Hello,
while I am getting towards the final stages of parts of my analysis I am wondering how to properly take care of multiple combinations in one event. By that I do not mean multiple combos from random tagger hits. This was well explained by Richard in a tutorial given at a Physics WG meeting. I mean the additional combos arising from allowing additional tracks or showers in the event. I think I brought this topic up a couple of times with various people and so far the consensus was always: This is not an easy topic and we should discuss this more.
I wrote a short document stating what the current recommendations are for uniqueness tracking and where I see problems with that. You find it here: https://halldweb.jlab.org/doc-private/DocDB/ShowDocument?docid=3993
Short summary of the problems (not complete but meant as a starting point to think about uniqueness tracking):
- histogram the phi angle of a proton to measure the (e.g. pi0) beam asymmetry:
what happens if you have two possible protons that survive all cuts. Which one should we take? Going by the current recommendations we should take both, but this inflates the signal (and messes up errors)
- histogram variables from kinematic fit four-vectors:
if there are additional tracks/showers all particles will have slightly different values based on which subset (i.e. combo) of tracks/showers was used for the fit
- histogram inv. mass in bins of momentum transfer (measure dsigma/dt):
Stay with gp->pi0 p plus an additional track that could be the proton as example. The inv. mass only depends on the photons from the pi0 decay but the momentum transfer can be calculated from the proton. So depending on which track is used as proton the event could even end up in different bins.
A possible solution (not tested, just an idea):
Don't do uniqueness tracking anymore. Instead always take all combinations. If after applying all cuts (i.e. using all the discriminatory information you have on a given event) n combos survive an event that do not originate from beam photons than weight the combo with 1/n
I admit this might not be a huge effect for some analyses but given that we really push the statistical uncertainties to less than 1% in some channels it might be something that biases the results and/or their errors. In any case it is surely something we want to get right.
I hope that this triggers a discussion that yields some sort of definite prescription of how to properly account for multiple combinations.
Hello,
while I am getting towards the final stages of parts of my analysis I am wondering how to properly take care of multiple combinations in one event. By that I do not mean multiple combos from random tagger hits. This was well explained by Richard in a tutorial given at a Physics WG meeting. I mean the additional combos arising from allowing additional tracks or showers in the event. I think I brought this topic up a couple of times with various people and so far the consensus was always: This is not an easy topic and we should discuss this more.
I wrote a short document stating what the current recommendations are for uniqueness tracking and where I see problems with that. You find it here: https://halldweb.jlab.org/doc-private/DocDB/ShowDocument?docid=3993
Short summary of the problems (not complete but meant as a starting point to think about uniqueness tracking):
what happens if you have two possible protons that survive all cuts. Which one should we take? Going by the current recommendations we should take both, but this inflates the signal (and messes up errors)
if there are additional tracks/showers all particles will have slightly different values based on which subset (i.e. combo) of tracks/showers was used for the fit
Stay with gp->pi0 p plus an additional track that could be the proton as example. The inv. mass only depends on the photons from the pi0 decay but the momentum transfer can be calculated from the proton. So depending on which track is used as proton the event could even end up in different bins.
A possible solution (not tested, just an idea):
Don't do uniqueness tracking anymore. Instead always take all combinations. If after applying all cuts (i.e. using all the discriminatory information you have on a given event) n combos survive an event that do not originate from beam photons than weight the combo with 1/n
I admit this might not be a huge effect for some analyses but given that we really push the statistical uncertainties to less than 1% in some channels it might be something that biases the results and/or their errors. In any case it is surely something we want to get right.
I hope that this triggers a discussion that yields some sort of definite prescription of how to properly account for multiple combinations.