Cal Cluster ID (a.k.a. Eflow)

Cal Cluster ID(a.k.a. Eflow)

Gary R. Bower, SLAC

Santa Cruz LCD Workshop

June 28, 2002

June 28, 2002 G.R.Bower - Santa Cruz LCD Workshop

2

Acknowledgement

• Ron Cassell has made many essential contributions to this project.

June 28, 2002 G.R.Bower - Santa Cruz LCD Workshop

3

Outline• This is a work in progress.• (Very preliminary) results first!

– Describe test data sets and testing methods.– Efficiencies and fake rates.

• Details (as time permits)– Approach to problem– Discriminator tools– Discriminator capabilities

• Summary• Next Steps

June 28, 2002 G.R.Bower - Santa Cruz LCD Workshop

4

Test procedure

• Three datasets of 1000 single particle events.– piminus, gamma, and K0L– 1-50 Gev momentum– In barrel, within 45o of perpendicular to beam

• Make contiguous hit clusters– Ignore clusters with energy < 0.5 GeV– Treat most energetic cluster as primary deposition– Treat second most energetic cluster as fragment.

• Test both primary and secondary cluster– Is it a gamma, piminus, K0L, and/or fragment?

June 28, 2002 G.R.Bower - Santa Cruz LCD Workshop

5

ID result:gamma piminus K0L fragment

Input:

gamma

piminus

KOL

fragment

June 28, 2002 G.R.Bower - Santa Cruz LCD Workshop

6

Multiple ID rates

June 28, 2002 G.R.Bower - Santa Cruz LCD Workshop

7

Philosophy of Technique

• The Eflow problem: a common approach:– solve it with a clever cluster builder but still

need to identify shower origin.– Assumes showers fragment badly.– Assumes showers overlap each other.

• We take a different approach: – work with (simple) clusters of contiguous hits.– distinguish origins based on cluster properties

June 28, 2002 G.R.Bower - Santa Cruz LCD Workshop

8

Fragmentation problem?input piminus

June 28, 2002 G.R.Bower - Santa Cruz LCD Workshop

9

“Secret” to solving fragmentation problem

• Combine EM and Had clusters using contiguous hits cluster builder by Ron Cassell.

• Caveat: For the occasional neutral hadron there will be significant fragments but we have a promising technique to find and associate them.

June 28, 2002 G.R.Bower - Santa Cruz LCD Workshop

10

Isolation of gammas

June 28, 2002 G.R.Bower - Santa Cruz LCD Workshop

11

Gamma shower characteristics

• Compact

• Standard cigar shape

• Shower initiates in first few EM layers

• Shower contained in EM (if deep enough)

• Many hits/much energy in first few layers

• Accurately point back to IP

June 28, 2002 G.R.Bower - Santa Cruz LCD Workshop

12

Piminus shower characteristics

• Diffuse in shape and energy spread.

• Some fragmentation.

• Min-I track begins in first layer.

• Few hits/little energy until first interaction.

June 28, 2002 G.R.Bower - Santa Cruz LCD Workshop

13

KOL shower characteristics

• Diffuse in shape and energy

• Some fragmentation

• First hit layer may be very deep

• Generally points back at IP

June 28, 2002 G.R.Bower - Santa Cruz LCD Workshop

14

Fragment shower characteristics

• Diffuse in shape and energy

• First hit layer may be very deep

• Generally do not point back at IP

June 28, 2002 G.R.Bower - Santa Cruz LCD Workshop

15

Measure cluster properties

• Form energy tensor– Energy (shape) eigenvalues– Energy axes– Center of Energy

• First and last layers with hits

• Energy/# of hits in first N layers

June 28, 2002 G.R.Bower - Santa Cruz LCD Workshop

16

Superb gamma location resolution

Resolve gamma direction to ~1/6 cell size using center of energy of hits

June 28, 2002 G.R.Bower - Santa Cruz LCD Workshop

17

Separating gammas

June 28, 2002 G.R.Bower - Santa Cruz LCD Workshop

18

Separating piminuses

June 28, 2002 G.R.Bower - Santa Cruz LCD Workshop

19

Separating K0Ls

June 28, 2002 G.R.Bower - Santa Cruz LCD Workshop

20

Summary

• Can find gammas with ~100% efficiency and ~few percent fakes.

• Can identify most pions without tracking

• Can identify majority of K0Ls.

• Have only sketched the power of the method.

June 28, 2002 G.R.Bower - Santa Cruz LCD Workshop

21

Next steps (arbitrary order)• Many details to cross-check.• Reconstruct π0s (dE/E~5%, loc res ~same as π±.)• Associate neutral hadron fragments (improve dE/E).• Work out special cases, eg, charge exchange.• Combine clusters between barrel and endcap.• Use neural net to improve results.• Test on signal events.• Test on physics measurement.• Release tools.