kali calo progress report dasha savrina (itep/moscow), vanya belyaev
TRANSCRIPT
Kali Calo
progress report
Dasha Savrina (ITEP/Moscow), Vanya Belyaev
Vanya Belyaev 2
Iterative p0 calibration
• The “standard” procedure • HERA-B
• Robust (as soon as p0 peak is vizible)• “Millipede-like” algorithms are fragile
• Rely only on “standard” reconstruction technique• No “dedicated” reconstruction
• Can be done “track-independent”
• Requires only limited information
• Can be rather fast ( “on-line” mode)
17 Feb 2k+10
Irina
Vanya Belyaev 3
• Rely on “multiplicative” calibration• 0 ~ Eprs <<Eecal, • the best contribution from Eseed ~ Eecal
• simultaneous Ecal/Prs calibration is difficult• Needed? Sensitivity to Prs is not large
• For physics: Eprs > E0 , for calibration Eprs < E1
• Contradiction?
Eprs > E0 : small background + large statistics
Eprs < E1 : large background + small statistics
17 Feb 2k+10
Vanya Belyaev 4
Eprs
Eprs > E0 : small background + large statistics
- fast convergency to “wrong” constantsEprs < E1 : large background + small statistics
- slower convergency to “correct” constants Combine!
• Few iterations with Eprs > E0
• Then switch to Eprs < E1
• (Intermediate scenario ? ) g1: Eprs < E1
g2: Eprs > E0
17 Feb 2k+10
Vanya Belyaev 5
Data Flow for Kali-p0 (I)
17 Feb 2k+10
Kali-p0 Job
ROOT NTuple/TTreeDST or DAQ
fmDST
Vanya Belyaev 6
Data Flow for Kali-p0 (II)
17 Feb 2k+10
ROOT NTuple/TTree
•Make histos using the current estimate for calibration
constants
• Fit histograms
• Get corrections for calibration constants
Iterate up to convergency
• produce the final set of calibration constants
Set of Calibration
constrants CondDB (?)
(optional)
• 2k+(4-5): 3-5 iterations are OK
• 2k+9 : 2-3 iterations are OK
Vanya Belyaev 7
Data Flow for Kali-p0 (III)
The secondary iterations
17 Feb 2k+10
Kali-p0 Job
ROOT NTuple/TTree
fmDST
CondDB (?)
(optional)
Set of Calibration
constrants
Vanya Belyaev 8
Current status/Progress report
• Samples has been studied:• Monte Carlo, min-bias : 1000M
• Study cuts & fit stability
• Monte Carlo. min-bias, miscalibrated: 200M• Make the real large scale excersize
• 2k+9 collisions : Reco06, Reco07• Apply simplified version to data
• Combine cells into groups
17 Feb 2k+10
Many thanks to Albert for kind help!
Vanya Belyaev 9
MC: bad fits & small statistics
17 Feb 2k+10
Small Statistics <100
p0/cell
Bad Fit
Vanya Belyaev 10
MC: bad fits & small statistics
17 Feb 2k+10
Vanya Belyaev 11
Fit quality
• Some smartness has been introduced:• For small statistics the background shape is
fixed from “per-area” histogram• For “crazy mass, width”, skip the cell from
*current iteration*, …
• Still a lot of room for improvement
17 Feb 2k+10
5953 from
6016
Vanya Belyaev 12
How to improve fit quality?
