Background subtraction in cov_tau leads to NaNs in regions with sign problem
In cov_tau there is currently the following line
OneBin(nt,no,no1) = OneBin(nt,no,no1) - &
& bins0(nb-n_skip,no)*bins0(nb-n_skip,no1)*cmplx(dble(Nunit),0.d0,kind(0.d0))&
& /Phase(nb-n_skip)
which makes it so that if any single bin has an average sign of zero, the analysis gives NaNs. This can make it fail for time-displaced quantities even in areas with mild sign problems. This could be fixed by computing the background observable also according to jackknife, so that it would be very unlikely to encounter division by zero, even in regions with a severe sign problem.
Edited by Emilie Huffman