Well, we're all here. But Peter's not. Saul and Alex are talking about errors with each other. I'm typing random things. I should have stayed in my office and kept trying to put together the pieces of the APM fitting, which are absolutely all over the floor of my office at the moment. It's stressful, because I'm out of town in two days, and I'm supposed to put the software all up for Alex Kim to try to install somewhere else next week while I'm out of town and offline. So I've got to get it working now, and I've got to write the Keck AO at the same time.
We seem to be miscellaneously talking about HST data.
Aha. Peter has arrived. Maybe we will actually go to our agenda isntead of the usual individual conversations that we have in group meeting which don't really need to happen in front of everybody else.
Oops, now Saul doesn't want to start because Don isn't here. Saul has gone to get Don. Hmm, so, 10 or 15 minutes later, I could have gotten some more APM repair done.
Peter discusses a paper recently written up primarily by Peter and Greg, with some help from Rob. It will be available in the next couple of days, and Peter says he will send out a web location. He's telling folks that they should look at it, and says that anybody who wants to be on it (and has legitimatly been involved in data taking) should contact him.
What they did was fit the supernovae using a template which has a quadratic rise that meets the Leibengut template at some t_join. There are two parameters in the template, t_exp (explosion date) and t_join. Then there were the usual parameters for each supernovae: t_max, stretch, date of max, zero offset. They did a grid search of this giant data space, computing a chisquare for each supernova at each point on this giant 6-dimensional grid. (The ultimate goal is to find which t_exp and t_join work best for our dataset as a whole.)
After this, they can then collapse down on any dimension or set of dimensions to figure out things.
Greg rewrote snminuit, producing something that instead does a grid search, but does it relatively quickly and efficiently. The grid search has the advantage that it won't get stuck in a local minimum. Peter wrote something from scratch that does a brute force grid hunt. Greg did a cleverer grid search that runs faster; it was checked against Peter's brute force code (the latter running on the T3E).
Gerson asks if the correlated errors were considered. Peter says that yes, they were considered (correlated photometry). Also, because the fit parameters are correlated, that's implicitly handled because it will warp the shape of the chisquare surface in the giant chisquare hyperspace.
Peter shows the chisquare surface for t_join and t_exp. I refer you to the paper on the group web page (when it gets there) to see the actual image. It shows points for the SCP data and Adam's data. (Not quite all of Adam's supernovae were used, since some data weren't published.) The bottom line is that the only difference between these results and Adam's results are that the error bars are way bigger. Adam's fit is within two sigma of the fit for the SCP data. (This in contrast to the 5.8-sigma that Adam quotes.)
There is some discussion about the sigma difference between our minimum and Adam's point, using our errors. Peter and Greg have 1.4 sigma as the difference, Gerson has 1.2 sigma as the difference. Should we worry about this discrepancy? Saul wants to know if these are exactly the same calculation; Peter says that they can't be given that one used snminuit, and the other used a grid search.
Other little fiddly discrepancies, much discussion.
Peter next talks about the test for systematics that he's done. He points at two supernovae (nearby supernovae done way back when; 86G and 94D) which have nice data over peak, and extend on to 75 days. If you match them up to 15 days after max, the late time temlpates differ a little bit. It's not by much, but it does differ. The difference is something like +/- 2% at 40 days after max.
"I was up at 6 AM putting oil based stain on my deck, and the fumes have clearly gotten to me."
--P. Nugent
Peter took up ungodly amounts of T3E time and started with these three templates. He generated thousands of fake supernovae, given the sampling from the 34 SCP SNe (from the 42 paper, selected for a B-R redshift, blah blah blah). He then went through for each template (Leibengut and the two different ones, but using a t_exp=20 and a t_join=10 for the beginning), and then fit the supernova to a Leibengut template (which is what we did for our 42 SNe paper) He found that using the different templates shift the values for t_exp and t_join.
The Leibengut template, happily, returned the same t_exp and t_join. The other two returned things with t_exp different by a couple of days. What this means: we have data out to 40 or more days, which has enough statistical weight to affect the rise measurements. Because we don't really know the template, there could be statistical things which futz up the measurement of the rise time. Peter says there aren't enough supernovae to do a real study with any sort of real population. However, it's just a warning that the potential range of systematic error is 1.5 days or more.
Peter then worried, well, how much did we screw up the cosmology if this is going on? He also looked at the mb_corr you get using supernovae generated from the three different templates but fit with just the Leibengut template. The difference in delta mb_cor was 0.04 and -0.02 for the two not Leibengut supernovae. This is less than the systematics we we already considering in the 42 SNe paper. If we ever want to do better than 0.4 magnitudes, we're eventually going to have to consider this issue more seriously. However, right now, these considerations don't substantively affect any of our conclusions from our 42 SNe paper.
Peter repeats that the paper will be posted, and says that people can tell him how he feels about it and whether they want to be on it. Peter says he will submit the paper to ApJ for a quick refereeing job. (It's too long for a letter.) He says he won't put it on LANL until after the paper has been refereed. He speaks very strongly against putting something up before it's refereed. (Peter notes that if he had received Adam's paper to referee, he would have ripped it up.)
Peter gives a crash course in supernova physics.