• Better background parameterization, less degrees of freedom:
• e.g. sqrt(m)*p2(Dm) instead of p3(m)
• Better initial setting from “per-area” histos
• Some constraints against “crazy” values• Skip per iteration with hope to recover for next
iteration
• Release ePrs cuts for “problematic” cells
• Smart: Use Albert’s trick with fake p0 to fix the background shape
17 Feb 2k+10
Puzzle:
200M miscalibrated events: more stable fits…
Pure technical, but important…
Vanya Belyaev 13
200M miscalibrated events
• No more news today.• Calibration coefficients are obtained for the
first iteration• For improvement & stability, the local
reprocessing of fmDST is required (“secondary iteration”)• (Can also send the uncorrected coefficients to Olivier)
• Some pure technical problem with reprocessing of fmDST • It looks like something has been changed
recently in global DaVinci configuration (packing/unpacking/…)
• Should be easy to fix ~1-2 days… 17 Feb 2k+10
Vanya Belyaev 14
2k+9 Collision Data: Kali with 12 zones
• 5799&5800
• 4.2M events
• Mass: 139.3±0.3 →135.0±0.3
• Sigma: 13.5 ±0.5 →11.2± 0.4
• 5842 & 5845
• 0.5M events
• Mass: 139.0±0.2 →135.1±0.2
• Sigma: 13.1± 0.3 →10.9± 0.3
17 Feb 2k+10
Vanya Belyaev 15
Note: p0 mass by Victor
17 Feb 2k+10
Vanya Belyaev 16
Calibration
17 Feb 2k+10
Reco06
Reco07
Vanya Belyaev 17
Instead of summary
• Kali works with MC and real data• Final check with mis-calibrated MC samples
• (fix the reprocessing)
• Kali for Reco06 and Reco07 are in agreement• Some puzzle with PT(p0)-dependence
• Stability of fits need to be improved• It must work in automated regime!• A lot of ideas how to do it
• Selection cuts *must* be reoptimizedThere is some confidence that Kali is
does the job: Ecal calibration with p0
17 Feb 2k+10
Vanya Belyaev 18
BACKUP
17 Feb 2k+10
Vanya Belyaev (Nikhef & ITEP ) 19
Cuts
• Standard Recontruction
• No Spd hit (???)
• ET(g) > 300 MeV
• EPRS < 10 MeV
• PT (p0) > 800 MeV the most discussed cut...
6 Oct 2k+9
Vanya Belyaev 20
Kali-p0: Few plots
17 Feb 2k+10
all
min(Eprs
) > 10 MeV
min(Eprs
)< 10 MeV,
max(Eprs
)> 10 MeV
max(Eprs
) < 10 MeV
Vanya Belyaev 21
Kali-p0 Job
• Regular Gaudi-based job• Actually “stripped-down” version of DaVinci• (optionally) apply constants to Ecal digits
• Calibrate/re-calibrate/mis-calibrate
• (re-recontruct) Calorimeter objectsClusters, Hypos, Neutral ProtoParticles, Photons
• LoKi-based algorithm that acts on LHCb::Particles• StdLooseAllPhotons• Find good p0→gg candidates with loose cuts• Fill n-tuple• (optionally) Destroy TES!
• Write femto-DST
17 Feb 2k+10
Vanya Belyaev 22
Kali-p0: fmDST
• Write only Spd/Prs/Ecal/Hcal digits that make contributions into “good” photons from “good” p0-candidates
• Write in TES-format:Raw/Ecal/DigitsRaw/Spd/DigitsRaw/Prs/DigitsRaw/Hcal/Digits
• 500k minimum bias MC09 events on input:• 380k evens with “good” p0 : 150MB of fmDST
• ~ 330 bytes/event, mainly due to Gaudi overhead• ~ 300GB for 109 available MC09 statistics
17 Feb 2k+10
“Natural” input for Kali Job
Easy to (mis)Calibrate!
Vanya Belyaev 23
Kali-p0: Summary
• (Some) progress in Kali(-p0) framework• Resurrect 2k+(4/5) code
• “Ready” for full-scale test with 109 events
• Few tiny (pure technical) aspects to be solved
• GRID is essential
• fmDST are very useful
My dream: on-line Kali-p0
17 Feb 2k+10
Vanya Belyaev 24
Calibration
17 Feb 2k+10
No preshower cut• min Eprs>10 MeV• min Eprs<10 MeV max Eprs > 10 MeV• maxEprs<10 MeV
• Upper left outer• 0.935 • 0.936 • 0.937• Upper right outer• 0.956 • 0.962 • 0.963• Upper left middle• 0.979 • 0.989 • 0.961• Upper right middle• 0.992 • 0.986 • 0.982• Upper left inner• 1.037 • 1.050 • 1.002• Upper right inner• 1.041 • 1.047 • 1.036• Lower left outer• 0.934 • 0.939 • 0.948• Lower right outer• 0.945 • 0.951 • 0.918• Lower left middle• 0.977 • 0.979 • 0.966• Lower right middle• 0.993 • 0.988 • 0.970• Lower left inner• 1.065 • 1.062 • 1.153• Lower right inner• 1.098 • 1.096 • 1.065