Kinematic fit over constrained in fitted pi0 channels?

[acschick]

I'm looking at events from the gp->pi0(p) and gp->pi0 pi0 (p) reaction filters from the 2017 run period. The ROOT trees I'm running over are located at
/cache/halld/RunPeriod-2017-01/analysis/ver62/tree_pi0missprot__B2/
and
/cache/halld/RunPeriod-2017-01/analysis/ver62/tree_pi0pi0missprot__B2/

My intuition tells me that the kin fit momenta of the two photons summed together should match the DecayingPi0 kin fit momentum--instead I see a large discrepancy. This occurs in both the single pi0 and double pi0 channels.

Apologies for non-consistent color scheme--the plots came from two different students.

Is this a bug? I'm just using the auto-generated DSelector code to pull out the momentum vectors and histogram them.
Example DSelector here: /work/halld/home/shong/channels/pi0missprot/BasicRunScripts/DSelector_pi0missprot.C

e.g. for top plot
RED:
TLorentzVector locPhoton1P4 = dPhoton1Wrapper->Get_P4();
TLorentzVector locPhoton2P4 = dPhoton2Wrapper->Get_P4();
double W2Photon = (locPhoton1P4 + locPhoton2P4).Mag();

BLUE:
TLorentzVector locDecayingPi0P4 = dDecayingPi0Wrapper->Get_P4();
double WDecayingPi0 = (locDecayingPi0P4).Mag()

BLACK:
TLorentzVector locPhoton1P4_Measured = dPhoton1Wrapper->Get_P4_Measured();
TLorentzVector locPhoton2P4_Measured = dPhoton2Wrapper->Get_P4_Measured();
double W2Photon_meas = (locPhoton1P4_Measured + locPhoton2P4_Measured).Mag();

With regards to the issue in the double pi0 channel maybe being caused by a user error in pairing the photons, here is the auto generated DSelector code which I think is pretty suggestive as to which photons has which pi0 as a parent particle:

[zihlmann]

Using Andrews code and running over 10 files then look at the two histogram
DecayingPi0_W and 2Photon_W do not compute because the DecayingPi0_W
is the mass constraint Pi0, hence a delta function, while the 2Photon_W is a distribution
which should not be because it is made from the two photons from the kin fit.
this second plot should also be a delta function.

the code snippet:
double locW2Photon = locPhoton1P4 + locPhoton2P4;
is what is going into the second plot and should be a delta function but it is not?????

Note that there is NO cut on unused energy so all events with at least 2 photons will contribute.

[jonzarling]

Definitely looks wonky. Maybe the interplay between mass constrained pi0 and missing proton is some edge case we haven't encountered before? Like maybe it forgets that a missing proton can still be mass constrained, and this somehow leaks into the calculation of gamma1 and gamma2. Maybe checking the hd_root.root folders with kinfit info or some sanity checks calculating the same quantity different ways will turn something up....

[zihlmann]

The missing Mass looks like the plot below requiring events with exactly 2 reconstructed photons. the red histograms
is after cutting 2 sigma around the pi0 invariant mass.

the invariant mass of the pi0 looks like this:

[zihlmann]

Yes, indeed, I also think that the code somehow does not contrain the pi0 mass the same way as usual when there is a missing proton involved. In the plot below the red histogram represents the measured values while the blue the kinfit values. there are only small changes best seen at the "edges".

[sdobbs]

I'll see if I have some time to look into this in the near future, but let me suggest you redo the fits with the "F1" option, which only imposes the 4-momentum constraints, and removes the vertex constraint from the fit - since there's no charged particle in the final state, there's no possible measurement of the vertex position, so that could be confounding the fit.

As a note, for the gp -> pi0 (p) fit, this should then be a 2C fit - one constraint for the pi0 mass in pi0 -> gg, and 1 constraint for the missing mass in the reaction being consistent with that of the proton. If the F1 fits still show a problem, the next step would be to look at these constraints and make sure they are not beating against each other in some strange way.

[jrstevenjlab]

One more thing to consider as a general analysis strategy: using a mass constraint on the pi0s in these channels means you will not be able to evaluate the non-pi0 background. So unless there is no non-pi0 background or you have some other way of estimating it, I would suggest you also consider the F1 and M7 fit options together to have 1 constraint on the missing mass being consistent with the proton, but allowing you to study the non-pi0 background.