Add o2-sim-geometry-doctor, a geometry against field audit by sawenzel · Pull Request #15683 · AliceO2Group/AliceO2

sawenzel · 2026-08-16T18:53:32Z

This commit provides a new tool o2-sim-geometry-doctor that reads a placed geometry and a magnetic field and reports where the two do not fit together: logical volumes placed both inside and outside the field, mothers whose own material straddles the boundary, daughter clusters that would need a container of their own, and media asking for straight-line transport while sitting in real field.

The tool was motivated from the idea to check the consistency of geometry and Geant4 media properties against the properties of the magnetic field, and to potentially optimize this assignment in a second step. It is related to these O2 JIRA tickets:

https://its.cern.ch/jira/browse/O2-174
https://its.cern.ch/jira/browse/O2-175
https://its.cern.ch/jira/browse/O2-181

In the future it can be extended to provide automatic repair PRs.

It:

builds an outer bound on the field support per threshold, phi sampled by arc length and every threshold crossing bisected, and checks the bound against the field before using it
concludes field-free from geometry only; sampling is used only to disprove
takes placement extents from the shapes, testing TGeoEltu before TGeoTube
verifies a field read from a file against stored reference vectors and repairs the polarity inversion re-initialisation introduces
reads the ifield and sensitivity flags from the geometry file itself, so it links no detector code and no simulation engine
ships the ALICE anchor verdicts for --verify-anchors

Example:

o2-sim-geometry-doctor --geometry-file o2sim_geometry.root --field-current -5

writes geometry-doctor-proposals.json (findings with suggested actions), geometry-doctor-placements.csv (one classified row per placement) and geometry-doctor-report.txt, and prints the report.

This commit provides a new tool o2-sim-geometry-doctor that reads a placed geometry and a magnetic field and reports where the two do not fit together: logical volumes placed both inside and outside the field, mothers whose own material straddles the boundary, daughter clusters that would need a container of their own, and media asking for straight-line transport while sitting in real field. The tool was motivated from the idea to check the consistency of geometry and Geant4 media properties against the properties of the magnetic field, and to potentially optimize this assignment in a second step. It is related to these O2 JIRA tickets: https://its.cern.ch/jira/browse/O2-174 https://its.cern.ch/jira/browse/O2-175 https://its.cern.ch/jira/browse/O2-181 In the future it can be extended to provide automatic repair PRs. It: - builds an outer bound on the field support per threshold, phi sampled by arc length and every threshold crossing bisected, and checks the bound against the field before using it - concludes field-free from geometry only; sampling is used only to disprove - takes placement extents from the shapes, testing TGeoEltu before TGeoTube - verifies a field read from a file against stored reference vectors and repairs the polarity inversion re-initialisation introduces - reads the ifield and sensitivity flags from the geometry file itself, so it links no detector code and no simulation engine - ships the ALICE anchor verdicts for --verify-anchors Example: o2-sim-geometry-doctor --geometry-file o2sim_geometry.root --field-current -5 writes geometry-doctor-proposals.json (findings with suggested actions), geometry-doctor-placements.csv (one classified row per placement) and geometry-doctor-report.txt, and prints the report. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0178JHAQ2biaX5ac9ThppBXi

alibuild · 2026-08-17T06:56:47Z

Error while checking build/O2/fullCI_slc9 for 4a455a8 at 2026-08-17 08:56:

## sw/BUILD/o2codechecker-latest/log
100% tests passed, 0 tests failed out of 1


## sw/BUILD/ThePEG-latest/log
collect2: error: ld returned 1 exit status


## sw/BUILD/O2Physics-latest/log
c++: fatal error: Killed signal terminated program cc1plus
c++: fatal error: Killed signal terminated program cc1plus
ninja: build stopped: subcommand failed.

Full log here.

