing MC calculations, it should not be critical. This path "
        "sampling will be rejected and replaced with a new one.");
  }
 
  // For stokes_dim = 1, we only need Z11=F11:
  pha_mat_lab(0, 0) = F11;
 
  if (stokes_dim > 1) {
    Numeric za_sca_rad = za_sca * DEG2RAD;
    Numeric za_inc_rad = za_inc * DEG2RAD;
    Numeric aa_sca_rad = aa_sca * DEG2RAD;
    Numeric aa_inc_rad = aa_inc * DEG2RAD;
 
    const Numeric ANGTOL_RAD = 1e-6;  //CPD: this constant is used to adjust
        //zenith angles close to 0 and PI.  This is
        //also used to avoid float == float statements.
 
    //
    // Several cases have to be considered:
    //
 
    if ((abs(theta_rad) < ANGTOL_RAD)          // forward scattering
        || (abs(theta_rad - PI) < ANGTOL_RAD)  // backward scattering
        ||
        (abs(aa_inc_rad - aa_sca_rad) < ANGTOL_RAD)  // inc and sca on meridian
        || (abs(abs(aa_inc_rad - aa_sca_rad) - 360.) < ANGTOL_RAD)  //   "
        || (abs(abs(aa_inc_rad - aa_sca_rad) - 180.) < ANGTOL_RAD)  //   "
    ) {
      pha_mat_lab(0, 1) = F12;
      pha_mat_lab(1, 0) = F12;
      pha_mat_lab(1, 1) = F22;
 
      if (stokes_dim > 2) {
        pha_mat_lab(0, 2) = 0;
        pha_mat_lab(1, 2) = 0;
        pha_mat_lab(2, 0) = 0;
        pha_mat_lab(2, 1) = 0;
        pha_mat_lab(2, 2) = F33;
 
        if (stokes_dim > 3) {
          pha_mat_lab(0, 3) = 0;
          pha_mat_lab(1, 3) = 0;
          pha_mat_lab(2, 3) = F34;
          pha_mat_lab(3, 0) = 0;
          pha_mat_lab(3, 1) = 0;
          pha_mat_lab(3, 2) = -F34;
          pha_mat_lab(3, 3) = F44;
        }
      }
    }
 
    else {
      Numeric sigma1;
      Numeric sigma2;
 
      Numeric s1, s2;
 
      // In these cases we have to take limiting values.
 
      if (za_inc_rad < ANGTOL_RAD) {
        sigma1 = PI + aa_sca_rad - aa_inc_rad;
        sigma2 = 0;
      } else if (za_inc_rad > PI - ANGTOL_RAD) {
        sigma1 = aa_sca_rad - aa_inc_rad;
        sigma2 = PI;
      } else if (za_sca_rad < ANGTOL_RAD) {
        sigma1 = 0;
        sigma2 = PI + aa_sca_rad - aa_inc_rad;
      } else if (za_sca_rad > PI - ANGTOL_RAD) {
        sigma1 = PI;
        sigma2 = aa_sca_rad - aa_inc_rad;
      } else {
        s1 = (cos(za_sca_rad) - cos(za_inc_rad) * cos(theta_rad)) /
             (sin(za_inc_rad) * sin(theta_rad));
        s2 = (cos(za_inc_rad) - cos(za_sca_rad) * cos(theta_rad)) /
             (sin(za_sca_rad) * sin(theta_rad));
 
        sigma1 = acos(s1);
        sigma2 = acos(s2);
 
        // Arccos is only defined in the range from -1 ... 1
        // Numerical problems can appear for values close to 1 or -1
        // this (also) catches the case when inc and sca are on one meridian
        if (std::isnan(sigma1) || std::isnan(sigma2)) {
          if (abs(s1 - 1) < ANGTOL_RAD) sigma1 = 0;
          if (abs(s1 + 1) < ANGTOL_RAD) sigma1 = PI;
          if (abs(s2 - 1) < ANGTOL_RAD) sigma2 = 0;
          if (abs(s2 + 1) < ANGTOL_RAD) sigma2 = PI;
        }
      }
 
      const Numeric C1 = cos(2 * sigma1);
      const Numeric C2 = cos(2 * sigma2);
 
      const Numeric S1 = sin(2 * sigma1);
      const Numeric S2 = sin(2 * sigma2);
 
      pha_mat_lab(0, 1) = C1 * F12;
      pha_mat_lab(1, 0) = C2 * F12;
      pha_mat_lab(1, 1) = C1 * C2 * F22 - S1 * S2 * F33;
 
