search

LSSTDESC / TJPCov

TJPCov is a general covariance calculator interface to be used within LSST DESC
https://tjpcov.readthedocs.io
MIT License
11 stars 1 forks source link

xim covariance not pos-def #4

Open chihway opened 4 years ago

chihway commented 4 years ago

I noticed that in the DES example xi- was not included. When including xi- in the cosmoDC2 example: https://github.com/LSSTDESC/txpipe-cosmodc2/blob/master/notebooks/PlotCov.ipynb it appears that that block make the covariance not pos-def. Is this something expected? I've not tried yet varying scale cuts, it could be mainly a small-scale problem...

sukhdeep2 commented 4 years ago

@chihway Sorry for the long delay in response. I cannot access the notebook you linked to. As you mentioned, it could be a small-scale problem. xi- is likely to be more sensitive to the choice of l_max when doing hankel or wigner transform. Try increasing the l_max by a factor of 2-5 to see if it helps. My back of the envelope estimates are: assume ell\theta~pi and hence ell_max~ pi/theta_min. Then use 2-5 X the value of ell_max obtained. A word of caution, using all ell upto a large ell_max can slow down the code and also cause memory issues. Subsampling of ell is supported in tjp_cov (in fact we only use user defined ell) but I haven't fully tested it yet.