Microlensing and TDE metrics

(Peter Yoachim) #1

Hi folks,

I’ve been working on some new metrics so we can evaluate how well different survey strategies perform for microlensing events and tidal disruption events. Both of these metrics distribute a sample of 10,000 events on the sky over 10 years, then we see how many of them would be detected by a given LSST cadence simulation.

For microlensing, we assume events are distributed as stellar density squared, and draw from a distribution of possible impact parameters and event duration.
So the input looks like this:
And we recover 16% of them:

For the TDEs, we use an isotropic distribution, draw a lightcurve from a set of example lightcurves. We apply dust extinction to the curves, and check which ones get recovered.

TDE input:
TDE recovered:

For both of these populations, I’ve set the criteria for “detected” to simply be 2 observations above the 5-sigma limiting depth pre-peak in any filter(s). The motivation is that should be enough to get a decent alert and preliminary classification so others can follow it up. We can add more detection criteria if people want to. We can also replicate this format for other transients if folks have different lightcurves and/or different spatial distributions they would like the check.

These run quite a bit faster than some other transient metrics (under a minute), but still have N=10,000, so I’m hoping that’s sufficient statistical sampling. We want to use these to help rank survey strategies, so the absolute number of events is not as important as how the fraction recovered varies.

Microlensing code: https://github.com/LSST-nonproject/sims_maf_contrib/blob/u/yoachim/microlensing/mafContrib/microlensingMetric.py
and notebook: https://github.com/LSST-nonproject/sims_maf_contrib/blob/u/yoachim/microlensing/science/Transients/Microlensing.ipynb

TDE code: https://github.com/LSST-nonproject/sims_maf_contrib/blob/u/yoachim/microlensing/mafContrib/TDEsPopMetric.py
and notebook: https://github.com/LSST-nonproject/sims_maf_contrib/blob/u/yoachim/microlensing/science/Transients/TDEsSlicer.ipynb

Tagging the possibly interested: @rstreet, @wadawson, @nsabrams, @fed, @CStubbs

(Sjoert Van Velzen) #2

Hi Peter,

Thanks for this post. Within the TDE working group we have already developed our own metric based on light curves, and this is also being applied to current opsims. Perhaps we can use your example here to make ours faster. We’ve been in touch with @ljones to tweak the final details.

One important difference is that we require more detections in multiple filters. With just 2 detections at 5 sigma you get a enormous sample, of not just TDEs, but also SNe and AGN flares. Two detections are not enough to for photometric classification. For the follow-up observations that you propose, you need another LSST-scale telescope to keep up.

LSST science is photometric science, we need to make sure our metrics for transients reflect this reality. Otherwise we end up optimizing the survey cadence for science projects that are not feasible.


(Lynne Jones) #3

I would like to encourage everyone to send us their metrics, as well as the parameters that are useful. While it is wonderful that people are running their own metrics (in this and in other science collaborations), it’s immensely useful for us to have access to run them ourselves as well – this lets us check deeper into runs we do for technical evaluations, or just when we try something new.

We have followed up with @sjoert via email, so in this case I hope we’re a little more clear, but I wanted to bring this up for everybody … it’s helpful to have an explanation of what the criteria we should use are, and why they are that way. Peter’s setup above was likely too simple for TDEs, but understanding that follow up facilities will be overwhelmed and thus identification/classification needs to be done with information from LSST alone is important too (this isn’t the case for all kinds of objects, or more specifically all levels of classification for all objects).