As reported by @ac2290 on GitHub and the mailing list, PR #372 (included in mumax3.11.1) set B_demag to zero outside the geometry (where Msat==0), preventing users from obtaining stray fields.

Fix

The (inverse) zero-masking by Msat applied in the previous fix was moved from SetDemagField to SetLLTorque. This...

• ... reverts the behaviour of the B_demag variable back to its pre-3.11.1 behaviour (see validation 2), i.e., it is once again nonzero outside the geometry.
• ... still prevents demag from affecting the m vector where Msat==0 (see validation 3). Indeed, recall from Fix #357: Msat=0 affects exchange #372 that cells with Msat==0 should not experience a torque, since this makes relax() converge incredibly slowly.

Validation

1. As expected, this fix does not change the results of test/msatzero.mx3.

2. The behavior of the script in the aforementioned mailing list issue returns to how it was in mumax3.11.0 and earlier. Indeed, the top-down view of B_demag000050.ovf generated by this mailing list script in different mumax versions is as follows:

   ≤3.11.0	3.11.1	After this PR

3. The script of the original issue (Minor) change in exchange nonmagnetic boundary behavior from mumax3.9.3 to Mumax 3.10 #357 (where the right half of the system has Msat=0) yields the following relaxed m in different mumax versions:

   3.9.1	3.10 & 3.11.0	3.11.1	After this PR