GCspectro likelihood

Run technicalities in [[MontePython Oneloop Runs]]

  1. Installation issues:
    1. Class Oneloop C-code not compatible with older gcc 9.4, it is compatible with gcc~>11.0
    2. Classy does not work with numpy >=2.0 .
    3. Montepython requires numpy>1.23.5 because of a scipy.optimize import fsolve
  2. w0, wa priors are not important in (Analytical Marginalized) AM likelihoods.
  3. w0, wa priors are important in full likelihoods, getting constantly errors like:
desinomarg.err:48:Error in Class: perturbations_init(L:796) :condition (w_fld_ini >= 0.) is true; The fluid is meant to be negligible at early time, and unimportant for defining the initial conditions of other species. You are using parameters for which this assumption may break down, since at early times you have w_fld(a--->0) = 1.406016e-03 >= 0
  1. AM likelihoods do not generate their own fiducials, one has to generate them using the full likelihoods and be very careful with fiducial parameters.
  2. MP likelihoods when reaching max number of Nsteps, finish (on the cluster at least) with FAILED, ExitCode 137 giving the wrong impression that the run has failed or there was a problem. They should exit with ExitCode 0
  3. When generating the fiducial, the MP likelihood returns :
256             warnings.warn(
257                 "Writing fiducial model in %s, for %s likelihood\n" % (
258                     self.fid_filename, self.name))
259             return 1j

This is a strange complex number return, because in the case of using the option --display-each-chi2 at the same time as -f0 to generate the fiducial, this return value gets added to the other chi2 of other likelihoods and one gets the wrong impression that something is going on. The logic here should be clarified.

  1. The public (Casas et al 2023) ISTF-like euclid_spectroscopic likelihood does not work for m_nu=0 because it automatically requests P_cb, f_cb and other cb quantities, which are not available in this case from class.
  2. The oneloop spectroscopic likelihoods do not contain P_cb explicitly, working with P_mm only.
  3. The runs of the MP validation paper were done in the Omega_m, sigma8 and camb basis. The oneloop runs have been done using the class basis and logAs making a comparison more difficult.