Cobaya LCDM+Neff+Yp chains

Use of Cobaya. camb, CLASS, cosmomc, compilers, etc.
Post Reply
Lennart Balkenhol
Posts: 1
Joined: June 11 2020
Affiliation: University of Melbourne

Cobaya LCDM+Neff+Yp chains

Post by Lennart Balkenhol » January 27 2021

Hi all,

I am trying to run Cobaya for a LCDM + Neff + Yp model using some CMB power spectra. While adding the two extension parameters individually produces reasonable results, freeing them simultaneously yields the monstrosity below.
neff_yp.png
neff_yp.png (34.63 KiB) Viewed 237 times
These chains have run for O(10h) on a multi-processor system and have >100k samples, so I don't think convergence is an issue here. Instead, I suspect this is due to some relation between Neff and Yp being imposed that I'm not aware of. In the .yaml file I simply add the two as free parameters ala

Code: Select all

(standard LCDM parameters)

    YHe:
        prior:
          min: 0.05
          max: 0.8
        ref:
          dist: norm
          loc: 0.245
          scale: 0.01
        proposal: 0.001
        latex: Y_\mathrm{P}
    nnu:
        prior:
          min: 0
          max: 10
        ref:
          dist: norm
          loc: 3.046
          scale: 0.5
        proposal: 0.001
        latex: N_\mathrm{eff}
In the CAMB settings I specify

Code: Select all

bbn_predictor: PArthENoPE_880.2_standard.dat
Does this automatically enforce BBN consistency? Adding

Code: Select all

bbn_consistency: False
does not work as the keyword is not recognised by Cobaya. Any help would be appreciated.

Thanks in advance!

Antony Lewis
Posts: 1709
Joined: September 23 2004
Affiliation: University of Sussex
Contact:

Re: Cobaya LCDM+Neff+Yp chains

Post by Antony Lewis » January 27 2021

Surely the convergence diagnostics are terrible? Looks to me as though you have chains stuck in different places - the scale for your initial nnu values is huge compared to your proposal width. I suggest you use a covmat from Planck to get things started (can use the GUI to make appropriate yaml file).

Post Reply