How will the difference imaging pipeline respond to marginally resolved sources?

I am trying to understand how the LSST pipeline will handle blended point sources (i.e., strongly lensed supernovae with marginally resolved multiple images).

To make things a bit more concrete, I’ve attached an example “filmstrip” that illustrates the effect I’d like to understand. The film strip (read like a book, starting at the top and going left to right, up to down. You may have to zoom in to get a good look…) shows simulated LSST subtractions of a z=1 quad-strongly lensed supernova with minion_1016 cadence and conditions. The panels are labeled MJD-band.

Technically there are four sources here - one for each image of the strongly lensed supernova. However they are blended together and LSST cannot resolve the detailed structure of the system. The ellipses show the results of running source extractor on these simulated images. On nights with good seeing and when the SN is near peak, source extractor resolves multiple “sources” (2 but not 4 – conditions are never good enough). On nights where the images are fainter or seeing is bad, they all blend together into one more point source like blob.

My question is: how exactly will the pipeline respond to cases like this? Will the DIAsources produced by this strongly lensed SN (probably one per night on some nights and multiple per night on other nights) be consolidated into one DIAobject or will they be turned into multiple DIAobjects on different nights? If the former, how will the photometry be consolidated? If the latter, what will the criteria be for creating a new DIAobject?

filmstrip.sub (1).pdf (443.4 KB)

Hi Danny, thanks for this question. I’ve written a description of source association that is based on the DM pipelines and algorithms described in DM documentation, mainly and Some of this description I’m sure you already know, but I include for completeness. I’ve also included references back to the DM documentation for posterity.

  1. Source detection is run on a difference image, generating either a single DIASource or multiple DIASources, depending on separation and seeing, as you’ve simulated. Additionally, the deblender might identify parent/child sources in some cases. (Source detection, Section 6.5, LDM-151. Deblending, Section 6.6, LDM-151).

  2. DIASources are associated with existing DIAObjects. The association will be probabilistic and account for the uncertainties within the positions. If no DIAObject is associated, a new one (or two, in the good seeing case) is created. (Section 3.2.5, LDM-151)

  3. Based on 1 and 2, the single lensed SN from your scenario could indeed conceivably end up with 3 separate DIAObjects: one at the centroid in poor seeing, and two at the centroids in good seeing.

  4. As time goes on, with more epochs of good and bad seeing, the positions for the DIAObjects may* be updated, as they are based on a a time windowed (default 30 day) average of the DIASources that make up the DIAObject (Section 3.2.5, LDM-151). When the position of the DIAObjects are updated, the DIASource associations are also updated* (i.e., the algorithm of association is recursive). *see comment from @MelissaGraham below

  5. The exact algorithm for source association remains to-be-determined, but an example is given in the DM documentation: “Probabilistic Cross-Identification of Astronomical Sources”, Budavari & Szalay, 2008, ApJ, 679, 1.

  6. It is my understanding that currently, a single DIASource will not be associated with more than one DIAObject, and DIAObjects will not be associated with each other. However, the DM documentation also states that (depending on the chosen source association algorithm) the probabilistic association could account for one-to-many and many-to-one associations and/or include associations between DIAObjects. (Section 3.2.5, LDM-151)

  7. Every DIAObject with a detection in the past ~12 months will also have forced PSF photometry performed in each new difference image. This means that the DIAObjects defined by the good seeing nights will have forced PSF photometry at their location in the poor seeing nights (and vice versa). These are called “forced sources” and are separate from DIASources. (Section 3.4.3, LDM-151)

I think the main take-away for a multiply-lensed case like what you’ve shown is that photometry for the individual point source appearances from good seeing epochs should spawn their own DIAObjects, which will have well-measured multi-band light curves. However, for a given multiply-lensed object it also seems possible for the separations, relative fluxes, time delays, epoch timing, etc., to all contrive to produce a situation in which the centroid in the poor seeing images moves around so much that, because the association is probabilistic, the DIASources from the bad- and good-seeing images all end up associated with a single DIAObject. But this wouldn’t necessarily always be the case.

I’ll stop here for now, but please just let me know if you’re looking for more, or different, information. I’m happy to answer another round of questions about source association, or anything else that might be useful to your work.


Regarding point 4 above, the original posting said “will be updated”, but this was changed to “may be updated” to reflect the fact that in most cases the DIAObject positions and the DIASource–>DIAObject associations will not change. However, for situations like this case of a lensed SN in variable seeing, such a change is not currently forbidden.