We use the framework discussed in paper I as the baseline for the line transfer problems discussed in this paper. In addition to the highly efficient parallelization of solid angle space, we have implemented a parallelization over wavelength space using the MPI distributed memory model. For static configurations (or for configurations with velocity fields treated in the Eulerian frame) there is no direct coupling between different wavelength points.
Our basic setup is similar to that discussed in paper I. We use a sphere with
a grey continuum opacity parameterized by a power law in the continuum optical
depth
. The basic model parameters are
The line of the simple 2-level model atom is parameterized by the ratio of the
profile averaged line opacity to the continuum opacity
and
the line thermalization parameter
. For the test cases presented
below, we have used
and a constant temperature and thus a
constant thermal part of the source function for simplicity (and to save
computing time) and set
to simulate a strong line, with
varying
(see below). With this setup, the optical depths as seen
in the line range from
to
. We use 32 wavelength points to model
the full line profile, including wavelengths outside the line for the
continuum. We did not require the line to thermalize at the center of the test
configurations, this is a typical situation one encounters in a full 3D
configurations as the location (or even existence) of the
thermalization depths becomes more
ambiguous than in the 1D case.
The sphere is put at the center of the Cartesian grid, which is
in each axis 10% larger than the radius of the sphere.
For the test calculations we use voxel grids with the same
number of spatial points in each direction (see below). The
solid angle space was discretized in with
if not stated otherwise. In the following
we discuss the results of various tests. In all tests we use
the LC method for the 3D RT solution. Unless otherwise stated,
the tests were run on parallel computers using 128 CPUs. For the 3D solver
we use
or
points along each axis, for
a total of
or
spatial points, depending on the test case. The
solid angle space discretization uses
points.