      //assert(!std::isnan(pha_mat_lab(0,1)));
      //assert(!std::isnan(pha_mat_lab(1,0)));
      //assert(!std::isnan(pha_mat_lab(1,1)));
      if (std::isnan(pha_mat_lab(0, 1)) || std::isnan(pha_mat_lab(1, 0)) ||
          std::isnan(pha_mat_lab(1, 1))) {
        throw runtime_error(
            "NaN value(s) detected in *pha_mat_labCalc* (0/1,1). Could the "
            "input data contain NaNs? Please check with *scat_dataCheck*. If "
            "input data are OK  and you critically need the ongoing calculations, "
            "try to change the observation LOS slightly. If you can reproduce "
            "this error, please contact Patrick in order to help tracking down "
            "the reason to this problem. If you see this message occasionally "
            "when doing MC calculations, it should not be critical. This path "
            "sampling will be rejected and replaced with a new one.");
      }
 
      if (stokes_dim > 2) {
        /*CPD: For skokes_dim > 2 some of the transformation formula 
            for each element have a different sign depending on whether or
            not 0<aa_scat-aa_inc<180.  For details see pages 94 and 95 of 
            Mishchenkos chapter in : 
            Mishchenko, M. I., and L. D. Travis, 2003: Electromagnetic 
            scattering by nonspherical particles. In Exploring the Atmosphere 
            by Remote Sensing Techniques (R. Guzzi, Ed.), Springer-Verlag, 
            Berlin, pp. 77-127. 
            This is available at http://www.giss.nasa.gov/~crmim/publications/ */
        Numeric delta_aa = aa_sca - aa_inc + (aa_sca - aa_inc < -180) * 360 -
                           (aa_sca - aa_inc > 180) * 360;
        if (delta_aa >= 0) {
          pha_mat_lab(0, 2) = S1 * F12;
          pha_mat_lab(1, 2) = S1 * C2 * F22 + C1 * S2 * F33;
          pha_mat_lab(2, 0) = -S2 * F12;
          pha_mat_lab(2, 1) = -C1 * S2 * F22 - S1 * C2 * F33;
        } else {
          pha_mat_lab(0, 2) = -S1 * F12;
          pha_mat_lab(1, 2) = -S1 * C2 * F22 - C1 * S2 * F33;
          pha_mat_lab(2, 0) = S2 * F12;
          pha_mat_lab(2, 1) = C1 * S2 * F22 + S1 * C2 * F33;
        }
        pha_mat_lab(2, 2) = -S1 * S2 * F22 + C1 * C2 * F33;
 
        if (stokes_dim > 3) {
          if (delta_aa >= 0) {
            pha_mat_lab(1, 3) = S2 * F34;
            pha_mat_lab(3, 1) = S1 * F34;
          } else {
            pha_mat_lab(1, 3) = -S2 * F34;
            pha_mat_lab(3, 1) = -S1 * F34;
          }
          pha_mat_lab(0, 3) = 0;
          pha_mat_lab(2, 3) = C2 * F34;
          pha_mat_lab(3, 0) = 0;
          pha_mat_lab(3, 2) = -C1 * F34;
          pha_mat_lab(3, 3) = F44;
        }
      }
    }
  }
}
 
ostream& operator<<(ostream& os, const SingleScatteringData& /*ssd*/) {
  os << "SingleScatteringData: Output operator not implemented";
  return os;
}
 
ostream& operator<<(ostream& os, const ArrayOfSingleScatteringData& /*assd*/) {
  os << "ArrayOfSingleScatteringData: Output operator not implemented";
  return os;
}
 
ostream& operator<<(ostream& os, const ScatteringMetaData& /*ssd*/) {
  os << "ScatteringMetaData: Output operator not implemented";
  return os;
}
 
ostream& operator<<(ostream& os, const ArrayOfScatteringMetaData& /*assd*/) {
  os << "ArrayOfScatteringMetaData: Output operator not implemented";
  return os;
}
 
 
void opt_prop_sum_propmat_clearsky(  //Output:
    PropagationMatrix& ext_mat,
    StokesVector& abs_vec,
    //Input:
    const ArrayOfPropagationMatrix& propmat_clearsky) {
  Index stokes_dim, freq_dim;
 
  if (propmat_clearsky.nelem() > 1) {
    stokes_dim = propmat_clearsky[0].StokesDimensions();
    freq_dim = propmat_clearsky[0].NumberOfFrequencies();
  } else {
    stokes_dim = 0;
    freq_dim = 0;
  }
 
