CosmoCoffee

[Josh Kable]

I am attempting to calculate the parameter covariance matrix from a Fisher matrix generated by finite difference differentiation of the TT Cls generated by CAMB; however, I seem to be running into a problem with the accuracy of the final digit outputted into the text file. Namely, my derivative power spectrum seems to have a blocky structure, particularly near the peaks, which is consistent with CAMB generated Cls that are not sufficiently accurate for the step size of my derivative. I have tried a larger step size, and that problem goes away; however, the calculated covariance matrix becomes very sensitive to changes in the step size indicating that it is not very accurate. What accuracy settings should I edit in the params.ini file to increase the accuracy of the outputted Cls? Do I also need to edit CAMB?

[Antony Lewis]

If you use the python wrapper you can avoid the text files. There are three accuracy parameters you can change, either in the .ini file or in via the python wrapper.

[Josh Kable]

What are the three accuracy parameters that you mentioned? What do each of them do, and do you have any suggestions as to which ones I should change in this case?

[Luke Hart]

Hey Josh,

The AccuracyBoost parameter is a generic accuracy parameter which affects a lot of stuff such as the k precision in the power spectra and the timestep precision when computing things like recombination and reionization.

The LAccuracyBoost parameter adds to the terms evaluated in the hierarchy and LSampleBoost increases the number of l values in the interpolation.

Readme for the CAMB Accuracy stuff