I ran the reconstruction with a different selection function to see if that made things better. The photometric distribution is shown here (workingDir is here: 'run2011530_1632'):
I'm still getting the issue that I am having trouble finding phibins that don't give me the error. But here are some reconstructions, the first one looks good-ish, the second one not so much:
Note that the label on the y-axis is not right, there are more objects in each bin than it says, I just had to normalize them to be on the same scale.
I'm running again with small angles to see if this helps. This smaller angle run is here (workingDir = 'run2011531_1451').
The angular correlation functions are freaking out at small angles. Not sure what that is about:
They look normal starting around 10^-3:
I tried doing the reconstruction both using all the angular bins, and just the angular bins that weren't freaking out, the look the same:
less angular bins:
all angular bins:
Here are some other reconstruction binning, disturbingly I am still getting the "feature" where if I change the # of bins slightly (from 15 to 16) I get very different reconstructions:
Because I am still not sure how doing this "non uniform" binning in r effects the reconstruction, I am running again, but with uniform binning. We'll see how that goes.
I think it's interesting that the reconstruction is doing better with more bins rather than fewer... I'm trying to understand that.
ReplyDeleteThose 2DCFs look wrong. We do expect the correlations to grow significantly at small separations (that's how the method works) but that bounce in the middle looks wrong, I'm afraid you'll need to track that down before attempting a reconstruction.
With irregular binning, you definitely need to have weight factors that indicate the bin widths, because we are approximating an integral as a sum, and the matrix multiplication is assuming all the bin widths are the same. I think there is a 1./binwidth factor in the code some place, you could just make that a 1/binwidth matrix...