nilsdeppe / spectre Goto Github PK

Implement the TCI on the DG grid

• reconstruct_work (or reconstruct_prims_work)
• Direction, DirectionMap
• Functions used solely by DG or FD elements

Currently we have many double persson_tci_epsilon which is set to the same value(=1e-18) around many spots in the code, so maybe cleanup these and make it as an option specifiable one?

The TCI-ing criteria of current implementation uses the following lines:

if (l2Norm(*filtered_component) <= zero_cutoff * l2Norm(component)) {
      continue;
}

if (std::log10(l2Norm(*filtered_component) / l2Norm(component)) >
    -alpha * std::log10(static_cast<double>(dg_mesh.extents(d) - 1))) {
  return true;
}

which may be simplified to its equivalent form without using log:

l2_norm_filtered * N ** alpha > l2_norm_component

where N = dg.mesh.extent(d)-1.

This way may have some computational speed-ups (however, up to the cost of pow function) as well as that we do not need the zero_cutoff (persson_epsilon) and the first if branch using it.