...
Richard's core of a proposal for how the data should be organized for access remains posted.
Likelihood analysis
From Jim:
"Fixed a bug in binned likelihood for point sources very close the edge of the map.
Refactored the Gaussian quadrature integration used by the IRF (and other) packages to handle error conditions more gracefully."No news, owing to my being late in asking. Jim is thinking through implementing a method for limiting accumulation of numerical rounding errors in the evaluation of the likelihood function for large data sets. The log(likelihood) is a sum of many terms, each typically of small magnitude. Andy Strong pointed out one way to do this, basically by grouping terms. This is not going to affect the significances (TS values) very much but Jim has pointed out that it will make a difference in the tail of the distribution of TS in the null hypothesis.
Analia and I spoke briefly last week about her offer to implement upper limit calculations in gtlikelihood. I think that the sensible methods that she proposed all depend on knowing the distribution of TS in the null hypothesis, and so I don't think that we can calculate meaningful upper limits yet.
...
"Corrected behavior of exposure_map to address a problem reported by Jean Ballet, in which the energy bins were not correctly determined in the case where the user does not supply a count map. The exposure_map application was changed to respect the bincalc parameter in this case, allowing the user to choose whether to use the centers or edges of energy bins. This means the bincalc parameter behaves the same way for cases both with and without an input count map."
Source Catalog
Met this week. Jean presented results from an investigation of the bright sources that were missed in the SC1 catalog pipeline analysis and described work on getting the 'bands' and 'time' pipelines going to calculate fluxes for several bands and time histories of fluxes for the SC1 sources. We had an interesting discussion about error matricies reported by Minuit which transitioned to discussion of the shape of the likelihood function near minimum.