CAMB: transfer_k_per_logint (possible bug?)

Use of Cobaya. camb, CLASS, cosmomc, compilers, etc.
Post Reply
Charles Shapiro
Posts: 24
Joined: February 05 2005
Affiliation: University of Portsmouth

CAMB: transfer_k_per_logint (possible bug?)

Post by Charles Shapiro » December 20 2006

Hi All,

I find that when I run CAMB (September 06) with transfer\_k\_per\_logint = 0 or 5, I get more than just different k-sampling - I get differences in the transfer function and matter power spectrum. The latter changes up to 3 percent near k=0.1h/Mpc (i.e. the baryon wiggles) with the settings I'm using. Is this a bug? Expected? My fault?

Here is my base parameter file

Here I plot the fractional difference in the linear power spectrum.

Here I vary Omega_K (curvature) by 0.02 and plot the fractional difference in the linear power spectrum for both samplings.

Chaz

Antony Lewis
Posts: 1941
Joined: September 23 2004
Affiliation: University of Sussex
Contact:

Re: CAMB: transfer_k_per_logint (possible bug?)

Post by Antony Lewis » December 20 2006

If you set

transfer_high_precision = T

the errors should be below \alt 0.5%.

Charles Shapiro
Posts: 24
Joined: February 05 2005
Affiliation: University of Portsmouth

CAMB: transfer_k_per_logint (possible bug?)

Post by Charles Shapiro » December 20 2006

Setting transfer_high_precision = T, I am still seeing a fractional difference of 3\% in the power spectrum for the two sampling methods.

Antony Lewis
Posts: 1941
Joined: September 23 2004
Affiliation: University of Sussex
Contact:

Re: CAMB: transfer_k_per_logint (possible bug?)

Post by Antony Lewis » December 20 2006

But I think the adaptive transfer_k_per_logint = 0 is correct? Compared to transfer_k_per_logint = 15 I have ~ 0.3% max error: transfer_k_per_logint = 5 is just a fixed spacing which may not be good enough to sample the baryon oscillations very well.

Charles Shapiro
Posts: 24
Joined: February 05 2005
Affiliation: University of Portsmouth

CAMB: transfer_k_per_logint (possible bug?)

Post by Charles Shapiro » December 23 2006

Yes - now I see. I was confused why the results should differ at identical wavenumbers... but since some splines are involved, the sampling would affect the accuracy. Thanks, Antony.

Chaz

Post Reply