Talked to Alexia. She said that probably there are several things going on that could explain the results from yesterdays post. First of all, I had reverted to using an older version of crl when doing this reconstruction, this older version didn't use interpolation, but instead assumed that the correlation functions are power-laws. This explains why they are just straight lines.
Also she thinks that perhaps there is something going wrong with the conversion between Mpc and Gpc which might be the cause of the offset.
She suggested I go back to using the latest version of the code (with the interpolation turned on) and see what these same plots look like.
The reconstruction using the latest version of the code doesn't match very well now, however there isn't a lot of variation when changing the binning between 39-41:
However, when I go to less bins (14-15), we do see a lot of variation in the reconstruction:
In terms of how ximat and sximat compare to the correlation functions, the main thing that I notice is that they don't overlap very many points. I would think we would want more points to overlap to properly do the conversion:
Okay. It is doing exactly what it's intended to do...extrapolate to a power law when the x values are outside the range of the data. However, the info is in the measured part of the spectrum, the points outside all contribute the same info as one another, namely just the normalization.
ReplyDeleteThere are two things that can be done here, 1 is to measure xi and cxi out to slightly higher radii, but more importantly, we need more of the ximat to fall in the region where we have data. This can be accomplished by (for example) changing the range of theta values being used, and making the binning in theta logarithmic rather than linear.
I'm a bit concerned that the slope of the power law is not matching well onto the end of the correlation functions, that seems off to me. What slope is it?
It sort of looks to me like the slope is the same regardless of what the data looks like. I'll check that. But yes, I'd like more of the cximat points to fall on top of the data.
ReplyDeleteThe slope in the extrapolated power law part is always the same -- it's a functional form hard wired into the code. The question is, is it the right functional form?
ReplyDeleteIn terms of overlap, see my comment about adjusting the theta range and binning.
also, are you actually plotting cximat as you labeled it, instead of ximat? I don't think you should be having a cximat in your code, I used to use that to generate fake 2DCFs. what you want to be plotting is the 3DCF, and ximat, the matrix of 3DCF values.
ReplyDeleteI'm plotting ximat vs sximat, and rlos vs xi. (not sure why I used c, sorry for that confusion)
ReplyDeleteximat is what I call what is in ./xmple_rv
sximat is the output from getximat:
sximat=crl.getximat(th,rlos,phibins,rs,xispec,xierr,interpolate=1)
The slope of the line is -2.
I'll play with the angles to see if I can get more overlap with the data.