Page 1 of 1

Bad Contours with CMB and BAO likelihoods

Posted: March 19 2019
by George Alestas
Hello everyone,

I'm new at using cosmomc and I have some problems with it. I'm currently using the 2015 version. My goal is to produce the plot of w - H0 for CMB and BAO likelihoods using a certain w(α) model and parametrization.

More specifically the test.ini file I'm using is,

Code: Select all

#general settings
#Bicep-Keck-Planck, varying cosmological parameters
#DEFAULT(batch2/BKPlanck.ini)

#Planck 2015, default just include native likelihoods (others require clik)
DEFAULT(batch2/plik_dx11dr2_HM_v18_TT.ini)
DEFAULT(batch2/lowTEB.ini)
#DEFAULT(batch2/lowl.ini)
#DEFAULT(batch2/lensing.ini)

#Other Likelihoods
DEFAULT(batch2/BAO.ini)
#DEFAULT(batch2/WiggleZ_MPK.ini)
#DEFAULT(batch2/MPK.ini)
#DEFAULT(batch2/WL.ini)

#general settings
DEFAULT(batch2/common.ini)

#e.g. to vary r in addition to standard 6:
#(for r>0 also need compute_tensors=T)
#compute_tensors = T
#param[r] = 0.03 0 2 0.04 0.04

#high for new runs
MPI_Max_R_ProposeUpdate = 30

propose_matrix= planck_covmats/base_BAO_TT_lowTEB_plik.covmat
And the params_CMB_defaults file has the following form (since I wanted to vary some fixed parameters such as w0),

Code: Select all

#param[omegabh2] = 0.0221
param[omegabh2] = 0.023 0.013 0.033 0.0001 0.0001

#param[omegach2] = 0.12
param[omegach2] = 0.12 0.001 0.99 0.001 0.0005

#param[theta] = 1.0411
param[theta] = 1.0411 0.5 10 0.0004 0.0002

#param[tau] = 0.09 
param[tau] = 0.09 0.01 0.8 0.01 0.005

num_massive_neutrinos=1
param[mnu] = 0.06
param[meffsterile] = 0

param[omegak] = 0
param[w] = -1 -2 0 0.001 0.001
param[nrun] = 0
param[nrunrun] = 0
param[r] = 0

param[wa] = 0
param[nnu] = 3.046
param[yhe] = 0.24

param[alpha1] = 0
param[deltazrei] = 0.5
param[Alens] = 1
param[Alensf]=-1
param[fdm] = 0

#param[ns] = 0.96
param[ns] = 0.96 0.7 1.3 0.004 0.002
#log[10^10 A_s]

#param[logA] = 3.1
param[logA] = 3.1 1.7 5 0.001 0.001
I also made changes to the files equations_ppf and halofit_ppf in camb, setting the new w and density equations in both of them. However the results were less than anticipated, and had the form,

Image

I'm open to any and all suggestions as to where my mistake is.

EDIT: Could it be because the default sample number is not enough for 2 likelihoods?

Thank you in advance

Re: Bad Contours with CMB and BAO likelihoods

Posted: March 21 2019
by Shouvik Roychoudhury
Default sampler is enough for two likelihoods or more, for usual DE parametrizations included in CosmoMC, from my experience.

Re: Bad Contours with CMB and BAO likelihoods

Posted: March 22 2019
by George Alestas
The thing is that I changed the DE parameterization. Could the sample number be at fault in this case?

Re: Bad Contours with CMB and BAO likelihoods

Posted: March 22 2019
by Antony Lewis
The problem is you think the contours should be smaller? Check your output .inputparams file to check all as expected, e.g. BAO actually being used.

Re: Bad Contours with CMB and BAO likelihoods

Posted: March 22 2019
by George Alestas
BAO likelihoods are definitely included as well as CMB (I checked the .inputparams file). I did another run, this time using the Metropolis-Hastings method with a sample number of 600000 and this time I almost got the contours I expected.

Image

The reason I changed it to Metropolis was because of the inclusion of BAO data, was my logic sound or were the results better just because of the higher sample number?

Re: Bad Contours with CMB and BAO likelihoods

Posted: May 10 2019
by Shouvik Roychoudhury
What was the R-1 achieved in both cases?

Re: Bad Contours with CMB and BAO likelihoods

Posted: May 12 2019
by George Alestas
Hi Shouvik,

The R-1 data for each case is unavailable at this point in time, but I'm sure that they had an order of magnitude of 10^-3 and that the runs were completed successfully. However, even in the second case where the contours have a far better form, they are in a significantly different position from the ones we're trying to reproduce (the best fit value is quite different). In our study we're using the ppf model for perturbations, since they don't describe theirs, could such differences be because of a different perturbation model? How likely is that from your experience?

Also as you can see there still remains a small bump in the Metropolis-Hastings contour, therefore, any suggestions on how to further improve it would be welcome.