1. Simulation & Reconstruction
Emanuele Leonardi
PADME Weekly Meeting - LNF April 27th, 2017

2. Investigation is ongoing
Simulation time EXPLODED
1 event → 25s
i.e. >10x wrt old MC
Possible culprits:
New complex geometry with a lot of extra material
New (much larger) magnetic volume
Outgoing beam particles tracked in air ← tested: 10% effect
Other (less obvious) possibilities:
hits, digis and I/O handling
in-simulation analysis
Investigation is ongoing
27/04/2017
E. Leonardi - PADME W.M. - Simulation & Reconstruction

3. E. Leonardi - PADME W.M. - Simulation & Reconstruction
Magnetic Field
The magnetic field is now handled by the MagneticField dedicated package
Used to be a subsystem of the magnet
It is defined in a magnetic volume that contains:
Vacuum chamber, Magnet, Target, P/E/HEPveto, TPix
ECal and SAC are outside the magnetic volume
E.M. shower simulation time would be affected
Magnetic field map does not include the measured survey yet
Constant field inside the magnet with gaussian tails before and after the magnet
BAD!!!
Effect of magnetic field gaussian tails: ionization e- from target spiralizes out of target region.
27/04/2017
E. Leonardi - PADME W.M. - Simulation & Reconstruction

4. Magnetic Field vs Vacuum Chamber
Tracking time is non-linearly affected by complex geometries
Replaced large magnetic volume outside vacuum chamber with smaller magnetic volumes inside vacuum chamber and target pipes
If magnetic field outside VC is relevant, it can be added.
Tracking times for MeV e+ in vacuum
Zero magnetic field
Realistic magnetic field
No vacuum chamber geometry
0.07 s/evt
0.20 s/evt
With vacuum chamber geometry
7.3 s/evt
Current simulation times for 5000
Standard simulation
All detectors, I/O active, World in vacuum
2.9 s/evt
No physical vacuum chamber
Combined effect of material and geometry
1.7 s/evt
With G4_AIR in World
Effect of tracking the outgoing beam
5.9 s/evt
27/04/2017
E. Leonardi - PADME W.M. - Simulation & Reconstruction

5. ECal Digitization – The theory (1)
A MC hit in ECal is a release of ionization energy Eh along a single tracking step
PreStepPoint: P1=( 𝑥 1 ,t1)
PostStepPoint: P2=( 𝑥 2 ,t2)
kBGOEh Scintillation photons are emitted along the step with distribution ng( 𝑥 0 ,t0,q0,f0,l, 𝑝 0 )
f( 𝑥 ) flat along P1P2
f(qf) flat over 4p
f(t), f(l), f( 𝑝 ) defined by the scintillation process P1 P2
tBGO=300ns
Scintillation emission spectrum of BGO
27/04/2017
E. Leonardi - PADME W.M. - Simulation & Reconstruction

6. ECal Digitization – The theory (2)
Each photon travels along the crystal before reaching the photocathode with probability Pg( 𝑥 0 ,q0,f0,l, 𝑝 0 ) at a time tg( 𝑥 0 ,t0,q0,f0,l, 𝑝 0 ) depending on:
optical characteristics of the crystal surface
absorption spectrum of BGO
active surface of the photocathode
The photocathode converts the photon into a photoelectron with quantum efficiency e(l)
PMT × e- ×
Photocathode
27/04/2017
E. Leonardi - PADME W.M. - Simulation & Reconstruction

7. ECal Digitization – The theory (3)
Each dynode of the PMT is characterized by:
Collection efficiency: probability that an e- from previous stage is collected
Gain: average number of e- emitted for each incoming e-
Transit time spread: time distribution of emitted e- (roughly Gaussian)
N.B. The first dynode is usually different from the others
From these parameters we can compute the average amplification and time delay of the PMT , APMT and DtPMT, but…
…fluctuations are dominated by the first amplification step and the final distributions are NOT Gaussian!
Each photoelectron will contribute to the final signal with qg(t)
27/04/2017
E. Leonardi - PADME W.M. - Simulation & Reconstruction

8. ECal Digitization – “Correct” method
Get E, P1 and P2 from GEANT4
Extract number of initial photons with Poisson(kE)
Assign 𝑥 0 ,t0,q0,f0,l, 𝑝 0 to each photon
𝑥 0 ,q0,f0, 𝑝 0 flat, t0 and l according to BGO distributions
Compute Pg( 𝑥 0 ,q0,f0,l, 𝑝 0 ), tg( 𝑥 0 ,t0,q0,f0,l, 𝑝 0 ), and e(l) to get the time distribution of photoelectrons
For each p.e. compute qg(t) from PMT characteristics
Compute Q(t) = Sgqg(t) and integrate over t
Take into account ADC time acceptance and saturation
27/04/2017
E. Leonardi - PADME W.M. - Simulation & Reconstruction

