Conversation
This widens the six angle arguments of o2::base::Detector::Matrix from Float_t to Double_t. - The helper only forwards to TVirtualMC::Matrix, which already takes Double_t, so the narrowing served no purpose. - Every detector that places a rotated volume goes through it, so the rotation of every placement was quantised to float. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> (cherry picked from commit 86e074bddc867b1d5278fdc5b6dc42062837cfc3)
This converts the dimensions, positions and angles of the ALICE space frame from float to double. - FrameStructure.cxx was already half converted: lbox, xxtru1..4, trd1par and dzFB were double while ptrap, ptrd1, ppgon and tpar were float. - B045cut subtracts a double-fed TGeoTrd1 from a float-fed TGeoTrap, so a single composite shape carried both precisions. - The float intermediates make algebraically identical surface expressions land about 0.1 nm apart, which VecGeom rejects with DistanceToOut = -1. - The near-coincident face pairs in the geometry fall from 2255 in 216 mothers to 2110 in 200. - createMaterials() keeps float: MaterialManager::Mixture, Material and Medium take Float_t, and initFieldTrackingParams takes float&. - makeHeatScreen, createWebFrame and createTOFRail are private and called only from this file, so their signatures move with it. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> (cherry picked from commit 4f954326933e7f58a4a0c48a7cda18ee809e16f4)
…ng planes This removes the sub-nanometre gaps between the layers of the TOF strips and converts the TOF volume tree from float to double. - The layer centres of a strip were summed in different orders, so faces that are the same plane came out up to 0.34 nm apart, and FPCB was gapped from FRGL by 0.037 nm. - The stack is now built by two running planes that meet inside the glass slot, which has a 625 um freon gap either side by design. - The Geo constants, the aligned-strip tables, their getters and the Detector helpers are double, and 45 static_cast<Float_t> inside geometry expressions are removed. - Material recipes, persisted hit variables and the Float_t* transforms used by the global tracking keep float. - TOF hits under VecGeom navigation go from 2 to 1509, against 1550 with TGeo. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
This makes the FT0 photocathode reachable by a navigator other than TGeo's and puts the reflector strips inside their own mother. - The layer thicknesses are Float_t and each layer's z was summed in a different order, so 0MTO and 0REG were 0.075 nm apart and 0REG and 0MCP overlapped by 0.24 nm. VecGeom abandoned 0REG with DistanceToOut = -1, so no Cherenkov photon was ever registered. - The stack is now built from one running plane, mirroring the addition the box shape uses for its own half-length. - 0TRE was declared one micron wider than the radiator while each reflector strip is four microns thick, so three quarters of every strip lay outside its mother. The geometry doctor reports those strips at 23.9 to 27.2 percent reachable. The container is now derived from its contents. - The strips are seated against the radiator in double, closing a 0.33 nm gap. - Neither change moves the light yield: 41783 +- 564 hits per event before and 40995 +- 534 after, with the beam aimed into FT0-A. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
This removes the gaps between the CPV frame and its active planes and converts the CPV geometry from float to double. - The frame and the active planes it touches were placed from two different float sums of the same plane, 38 nm apart in twelve pairs, and CPVF and CPVAr were 0.11 nm apart. - The GeometryParams members and getters, the shape half-lengths and the placement coordinates are double; the class version is bumped. - The twelve near-coincident face pairs in the CPV mother are gone. - In a PIPE plus CPV geometry, CPV hits under VecGeom navigation go from 0 to 6876, against 7392 with TGeo. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
This converts the TRD volume tree and the constants of TRDBase/GeometryBase.h from float to double. - createVolume forwards to the Double_t* overload of Gsvolu, so the 22 shape arrays and the placement locals move with it. - Derived constants such as CH, CHSV, ANODEPOS, FLENGTH and XETHICK were summed in float while Geometry.cxx rebuilds the same quantities in double; the constants and their 23 constexpr getters are now double. - rotateBack, kTiltAngle, the pad-plane locals and the PadPlane accessors keep float: they are reconstruction values. - Zero-length exits under VecGeom navigation fall from 307310 to 47851 over the same events, and TRD hits go from 96 % to 102 % of TGeo. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
