sphenix-hcal-l AT lists.bnl.gov
Subject: sPHENIX HCal discussion
List archive
Re: [Sphenix-hcal-l] comments on HCAL simulation study slides.
- From: Edward Kistenev <kistenev AT bnl.gov>
- To: John Lajoie <lajoie AT iastate.edu>
- Cc: sphenix-hcal-l AT lists.bnl.gov
- Subject: Re: [Sphenix-hcal-l] comments on HCAL simulation study slides.
- Date: Wed, 7 Oct 2015 08:43:38 -0400
Hi, John,
I have couple comments to this study
- the estimator for the nongaussian tail is good, we finally have something qualitative to judge the outcomes;
- the tilt angle was chosen to make sure the overlap between towers is no more and no less then one tower. Larger overlap artificially spreads showers, smaller overlap introduces nonuniformities into many aspects of pattern recognition. It also preludes using energies seen in neighboring towers in a single calorimeter section for LCG estimates (when impact point is known);
- the HCal as it is today already has sampling fraction correction - it consists of two section with different “balk” sampling fractions;
- the High/Low tails appearance/disappearance is the feature of e/h different from one, not much to be done except probably - either measuring shower profile in timing domain (DRS on every channel) or further degrading the sampling (this is already done by design - higher energy showers penetrating deep do see the low SF structure - e/h->1 - so they are less affected by e/h but do have larger uncorrected leakage);
- the energy as it is estimated (with or without nongaussian tails) is not corrected for leakage. Unfortunately this part of the algorithm is prone to autocorrelations and what not (especially in jets). We may ignore it just saying that our energy estimates are already good enough for physics we are doing or try to use calorimeter design to better it (use azimuthal tilt and complementary shower measurements in more then one tower). This is a bit speculative - I never tried to implement such thing.
Other then that - 50 years of calorimetry in physics left us with a lot of blind spots to explore and I am still struggling to understand our T1044 data - were they simply affected by bad calorimeter alignment in the beam (it was slowly descending vertically while taking data) or there was more to it. Something for the planners.
Edward
On Oct 6, 2015, at 9:32 PM, John Lajoie <lajoie AT iastate.edu> wrote:
Dear HCAL Folks,
My apologies for not being at the HCAL meeting today, something came up that I had to deal with on short notice. I wanted to send around a few words to go with the slides I posted on the Indico agenda:
https://indico.bnl.gov/conferenceDisplay.py?confId=1423
to try to keep the conversation going and let people know what I've been working on, especially since there's not a lot of words on the slides.
Slides 3-4 show the resolution and non-Gaussian tails for pi- in the G4 simulation of the reference design. The error bars are small because I have been using the high-statistics samples that Chris has recently added. To make these plots I did the following:
-> Use the combined 10-50GeV events to determine the average sampling fraction in the EMCal, inner and outer HCAL
-> Using these sampling fraction I generated the reconstructed energy distributions for each energy and pseudorapidity and fit the distributions to a Gaussian.
-> The sigma and mean of the Gaussian were used to generate the resolution plots
-> Using the Gaussian sigma I determined the fraction of events above and below 2-sigma (2.5% for each would be a true Gaussian). This is on quantitative way to try to characterize the tails.
On slide 4 you can see that at low energies there is a small but significant non-Gaussian high side tail that decreases with energy. There is a more significant low-side non-Guassian tail that increases with energy.
Slide 5 attempts to explain the high-side tail at low energies by examining 1GeV events. The high side tail seems to originate from events that are on the high side of the sampling fraction distribution in the inner HCAL - they are oversampled, or they put significantly more energy into the scintillator.
Slide 6 attempts to examine the low-side tail that develops with energy. This is more complicated as it appears to have two sources. The first is *undersampling* in the outer HCAL, and the second is significant leakage for some events that shower deep.
Slide 8 is a look at the sampling fraction vs. radius in the inner and outer HCAL. This plot was generated by collecting the visible energy in 2cm bins and in groups of 10k events, dividing by the total energy deposited in the 2cm bin. The event averaging is necessary because event-by-event fluctuations make it very difficult to determine the sampling fraction at the furthest extent of the outer HCAL.
Slide 9 shows linear fits to the sampling fraction vs. distance along the scintillator tile for the inner and outer HCAL, along with the correction factor for each. The correction factor is a target sampling fraction (the sampling fraction at the furthest extent of the tile) divided by the linear fit. The target SF in the inner HCAL is 0.058 and 0.028 in the outre HCAL.
Ideally this correction would be worked into the G4 code and applied as a correction to the visible light for each G4 step. As a first pass I applied the correction externally to each G4 hit, using the correction factor at the midpoint of the hit. Slide 10 just shows that this results in a flat SF in the inner and outer HCAL.
Slides 11/12 show the resolution and non-Gaussian tails with the modified sampling fractions. Compare these to slides 3/4 without the corrections - they are almost indistinguishable. Slide 13 shows the energy distributions for 50 GeV pions. *This needs to be repeated with the correction factor at each G4 step, but what I take away from this is that the light yield correction to even out the sampling fraction has no effect at all on the energy resolution OR the non-Guassian tails, at least in the MC.* This is very consistent with earlier results from Liankun.
If this holds up it should give us pause to consider if spending R&D time and resources on graded tiles is a worthwhile exercise....
Slides 15+16 give some basic information about the entry angle of particles into the inner and outer HCAL, and the conversion between "crossings" in the G4 studies and the approximate tilt angle. Note that in these studies BOTH the inner and outer HCAL were modified at the same time.
