Problem Description

Both the Sweep and Pencil Method Integrators experienced improper AMR child-grid nesting in r394. These errors typically occurred near domain decomposition boundaries. This issue may be general to the code or could simply be constrained to 3-D simulations. These errors haven't been fully worked out of the code for general use as of r408.

Intermediate Troubleshooting / Code Improvements

We had made a variety of changes to bearez.f90 and InfoFieldUtils.f90 in an attempt to eliminate the improper nesting. However, while these changes have drastically reduced the occurance, there are still rare occasions of improper nesting. We believe that one of the underlying reasons for the problem stems from refinement starting from the root grid down.

Stop-Gap Solution

Instead of continuing to rework the AMR algorithm near domain decompositions, which would take some time and serious testing to ensure the quality of the results, Jonathan divined another answer. iBuffer in bear.data should be set to 2 generally, but if we increase that number for the coarser refinement levels (increasing the buffer zone necessary to refine) then this should prevent boundaries from being unrefined on one side and highly refined on the other. Additionally, field variable tolerances should be set to lower values for the coarser levels to make them more sensitive to gradient changes and thus more likely to refine near domain decompositions. These steps have thus far prevented any further improper nesting issues.