  // old abs_vecInit
  abs_vec = StokesVector(freq_dim, stokes_dim);
  abs_vec.SetZero();
 
  ext_mat = PropagationMatrix(freq_dim, stokes_dim);
  ext_mat.SetZero();
 
  // old ext_matAddGas and abs_vecAddGas for 0 vector and matrix
  for (auto& pm : propmat_clearsky) {
    abs_vec += pm;
    ext_mat += pm;
  }
}
 
 
PType PTypeFromString(const String& ptype_string) {
  PType ptype;
  if (ptype_string == "general")
    ptype = PTYPE_GENERAL;
  else if (ptype_string == "totally_random")
    ptype = PTYPE_TOTAL_RND;
  else if (ptype_string == "azimuthally_random")
    ptype = PTYPE_AZIMUTH_RND;
  else {
    ostringstream os;
    os << "Unknown ptype: " << ptype_string << endl
       << "Valid types are: general, totally_random and "
       << "azimuthally_random.";
    throw std::runtime_error(os.str());
  }
 
  return ptype;
}
 
 
PType PType2FromString(const String& ptype_string) {
  PType ptype;
  if (ptype_string == "general")
    ptype = PTYPE_GENERAL;
  else if (ptype_string == "macroscopically_isotropic")
    ptype = PTYPE_TOTAL_RND;
  else if (ptype_string == "horizontally_aligned")
    ptype = PTYPE_AZIMUTH_RND;
  else {
    ostringstream os;
    os << "Unknown ptype: " << ptype_string << endl
       << "Valid types are: general, macroscopically_isotropic and "
       << "horizontally_aligned.";
    throw std::runtime_error(os.str());
  }
 
  return ptype;
}
 
 
String PTypeToString(const PType& ptype) {
  String ptype_string;
 
  switch (ptype) {
    case PTYPE_GENERAL:
      ptype_string = "general";
      break;
    case PTYPE_TOTAL_RND:
      ptype_string = "totally_random";
      break;
    case PTYPE_AZIMUTH_RND:
      ptype_string = "azimuthally_random";
      break;
    default:
      ostringstream os;
      os << "Internal error: Cannot map PType enum value " << ptype
         << " to String.";
      throw std::runtime_error(os.str());
      break;
  }
 
  return ptype_string;
}
 
 
void ConvertAzimuthallyRandomSingleScatteringData(SingleScatteringData& ssd) {
  // First check that input fulfills requirements on older data formats:
  // 1) Is za_grid symmetric and includes 90deg?
  Index nza = ssd.za_grid.nelem();
  for (Index i = 0; i < nza / 2; i++) {
    if (!is_same_within_epsilon(
            180. - ssd.za_grid[nza - 1 - i], ssd.za_grid[i], 2 * DBL_EPSILON)) {
      ostringstream os;
      os << "Zenith grid of azimuthally_random single scattering data\n"
         << "is not symmetric with respect to 90degree.";
      throw std::runtime_error(os.str());
    }
  }
  if (!is_same_within_epsilon(ssd.za_grid[nza / 2], 90., 2 * DBL_EPSILON)) {
    ostringstream os;
    os << "Zenith grid of azimuthally_random single scattering data\n"
       << "does not contain 90 degree grid point.";
    throw std::runtime_error(os.str());
  }
 
  // 2) Are data sizes correct?
  ostringstream os_pha_mat;
  os_pha_mat << "pha_mat ";
  ostringstream os_ext_mat;
  os_ext_mat << "ext_mat ";
  ostringstream os_abs_vec;
  os_abs_vec << "abs_vec ";
  chk_size(os_pha_mat.str(),
           ssd.pha_mat_data,
           ssd.f_grid.nelem(),
           ssd.T_grid.nelem(),
           ssd.za_grid.nelem(),
           ssd.aa_grid.nelem(),
           ssd.za_grid.nelem() / 2 + 1,
           1,
           16);
 
  chk_size(os_ext_mat.str(),
           ssd.ext_mat_data,
           ssd.f_grid.nelem(),
           ssd.T_grid.nelem(),
           ssd.za_grid.nelem() / 2 + 1,
           1,
           3);
 
  chk_size(os_abs_vec.str(),
           ssd.abs_vec_data,
           ssd.f_grid.nelem(),
           ssd.T_grid.nelem(),
           ssd.za_grid.nelem() / 2 + 1,
           1,
           2);
 
