Embedding problem, hit reconstruction issue

Speaker : Gene V. Buren ( BNL )

Talk time : 12:20, Duration : 00:15

See BT #2378 and this post for a summary.


Reviewing some of Kolja's slides from his presentation at the 2012-09-24 S&C meeting:

Different days have different DCA (2D, unsigned) distributions vs. phi:

<DCA> vs. day shows some differences between real data (left) and embedded pions (right):

I was able to find the problem in sector 24 for day 53. On the left are hits from real data in sector 24 (using the Offline QA plots), and on the right are the simulated hits:

Bug: By design, dE/dx is not calcuated for hits of poor quality, decided in the StTpcdEdxCorrection class. This includes hits under anode channels with reduced gain due to reduced voltage (done purposefully to prevent tripping of the voltage supplies). However, TpcRS uses the same class to provide dE/dx of simulated TPC hits. And without dE/dx, there is nothing to embed. The result is that channels with sufficiently low anode voltage will not have any simulated hits. This is fine if the voltage is so low that the hits aren't substantial enough to be included on tracks (i.e. in the track fit), but can be a problem if the voltage is at an intermediate level such that the hits don't have dE/dx but do contribute to the track fit.

The window for accepted TPC anode voltages is set in the Calibrations_tpc / tpcGainCorrection table. This table has no entries before Run 10. For Run 10, the acceptablle voltage windows are set to (-50 V, +30V) from nominal (1170 V) for inner TPC sectors, and (-30 V, +10 V) from nominal (1390 V) for outer TPC sectors. For Runs 11 and 12, the windows are set to (-100 V, +100 V) for all. The windows used in Run 10 are definitely too small, but even at -100 V, we expect the gain to be ~35-40% of nominal, which will be sufficient to create hits for tracking, especially for high energy loss particles such as low pT protons!

Fix: Yuri has modified StTpcRSMaker not to follow the same restrictions for anode voltage, allowing a much, much larger window of acceptable voltages.

Excluding these hits (obviously) reduces the number of hits on tracks (a likely cut in analyses), and degrades the quality of tracks (as seen in the above <DCA> plots from Kolja), which may also be used in cuts. Further, the degradation can be even worse than simply expected because large gaps in the hits can cause the tracker to have difficulty connecting tracks across the gaps. This is demonstrated in the following plot of padrows with hits on tracks in sector 24 using an embedding sample of just 30 events on Run 10, day 53. The cyan and magenta histograms are real hits (which overlap so closely that they cannot be clearly separated on this plot), and blue and red are simulated hits.  The magenta and red are before the bug is fixed, and the cyan and blue are after. We can see from the red histogram that when the hits in padrows 14-21 are missing, the tracker has difficulty putting hits on padrows 1-8 on the tracks due to the large gap.

We have yet to get feedback from those involved in embedding work on the impact to their analyses, but there are some things we can learn by scanning the database...(note, these do not, as far as I know, reflect actual problems with the database records, just that these are records where we may see a problem in the embedding)...

  • Run 10
    I estimate that more runs are affected than are not, and there aren't large groups of unaffected data from what I can tell (I found 2206 potential problem records in the database spread widely over the Run year, and most problem entries were valid for thousands of seconds). If I understand correctly, we use SL10h_embed for the 7.7 and 11.5 GeV productions, and SL10k_embed for the rest. I think we should definitely update SL10k_embed (though it would be good to have people consider whether they need to re-produce their embedding). We actually haven't applied a few TpcRS fixes for SL10h_embed and would need to apply several patches to get it back up-to-date. Does anyone who worked with SL10h_embed think we need to do so?
  • Run 11
    Run 11 is all in SL11d_embed. Here are the few problem areas I found:
    • I found two pp500 runs (12038113 and 114) where the voltage was set low on all inner anodes. While these appear to have been test runs for Alexei according to the ShiftLog, they were produced in P11id. Representing only a small fraction of all data (~146k events), I would suggest we mark these two runs as bad. Any other viewpoints?
    • pp500 runs 12107008-038 (April 17, 2011) exclude channel 23-7. These runs include several million events.
    • Many AuAu27 runs on days in late June (23,24,28) exclude channel 20-5. The voltage was really low (~25-30% of nominal gain?), and sector 20 has some other systematic issues as well (unusual GridLeak distortion). Is this data and channel a concern for anyone?
  • Run 12
    I found lots of places where the voltage was low enough to kill the simulated hits (254 potential problem records), but it would've only been for a few seconds, and probably most of these situations were NOT during TPC runs. However, I found 3 run ranges where the situation is pretty clear:
    • pp200 runs 13043034-36 dropped channel 20-5.
    • pp500 runs 13085029-32 dropped channel 3-4.
    • pp500 runs 13087014-38 dropped channel 18-3.
    The pp200 data will likely be produced in SL12d, so we'll make sure SL12d and any new libraries to come get updated with the code fix. The pp500 runs were only produced in the W preview physics production.