This converts the dimensions, positions and angles of the muon station 1 quadrant geometry from float to double. - Station1Geometry.cxx builds its volumes with TGeo directly, whose API is double throughout, so the float constants and locals were the only narrowing in the chain. - SQM1 and SQM2 each carried 112 near-coincident face pairs at 38 nm, which VecGeom rejects with DistanceToOut = -1. - The near-coincident face pairs in the geometry fall from 1279 in 156 mothers to 1045 in 148. - The eleven TRAP volumes go through TGeoManager::Volume, which has a Double_t* overload. - Materials.cxx keeps float: Material, Mixture and Medium take Float_t, and initFieldTrackingParams takes float&. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> (cherry picked from commit dadef7493cd112317234e3dab392547019e90a35)
sawenzel
requested review from
a team,
afurs,
andreasmolander,
bazinski,
coppedis,
f3sch,
jotwinow,
kharlov,
noferini,
peressounko,
sahilupadhyaya92,
shahor02 and
wille10
as code owners
September 29, 2026 09:35
This converts the PHOS GeometryParams singleton and the volume tree built from it from float to double. - PTII carried 160 near-coincident face pairs and PTIH 80, at 19 nm, which VecGeom rejects with DistanceToOut = -1. - 24 of the Float_t in Detector.cxx are const Float_t* handles onto the GeometryParams tables, so widening the members alone would have narrowed them straight back. - The near-coincident face pairs in the geometry fall from 1045 in 148 mothers to 785 in 142. - GeometryParams is a process-local singleton reached through GetInstance and no instance is ever written, but its class version is bumped since the member layout changes. - PHOSBase/Geometry keeps float: it is the reconstruction cell mapping, shares no number with GeometryParams and places no volume. - CreateMaterials and the hit variables keep float. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> (cherry picked from commit 6fcb5933ff8e01c86e9567358c07756d52fa923a)
This converts the boxpar array of the ZDC geometry from float to double. - The rest of the module was already double: the Geometry.h constants and the tubpar, tubspar and conpar arrays. - boxpar was mixed into double placements, as in Gspos(Q12T, ..., tubspar[1] + boxpar[0], ...), so a double seat was computed from a float half-length. - The near-coincident face pairs in the geometry fall from 687 in 80 mothers to 683 in 79. The effect on ZDC hits is not measured: a box gun at central rapidity produces no ZDC hits at all, so this needs a generator that populates the beam line. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> (cherry picked from commit 62df60a2874d17d9e8b71a080f809ab863e224b3)
sawenzel
force-pushed
the
swenzel/geometry-float-rounding
branch
from
September 29, 2026 09:42
1191b63 to
11be4aa
Compare
bazinski
approved these changes
Sep 29, 2026
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This PR prepares the ALICE geometry for VecGeom navigation by removing nanometre-scale gaps and overlaps between volumes that are meant to touch in the TOF, FT0, CPV, TRD, PHOS, MCH station 1, ZDC and space-frame geometries.
These artefacts arise from
floatconstants and intermediate calculations, which can cause algebraically identical surfaces to differ by a fraction of a nanometre up to 38 nm. While TGeo tolerates these discrepancies, VecGeom's tighter surface tolerance can lead to zero-length steps and missed hits in sensitive volumes.The geometry is now constructed consistently in
double, and calculations of shared boundaries have been aligned so that touching faces coincide exactly. This reduces near-coincident face pairs (<100 nm) from 2255 in 216 mother volumes to about 680 in 79.The effect on VecGeom navigation is substantial:
Note that the changes are not intended to alter the geometry itself. With TGeo navigation, 500 paired Pythia8 pp events give the same hit counts and energy deposits as
devwithin 1σ for all affected detectors. The FT0 stack and radiator-wrap corrections also leave the light yield unchanged within statistical uncertainties.Overall, this makes the geometry independent of a navigator absorbing nanometre-scale rounding artefacts and provides the numerical consistency needed for VecGeom navigation.
Assisted by Claude Code.