LCD Weekly "Reconstruction and Analysis" 04/26/2005 Attendees: NG (Norman) TJ (Tony) JM (Jeremy) SM (Steve) WM (Wolfgang) JS (Jan) MC (Matthew Charles) WL (Willy) MD (Mark) NTM (Niels) NS (Nik Sinev) MA (Makoto Asai) TK (Tatsumi Koi) RC (Ron Cassell) CJ (Chen Jin) TB (Tim Barklow) -- NG: TB to present on Whizard TB: [slides] MC production Put on SLAC mass storage. Users request combos of polarization, can produce them for the users. For Snowmass, put large sample on disk. 100% electron and positron polarization. 0-fermion, 2-fermion, and 4-fermion datasets. 11 process groups. 5 steps for complete MC dataset. 1. generate executable 2. submit MC ingegration jobs 3. repair mc ingegration jobs 4. submit first set of event generation jobs 5. submit more event generation jobs Schedule. [see slides] TB: Includes fragmentation and decay. NG: Fast MC? In addition to complete sample, diagnostic single particle events exist. Re Whizard, don't produce final state processes as resonances -- produced as final fermion states. Specific signal samples for people to develop their algorithms on. TB: Find use Whizard samples when analysis is close to being finished. Included all backgrounds. Usually just produce signals using PYTHIA. Same Guinea Pig input can be input to PYTHIA. Have in own directory. NTM: Guinea Pig used for beam backgrounds? TB: Can be, but we are using to produce e+e- energies for annihilation. NG: Replaces old system using Circe. TB: Not easy to include linac energy spread using Circe. So far, appears to be best way to include beam energy spread. Can be used in the future for more sophisticated e+e- energy distributions. Easy to make the two beams different. NG: nb samples, but not run through full sim. Run some subset through full G4 simulation. Easiest way is to use Fast MC as trigger and only run subset. TB: Benchmarking Action Plan [slides] Critical Questions 1. Physics Error vs Calorimeter Parameters -cannot directly vary B, R, etc. until full Calorimeter Simulation & Reconstruction is more fully developed -physics error vs delta(Ejet)* can be calculated before full simulation and reco software ... 2. Physics Error vs VXD, Tracker Parameters ... 4. Fast MC vs Full MC -use LCIO analysis objects as inputs -some physics analyses using the Full Mc before Snowmass so can evaluate the quality of the Fast MC approach Simulation Tools. MC Programs for Generating Physics Events MC datasets of all SM processes at Ecm=350, 500, 1000 GeV Fast detector MC with Reco Particle LCIO output -reco particle = E, p, impact params, charge, id(e-, u-, pi+, gamma, K0_L) and errors Full MC with Reco Particle LCIO output WL: Ignoring Lelaps. TB: Really need the reconstructed objects. No sense in every analysis duplicating creation of the analysis objects. NG: Lelaps does provide intermediate of just smearing the 4-vectors. Can do pattern reco on output at the hits level. TJ: Doesn't address this issue of writing the PID code. NG: Fast MC for lcsim, take tracks and clusters that are made and "stuff" into the reconstructed particle object. TJ: Simdet? TB: Call every electron/muon. Minimum momentum for electron/muon ID. TJ: Add tables of cutoffs to existing MC. Just making reconstructed particles with crude cuts should be straighforward. NTM: TESLA very basic. Just throws away every Xth electron. Doesn't account for charge. TB: Momentum dependence on charge? NTM: No. NG: Fiducial cuts on angle or momentum. TJ: Efficiency vs. momentum. NG: Confusion terms, track reconstruction efficiency vs. local track density = harder. TB: Need neutral energy flow objects. Realistic -> need crude estimation of overlap between charged and neutral objects. SM: Assume that neutrals that the hadrons only interact in the hcal? TB: Assuming from Simdet. SM: Full gives how much of neutron in ecal vs. hcal. TB: Simdet does 0 longitudinal dev of shower. SM: Smear photons with ecal res and hadrons with hcal res. But no exactly correct in the end. TB: Goal is to get SiD MC started. Bring it up at least to simdet. Begin process of improvement. NG: Full simulation. Makoto or Tatsumi? MA: Shorten production threshhold -> more energy in sensitive part of calorimeter. Actually increases ~30% which is an "incredible amount". Where tiny eneries appear. Shorter range cuts gives the best results. Comes from problem of multiple scattering. GLAST people looking at it. Michelle interested in seeing if the problem persists since major revision of EM since 5.2. If see same problem, then can locate the problem. Should still help if same. Maybe problem inside multiple scattering or connecting multiple scattering w/ ionization. RC: GLAST found multiple scattering problems in earlier versions. MA: Sorry couldn't come up with quick fix. Highest prio in EM package for G4 devs. NG: Discrepancy in Eres. ~20% with Geant4 out-of-the-box. JM: [SLIC recap] NG: Compact description to expand into complete format should have info for reco as well. Tracker layer # into tracker radius. Working on getting SiD into the compact geometry, so we can model this detector. LDC and GLD also interested in Full MC tools. TJ w/ Akiya got first iteration of compact description for GLD. LDC can be modelled as well in the same environment. RC: Have all SiD params? NG: Not all "signed-off", but have what we need. RC: Cylinders? NG: Yes, main engine for our studies. As closer w/ subdetectors, then we can put them in. For Snowmass, main detector response with cylinder-based detector. NS wrote hep.lcd convertor for VXD and tracker. NS: Need to be able to read geometry using disk CCDs. Question about SIO files for Feb05 -- generated with LCDG4? NS: Minor problem. Assigned same MCID for compton electron as for initial gamma. Charged and initial. Define as secondary -- effects efficiency calculation. RC: Solved generally in LCIO. TJ: Move digi to lcsim? NS: Probably not before Snowmass -- don't know how much work it would be. RC: Object to someone else looking at it and seeing what it would involve? NS: Port -- don't know if anyone will volunteer. NG: done the "heavy lifting". Code and interface should be there. Similarly for Si microstrip trackers -- Santa Cruz group and Tim. Have basics for clustering and track finding. Clustering, there are already three algorithms -- fixed cone, nearest neighbor and cheater. WM working on minimum spanning tree. WM: Make an implement for LCIO with RC. Depends on RC's judgement. RC: Offered to start with clusters and look at implementing features from hep.lcd. Don't want to do like before. Diagnostics to look at clusters needs a lot of work. Start on this week. WM: Should put effort into existing algorithms in hep.lcd or port to lcsim/LCIO? Follow RC's lead. RC: Answer in next 2 weeks. WM: Busy with Babar. NG: RC will look at it this week -- what is needed to fulfill obligations of clusters in org.lcsim. Suggestion to try and implement in lcsim. This is pathway for future dev. MA: Use GPS for rho, other short-lived particles? Currently, no short-lived with GPS. JM/NG: Ok with Stdheps. NG: Track cheater -- discuss with LCIO devs. What other formats for tracks are needed? Once have tracks and clusters, can start to do the associations. SM working on this. SM: [slides] NG asked to talk about goals for PFlow dev. move existing code from hep.lcd to org.lcsim (next 2 weeks) -Helix track swimmer -exists in org.lcsim, but not the one we used -interaction layer in Cal -track/Cal object association -utilities for PFA evaluation complete PFA for Snowmass (for the CD) -incorporate photon finder -finish neutral finder(s) improve/tune PFAs (ongoing) use for LC Detector development/evaluation -SiD, LDC, GLD, etc. track extrapolation PFA algo 1. track/Cal cell association algorithm 2. photon finder algorithm 3. neutral finder algorithm 4. jet algorithm -tracks + photons + neutral clusters used as input to jet algorithm Mip reconstruction / shower reconstruction WM: Mip peak more prominent in my analysis? SM: All tracks with > 1 MeV -- otherwise, they are loopers. Will get pileup somewhere. e+e- -> Z w/ SDFeb05 NG: Tungsten absorber in hcal. SM: Think using SDFeb05. SM: [histos on slides] WM: Have a tail for the perfect Pflow? SM: Sometimes "get the wrong hits". SM: Results = 28% / sqrt(E) for true detector PFA Real PFA results for best fit = 34% / sqrt(E) SM: Get interaction layer portion into lcsim ASAP. NG: One other piece after found jets/tracks/clusters -> PID. Way we've decided to do this, at least for quark tagging -> look at displaced vertices. JS: Work on porting from hep.lcd to lcsim. hep.lcd was working pretty well, because original author of SLD code was using it. "Black box" -- just copying from hep.lcd to org.lcsim is tricky. Uses its own track representation, own fitter, own swimmer -- standalone package. Want to make more modular and integrate into org.lcsim. Currently working on getting the fitter into org.lcsim from hep.lcd. Hope ~1 week to have something that works for the PID. By end of May, have first working version of Zvtop. Starting debugging by passing tracks, getting vertices. Collaboration with LCFIDG (UK Flavor ID Group). Sent them JavaDoc. They want C++ implementation. Validate C++ and Java implementations on old SLD data by running old FORTRAN analysis package. By Snowmass = reasonable? NG: End of May for tagged release of org.lcsim and burn to CD? JS: Alpha version. NG: Alpha okay as long as indicate this. JS: More "beta testers" this way. MA: Meeting with accelerator guys for MDI? NG: Yes. MA: Couple contacts with machine guys switching to Geant4. NG: Spoke to Ilya who asked for a few more weeks with BDSim. Look at best way to merge in geo info from MDI data. NG: Thurs at Goose for beer. Keep Tues meeting at 1:30.