CosmoCoffee

Pavel Motloch · Joined: 14 Oct 2016 Posts: 6 Affiliation: U Chicago

Lenspix obtained from http://cosmologist.info/lenspix/submit.html is fairly outdated (missing at least the toms760.f90 update from June 2014). Is there any other significant difference between the current (git) version and what is on the cosmologist.info website?

I have a slight bias (relative difference 1e-4 in TT, EE around ell = 3000 for nside 4096, lmax 8000, interp_fac = 2) in my lensed power spectra when compared with CAMB and am trying to figure out whether that is just interpolation error or some systematic issue.

Thanks,

PM

Pavel Motloch · Joined: 14 Oct 2016 Posts: 6 Affiliation: U Chicago

[Based on one quick example it looks like except toms760.f90 there are no other changes affecting SimLens.f90 and related routines]

Antony Lewis · Posted: February 24 2017

10

-4

sounds quite good...

You can see the commit history at
https://github.com/cmbant/lenspix/commits/EBsep

but I don't remember fixing anything recently. (should update the download link to point to github)

Pavel Motloch · Joined: 14 Oct 2016 Posts: 6 Affiliation: U Chicago

Thanks.

Yeah, it is really good. But it is a systematic bias, so it shifts your best likelihood point for CMB S4-like specs in a manner which we are not entirely happy about. Increasing interp_fac would resolve this, but then the memory demands increase. Presumably larger nside would also help, but that is getting slow.

Antony Lewis · Posted: February 25 2017

CAMB is of course only designed to be designed to be accurate at the 1e−3 level with default setting. You could also try use the lensing functions in pycamb's correlations module to do a more accurate lensing calculation from the input power spectra (though the analytic result is still truncated at O(2) in C

gl,2

, though I think that's OK to about 1e−4).

Pavel Motloch · Joined: 14 Oct 2016 Posts: 6 Affiliation: U Chicago

Thanks, those are good suggestions. Previously I did not mention that when you run Lenspix without any lensing potential, you get roughly the same bias between the sims and input unlensed Cls in Lenspix alone as you get between CAMB and Lenspix. So we believe it is related to the interpolation in Lenspix.

Will take a look at pycamb lensing, did not know it is different than what CAMB is doing.