Last week  results of W-algo (ver 4.1) from Run 9 data have been presented to spin PWG. Since then one error has been fixed and W-algo was  improved.

Fig 1. W-spectrum , Run 9, algo ver. 4.2

Fig 2. Typical W-event , Run 9 data, algo ver. 4.2

Section 1)  Changes in new  W-algo ver 4.2 are discussed  in attachment A).

All changes:

1. correct phi location of away-cone for BTOW 
2. separate code computing near, away ET from code selecting Ws (before logic was not bullet proof)
3. remove any code related to Z-search
4. replace method of summing ET for near & away BTOW cone from counting towers in 20xN patch to actual computation of delta-Eta, delta-phi  for every hit tower in the event ref frame.
5. relax cut on 2x2ET/near-cone ET from 0.90 to 0.88

W-algo ver 4.2 params: 
TPC: nPileupVert>3, vertex |Z|<100.0cm, globEleTrack: nFit>15, hitFrac>0.51 Rin<90.0cm, Rout>170 cm, PT>10.0GeV/c
 BTOW ADC: kSigPed=3 AdcThr>8 maxAdc>200 clustET>15.0 GeV  ET2x2/ET4x4>0.95 
 dist(track-clust)<7.0cm, nearDelR<0.7  ET2x2/nearTotET>0.88
 W selection  awayDelPhi<0.7rad awayTotET<8.0GeV , highET>28.0

Section 2) Full set of data reply plots are shown in attachments:

• B : Run 9 data
• C: Pythia QCD events, filtered at 20 GeV
• D: Pythia W-events
• E: Pythia Z-events

Examples of single event display   are  in attachments:

• F : Run 9 data (all ~1000 events)
• G: Pythia QCD events, filtered at 20 GeV, sample of 10 events
• H: Pythia W-events , sample of 10 events
• J:  Pythia Z-events , sample of 10 events

Section 3) Full code for ver 4.2 is posted at /star/institutions/mit/balewski/freezer/2009-W-algoVer4.2/

Section 4) Ascii file with list of all W-candidates from Run 9 setC is in attachment K). Content:

printf("run=%d daqSeq=%d eveID=%7d vertID=%2d zVert=%.1f prTrID=%4d  prTrEta=%.3f prTrPhi/deg=%.1f globPT=%.1f hitTwId=%4d twAdc=%.1f clEta=%.3f clPhi/deg=%.1f  clET=%.1f", .... names of variables

where track refers to the one identified as electron track, hit tower is the one selected by this track,
cluster ET,eta,phi is in event reference frame.
By providing run# and daq sequence # it will be trivial to run our codes again by hand  a single muDst to 
investigate individual events.

To reduce the  turn around time lets  at first generated such W event ascii file for data setC, 
runs: [10096139 - 10101020], about 1/4 of LT.
 

Section 5) Answers to questions received so far , please refer to PDF-file in Attachment A)

Carl:

Q: why away ET (TPC+BTOW) had large peak above 10?
A: it was a bug in summing away BTOW ET, fixed, see page 2

Q: what is PT cut off for TPC on the eta-phi event-display plot
A: Was none. The only requirement is : flag=301, nFitP/nPoss>0.51. Pt of all 
prim tracks for given vertex are added to this histogram 
Now I changed the event display and there is minimal ET or PT cut set to 0.3 GeV,
 also the max ET, PT is set lower.
Please have a look at collection of all run9 events passing W-cut  
in attachments : F,G,H,J.

Q: properties of vertex for reoc W events.
A: see slide 4 . In short 
* input events have 8e-4 fraction of 1-track vertices with rank in [0,1e6]. 
The W-events have 70 such vertices out of 487 total. This is 14% . This extension 
of PPV to store 1-track vertices was important for the W program
* W-events were found in non-first vertex in event only for 8 out of 487 events. 
This is 1.6%. Using of all vertices instead of just the 1st one has marginal effect.


Bernd, follow up:
Q: how ET spectrum looks like for events with 1-track vertices.
A: see slide 5. Statistically there is no difference vs. events with multi-track vertex.

Joe: 
Q: how 2x2 ET spectrum reduces with applied cuts
A: see page 6.

Jan:
Q: W yield stability vs.  time
A: see slide 8. In short : yield for sets B,C,D is consistent with seen # of
 BHT3 triggers. Set A has different BHT3 DSM threshold and will require boot 
strapping from set B.

Jim:
Q: estimate algo effi of the W-algo as implemented
A: see slide 9. I made up most of the values. In short W-algo EF=0.6 * 0.67 =0.40 , 
where 0.6 is probability of reconstructing a 10 GeV track in TPC,
 0.67 is probability that this track will be not destroyed by the W-algo due 
to various matching cuts cleaning the QCD background.

Jamie 1:
Q:On the gap [between e+ & e-] : have you made sure in MC that there isn't
 some subtle cut in the tracking algorithm?  i.e. throw 40, or 50, or to make 
it really hard, 100 GeV tracks at the tracking algorithm and make sure that at
 some point the peaks merge.  Should be pretty simple.

A: How about throwing only W+ or only W- :)? See slide 10 (W-algo-ver4.2.pdf)
Looks like Sti does e+, e- separation rather well. For data I used plot form 
older version of the code (forgot to enable the plot for the new data pass 
- makes not big difference)

Jamie 2:
I am a bit concerned, though, by a comparison of slide 31 (Q/pt for primaries
vs. Et) and slide 35 (signed DCA plots).  It appears that the signed DCA's are off by
~2-3 mm on the mean. Is this a 2d DCA?  (if not, please make it
one; the Jacobian for a 3d DCA will cause an artificial dip and peaks
at ~+/- resolution, which is ~2 mm.
helix- > geometricSignedDistance(vx,vy) should work; might be something
fancier stored in the MuDst.  You really want to do it for a circle and a point
to avoid Jacobian problems, and get a Gaussian that should be centered
at zero.)

A: Slide 35 [from last week], right & bottom shows signed DCA: glMuTrack->dcaD().
This is 2D measure, you asked for, right? Let me know.

An independent refit of the circle based on final track is on the long term goals 
for FGT. I was hoping to engage this type of machinery, since it would require 
working w/ StEvents. Perhaps we will need to do this.