CosmoCoffee

[Antony Lewis]

This seems to be a nice paper looking at the effect of various inhomogeneous reionization scenarios on the CMB. In particular they claim a potentially very significant contribution to the C_l at l < 2000, enough to significantly bias parameters with Planck at l > 1000 if not included in the model. This is of course well known (e.g. astro-ph/9805012, astro-ph/0305471), but their estimates are rather higher than previous ones.

I note that they are using for their simulations, both choices giving small scale power at the high end of what is currently allowed. How sensitive are the results? (e.g. would give a much smaller signal due to a high-power scaling with σ₈?)

I would also be interested to see Table 2 for a more sensible cut at rather than at l = 4000 - ie. how wrong are people (like me) who usually cut at in the hope that by doing that problems with (non-lensing) non-linear physics and non-Gaussianity are mostly avoided. Also presumably the lensed CMB power spectra should be used for Table 2, since lensing increases the small scale power at l > 2000 and hence decreases the fractional effect of kSZ.

Finally, if, as they suggest, the l > 1000 temperature is not used (using only polarization at l < 2000), I'd have thought parameter uncertainties are likely to be more biased by the uncertainty in the recombination history (c.f. http://cosmocoffee.info/viewtopic.php?t=174) which has a larger effect on the polarization.

[Oliver Zahn]

Thanks for those comments,

the reason why our parameter biases are larger than the results of 9805012 is that their patchy power spectrum peaks at around l=4000 as opposed to at around l=2000 in our model. In 0305471 on the other hand, the signal peaks at a similar angular scale, and it has a larger amplitude. The reason why their biases are slightly smaller than ours should be that they use parameters in their Fisher analysis that are degenerate with the patchiness, such as running of the spectral index, the primordial Helium fraction, massive neutrinos, and the equation of state parameter.

Concerning the second point, that sigma8=0.9 and n_s=1 in the lss simulations we had are on the high side of what is currently favoured by experiments (at least sigma8=0.9 is fine with current CMB experiments). The dependence of Doppler anisotropies induced by patchy reionization will not be the same as the dependence of kSZ/OV induced by density fluctuations (e.g. sigma8⁴−6), and it would be interesting to look into that. Most changes can probably be compensated by varying the (unknown) ionization efficency zeta (compare equation 4 of the paper).

About your third point, if I do the analysis of Table 2 by using Cls only out to l=2000, the parameter errors stay almost the same, while the biases go down 20−25%. The bias from patchiness at l=2000 is already significant because this is roughly where our signal peaks. Plancks constraints will anyway not be affected much by what is going on on smaller scales.

With respect to the fourth point, it may well be that the best strategy is to incorporate patchy reionisation as an extra parameter (as we did at the end of section 5.2). This would be most accurate if the epoch is understood well from simulations as well as from experiments, for example by measuring on the smallest scales with ACT, SPT, or APEX, which should happen at around the same time as Planck. Furthermore our understanding of recombination should improve with intermediate scale polarization data coming out in the next couple of years.

All the best, Oliver