FCAL Reconstruction in b1pi events 03/14/2012

From GlueXWiki
Jump to: navigation, search

This is mostly an update of FCAL Reconstruction in b1pi events 03/06/2012, but using svn rev 8894, which includes changes to the cluster to charged track matching code.

50,000 b1pi events. No EM background.

Thrown photons

PhotonEvsTheta thrown FCAL Feb12.png

Reconstructed photons

(after vetoing clusters matched to charged tracks)

PhotonEvsTheta recon FCAL.svn8894.png


  • Why are so many "photons" reconstructed? How can we reduce this?
  • What is the actual reconstruction efficiency?
  • How many photons reach the FCAL without converting?
    • Efficiency for non-converting photons?
    • Can we recover converted photons?
  • Energy resolution, etc.

1 GeV photon events

Mihajlo in GlueX-doc-823 (2007) says for 1 GeV photons:

  • 37% photon conversion before FCAL
  • 99.6% efficiency for photons that reach FCAL
  • Taking into account single clusters from photons that converted close to the FCAL face, the overall photon reconstruction efficiency was estimated to be 77%
  • 4% energy resolution at 1 GeV


Energy distribution of thrown photons:

  • Black=all thrown photons
  • Red=photons which reach the FCAL without converting (determined using DFCALTruthShower object)
  • Blue=photons which reach the FCAL without converting AND are reconstructed successfully (fabs(theta_thrown - theta_recon) < .006 && fabs(phi_thrown - phi_recon) < .15)

PhotonE thrown FCAL all truthshower truthshower+recon.svn8894.png

Red/Black=% of photons that reach FCAL without converting:

Photon FCAL truth shower.png

Roughly 20% of photons convert before reaching FCAL. Relatively constant across wide range of energy. There is an angular dependence, not shown here. Big improvement since 2007 (geometry change?)

Unconverted photons


Photon FCAL recon eff.svn8894.png

Above 1 GeV, 94% of unconverted photons are reconstructed. This is less than the 99.6% reported for single photon sample; this difference is due to photon showers rejected due to overlap with charged particle shower.

Below ~600 MeV, efficiency drops. This is due to showers contained within a single cell, which means no cluster is created.

Overall FCAL reconstruction efficiency considering only unconverted photons: 70%. Could be higher after trying to recover photons that convert.

After conversion

For photons that do convert, we use different criteria to determine if a photon is reconstructed "successfully": (E_recon - E_thrown)/E_thrown < 3.5*sigma_E && fabs(theta_thrown - theta_recon) < .01 && fabs(phi_thrown - phi_recon) < .2), where sigma_E=0.062/sqrt(E) as calculated below.

Photon FCAL recon eff all.svn8894.png

Total FCAL efficiency: ~77%

More can be done here? Where do photons convert? Can split clusters be re-merged?

Extra photon problem

Use a timing cut (t_shower-t_flight) < 1.2 ns to cut out of out-of-time particles:

PhotonTproj recon FCAL.svn8894.png

Timing cut reduces average number of "photons" per event from 1.44 to 1.16. (actually 0.77 photons should hit the FCAL per event)

Timing cut has negligible effect on efficiency.

Energy resolution

Unconverted photons

Look at distribution of (E_recon - E_thrown)/E_thrown as a function of E:


Photon FCAL E res sigmas.uncoverted.svn8894.png

Fit is 0.062/sqrt(E). Resolution is worse than nominal value (~5%/sqrt(E)?). Why is this? Is this really better for a single photon sample?


Photon FCAL E res means.uncoverted.svn8894.png

Some trouble at low energies?

All photons

This will be biased since we cut on (E_recon - E_thrown)/E_thrown before making a correspondence between a thrown and reconstructed photon, in the case that the photon has converted before the FCAL.

Photon FCAL E res sigmas.all.svn8894.png


Photon FCAL E res means.all.svn8894.png

issues here...

Single block "clusters"

There is a tunable parameter in the FCAL code:


At lower energies, reconstruction efficiency is lower because often only one block has energy deposited in it (above the 20 MeV threshold in mcsmear).

Try setting MIN_CLUSTER_BLOCK_COUNT = 1 and see what happens.


Photon FCAL recon eff.minBlock1+minBlock2.svn8894.png


Clearly a big improvement in efficiency below ~500 MeV.

Overall efficiency increases by ~6%. (from 77% to 82%)

Total FCAL photon count (after timing cut) increases by 16% (1.16 photons/event to 1.34).

Energy resolution

Issues with energy resolution.

Photon FCAL E res means.minBlock1.all.svn8894.png

Energy is underestimated, especially at lower energies.

Energy resolution gets worse.


  • Redo energy correction/calibration
  • Deal with 1 block clusters as a special case?

Do hdgeant/mcsmear model FCAL correctly?

Identifying hadronic clusters by shape?

Hadronic showers often "look" different that EM showers:

FCAL face.b1pi event.png

How can we distinguish between the two?

DFCALCluster computes several size parameters:

		double fRMS;                 // cluster r.m.s. size (cm)
		double fRMS_x;               // cluster r.m.s. size along X-axis (cm)
		double fRMS_y;               // cluster r.m.s. size along Y-axis (cm)
		double fRMS_u;               // cluster r.m.s. size in radial direction (cm)
		double fRMS_v;               // cluster r.m.s. size in azimuth direction (cm)

Plot RMS size for three classes of reconstructed photon:

  • Black: photons unmatched to thrown photon (more likely to be hadronic)
  • Red: photons matched to a thrown photon (more likely to be EM)
  • Blue: photons matched to a thrown photon that converted before reaching the FCAL

Photon FCAL RMS.svn8894.png

Can cut out photons with RMS size outside the green lines. This cuts out 3792/57850 of photons that survived the timing cut.

Look at size vs energy.