InterfaceΒΆ
The primitive recovery is provided in form of a function object
con2prim_mhd
which stores all fixed parameters such as EOS, required accuracy, and
error policies. Once created, it can be used repeatetly as a function
to recover primitives. The evolved variables are passed as a
structure cons_vars_mhd and the
recovered primitives are stored in a structure
prim_vars_mhd.
The recovery also requires the metric, which is passed in another object. The framework provides objects representing tensors (not tensor fields), used here to represent co- and contravariant 3-vectors and the 3-metric. We use the convention that the 3-metric has positive signature.
If the evolved variables are invalid, the code checks whether the given error policy allows to apply corrections to the conserved variables to make them valid. If yes, the structure with the evolved varibles is modified to be consistent with the corrected primitives. The conserved variables and metric are checked to ensure all values are finite (not NAN or INF) and that the metric determinent is positive.
Whether or not the evolved variables are valid after allowed corrections
is reported via another object
c2p_mhd_report.
If the evolved variables violate the error policy, this also contains
the type of violation. In addition, it reports if atmosphere
was enforced, and if the evolved variables were corrected.
The object also provides a method to assemble a string with a
diagnostic message.
If the recovery fails for any reason the conserved and primitive variables are all set to NAN. Otherwise, they are consistent up to the root solving accuracy. The quantities \(\rho,W, v^i, D\) are always consistent to machiene precision.
A minimal example can be found in example/minimal.cc.
Note
Since the function operates pointwise and does not need to know anything about numerical grids, it is usable in any evolution code based on (or able to link to) C++.
Note
It is save to use the same function object in multiple threads. Copying the function object inside an innermost multithreaded loop should be avoided for performance reasons, since copying the EOS is an atomic operation.