search

HISKP-LQCD / sLapH-contractions

Stochastic LapH contraction program
GNU General Public License v3.0
3 stars 3 forks source link

Normalization of correlation functions #44

Closed martin-ueding closed 6 years ago

martin-ueding commented 6 years ago

Bartek suggested that there the V^\dagger V objects are not properly normalized, a 1/L^3 factor is missing for each of them. This means that the correlation functions are not properly normalized.

For the GEVP this should not have any effect, but we should still make sure that the normalization is correct.

kostrzewa commented 6 years ago

@maowerner @martin-ueding

So, after https://github.com/HISKP-LQCD/sLapH-contractions/commit/a4f909ef5929fb09d09a2c929d6c29e4ed7e280e#diff-7757c5b9342e11601001841afc6c0133, it seems to me that the individual diagrams have the correct averaging normalisation (division by the number of summands) and there are no hard-coded special factors around. The correlation functions are normalised further by the number of source time slices, which is appropriate.

Beyond that, it seems to me that the normalisation as it is should be correct, in the sense that it will result in a GEVP with the correct relative weights of the various correlation functions. There are, however, some odd factors of "2" and "3" which I've pointed out.

The absolute volume- and kappa-dependent normalisations would only need to be performed if one wanted to directly compare correlation functions from different volumes with very well matched smearing radii, which we don't want to do at this point in time.

maowerner commented 6 years ago

I answered your questions in the inline comments. The short version is:

The commit you looked at is the one where we changed the normalization. These fudge factors where previously wrong where 5 random vectors where hardcoded. We put them in to see whether our tests worked and we understood the discrepancies after redoing the normalization. They were deleted the very next commit

https://github.com/HISKP-LQCD/sLapH-contractions/commit/a96ebbfa5f7317f213b81078668829cf5ff3bf03

kostrzewa commented 6 years ago

Indeed, I missed the following commit. Should have checked against the current code :)

maowerner commented 6 years ago

Are all normalization coefficients covered or is there still something we need to work out? Otherwise we might close this issue.

kostrzewa commented 6 years ago

I think we should be fine.