  // Now that we are sure that za_grid is properly symmetric, we just need to
  // copy over the data (ie no interpolation).
  Tensor5 tmpT5 = ssd.abs_vec_data;
  ssd.abs_vec_data.resize(tmpT5.nshelves(),
                          tmpT5.nbooks(),
                          ssd.za_grid.nelem(),
                          tmpT5.nrows(),
                          tmpT5.ncols());
  ssd.abs_vec_data(joker, joker, Range(0, nza / 2 + 1), joker, joker) = tmpT5;
  for (Index i = 0; i < nza / 2; i++) {
    ssd.abs_vec_data(joker, joker, nza - 1 - i, joker, joker) =
        tmpT5(joker, joker, i, joker, joker);
  }
 
  tmpT5 = ssd.ext_mat_data;
  ssd.ext_mat_data.resize(tmpT5.nshelves(),
                          tmpT5.nbooks(),
                          ssd.za_grid.nelem(),
                          tmpT5.nrows(),
                          tmpT5.ncols());
  ssd.ext_mat_data(joker, joker, Range(0, nza / 2 + 1), joker, joker) = tmpT5;
  for (Index i = 0; i < nza / 2; i++) {
    ssd.ext_mat_data(joker, joker, nza - 1 - i, joker, joker) =
        tmpT5(joker, joker, i, joker, joker);
  }
 
  Tensor7 tmpT7 = ssd.pha_mat_data;
  ssd.pha_mat_data.resize(tmpT7.nlibraries(),
                          tmpT7.nvitrines(),
                          tmpT7.nshelves(),
                          tmpT7.nbooks(),
                          ssd.za_grid.nelem(),
                          tmpT7.nrows(),
                          tmpT7.ncols());
  ssd.pha_mat_data(
      joker, joker, joker, joker, Range(0, nza / 2 + 1), joker, joker) = tmpT7;
 
  // scatt. matrix elements 13,23,31,32 and 14,24,41,42 (=elements 2,6,8,9 and
  // 3,7,12,13 in ARTS' flattened format, respectively) change sign.
  tmpT7(joker, joker, joker, joker, joker, joker, Range(2, 2)) *= -1.;
  tmpT7(joker, joker, joker, joker, joker, joker, Range(6, 4)) *= -1.;
  tmpT7(joker, joker, joker, joker, joker, joker, Range(12, 2)) *= -1.;
 
  // For second half of incident polar angles (>90deg), we need to mirror the
  // original data in both incident and scattered polar angle around 90deg "planes".
  for (Index i = 0; i < nza / 2; i++)
    for (Index j = 0; j < nza; j++)
      ssd.pha_mat_data(
          joker, joker, nza - 1 - j, joker, nza - 1 - i, joker, joker) =
          tmpT7(joker, joker, j, joker, i, joker, joker);
}
 
 
ParticleSSDMethod ParticleSSDMethodFromString(
    const String& particle_ssdmethod_string) {
  ParticleSSDMethod particle_ssdmethod;
  if (particle_ssdmethod_string == "tmatrix")
    particle_ssdmethod = PARTICLE_SSDMETHOD_TMATRIX;
  else {
    ostringstream os;
    os << "Unknown particle SSD method: " << particle_ssdmethod_string << endl
       << "Valid methods: tmatrix";
    throw std::runtime_error(os.str());
  }
 
  return particle_ssdmethod;
}
 
 
String PTypeToString(const ParticleSSDMethod& particle_ssdmethod) {
  String particle_ssdmethod_string;
 
  switch (particle_ssdmethod) {
    case PARTICLE_SSDMETHOD_TMATRIX:
      particle_ssdmethod_string = "tmatrix";
      break;
    default:
      ostringstream os;
      os << "Internal error: Cannot map ParticleSSDMethod enum value "
         << particle_ssdmethod << " to String.";
      throw std::runtime_error(os.str());
      break;
  }
 
  return particle_ssdmethod_string;
}