[Halld-cal] [EXTERNAL] FCAL calibration follow-up

Thu Apr 30 18:11:42 EDT 2020

Thank you Matt for confirming the procedure.

tks ig.
________________________________
From: Shepherd, Matthew <mashephe at indiana.edu>
Sent: Thursday, April 30, 2020 4:33 PM
To: Colin Gleason <gleasonc at jlab.org>
Cc: Igal Jaegle <ijaegle at jlab.org>; Alexander Somov <somov at jlab.org>; halld-cal at jlab.org <halld-cal at jlab.org>
Subject: Re: [Halld-cal] [EXTERNAL] FCAL calibration follow-up

Colin and Igal,

I think this iteration between gains and non-linear correction is not so intentional in the existing procedure but happens as a byproduct of what is done.  The plug-in uses corrected showers in its construction of the pi0 mass on a block by block basis.  Those corrections are based on the previous calibration, which is then revisited at the block-by-block gains are set.

I think this is a small effect overall, but potentially visible on the ~< 1% level things that Igal is attempting to prob.

To specifically answer Igal, I would take those numbers, C,D,E and put them into nonlinear correction.  Then reiterate the gain balancing until it converges.  Then recheck and see how much the values of CDE change.  If significant, consider updating and re-iterating gain balancing.  Then check for incremental changes in gain.. probably tiny.

In doing this you remove the effect of the correlation between position and average energy for pi0 gammas, but I suspect this is very mild so few iterations are needed.

Matt

On Apr 30, 2020, at 4:15 PM, Colin Gleason <gleasonc at jlab.org<mailto:gleasonc at jlab.org>> wrote:

Historically, I wait until the "end" of the calibration procedure to implement the energy dependence correction. By the "end", I mean when the pi0 mass and width stabilize at a fixed value. The pi0 mass generally stabilizes close to the pi0 mean, after 3-4 iterations. The width will eventually stabilize around ~8 GeV and takes a few more iterations to become flat, but is generally close to the stable value around 5 iterations.

After the mass and width have stabilized, I then do the energy correction and see that the pi0 mass will be aligned with the true value, and the width will remain constant. Some one correct me if I'm wrong, but I always thought of the gain calibration as the method to bring the width to its most consistent value, then the energy correction is used to shift the pi0 mass to where it should be. I do not know what the effect of implementing the energy calibration after each iteration is, but I guess it would generally speed up the calibration process?

On Thu, Apr 30, 2020 at 4:00 PM Igal Jaegle <ijaegle at jlab.org<mailto:ijaegle at jlab.org>> wrote:
Colin&Matt,

Thank you for your help. I came back to iteration 0 i.e. no correction applied and extracted following the IU procedure: C=0.95502, D=0.0479489, and E=0.0132508. Now if I understand, I have to apply the IU energy dependence correction, iterate the gains again, and then re-run the IU energy dependence correction, check if I need to iterate the gain again and basically do this gains calibration and extract energy dependence until there is no change in the gains and/or the energy dependence correction. And only then look for the ring dependence if any. Let me know if I got it right?

tks ig.
________________________________
From: Shepherd, Matthew <mashephe at indiana.edu<mailto:mashephe at indiana.edu>>
Sent: Thursday, April 30, 2020 2:51 PM
To: Colin Gleason <gleasonc609 at gmail.com<mailto:gleasonc609 at gmail.com>>
Cc: Igal Jaegle <ijaegle at jlab.org<mailto:ijaegle at jlab.org>>; Alexander Somov <somov at jlab.org<mailto:somov at jlab.org>>; halld-cal at jlab.org<mailto:halld-cal at jlab.org> <halld-cal at jlab.org<mailto:halld-cal at jlab.org>>
Subject: Re: [Halld-cal] [EXTERNAL] FCAL calibration follow-up

Colin,

Igal just needs to edit one or two lines of code and he will have the baseline GlueX approach and can easily explore a ring dependence.

I would advocate for leaving the initial energy dependent correction in for the first iteration on gain balancing as it removes the potential for bias and apparent "ring dependence" that I noted in my message.  One can then recheck/iterate after the first round.

