Theodore Papadopoulo

OK. I made a comparison with OpenMEEG (Head3 and a rather deep dipole on a 1 single layer model) and the Sarvas formula. There seems to be a localized error....

Well with Git it's easy. I may even have old binaries on my computer (but the problem is to find the exact version). Brainstorm binaries (if available on linux) is...

OK it is not my big changes of last year. The problem already appears for commit a7528400ff993 (22-08-2018). It's not OpenMP either. Next step, before the introduction of non-nested geometries...

OK. It is not so easy to recompile old versions... The compiler is more picky and some old packages are no longer there. I used the old OpenMEEG 2.1 release...

OK. OpenMEEG solution is dominated by the src2meg term. See the new curve in https://gist.github.com/papadop/a09b30d49805aae38075763845588381.

> So it sounds like FieldTrip, MNE-Python, and Sarvas all agree and OpenMEEG is the outlier here. If you can write the Sarvas values you computed to some format (.h5,...

The non-source term is at best 3 orders of magnitude lower than the source term.

Several things: - I had a small mistake in my Sarvas code. But this does not change the conclusions. - I think the error is in the computation of the...

Why not. Needs to be tested. Still computations need to be done in double (need to recheck but last time I checked computations in floats were slower !), so there...

OK, I will dig in the history to re-add these old tests.