SCAN correlation energy gives NaNs with unrestricted densities
Originally reported by Susi Lehtola:
The SCAN correlation energy evaluation gives NaNs? for spin-polarized calculations.
I've tracked the issue down to the PBE-style correlation part. The NaNs? appear on line 194 upon
f1 = 1.0 + 4.0*A*t2; f3 = POW(f1, -1.0/4.0);
It appears A can have very large negative values, which make f1 also very negative and f3 NaN.