Minutes of E895 Meeting - Aug. 12, 1997 --------------------------------------- In attendance: Hans-Georg Ritter (HGR), Gulshan Rai (GR), Nathan Stone (NS, on minutes), James Symons (JS), Bill Caskey (BC), Jessica Kintner (JK), Dieter Best (DB), Heng Liu (HL), Jerry Chance (JC), Sergei Panitkin (SP), Randy Wells (RW), Mike Lisa (ML), Mark Gilkes (MG), Chris Pinkenburg (CP) GR - discussed laser beams with Jim Draper, specifically one which showed a discontinuity - 2 pass running makes discontinuity in laser beam go away - field on/off changes slope of the diagonal beam CP - propose that we check the lambda peak to validate the distortion correction ML - propose that we look at residuals at TPC edges to see what difference the distortion correction makes, since it should be greatest at the edges BC - diagonal laser beam was rotated before applying corrections. - horizontal laser beams do not undergo any such rotation NS - would like to see interaction trigger comparison of with/without DISTN DB - I did that and didn't see much difference - BTF now handles distorted hits "correctly". Meaning, there is a final track fitting using distorted hits coordinates, which follows the Kalman track fit/finding using un-distorted coordinates ML - reminds us that 2DH only does DISTN on "last" pass (1st of 1, or 2nd of 2) NS - we still have not verified the usefulness of 2nd pass tracking (ie. HIT-TRK-HIT-TRK vs. HIT-TRK-HIT) ML - there was a problem with this at the collaboration meeting, but that is all fixed NS - XDF was released, one bug to go... - that bug affects only re-opening XDF output streams, after closing them. All other operating modes "are fine". - EDITOR'S NOTE: As of this date, other bugs have been found, some fixed, relating to a conflict with opening BDF files. HL - HITS packing/unpacking is done. Released to Nathan for TRKS packing - there is a 3-line correction that needs to be applied to 1DH in the distortion routine NS - will send the recommended fix to ML/CP for an opinion HL - we have processed 40 tapes of 2/4 GeV data to extract Vdrift. CP - what about the (10-15) missing ExaByte tapes? HL - LBNL is missing 2-3 tapes, that we know of JC - we have a few tapes at UCDavis which were never copied. Will check which ones they are. DB - worked on including DISTN into track fit. can submit this soon - regarding residuals (which were large and not centered at zero) - now they are calculated _during_ track finding, not after - plots are now different: symmetric, centered at zero, not "large" - regarding dE/dx resolution - sigma(dE/dx)/(dE/dx) vs. Nhits doesn't look much better (still flat, and ~40%, even when excluding hits used more than once) - plotted 1D plot of dE/dx values for negative pions for BTF/JTF they look basically the same (centroids and widths) - regarding Arun's splitting/merging analysis: - running it now successfully, and can reproduce this "60% efficiency" result by running with min hits per track = 10 (5 or 10 both give ~60%) - 60% efficiency number is almost exactly the same for all particle types (?) - Also investigated filtered RQMD events - BTF shows roughly the same performance for simulated data as for real data - JTF shows much better results for simulation than for real data, as measured by, e.g., hit removal efficiency (60% for data --> 85% for sim) - HITS.ntrks is now filled too (number of tracks to use a hit) ML - why choose min hits per track = 5 in the first place? DB - it is a minimum condition (need at least 5 to determine a helix) ML - but using these hits for short tracks may be hurting you (stealing from "better" tracks) DB - yes, I am working on removing multiple use of hits from BTF CP - what about these particles with Pz<0? What are they? DB - they are particles which do not cross the target plane (e.g. decays whose complete spiral is inside the volume of the TPC) CP - could we flag them via Px/Py values? NS - no. that throws away meaningful information ML - we should create some other flag so that we can identify them - what about Pz<0? How can you distinguish Pz<0 from Pz>0? DB - dca? CP - nope. dca distributions are continuous, so there is no obvious cut you can use DB - I guess we can't. should look into this, perhaps make the flag JK - UCD gave the vertex module toathan NS - I gave it to Dieter, it looks ready, but has some inconsistencies... BC - will substitute VTX table for VERTEX_LIST, but the hard-coded target position is only a first-guess anyway. Changing it does not change the output. PC - MUSI pulse-shape fitting work has been done, and now requires TAS to read in some data files upon TAS-startup. - I will release this data file to the collaboration ML - is BTF coming soon? DB & GR - Today! ML - are the BTF switches "mature" or only guesses? DB - they are semi-final CP - do they work for all energies? DB - they are tuned on 2 GeV, but they should not change much for other energies GR - how long will it take to find suitable values for all energies? DB - it is generally a long but non-distinct process, but the current values should already be very close to those used for all energies CP - I ran BTF/JTF with min hits per track = 5 (for both). Then, plotted dE/dx vs. Ptot for each, and subtracted the 2D histograms from each other. This shows me that JTF seems to yield more tracks. ML - I insist that people begin using 2DH. If we are going to adequately test and prepare for DST production, we need many users who understand TAS to be putting it through its paces. DB - I will run 2DH in my future studies. BF - I would like for us now to determine what is the best approach to making a clear choice between BTF/JTF. ML - my observation is that BTF is now better than JTF. CP - well, I am looking at PID resolution, and would like to compare invariant mass distributions JS - we should keep in mind that the best "track"er may not be the best "PID"er, as regards the small differences we are seeing. ML - except that these are big effects... NS - well, as regards specific observables, I propose we should stop looking at or using this sigma(dE/dx)/mean(dE/dx) variable as a diagnostic until we at least understand why it has such strange values (40%). ML - when is zero-suppression and gain correction done? in that order, or the reverse? GR - there is a "hit threshold" set by run control which is applied before data is sent to the receiver. Then comes zero suppression and then gain correction. ML - so this could explain how we get raw data values < 15... GR - I should check with HGR to be sure, since he wrote the DSP. ML - Pentium studies - has anyone looked at platform dependance of output? - someone should (James?) - we will forward our findings (to date) to SUNY/Purdue - I am planning on purchasing of order 8 Pentiums in the next 2 weeks. - performance is something like 1 Pentium --> 4 SUN4s / 3 HPs (8 pentiums could replace ~ 40 PDSF Suns) GR - LBNL will buy Pentium's too. - of order 5-10, depending on the contributions of the collaborating institutions NS - So, what I am hearing is that Solaris/PRO is the E895 "chosen platform"? ... and everybody said: basically, yes.