9. ECal Digitization – 0th order method
Compute distance d of hit from photocathode
Compute a number of photoelectrons Npe=Poisson(kE(MeV)×f(d))
k summarizes the average E → ng process
f(d) summarizes Pg dishomogeneities (f(d)<=1)
Compute final integrated signal Q = qNpe
In this method all E resolution fluctuation come from Npe
sE/E = 2%/√E(GeV) → k = 2.5
Other values used: f(d)=1, q=1 i.e. Q=2.5E with 2%/√E(GeV) resolution
f(d) can be estimated using cosmics
q can be computed from PMT characteristics
27/04/2017
E. Leonardi - PADME W.M. - Simulation & Reconstruction

10. ECal Digitization – Time estimation
Time of the digi tdigi: instant when Q(t) passes a given threshold
Dominated by q(t) of the first photoelectron
tdigi≈t0+d/(c/nBGO)+DtPMT+Dtcables
Assumes that the number of photoelectrons is large
i.e. at least one reaches the photocathode along the shortest path
Should use a distribution for the timing of the first photoelectron
t=tBGO/√Npe(??? Can’t remember)
Currently using:
DtPMT = 23 ns (from XP1911 datasheet)
Dtcables = 0 ns
27/04/2017
E. Leonardi - PADME W.M. - Simulation & Reconstruction

11. E. Leonardi - PADME W.M. - Simulation & Reconstruction
SAC Digitization
Very similar to ECal, but…
Optical photons are produced by Čherenkov effect, i.e.
f(qf) is not flat over 4p: photons’ direction depends on e.m. shower development and must be evaluated from MC
f(t) = d(t), i.e. optical photons are emitted without delay
The number of photoelectrons per MeV is smaller than ECal
Using k=0.5/MeV
Readout PMT is faster: DtPMT = 9.1 ns
Final digi is not integral: simulate the ADC readout waveform
GHz (200 ps per sample)
27/04/2017
E. Leonardi - PADME W.M. - Simulation & Reconstruction

12. E. Leonardi - PADME W.M. - Simulation & Reconstruction
Optical Simulation
GEANT4 allows a detailed simulation of the propagation of optical photons inside a crystal
Useful to estimate digitization parameters for ECal and SAC
Need to correctly define all optical properties of the crystal and of its surface
Need to know material’s light absorption length as f(l)
Surface modeling is the most difficult part
Use different strategies for ECal and SAC
27/04/2017
E. Leonardi - PADME W.M. - Simulation & Reconstruction

13. ECal Optical Simulation
Scintillation photons are emitted isotropically around the energy release point → no need to simulate scintillation
Evaluate Pg( 𝑥 0 ,q0,f0,l, 𝑝 0 ) and tg(t0, 𝑥 0 ,q0,f0,l, 𝑝 0 ) averaging over q0,f0, 𝑝 0 to get a map depending only on 𝑥 0 ,l
Choose binning wisely (x,y dependence is different from z)
Use map to compute E → ng(t)
Cosmic data can be used to tune the simulation
27/04/2017
E. Leonardi - PADME W.M. - Simulation & Reconstruction

14. SAC Optical Simulation
Cherenkov photons are emitted according to the charged particle direction → dependence on e.m. shower development
Need to simulate the full process
g → e.m. shower → Cherenkov process → optical g’s collection
Simulate full signal for each incoming g depending on E(g), position and angle on crystal surface
Analogous to traditional e.m. shower libraries
27/04/2017
E. Leonardi - PADME W.M. - Simulation & Reconstruction

15. E. Leonardi - PADME W.M. - Simulation & Reconstruction
OpNovice
OpNovice, one of the GEANT4 “novice” examples, implements all we need to make a full simulation of the optical properties of ECal and SAC
Started tweaking it to our goals:
Removed original geometry
Added BGO crystal of 21×21×230 mm3
Refractive index depending on l
Absorption length set to 3 m
Spike reflection surface model (dielectric_metal) with 90% reflectivity
27/04/2017
E. Leonardi - PADME W.M. - Simulation & Reconstruction

16. E. Leonardi - PADME W.M. - Simulation & Reconstruction
PadmeMC now produces a ROOT file containing hits/digis for each detector
User can select which formats are saved, e.g.
/output/EnableDetectorIO ECal:HD for ECal save both hits and digi
/output/EnableDetectorIO Target:D for Target save only digi
/output/DisableDetectorIO SAC for SAC do not save anything
The output file can be read by PadmeReco which then calls the reconstructor for each detector found in the data
Currently a simple printout of all hits and digi found
27/04/2017
E. Leonardi - PADME W.M. - Simulation & Reconstruction

17. E. Leonardi - PADME W.M. - Simulation & Reconstruction
ECal clustering
A set of clustering classes have been included in PadmeReco and have been interfaced to the Ecal digis.
27/04/2017
E. Leonardi - PADME W.M. - Simulation & Reconstruction

18. E. Leonardi - PADME W.M. - Simulation & Reconstruction
Conclusions
The simulation time mystery is solved
Fixed by reducing the magnetic volume
A poor man version of the ECal digitization is in place
Needs many improvements
Started working on detailed optical simulation of ECal and SAC
The reconstruction program can now read and process data from MC simulation
ECal reconstruction includes clustering code
Can be used for resolution studies
27/04/2017
E. Leonardi - PADME W.M. - Simulation & Reconstruction