Matt

On Apr 30, 2020, at 2:45 PM, Colin Gleason <gleasonc609 at gmail.com<mailto:gleasonc609 at gmail.com>> wrote:

If you think it would help, I can run the procedure I have for GlueX production over one or two of the run ranges to establish a "baseline" that could act as a reference point. I can set the energy calibration to 0 as Igal did and iterate from there. If I were to start today, I could probably have a decent estimate of what the gains, correction, and pi0 width would be by Monday/Tuesday. I could also implement the crystal ball + exp(polynomial) fit function and compare the current function I have to this one. We have runs  60819-61751 on the IU cluster, so I wouldn't have to wait on the farm to get files and process jobs.

-Colin

On Thu, Apr 30, 2020 at 1:17 PM Shepherd, Matthew <mashephe at indiana.edu<mailto:mashephe at indiana.edu>> wrote:

Hi Igal and Sasha,

I'll start the discussion we weren't able to have today....

I have two concerns with Igal's proposal:

1) The outer ring gain calibration is dubious.  There is no reason that the blocks around (row,col) = (27,0) should have dramatically higher gains on average than those at (20,20).  The detector is cylindrically symmetric as far as I know.

I don't see how anything meaningful can be extracted from the plots on slide #5.  When the detector was commissioned, the HV was set on the outer blocks in the very same way that it was on the rest of the detector, using bench measurements of PMT gains.  We found gain variations were minor in the middle of the detector once we started studying real pi0's.  And therefore we expect the outer rings, where we have maintained constant gains and set HV in the same way, to behave in the same way.  It is highly unlikely they were all systematically set in high or low groups in the strange pattern that your analysis suggests.

2)  I'm not convinced there is any evidence of any ring dependence in the detector response (aside from perhaps some small leakage into the beam hole).  Compare Igal's ring 2 plots to ring 15.  The shape of the background is dramatically different.  It is not evident that the 1-2% variation between these rings observed on page 9 is real effect.  I think some of the trends you see on slide 9 are related to systematic effects in parameterizing the background in the fits.  For the plots on slide 11, 13, or 15, try varying the fit range substantially or increase the order of the polynomial for the background and see how much the pi0 mean varies.  You need to convince us that the *systematic* uncertainties on the mean are less than 1%-2% across a dramatic variation of background shapes.

If there were any ring dependence in the response, then the gain balancing procedure, if it used fixed energy photons, would remove this because it fixes the pi0 mass on a block-by-block basis.  If the gain balancing is done with all photon energies that are not corrected for non-linearities, then the block-by-block gains are susceptible to being biased by energy non-linearities because the average energy in each block probably depends on distance from the beamline.  Indeed these are intertwined and in the absence of a nice mono-energetic sample one may determine a non-linear correction and then go back and redetermine the gains, and then revise the non-linear correction.  This type of iteration has effectively happened over time in the standard procedure.

The existing calibration function addresses concerns about extrapolation to high energy as it is constructed to both match the observed performance and have a stable asymptotic behavior.  Igal's function on slide 9 uses a 5th order polynomial in energy, and such polynomials tend to be incredibly unstable in any extrapolation.

I'm certainly not claiming that the existing approach is complete or, most importantly, meets the precision needs of PrimEx.  However, it would be really nice if you would first demonstrate the existing approach does not work or breaks down at some desired level of precision.  It is trivial to take the framework and data sample you have and just "turn the handle" with the existing function we have used for production up to now, iterate, and study the ring dependence of the response.  Why not do that first?  Once you do this, then we can realize where it is deficient and address those deficiencies directly.  It is much easier to iteratively improve a strategy than evaluate something entirely new from scratch.  Also, this new proposal you have increases the number of calibration constants from 6 to 138.  It would be nice to be sure such enhanced complexity actually resulted in real improvements.

I'm happy to talk about this more and maybe it is easier to do it in a call than email.  If you want to have a dedicated informal meeting to discuss just this, we can, and we can invite anyone else who has input to this conversation.

We have already a JEF meeting tomorrow morning and I have two other meetings tomorrow.. and your work has now come up at two consecutive working group meetings spanning almost 3 hours and we have yet to have the time to really dive into the details..  I'm not sure the PrimEx meeting tomorrow is going be an opportunity to do that.

Cheers,

Matt

_______________________________________________
Halld-cal mailing list
Halld-cal at jlab.org<mailto:Halld-cal at jlab.org>
https://mailman.jlab.org/mailman/listinfo/halld-cal

--
Colin Gleason
Postdoctoral Fellow
Indiana University
Department of Physics

--
Colin Gleason
Postdoctoral Fellow
Indiana University
Department of Physics

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.jlab.org/pipermail/halld-cal/attachments/20200430/16f75028/attachment-0001.html>