1. 1 Update on Photons Graham W. Wilson Univ. of Kansas 1.More on  0 kinematic fit potential in hadronic events. 2.Further H-matrix studies (with Eric Benavidez).

2. 2  0 kinematic fit potential See Vancouver talk re intrinsic  0 energy resolution improvement given correct pairing of well measured photons. Today, characterize better the multi-photon issues in Z → uu, dd, ss events. Define prompt photons as originating within 10 cm of the origin (NB differs from standard c  cm definition)

3. 3 On average 19.2 GeV (21.0%)

4. 4 On average, 2.1 GeV (2.3%)

5. 5 Intrinsic prompt photon combinatorial background in m  distribution assuming perfect resolution, and requiring E  > 1 GeV. With decent resolution, the combinatorics are not so horrendous … Especially if one adopts a strategy of finding the most energetic and/or symmetric DK ones first. Next step: play with some algorithms

6. 6 Conclusion on  0 kinematic fitting Still very promising Plan to work on developing algorithm for the photon-pairing problem Non-prompt photons (K 0 S ) are an important second order effect (certainly in s-sbar events !)

7. 7 Next 3 slides are from Snowmass 05 (As a reference for “standard usage”) H-matrix

8. 8 Standard Longitudinal HMatrix Developed by Norman Graf. Compare observed fractional energy deposition per layer with the average behavior of an ensemble of photons including correlations. Current default implementation has a measurement vector with 31 variables: 30 fractional energies per layer and the logarithm of the energy. Method: calculate,  2 = D T M -1 D where D is the difference vector, D = (x i – x ave ) (i=0, 30) and M is the covariance matrix of the 31 variables. We were using FixedCone Clustering with  =60 mrad. Used sidmay05 with low energy photons to avoid containment and issues regarding change in sampling (with 20+10 geometries).

9. 9 Hmatrix Performance 5 GeV photons, 90 0, sidmay05 20 GeV neutrons, 90 0, sidmay05 Not perfectly distributed …………….. but a lot of discrimination These photons used for evaluating the expected fractions and the covariance matrix, M.

10. 10 Hmatrix Performance 5 GeV photons, 90 0, sidmay05 20 GeV neutrons, 90 0, sidmay05 Eg. cut at p > 10 -10 => eff (  = 99.2%, eff (n) = 9.3% p > 10 -5 => eff (g) = 98%, eff (n) = 4.6%

11. 11 Perceived limitations of standard method Chi-squared probability distribution is not flat. Matrix variables mix energy fractions with cluster energy –Can cause technical difficulties –Implicitly uses the overall energy in the cuts –Gives some scope for a “one-size fits all” solution – but unlikely to be the best possible solution. Matrix averages over the conversion layer Number of actual layers with significant energy deposits can be  2 not correctly normalized)

12. 12 New Strategy Use an H-matrix containing ONLY the cluster energy fractions per layer. –=> cluster energy is something that can be used separately. Use separate H-matrices depending on the layer with the first significant energy deposit. –Eg. for acme0605, we have H 30, H 29, H 28, …. –This has the additional benefit that longitudinal changes in sampling fraction can be treated “seamlessly”. –=> conversion point is something that can be added in afterwards as a further discriminant. Disadvantage: need more MC statistics …

13. 13 More details Apply a cut of 50 keV per cell. (MIP gives 124 keV in 320  m Si). Use number of layers with non- zero number of cells in normalizing the  2. In order to avoid photon fragments, have required clusters to have ncells > 5 and raw cluster energy > 0.03 GeV (cf. mean of 0.08 GeV for 5 GeV photons) 5 GeV photon 90° acme0605

14. 14 Cuts Require that the photon converts in the ECAL (r > 1260 mm) in the training samples (rejects conversions in the tracker) Interaction radius (mm)

15. 15 5 GeV photons 90° acme0605 Resolution: (19.0  0.2%)/  E

16. 16 Probability Distributions Flatter, but still spike at zero.

17. 17 Why is probability distribution not uniform ? layer 9 layer0 layer23 (10 GeV photons, 90°, acme0605) Response function is only Gaussian near shower max. Maybe a likelihood approach would have more potential ….

18. 18 5 GeV Photon  2 /dof = 37.1/23

19. 19 Performance Currently battling floating point errors associated with neutron clusters which are not at all photon-like.

20. 20 10 GeV neutron  2 /dof = 1562/25

21. 21 10 GeV photons and 10 GeV neutrons Photon efficiency Photon purity (assuming n  =n n ) I suspect this is worse than actual performance due to FP issues

22. 22 Photon purity (assuming n  =n n ) Photon efficiency

23. 23 Conclusions on H-matrix H-matrix work still a work in progress, but new approach looks to be promising. Probably should include some simple preselection cuts which discard really un-photon like events from the background samples. –Suggestions on what to use as a performance metric appreciated. Interested in looking into likelihood approach in the future.