CAMB failing ABS(TE)/SQRT(TT*EE) < 1 requirement

Use of Cobaya. camb, CLASS, cosmomc, compilers, etc.
Post Reply
Alex Conley
Posts: 11
Joined: February 08 2005

CAMB failing ABS(TE)/SQRT(TT*EE) < 1 requirement

Post by Alex Conley » September 14 2007

Hello All,

I've been trying to get some constraints in a vanilla universe + variable (but
constant) w and allowing for non-flatness using only the WMAP plus SDSS mpk
results from COSMOMC. I find that these fits will inevitably crash with the
complaint:
unphysical input: TE/sqrt(TT*EE) > 1 at l= 23
which is an error message from the WMAP 3 likelihood routine.

After hacking COSMOMC to output more information, I find that a set of
parameters where this happens is:
omb: 0.1206625
omc: 0.4612063
omv: 0.4134928
omnu: 0.0000000E+00
omk: 4.6384116E-03
omdm: 0.4612063
ombh2: 2.1580001E-02
omch2: 8.2484923E-02
omnuh2: 0.0000000E+00
omdm2: 8.2484923E-02
zre: 13.56100
nufrac: 0.0000000E+00
h: 0.4229019
H0: 42.29019
w: -5.2179843E-02

And, indeed, if I plug these into CAMB by itself, I get
TT 0.16622E+04
EE 0.12462E-02
TE 0.14969E+01
which gives |TE|/SQRT(TT*EE) = 1.04

Any ideas what is going wrong here that CAMB is producing unphysical
outputs that are making it through into the likelihood routines? I'm running
an unmodified (except for these outputs) version of CAMB+COSMOMC
using the Intel F90 compiler 9.1 and version 8.1 of the MKL.

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

Re: CAMB failing ABS(TE)/SQRT(TT*EE) < 1 requirement

Post by Antony Lewis » September 24 2007

CAMB can produce negative outputs because of the interpolation (this might be an arguement for interpolating in the logs instead). If you increase the l-sampling the problem should go away; or probably just change the accurate_reionization parameter.

Post Reply