load balance of adaptive refinement in libmesh

[YHDAUT]

load balance of adaptive refinement in libmesh

[roystgnr]

We don't have anything built-in that does AMR without hanging nodes (or their equivalent with hierarchical p-refinement: edge functions constrained to be 0).

But regardless of whether you use built-in or 3rd party refinement, the load balancing gets done whenever the mesh is repartitioned. By default that gets done every time a changed mesh is reprepared for use, i.e. after every refinement step, but some user codes do it less frequently, on the theory that the calculation and communication involved is only worth it when the load imbalance becomes sufficiently large.