Cobaya sampler minimizer options/outputs

[Helena Garcia]

I am running the minimizer sampler in the standard LCDM case with CAMB using cobaya with the following parameter selections:
I have the following questions:

1.- Every time I ran the minimizer I got a different chi2, the delta chi2 can be as high as 6, is that normal?
2.- I tried to change the best_of parameter but if I set it to a number >10 the minimization breaks before giving an output. Can this parameter help reduce the error bars in the chi2 results? How can I get it to work for a best_of >10 run?
3.- The output file that I get has a chi-sq at the beginning that doesn't match the sum of chi2__CMB+chi2__BAO that I find at the end of the file, where does the difference come from?

Thanks!

Here are some of the input .yaml file block settings I am using:

Code: Select all

sampler:
  minimize:
    method: bobyqa
    ignore_prior: true
    max_evals: 1e6d
    best_of: 2
    confidence_for_unbounded: 0.9999995
    seed: null
    override_scipy: null
    override_bobyqa:
      rhoend: 0.05
    covmat: null
    covmat_params: null
    version: null
    fallback_covmat_scale: 4

theory:
  camb:
    extra_args:
      halofit_version: mead
      bbn_predictor: PArthENoPE_880.2_standard.dat
      lens_potential_accuracy: 1
      num_massive_neutrinos: 0
      nnu: 3.044
      theta_H0_range:
      - 20
      - 100
likelihood:
  planck_2018_lowl.TT: null
  planck_2018_lowl.EE: null
  planck_2018_highl_plik.TTTEEE: null
  planck_2018_lensing.clik: null
  bao.sdss_dr16_baoplus_lrg: null
  bao.sdss_dr16_baoplus_lyauto: null
  bao.sdss_dr16_baoplus_lyxqso: null
  bao.sdss_dr16_baoplus_qso: null
  bao.sdss_dr16_baoplus_elg: null

[Antony Lewis]

The minimizer can be a bit unstable. It will likely work better if you provide a covmat from a chain, then it will know about degeneracies etc.

The difference in "chi2" is likely priors.

[Helena Garcia]

Ok!
Would increasing the "best_of" parameter in the minimizer block reduce this uncertainty?
If I modify the "best_of" parameter in the minimizer block , should any other parameter be modified to achieve a successful minimization?
Thanks