Roundtable
Paul
Instrumenting. detray with outputs - not sure yet where material conversion goes off rail. Library is still turned on/off with a macro instead of a library.
Geometry building does some index manipulation, not sure yet where it happens.
During the search we need to add more neighbours but not sure why on how many.
We need surface-svg-printing in.
Work on final preliminary report.
Need to catch up with merge conflicts on layer patterns.
Dimitra
Gen3: progress on building muon-system and ITk together with the blueprint. ITk is now a child of the muon-system.
Andreas Salzburger: synchronise beam pipe with ITk. Beam pipe then collapses down to a single portal.
Paul: Depends on the question if we need the beampipe without ITk, and the answer is probably no.
Included builders to configure the blueprint construction. Merge request is already opened.
Carlo
Seeding in CKF: trying to optimise some things, no huge progress yet. Trying now to identify failing seeds earlier.
Which version of Athena will be used for the report? -> Attila: we just mention the used version in the report, but probably no other confinements.
Attila
We need to combine forces on what Carlo is working on. It was surprising how much we are dominated by seeding on the GPU.
The difference comes mainly from the cuts. The ODD works well but the problem is on the ITk.
Andreas Salzburger
Continuing with Python bindings. Hopefully it will look simpler in the future and we can skip the context objects.
Not sure yet, if python has a general way of doing what we want to do.
Paul: python import * seems to be the way to go. Usually we don't want to pollute the namespace but in this case we actually want to do it.
Single libraries for Core/Plugins/Examples. With Geant some have weird dependencies. In the end it gets a snapshot on what to build the library on.
Also planning to work on the alignment.
Neza
Parameters for seeding: traccc used 3 but they were the wrong ones so no optimisation took place.
traccc wants measurements to be in annular bounds. mistake in r-phi vs phi-r. switching that made a huge improvement.
Attila: Maybe we create measurements with a slightly different order which creates the mess-up? But somehow the hand-made measurements work better.
Figured out how to make truth data available for traccc standalone. Now we can do the parameter training.
Stephen
Seeding: It is very slow but made some improvements. This will also affect the ITk-performance. This will also improve the trackfinding since fewer seeds are generated in total. Physics- and Computing performance looks promising.
Plan to work on the weighting-system. Currently, we take the negative of the impact factor and add 200 from any confirmationseed. Should be replaced by 2 new separate systems. Maybe we can even flip the problem and use the confirmation seeds to make more cuts.
Doga
Dataset for flight and looking on correct assignment on datasets. Good discussions. Figured out with PF that maybe a lot of the vertexes come from material interactions.
Paul: Physics people: build material map to figure out where secondary vertices are. Maybe we can do basic RG-regions. This needs to be done for each layer.
Paul: For this project we don't need to be super accurate and it is enough to identify peaks.
PF: what is the scope? -> find vertices from particle decays.
Paul: jet clustering works on simulated particles, labeling on hepMC. from the combination, it should be possible to figure out the end-point.
Postpone discussion for offline.
Joana
Looking at KF in traccc. Checking if it finds the correct surfaces. It seems to be sensitive to our noise model. Now looking at the covariance estimation. Now running on ODD with geant4 data. Setting the covariance gives a slight improvement. This still needs some improving. Time spent trying to tune it. Tried opening the mass tolerances but generated more holes.
Stephen: On Thursday there might be also a talk by Yuki in the ACTS ITk meeting, that might be interesting.
Plan to fix detray IO for material.
PF
Efficiencies for muons and ttbar. Looking at center part with |eta| < 1.2. We suffer a few percents in this region.
Started to look at predicted covariance distribution. Legacy covariance is larger than in ACTS. This doesn't mean it is good or bad in general.
Used material in front of the layers. The noise from the last pixel layer will be propagated to the material. Basically the filtered covariance is propagated.
Switching the material to the back improved the chi2 by a factor 2.
Measurements covariance: ACTS uses only the pitch. Somewhere a sqrt(12) is still in there. Trying to change the errors on the strips on move on from there.
Alex
Ellipses on the weekend for Andreas Stefl.
