Roundtable
Next meeting on 31.3.2025 in differrent room: 40/R-C10
Alex
Started with refitting with GX2F. Relies on direct fitter. Tried hacky approach without direct navigation, because not implemented yet and also don't want to have weird KF surfaces, that were dropped.
Doga
ADDT: truth association seems to work. Need to look deeper into it.
Paul
Converted podIO output can be stored and converted later. There is a threadsafe writer in podIO but in the end probably mutex. Conversion and writing was in the same class, so multiple objects wanted to write to the same file. Examples have already primitive conversions.
Attila: Why not provide a single output file to everything? -> podIO outputcollection is attached to a frame and then written.
ODD dataset: part of it we want to write in dd4hep
rewriting some old hepMC conversion code to be more general. Goal is to write our own hepMC-files from everything.
hepMC 3.0.4(?) is used.
sPHENIX meeting with Joe
Andreas Salzburger: Their detector is far away from CAD-drawings. Inner layers are shifted by 6mm which makes large difference. We tried to find a few steps to mitigate.
key4hep tracking meeting: Berkley involvement now. Shared Lorenzo's input: Have extra code for each type of detector.
key4hep tracking: Started with Doga a few years ago. Now building different configurations. A bit difficult to make everything interact with different geometries. Idea is, that ACTS is used inside.
Attila
no tracking news from last week.
ATLAS IO. Short-comings with ntuples.
Noemi
Produced material map with sensitive on the back. Plots already exist on the grid but still need to be looked into.
EF-tracking discussions:
- Template will be updated and shortened.
- Final performance numbers will be done by experts
Stephen
traccc: bug in truth seeding. After reading, they are in an arbitrary order. This was the reason, why seeds were picked up weirdly. Now it should work.
Now we pick up the same measurements, not sure yet, if this is the goal but a reasonable start.
Neza
traccc memory question: timing does not change.
Attila: Device memory allocation needs double checking. Maybe this wasn't a bottleneck in Athena anyhow.
We need to limit measurements but only want to show that tracking works anyhow. Track reduction doesn't work as wanted yet. Ambiguity solver is insanely slow currently.
Dimitra
Tracking geometry validation: outliers in odts and it seems clearer now where they are coming from.
changed the propagator a bit because we couldn't reach the last portal and couldn't run the full sample.
Some outlier seem to be related to the navigation. Several sim-hits on the same cube-layer. When bending is large enough surfaces might be hit that are not part of the navigation. Currently struggling with hepMC-linking.
Plan: fix the detector navigator and run with high energy particles to understand outliers better.
Andreas Salzburger
Lorenzo shifting now to Gen3.
component wise digital clustering https://github.com/acts-project/acts/pull/4142
wrote digitisation file and found weird shoulders.
Naive digital clustering might pull into wrong direction. This works with analog clustering, because the pathlength is considered. Considering this solves the problem.
Now we have a nice digitisation file for the ODD without the extra bit.
New grid description: Open PR with 2 follow ups. Also adding converter for grid2bin and bin2grid. Maybe remove bin in future?
Andreas Stefl
ODD performance for thesis:
Simulation problems for ttbar and singles. High energetic particles create a lot of showers. This overruns the barcode in ACTS and creates too large log-files. Quick-fix is to ignore secondary propagation. This is problematic for physics, but a good approach.
ttbar took too much memory. (15GB ram/thread).
Paul: there is also high memory consumption in ATHENA but not like this.
Problem is probably that we generate too many particles in memory. (ATHENA has particle killers)
Noemi: For secondaries below threshold, we could kill all the truth.
Tracking efficiency: drops in seeding efficiency with triplet seeder between 2-3 eta. Recovered by enlarging deltaR. Still inefficiencies but better already. Mayber also other parameters need to be tweaked?
Noemi: We often discussed if we need more tuning. But no solution yet.
Loosening the cuts gives around 20% more seeds. But also more fakes accumulate probably.