Slide 17 shows the sampling fraction vs. the number of crossings extracted from the files that Chris prepared. The inner HCAL sampling fraction increases rapidly and seems to level off by 10 crossings, while the outer HCAL SF rises slowly.
Slides 18-33 show the resolution and tails for the different crossing angles. They are arranged in the slides so you can flip between them to see how things change. These are lower statistics files, but within the statistics you can see that:
-> The resolution doesn't change dramatically, but seems to worsen for eta=0.9 at 10-15 crossings.
-> The high side tail at low energy seems worse at 2-3 crossings, then decreases
-> The low-side tail at high momentum evolves slowly, and seems worse than the reference design at 10-15 crossings.
Slides 35-53 repeat the tilt angle study for protons, same message.
From the tilt study I conclude that there is nothing special about the tilt angles we have chosen for the inner and outer HCAL, but nothing to indict them either....
Regards,
John
Contact me: john.lajoie
My apologies for not being at the HCAL meeting today, something came up that I had to deal with on short notice. I wanted to send around a few words to go with the slides I posted on the Indico agenda:
https://indico.bnl.gov/conferenceDisplay.py?confId=1423
to try to keep the conversation going and let people know what I've been working on, especially since there's not a lot of words on the slides.
Slides 3-4 show the resolution and non-Gaussian tails for pi- in the G4 simulation of the reference design. The error bars are small because I have been using the high-statistics samples that Chris has recently added. To make these plots I did the following:
-> Use the combined 10-50GeV events to determine the average sampling fraction in the EMCal, inner and outer HCAL
-> Using these sampling fraction I generated the reconstructed energy distributions for each energy and pseudorapidity and fit the distributions to a Gaussian.
-> The sigma and mean of the Gaussian were used to generate the resolution plots
-> Using the Gaussian sigma I determined the fraction of events above and below 2-sigma (2.5% for each would be a true Gaussian). This is on quantitative way to try to characterize the tails.
On slide 4 you can see that at low energies there is a small but significant non-Gaussian high side tail that decreases with energy. There is a more significant low-side non-Guassian tail that increases with energy.
Slide 5 attempts to explain the high-side tail at low energies by examining 1GeV events. The high side tail seems to originate from events that are on the high side of the sampling fraction distribution in the inner HCAL - they are oversampled, or they put significantly more energy into the scintillator.
Slide 6 attempts to examine the low-side tail that develops with energy. This is more complicated as it appears to have two sources. The first is *undersampling* in the outer HCAL, and the second is significant leakage for some events that shower deep.
Slide 8 is a look at the sampling fraction vs. radius in the inner and outer HCAL. This plot was generated by collecting the visible energy in 2cm bins and in groups of 10k events, dividing by the total energy deposited in the 2cm bin. The event averaging is necessary because event-by-event fluctuations make it very difficult to determine the sampling fraction at the furthest extent of the outer HCAL.
Slide 9 shows linear fits to the sampling fraction vs. distance along the scintillator tile for the inner and outer HCAL, along with the correction factor for each. The correction factor is a target sampling fraction (the sampling fraction at the furthest extent of the tile) divided by the linear fit. The target SF in the inner HCAL is 0.058 and 0.028 in the outre HCAL.
Ideally this correction would be worked into the G4 code and applied as a correction to the visible light for each G4 step. As a first pass I applied the correction externally to each G4 hit, using the correction factor at the midpoint of the hit. Slide 10 just shows that this results in a flat SF in the inner and outer HCAL.
Slides 11/12 show the resolution and non-Gaussian tails with the modified sampling fractions. Compare these to slides 3/4 without the corrections - they are almost indistinguishable. Slide 13 shows the energy distributions for 50 GeV pions. *This needs to be repeated with the correction factor at each G4 step, but what I take away from this is that the light yield correction to even out the sampling fraction has no effect at all on the energy resolution OR the non-Guassian tails, at least in the MC.* This is very consistent with earlier results from Liankun.
If this holds up it should give us pause to consider if spending R&D time and resources on graded tiles is a worthwhile exercise....
Slides 15+16 give some basic information about the entry angle of particles into the inner and outer HCAL, and the conversion between "crossings" in the G4 studies and the approximate tilt angle. Note that in these studies BOTH the inner and outer HCAL were modified at the same time.
Slide 17 shows the sampling fraction vs. the number of crossings extracted from the files that Chris prepared. The inner HCAL sampling fraction increases rapidly and seems to level off by 10 crossings, while the outer HCAL SF rises slowly.
Slides 18-33 show the resolution and tails for the different crossing angles. They are arranged in the slides so you can flip between them to see how things change. These are lower statistics files, but within the statistics you can see that:
-> The resolution doesn't change dramatically, but seems to worsen for eta=0.9 at 10-15 crossings.
-> The high side tail at low energy seems worse at 2-3 crossings, then decreases
-> The low-side tail at high momentum evolves slowly, and seems worse than the reference design at 10-15 crossings.
Slides 35-53 repeat the tilt angle study for protons, same message.
From the tilt study I conclude that there is nothing special about the tilt angles we have chosen for the inner and outer HCAL, but nothing to indict them either....
Regards,
John
Contact me: john.lajoie
Sphenix-hcal-l mailing list
Sphenix-hcal-l AT lists.bnl.gov
https://lists.bnl.gov/mailman/listinfo/sphenix-hcal-l
-
[Sphenix-hcal-l] comments on HCAL simulation study slides.,
John Lajoie, 10/06/2015
- Re: [Sphenix-hcal-l] comments on HCAL simulation study slides., Edward Kistenev, 10/07/2015
Archive powered by MHonArc 2.6.24.