search

fmihpc / vlasiator

Vlasiator - ten letters you can count on
https://www.helsinki.fi/en/researchgroups/vlasiator
Other
48 stars 39 forks source link

DPF unreliable? Unexplained differences in merging AMR2 #72

Open galfthan opened 9 years ago

galfthan commented 9 years ago

When merging the AMR2 branch there were surprisingly large differences between the master and branch versions when using DPF. When using SPF the differences were very small.

See discussion at end of https://github.com/fmihpc/vlasiator/pull/68

Old master: c872cfd merged master with AMR2: a57f9ea471a483265d63d0823e04d8047da1521f

galfthan commented 9 years ago

When merging removed_fx branch I looked again at this issue. The differences are surprisingly large between AAS run in november before this merge, and after.

Diff of AAS rho after 5 just seconds (SPF):

vlsvdiff_DP bulk.0000010.vlsv ../AAS_remake_pqm/bulk.0000010.vlsv rho 0
INFO Reading in two files.
Statistics on file: size 405000 min = 0 max = 1.03881e+06 average = 149783 standard deviation 287.834
Statistics on file: size 405000 min = 0 max = 1.03774e+06 average = 149783 standard deviation 287.835
The absolute 0-distance between both datasets is 3839.76
The relative 0-distance between both datasets is 0.0036963
The absolute 1-distance between both datasets is 4.81171e+06
The relative 1-distance between both datasets is 7.93199e-05
The absolute 2-distance between both datasets is 48543.7
The relative 2-distance between both datasets is 0.000322371

Of course, the diff does not say which of these are better or more correct, but it does point to some subtle change.