[Halld-pid-upgrade] thoughts after PID session

[Yi Qiang]

Hi,

I agree that we need to have as many combinations as possible when doing 
the kinematic fit and use the Chi2 (or likely-hood if available in the 
future) to determine the best guess. Furthermore, since our detector 
coverage and efficiency are not perfect, with more particles in the 
final states, we should start to think about the possibility to have 
pions or kaons missing in the final reconstruction as well. This will 
certainly make the list longer.
Regarding the proton reconstruction. In addition to the evaluations of 
the PID performance with proton reconstructed or ignored, I think we 
need to include the proton reconstruction efficiency as a weighting 
factor to make our decision. At the end, we may want to use the proton 
information as long as its available and only assume it missing in the 
fit when the proton can not be reconstructed.

Cheers,
Yi
On 10/8/2012 2:02 PM, Michael Williams wrote:
> Hi Matt,
>
> Sounds good.
>
> Paul, did I hear correctly that material interactions are not included in the study so far?  Pretty important ... if not b/c of time constraints so far that's totally understandable ... but pretty important to have this in before making any decisions (or showing to any external people).
>
> For MC vs data resolution I guess the following two "laws" hold: (1) the real alignment will never be perfect, especially for one subsystem w.r.t. another and (2) there is always more material where material isn't measured then expected.   Law (1) can probably be expected to reduce our resolution by 10%.  Law (2) seems to always hold for things like glue where nobody measures it when using it and simulations tend to be based on some ideal minimum amount required to hold things together (although why one expects the undergrads doing the glueing to achieve this minimum I don't know).    I don't think anybody would complain if we simply assumed 10-20% less than ideal resolution for the PID studies ... we can say "based on all experiments ever run" ;-)
>
> Anyway, I totally agree that more channels is better.  There will be correlations b/t what's in the final state and how important the PID is and it'd be good to know what these are.   Are you including KS, Lambda, Sigma in this list?  The KS will be used I'm sure.   If we're interested in hybrids with open strangeness I would assume they are often times produced with Lambdas and Sigmas.    These topologies help on their own (displaced vertex) but still worth looking at how PID helps here if we're interested in these channels.
>
> Another question I had (maybe this is obvious if you've been looking at GlueX MC for years) was how much different (mainly in "flatness") the efficiencies are when the proton is required vs ignored?  Does it matter?  For PWA S/B isn't the only metric.    Also, for honesty's sake, does requiring that the kaon is hard enough to pass the PID hurt the uniformity of the efficiency?   The channels we've looked at have a very hard kaon nearly 100% of the time so not an issue there but it is something worth considering.
>
> Cheers,
>
> Mike
>
>
> On Oct 8, 2012, at 12:53 PM, Matthew Shepherd wrote:
>
>> Hi all,
>>
>> It would be a good idea to establish a PID-upgrade mailing list that we can send all mailings to.  Can someone do this?  I expect the emails will only increase in importance over the next month and everyone should be in the loop.
>>
>> Now that people are aiming to produce a large Pythia sample and we have easy tools for analysis, it would be very interesting to write a generic algorithm to reconstruct as many exclusive final states as possible.  Just go down the list with K, pi and maybe eta and include as many combinations with recoil proton or neutron that conserve strangeness.  For each, do a kinematic fit, vertex, fit, and maybe constrain the pi^0 mass.  For those with a recoil proton you can study requiring or not requiring it.  You can come up with some rudimentary cuts on chi^2, track multiplicity, PID, etc. that work OK.  Then with just the Pythia sample study the signal purity for each of those final states under different PID options.  So, instead of trying to put in some resonance and hand tuning the analysis, you are taking the Pythia-predicted cross sections for the various topologies and seeing how pure the reconstruction is.  This may allow you to notice more broad patterns in PID behavior (e.g., channels with pi0 are worse, reconstructing the proton is always better, etc.)  Even though we do not optimize any one analysis we may be able to compare the benefits of various PID solutions for a wide range of topologies.
>>
>> (Side note:  Ryan has some code that he has used on CLEO and BES data to reconstruct pretty much every reasonable permutation of ten or less stable hadrons, do kinematic fits, and generate plots.  We have these huge catalogs of ~1000 sets of plots for any final state in various exclusive decays.  This is great when trying to understand detector performance or look for interesting channels to analyze.  Maybe his code can be adapted to Paul's framework.  Ryan also wrote his own framework for BES and CLEO, but hadn't got around to adapting it for GlueX yet.)
>>
>> Finally, Eugene pointed out that it may be helpful to explore trying to degrade our resolution in the MC somehow.  My instinct is that the signal purity Paul gets by reconstructing the proton and doing kinematic fitting is too good.  (Every experiment I've worked on required that the resolution be degraded in MC to get chi^2 distributions that match data for kinematic fitting.)  The danger is that people conclude we don't need to PID system because our signal purity is so good with just kinematic fitting.  This is really bad if our MC is too good.  We have to think of some justifiable quantitative way of degrading the MC resolution so it doesn't look like we're just trying to make an excuse for building a PID system when our simulation is telling us otherwise.
>>
>> Matt