---

Isolating blocks that cause seams

---

In Image1 an iso surface of density at 1e+09 is shown with the amr grid. Three fine to coarse block boundaries are circled. The blue circle is a case of where blocks that are at the second highest level of refinement in this data that have a neighbor to the left in X that is at one level lower. The blue circle is the site of a block boundary that causes a seam. The yellow circle is the same case, but no seam is produced. The red circle is a case of where blocks that are at the highest level of refinement in this data that have a neighbor to the left in X that is at one level lower. Since these are both problems involving copying data to a high level block from a lower level neighbor that is to the left in X, I searched through the dataset for both of these cases (for the highest level of refinement and the second highest level of refinement) and plotted the difference between sets of 4 higher level blocks and lower level neighbor to the left in X. Graphs of all the blocks in case 1 (second highest level of refinement) are here: http://flash.uchicago.edu/~jbgallag/seams" for case 2 (highest level) here: http://flash.uchicago.edu/~jbgallag/seams2 . Looking through these graphs for blocks whose values contain 1.0e+09 I was able to isolate each of the three circled cases and show that the seams correlate to larger differences in the data on a higher level block and that of a lower level neighbor to the left in X.




In each case there is a lower level block from which 4 higher level blocks will copy data. Since there was a small amount of graphs that fit the search criteria and an even smaller amount that had cell values that contained the iso surface value we're looking at I was able to try each case to explicity skip the loading of the lower level block so that it is not drawn. This causes an artifact at the expected blocks in each case. I used a flash produced corner data file, which I also hae a cell centered version of, when making images with lower level neighbor blocks removed.

CASE #1 (Blue Circle, site of a seam)



Each block number below copies data from the lower level block 7067 which isn't drawn in the above picture.

Block 7068	Block 7071
Block 7073	Block 7075

CASE #2 (Yellow Circle, no seam)



Each block number below copies data from the lower level block 5172 (red line) which isn't drawn in the above picture. Note that in this case block 6079 has a noticeable difference in the range of interest, but block 6077 does not. Apparently this discrepency is not big enough to cause a seam

Block 6072	Block 6075
Block 6077	Block 6079

CASE #2 (Red Circle, site of a seam)



Each block number below copies data from the lower level block 8190 (red line) which isn't drawn in the above picture.

Block 8239	Block 8241
Block 8243	Block 8245