shot noise removal on cross about nbodykit HOT 4 CLOSED

I just made it up so that it reduces to auto when data1=DATA2. do you know
if zero is correct ? Is there a reference? I remember Marcel said it was
complicated.

YU
On Jul 3, 2015 2:09 PM, "Nick Hand" [email protected] wrote:

Assigned #14 #14 to @rainwoodman
https://github.com/rainwoodman.

—
Reply to this email directly or view it on GitHub
#14 (comment).

from nbodykit.

Well if we are considering the classic Poisson shot noise, which is usually what is subtracted out, then I think it should just be 1 / number density in the auto case and zero in the cross case, if the tracers are non-overlapping. It comes from the correlation of objects with themselves, which why it is important if the samples overlap at all.

Section 2 of this paper has a good description (see eq 27). And section II of this paper has the derivation of Poisson noise as "self-pairs" .

On a slightly related note, I was testing the cross-correlation code using different halo mass bins and noticed that the power was asymptoting to a constant at high k, like a Poisson noise term was present in the cross-correlation. It turns out this was due to a bug in the cross-power logic (see 58456f8 and 30ce79c). I am now finding that the cross-power spectra asymptote to zero in the high-k limit, as I expected.

Nick

from nbodykit.

DId you commit the changes, if so let's close this.

from nbodykit.

shot noise now zero in the cross case. closed by ad35bcc

from nbodykit.