sum rules for hessian PDFs

[RoyStegeman]

When computing the uncertainty on the sum rules for hessian PDFs the standard deviation is taken over the moments computed per eigenvector

nnpdf/validphys2/src/validphys/sumrules.py

Line 214 in 050d723

d["std"] = np.std(arr)

[comane]

Hi @RoyStegeman, can you perhaps give more details on why this is a bug.

Should a different formula be used for the computation of how much the sum rules for each eigenvector of a hessian set vary?

[RoyStegeman]

See e.g. eqs. 3 and 4 in https://arxiv.org/pdf/1101.0536.