For the past two years I’ve been quite content to let most of the mundane details of coding stuff slip quietly into obscurity. I could probably have given quite a bit of insight into the process we went through designing and coding the search software, and who knows, maybe I’ll post a condensed review of that up here sometime soon. While not quite as romantic as new scientific results, it is nevertheless a significant and necessary portion of my work, so in my new effort to highlight research time, I’ll give a few glances into that in the coming weeks. I’m working on another supernova-related paper to follow up “That Paper”, but I’m putting that on hold until I can get a chance to talk to the observers among my MPA colleagues in a few weeks.
Yesterday (Monday) I returned to the pretty basic question of how to schedule our fields for observation with the telescope. Now that the search software is, as I said, in good shape, it’s probably past time to have a strategy for telling the (now fully robotic) SkyMapper telescope where and how to point and shoot. I already have a piece of hastily written code which can write schedules for the telescope, but I never intended it to be a permanent part of what we do — I just wrote it because the problem of writing test schedules was, how do you say, “in my way”. It’s time I cleaned this code up and put it into full-time production.
The basic idea is to return to a known set of fields repeatedly and regularly, every 3-4 days. But in cases where the weather is bad, or (heavens!) when it is too good for bad-seeing observations, or when telescope failures take out parts of the night, it’s good to have a general strategy for which fields to pick and which to skip encoded in some piece of software which can adjust the schedule, on the fly if necessary. I talked to Patrick Tisserand (SkyMapper postdoc across the courtyard from my office) about the best way to plug a field prioritizer for the search into the main telescope schedule control software (which he maintains).
Other esteemed colleagues of mine who have easy access to expert computer scientists have written nonlinear optimizers for their surveys — kind of like the algorithms that plan airline schedules, only built to wring every last bit of value out of their telescope time. It sounds great, but I don’t have time to do that myself; I’m betting we can get 90% of the way there with a few well-placed assumptions.