Cross correlation peaks inverted for quasars

[ScottCroom]

Dear Sam,
we are using Marz to redshift quasars and other objects in AAOmega data form the AAT for an eROSITA project. We are finding that the cross-correlation function for quasars often appears to be inverted, so that the peaks corresponding to good redshift matches are actually negative dips. This means that they are not found as the best match. A screen shot of an example is shown below.

The data used to make this can be found here: http://www.physics.usyd.edu.au/~scroom/outgoing/SEP4_erosita_run009_v19_20.fits

It has been reduced using the ozDES pipeline.

If you have any ideas why the cross-correlation should look like this, and how to fix it, that would be very helpful.
Thanks,
Scott

[Samreay]

Hi Scott, this is super weird, I've never seen it actually get strong negative xcorr before like this. Before I try and jump into the code or the file (it's been years since I've delved into this project), I'll reach out to Chris Lidman . He's done a truly unfathomable amount of quasar matching with Marz on OzDES data, and if anyone has seen this before and knows the quick fix, it's him. I'll ping him and direct him here to see if he has any thoughts :)

[ScottCroom]

thanks for the prompt response. Will be interested to see if Chris has found this sort of thing.

[Samreay]

Looks like Chris is on holidays until the 5th, I've also cc'd Tamara Davis who has done a lot of quasar matching as well. Will let you know what they say :)

[Samreay]

Hi Scott, sorry for the delayed response, I've just returned from my own holidays and saw Chris's response in my inbox. Unfortunately, neither he nor Tamara have ever seen something like this with their quasars. In terms of things we could try to debug this, I'd recommend generating a FITS file where the uncertainty is an array of all ones, as most of the weirdness I've seen in other results comes from unusual features introduced when dividing flux/uncertainty, and setting uncertainty all to one helps us figure out if that might also be the case here.

[ScottCroom]

Hi Sam, Thanks for this. There is indeed an issue with the variance array. There are a small number of pixels (~10 over the whole 2D image ~400x5000) that have negative variance. This is introduced in 2dfdr at the splicing of the red and blue arms. Setting these to NaN fixes the problem. However, its odd that these few pixels throw out the scaling of the cross correlation, and particularly it only seems to be for the quasars (and not just for the spectra where the bad pixels are). Anyway, pre-processing to remove -ve variances is the quick short-term fix, although it might be good to make MARZ a little less sensitive to these sort of issues (no urgency though). Thanks, Scott

…

------------------------------------------------------------------------- PROFESSOR SCOTT CROOM Sydney Institute for Astronomy School of Physics | Faculty of Science THE UNIVERSITY OF SYDNEY Rm 351, School of Physics, A28 | The University of Sydney | NSW | 2006 T +61 2 9036 5311 | F +61 2 9351 7726 E ***@***.******@***.***> | W sydney.edu.au/science/people/scott.croom<http://sydney.edu.au/science/people/scott.croom> CRICOS 00026A This email plus any attachments to it are confidential. Any unauthorised use is strictly prohibited. If you receive this email in error, please delete it and any attachments. Please think of our environment and only print this e-mail if necessary. From: Samuel Hinton ***@***.***> Date: Wednesday, 22 February 2023 at 11:44 pm To: Samreay/Marz ***@***.***> Cc: Scott Croom ***@***.***>, Author ***@***.***> Subject: Re: [Samreay/Marz] Cross correlation peaks inverted for quasars (Issue #153) Hi Scott, sorry for the delayed response, I've just returned from my own holidays and saw Chris's response in my inbox. Unfortunately, neither he nor Tamara have ever seen something like this with their quasars. In terms of things we could try to debug this, I'd recommend generating a FITS file where the uncertainty is an array of all ones, as most of the weirdness I've seen in other results comes from unusual features introduced when dividing flux/uncertainty, and setting uncertainty all to one helps us figure out if that might also be the case here. — Reply to this email directly, view it on GitHub<#153 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AOPKIDLJTCAWXADLQEGKKCLWYYC4BANCNFSM6AAAAAAUC3CYWE>. You are receiving this because you authored the thread.Message ID: ***@***.***>