Effect of commits for RT3243

For RT #3243, Victor committed a change to Sti/StiTrackNodeHelper.cxx, and also included other changes in the commit.

I examined 306 events from run 14150005 (Run 13 pp510) using track-by-track comparison (mode 52).

- When including the line to check on dx near 0, there are no unmatched tracks. This was the modification specifically to address the RT ticket.
- When futher including the changes to check on sinCA2 near 1, there are tracks lost and found...

Histograms below are matched tracks (black, 233k tracks), matched but altered in some way tracks (green, 1469 tracks, ~0.6% of tracks), unmatched from having only the check on dx near 0 (blue, 630 tracks that are lost when the sinCA2 cut is changed), and unmatched from having the changed check on sinCA2 near 1 (red, 622 tracks that are gained when the sinCA2 cut is changed).

About 0.3% of tracks appear/disappear integrated over all pT (I used log10(pT) for the plot axes), but appears to have two components:
(1) at low pT (0.10-0.16 GeV/c), more than 1% are lost, and closer to 10% are modified
(2) at other pT, it's a little less than 0.3% that are lost, and less than 0.1% are modified.
The lost tracks are peaked a little bit towards low numbers of hits, while the modified tracks seem reasonably well distributed across the number of hits.

The following plots show the correlation between the number of hits and log10(pT) for the three categories:
first: matched but altered (green above)
second: unmatched from only the dx cut (blue above)
third: unmatched from all modifications (red above):

These plots show clearly that the low-pT group (1) extends over a wide range of hits, and the higher-pT group (2) is mostly at low number of hits. The low-pT group is almost certainly what was expected from the change to sinCA2, but the other group isn't as clear.

For the modified matched tracks (green above), here are the changes to the number of hits on tracks for the low-pT group (green) and the higher-pT group (red), showing that neither really has much systematic effect on the number of hits:

And here is Δ(pT)/pT vs. log10(pT) for the modified matched tracks (green above), showing that there was a systematic increase in pT (ranging from 0 to about 1.5%) from applying the sinCA2 modification for only the low pT tracks:

Also, to document, the fix that was specifically to address RT #3243 (the cut on dx) had very, very minor effects (same colors as before):


Victor's changes also affected the reconstruction time. Using event-by-event reconstruction times (not just tracking), and excluding the first event, we see the following effects...

1) percentage change due to cut on dx is essentially zero:

2) percentage chang due to modified sinCA2 cut essentially zero:

3) percentage change due to removal of checks is about -1.8% ± 0.1%.