A tracking medium belongs to a logical volume, but sitting inside the magnetic field is a property of a placement. The beam pipe makes the conflict concrete: the same warm module and the same bellow are installed inside the barrel, where the solenoid fringe is still several kGauss, and again ten metres down the beam line, where the field map returns exactly zero. One set of volumes cannot state both, so a module author picks the answer that is right more often and the remaining placements are transported either wrongly or expensively. The bug fix in fix-nf-media-in-field settles that conflict in favour of correctness and pays for it at the far end of the beam line, where a field that is genuinely zero is now integrated rather than skipped. This series gives those placements volumes of their own and takes the shortcut back for them alone. This commit gives passive modules the second logical volume they need. MaterialManager::cloneSubtreeWithMediumSuffix copies a subtree and remaps every medium to a field-free variant. - Shapes and placement matrices are shared with the original, so a clone is the same geometry differently flagged and adds no placements. - fieldFreeVariantOf prefers the hand-written counterparts the passive modules already ship (PIPE_INOX_NF beside PIPE_INOX, PIPE_VACUUM_NFHC for the high-cut flavours) and derives one only where none exists. - Clones are cached per volume and suffix, so a subtree reached twice yields one copy rather than two volumes of the same name. - Cloning refuses outright if the subtree contains a sensitive volume: hits are recorded against geometry paths, and a second set of paths for the same detector would silently change what alignment and digitisation look at. The placements that need this, and the media that are wrong without it, were found with the o2-sim-geometry-doctor tool (AliceO2Group#15683). Related tickets: https://its.cern.ch/jira/browse/O2-174 https://its.cern.ch/jira/browse/O2-175 Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0178JHAQ2biaX5ac9ThppBXi

… field Not every case needs a second logical volume. The short RB24/2 tubes, the RB26/5 compensator with its flange, and the SAA2/SAA3 shielding each have a single placement, or several placements that are all in the clear, so their media are simply wrong wherever they are used rather than right in one place and wrong in another. This commit switches them to field-free media in place. The pipe modules have hand-written counterparts to move to; the shielding does not, so SAA3 asks MaterialManager to derive one -- same material, same cuts, no field -- which is also the first exercise of that path, as the pipe changes never take it. RB26/4 sits partly inside the muon dipole and is deliberately untouched, as is the near copy of every RB24 module. Field calls in the reassigned volumes go to zero. Found with the o2-sim-geometry-doctor tool (AliceO2Group#15683). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0178JHAQ2biaX5ac9ThppBXi

This commit provides a helper MaterialManager::cloneSubtreeWithMediumSuffix for passive modules. A tracking medium belongs to a logical volume, but being inside the magnetic field is a property of a placement. The beam pipe places the same modules inside the solenoid fringe field and again ten metres down the beam line where the field is zero. One logical volume cannot describe both cases. The helper clones a subtree and remaps every medium to a field-free variant, so that field-free placements can use the clone. Details: - shapes and placement matrices are shared with the original - existing hand-written _NF media are preferred, new ones are derived only where none exists - clones are cached per volume and suffix - cloning refuses subtrees containing sensitive volumes, because hits are recorded against geometry paths This builds on the bug fix in fix-nf-media-in-field and prepares restoring the optimization it gave up. The affected placements were found with the o2-sim-geometry-doctor tool (AliceO2Group#15683). Related tickets: https://its.cern.ch/jira/browse/O2-174 https://its.cern.ch/jira/browse/O2-175 Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0178JHAQ2biaX5ac9ThppBXi

This switches modules whose placements are all outside the field to field-free media in place. The short RB24/2 tubes, the RB26/5 compensator with its flange, and the SAA2/SAA3 shielding need no second volume, because none of their placements is in the field. The pipe modules move to their existing hand-written counterparts. SAA3 has none, so a field-free medium is derived by MaterialManager. RB26/4 reaches into the muon dipole and is left unchanged. Field calls in the reassigned volumes go to zero. Found with the o2-sim-geometry-doctor tool (AliceO2Group#15683). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0178JHAQ2biaX5ac9ThppBXi

This fixes a problem in the beam pipe media assignment, diagnosed with the o2-sim-geometry-doctor tool (AliceO2Group#15683). 22 volumes of the RB24 warm module, its ion pump and its B1 bellow, and the air shell at r = 79-80 cm, carried media labelled out-of-field, whereas their placements at z = 400-490 cm are inside the 4.2-4.6 kGauss solenoid fringe field. Geant4 honours the label and moved particles through these volumes on straight lines, about 171000 steps per min-bias event. This is now fixed by assigning the in-field counterpart media at the construction site of these volumes. After the change no step in a field-free volume sees a non-zero field. The same volumes are also placed far down the beam line where the field really is zero. These placements now integrate the field too, which is correct but slower. A follow-up gives them cloned volumes of their own and restores the shortcut. Related tickets: https://its.cern.ch/jira/browse/O2-174 https://its.cern.ch/jira/browse/O2-175 Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0178JHAQ2biaX5ac9ThppBXi

