[Shape-univ] Re: bleee 20th hour

[Boud Roukema]

Hi Bartek,

On Mon, 14 Feb 2005, Bartosz Lew wrote:

> eemmm, I'm about to finish my own versio of a program that calculates
> circles on the sky for dodecahedron. :) I was just wandering about
> the correlator you used. I have 2 remarks:
>
>
> 1) circles of different sizes are represented by different number of
> pixels. so naturally bigger circles will have bigger S value that the
> smaller because there is just simply more terms to sum over. like T_i*T_j
> from 'upper' and 'lower' circle. I'm trying to use the same correlator but
> normalized to one pixes - i.e. I divide each S value (for each individual
> circle) by the number of pixels that go into it. What do you think about
> that ?

See eq.(9) astro-ph/0402608.  The only difference is the normalisation.


> 2) I'm a bit worried about the fact that you don't use absolue values in
> the \delta T_i and \delta T_j. Imagine a fluctuations aroud the circle in
> shape of just a linear function T(dist_along_circle=d) ~ d and indentical
> in the opposite circle.  whaen you correlate this you get zero - that's not

You don't get zero.  \int_0^L d*d  dd  = L^3/3 - 0 = (L^3)/3
which is bigger than zero.

> much. I know it's not a prove but looks a bit worrying to me,since

It's wrong, so IMHO it's not a proof.

> probably some , any random fluctuations can generate S > 0, while for the
> perfect match we get S=0. So I indent to use abs(T_i)*abs(T_j) (except
> from the normalization thing).

That will increase the false signal.
ok, next two emails...

pozdr
boud