Closed karimpsi22 closed 1 year ago
Hi Karim,
Thanks for the report. @BStoelzner any thoughts on this?
Best, Thejs
Hi Karim,
The reason for the opposite sign in the Dz posterior is that the cosmosis pipeline used in Asgari et al. defines delta_z= z_true - z_est while the MontePython likelihood uses delta_z= z_est - z_true (this definition was also used for example in https://arxiv.org/abs/1909.09632; see Table 3). So, internally everything should be handled correctly; you just need to be aware of the relative minus sign when making comparison plots. @brinckmann maybe it's worth putting at comment on this in the .param file and/or the readme.
Regarding the delta_c parameter: It has been a long time since I last ran the 2PCF likelihood (since our preferred summary statistics for KiDS are COSEBIs and BandPowers) and I don't remember the exact details. But you're right: the prior on this parameter should be a Gaussian centred at 0 ± 2.3×10−4, as reported in Asgari et al. I guess that the values in the .param file are some old ones that didn't get updated since we never used the 2PCF likelihood extensively (and it doesn't seem to have a big impact on the cosmological parameters).
Thanks for letting us know!
Best, Benjamin
Hello,
I just want to report an issue that I found in KiDS-1000 likelihoods. By using CLASS I was trying to obtain the Fiducial analysis from Asgari et al. paper: https://arxiv.org/pdf/2007.15633.pdf. Moreover, I considered the priors from
K1K.param
file and the CorrelationFunctions module, however, when I compare my posteriors to their results I am getting several differences in the nuisance parameters.Firstly, I got an opposite sign in the D_z's posteriors if I directly change the sign of my Dz's chains then everything is consistent with their results. This also is happening in the others modules (BandPowers and COSEBIs).
Secondly, in CorrelationFunction as it mentioned that delta_c is needed for this module. The prior provided in the input file might be wrong and is not the same reported in the Table 2 from the paper. My posterior of delta_c is inconsistent with KiDS-1000 results.
I am attaching 2 plots of CorrelationFunction comparison with KiDS-1000 results, the first in teal is my unaltered results and second in red I introduced a minus sign in my Dz's chains, so the Dz's posteriors are in agreement.
Regarding delta_c parameter issue I think it might be fixed if we consider the prior range from the paper which is delta_c = 0 ± 2.3×10−4.
That's it all I want to share.
Cheers, Karim
unaltered_class_2PCF_vs_K1K_data.pdf
altered_class_2PCF_vs_K1K_data.pdf