This commit provides a helper MaterialManager::cloneSubtreeWithMediumSuffix for passive modules. A tracking medium belongs to a logical volume, but being inside the magnetic field is a property of a placement. The beam pipe places the same modules inside the solenoid fringe field and again ten metres down the beam line where the field is zero. One logical volume cannot describe both cases. The helper clones a subtree and remaps every medium to a field-free variant, so that field-free placements can use the clone. Details: - shapes and placement matrices are shared with the original - existing hand-written _NF media are preferred, new ones are derived only where none exists - clones are cached per volume and suffix - cloning refuses subtrees containing sensitive volumes, because hits are recorded against geometry paths This builds on the bug fix in fix-nf-media-in-field and prepares restoring the optimization it gave up. The affected placements were found with the o2-sim-geometry-doctor tool (AliceO2Group#15683). Related tickets: https://its.cern.ch/jira/browse/O2-174 https://its.cern.ch/jira/browse/O2-175 Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0178JHAQ2biaX5ac9ThppBXi

This switches modules whose placements are all outside the field to field-free media in place. The short RB24/2 tubes, the RB26/5 compensator with its flange, and the SAA2/SAA3 shielding need no second volume, because none of their placements is in the field. The pipe modules move to their existing hand-written counterparts. SAA3 has none, so a field-free medium is derived by MaterialManager. RB26/4 reaches into the muon dipole and is left unchanged. Field calls in the reassigned volumes go to zero. Found with the o2-sim-geometry-doctor tool (AliceO2Group#15683). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0178JHAQ2biaX5ac9ThppBXi

The L3 magnet structure, the compensator coils and their supports sit outside the region the field map covers. They were built from media that declare a magnetic field, so Geant4 integrates a field there that always evaluates to zero. Compensator.cxx already declared a field-free flag for exactly this case and never used it. Seventeen volumes are now tracked without a field. Five media whose every placement lies outside the field are switched in place. Three volumes share their medium with volumes that do reach the field, so they get a field-free variant of that medium instead: the horizontal compensator coil, the L3 yoke and the L3 crown. A new medium also needs its own line in the module's simcuts file. Without one it falls back to the global default cuts, and because Geant4 assigns production cuts per material rather than per medium, the material's other medium falls back with it. COMP 18 therefore repeats the cuts of COMP 17. The volumes were found with the o2-sim-geometry-doctor tool (AliceO2Group#15683). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0178JHAQ2biaX5ac9ThppBXi

This changes the media of volumes that sit outside the field map, diagnosed with the geometry doctor. 17 volumes were built from media that declare a magnetic field, whereas the field there is always zero. This is now fixed by switching five media in place, and by giving a field-free variant to the three volumes whose medium is shared with volumes that do reach the field. The new copper medium also needs its own line in simcuts_COMP.dat, because Geant4 assigns cuts per material and it would otherwise take the default ones. AliceO2Group#15683 Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

sawenzel requested a review from a team as a code owner August 16, 2026 18:53

sawenzel requested a review from amorsch August 16, 2026 19:00

sawenzel mentioned this pull request Aug 17, 2026

Fix wrong field-free media assignment in the beam pipe #15685

Closed

sawenzel merged commit 7c7b0ae into AliceO2Group:dev Aug 18, 2026
9 of 10 checks passed

sawenzel mentioned this pull request Aug 19, 2026

Use field-free media for the L3 magnet and the compensator #15694

Open

Navigation Menu

Uh oh!

Add o2-sim-geometry-doctor, a geometry against field audit - #15683

Add o2-sim-geometry-doctor, a geometry against field audit#15683
sawenzel merged 1 commit into
AliceO2Group:devfrom
sawenzel:geometry-doctor

sawenzel commented Aug 16, 2026 •

edited

Loading

Uh oh!

alibuild commented Aug 17, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

sawenzel commented Aug 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

alibuild commented Aug 17, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Milestone

Development

Uh oh!

2 participants

sawenzel commented Aug 16, 2026 •

edited

Loading