How to include BKP likelihood in Cosmomc?

[Jack Sparrow]

Hi, all. I have a question about BKP likelihood. I am trying to constrain the tensor amplitude in my model. I find a very strange thing. When I include all Planck likelihood (all the four likelihoods of Planck) AND the BKP likelihood, the tensor amplitude has a non-zero preferred value. This is not reasonable and I think all the model should prefer a zero tensor amplitude currently. When I include Planck likelihood OR BKP likelihood separately, the result is reasonable and a zero value tensor amplitude is more preferred. So, why a non-zero value is preferred when I combine Planck and BKP? Is there a conflict between them? Or are there non trivial settings in the .ini file? BTW, does the covmat matter in the settings?

Thanks a lot! Can some experienced persons explain it or give me some instructions? Thanks again.

[Sunny Vagnozzi]

Dear Jack,

first of all, what do you mean that you are including all four Planck likelihoods? Do you mean plik_dx11dr2_HM_v18_TT.ini , plik_dx11dr2_HM_v18_TTTEEE.ini , lowTEB.ini , and lowl.ini? If this is the case, that would be incorrect since you would be double-counting information, I believe you should only add two of these at a time, e.g. for what Planck calls PlanckTT+lowP, you would add plik_dx11dr2_HM_v18_TT.ini and lowTEB.ini, etc. This is the first point.

Second, I don't think a (small) preference for a non-zero tensor-to-scalar ratio is a big problem. As long as it is preferred at only say 1 sigma or so, I think that's not a huge problem.

Finally, in your distparams.ini, did you set limits[r] = 0 N? Because if you didn't do so, that would easily explain your error. You must do this for any parameter for which the prior cuts at a value where the posterior is non-negligible, and the tensor-to-scalar ratio is an example of these. Please see th CosmoMC README for more information.

Cheers,
Sunny

[Sunny Vagnozzi]

I would also suggest having a read at batch2/BK14_README.txt

[Shouvik Roychoudhury]

Doesn't the Nov 2016 version of CosmoMC automatically add the limits[r] = 0 N? (In the .derived_ranges files (in /paramnames folder), see Version History section)

[Sunny Vagnozzi]

Yes it should

[Sunny Vagnozzi]

The default distparams.ini I believe also has this setting