Open NiallMac opened 2 years ago
It is an interesting question; the current asymmetric estimator includes only "lens" to compute the estimator of https://arxiv.org/pdf/astro-ph/0701276.pdf, https://arxiv.org/pdf/1802.08230.pdf. Do we need also to consider an asymmetric estimator for sources?
Ah ok I didn't realize the current qtt_asym only included "lens".
Yes, I believe when doing bias-hardening for the asymmetric case (and I think we'll want to do this - Omar's paper showed tSZ cleaning + source-hardening was a useful option), we'll also need the source normalization, source N0 and lens-source response for the asymmetric estimator.
I thought that an optimal way is to estimate the source fields with the symmetric quadratic estimator and use it to bias-harden the asymmetric kappa estimator. But, in any cases, we need to implement the response and N0 for such estimator.
anyway, I create a pull request where I add asymmetric source estimator / response / noise spectrum
I thought that an optimal way is to estimate the source fields with the symmetric quadratic estimator and use it to bias-harden the asymmetric kappa estimator. But, in any cases, we need to implement the response and N0 for such estimator.
Hmm...I think to do this you'd need to assume some SED for your sources? Because you'd just be estimating their amplitude from the non-cleaned (e.g. ilc) map T_ilc, but to correct the asymmetric estimator, you need to know how they contaminate the product of the cleaned and non-cleaned maps
anyway, I create a pull request where I add asymmetric source estimator / response / noise spectrum
🙌 🙌 🙌
Hmm...I think to do this you'd need to assume some SED for your sources? Because you'd just be estimating their amplitude from the non-cleaned (e.g. ilc) map T_ilc, but to correct the asymmetric estimator, you need to know how they contaminate the product of the cleaned and non-cleaned maps
.
My assumption was too simple. It is much more accurate to do with the asymmetric one.
I think for bias-hardening with the asymmetric temperature estimators, the only extra thing we need is the cross-response i.e. a version of norm_general.xtt (or perhaps norm_lens.stt - is this just a special case of xtt?) that can take two sets of (lmin, lmax, Cl^obs), as qtt_asym does.
I think then it's straightforward to construct the bias-hardened estimator using
qtt_asym("lens", ....)
andqtt_asym("src", ...)
to get the lens and source normalisations, in addition to this cross-response term. @toshiyan does this sound correct to you?