Comparing relative error between current and old CAMB version

Use of Healpix, camb, CLASS, cosmomc, compilers, etc.
Post Reply
Cristhian GQ
Posts: 10
Joined: March 27 2019
Affiliation: University of Texas at Dallas

Comparing relative error between current and old CAMB version

Post by Cristhian GQ » February 02 2020

Hello,

I would like to compare the relative error between the current camb version and an old version in for example, CMB or matter power spectra. However, when I set the values of the cosmological parameters (H0, ombh2, etc...), it seems to me that these versions output a different results for quantities such as Om_Lambda, 100*theta, etc... Then, I am not comparing exactly the same model. So, is there a easy way to set exactly the same values by changing some settings in params.ini or this issue is unavoidable with the new modifications?

Best regards,
Cristhian Garcia.

Antony Lewis
Posts: 1609
Joined: September 23 2004
Affiliation: University of Sussex
Contact:

Re: Comparing relative error between current and old CAMB version

Post by Antony Lewis » February 03 2020

If you set parameters consistently they should agree well for background quantities (down to 1e-4 level. where it matters for example whether or not you include the photon density when working out $\Omega_\Lambda$). Check things like the Helium density and neutrino model are the same.

Cristhian GQ
Posts: 10
Joined: March 27 2019
Affiliation: University of Texas at Dallas

Re: Comparing relative error between current and old CAMB version

Post by Cristhian GQ » February 11 2020

Hello A. Lewis,

Just going back to this issue, I run camb by using the current version of CosmoMC which is in the github (I called CosmoMC-2020) and also run camb with the version of CosmoMC called CosmoMC-Planck2018.

Then, I set the background quantities and as you said, they agree well within 1e-4. However, I can see that by using the default settings, both version of cosmomc give a relative error of around 1e-3 for C^{TT}_{ell}. For the other C'ells I can reduce the relative error by changing the accuracy settings, however for TT I cannot, is this contemplated after updating the previous version? or did I make some omission?

I attach two plots (the TT I am talking about and the comparison of the P(k) which agreed very nicely).

https://drive.google.com/open?id=15t0O7VSmIeR4_gciseSpp57iG-ZUJMov

https://drive.google.com/open?id=1WwaY5id-mUtDVNt1cjZFUHunVoDrFVBr

Best regards,
Cristhian Garcia.

Antony Lewis
Posts: 1609
Joined: September 23 2004
Affiliation: University of Sussex
Contact:

Re: Comparing relative error between current and old CAMB version

Post by Antony Lewis » February 11 2020

There will be some numerical changes due to CAMB updates, what you plot looks fine - on small scales the difference is well below 0.1% (on large scales 0.1% is fine because the cosmic variance is large).

Cristhian GQ
Posts: 10
Joined: March 27 2019
Affiliation: University of Texas at Dallas

Re: Comparing relative error between current and old CAMB version

Post by Cristhian GQ » February 16 2020

Hello Anthony,

thank you! you explanation makes sense actually.

Best regards,
Cristhian Garcia.

Post Reply