September 22, 2014, Physics Working Group
Meeting Time and Place
The meeting will be on Monday September 22, 2014 at 11:30am EDT. For those people at Jefferson Lab, the meeting will be in room F326.
- To join via a Web Browser, go to the page  https://bluejeans.com/115815824.
- To join via Polycom room system go to the IP Address: 126.96.36.199 (bjn.vc) and enter the meeting ID: 115815824.
- To join via phone, use one of the following numbers and the Conference ID: 115815824.
- US or Canada: +1 408 740 7256 or
- US or Canada: +1 888 240 2560
- More information on connecting to bluejeans is available.
- Collaboration Meeting
- Recent Changes to the Simulation & Analysis Library (Paul Mattione)
- Offline Analysis Plan
- Solenoid Current Studies
There appears to be a discrepancy between some of the more-recent work and what Ryan Mitchell presented in August. For example, his study for the proton-five-pion final state found a total reconstruction efficiency of 4.6% with 86% purity. Michael's numbers for 90% purity seem to be about 1% efficiency. It would be very good to understand what is driving the discrepancies.
In a follow-up email from Matt Shepherd, we get: Our suspicion is that the discrepancy is not in reconstructing signal but that Michael has many more background events than Ryan. It probably means that for similar purity the efficiency appears much less.
As best I can tell Michael isn't tossing out any background events. Ryan requires kin fit FOM chi^2 / dof < 5. That will probably eliminate a lot of Michael's extra combos. Ryan also makes other loose cuts, but they are probably 100% efficient for chi^2 / dof < 5. Michael should implement this.
You can require: KinFitFOM > 1E-5 which is about right.
I think Michael is working under the assumption that the BDT takes care of this since it can select on KinFitFOM, but it is probably drowning in garbage. There is very little with KinFitFOM < 1E-5 that we want to keep for a physics analysis. These are all many track events, each which contribute many combos, none of which conserve four-momentum.
I was able to replicate Ryan's efficiency with Paul's system, so there is nothing about the analysis framework that is causing the problem. However, I implemented the cut on KinFitFOM.