Status of TFluka: geometry and validation Andrei Gheata ALICE Off-line week, 21 Feb. 2005
Outline Geometry in TFluka TGeo developments related to TFluka Validation: TFluka vs. FLUKA Conclusions
Geometry FLUKA native geometry cannot represent a G3-like description at ALICE scale FLUGG: an interface to GEANT4 geometry – dropped due to instability TGeo interface –creating/working with the same geometry used with GEANT3 Geometry building interface implementing VMC methods: Material, Mixture, Medium, Gsvolu, Gspos, Gsposp, Gsdvn, … Using common interface TGeoMCGeometry in most cases Methods with FLUKA-specific implementation in TFlukaMCGeometry Navigation interface implemented via specific wrapper functions No documentation -> “interface discovery” done by looking into FLUGG Implementation much simplified compared to FLUGG Needed building specific features directly in TGeo
Geometry in VMC VMC TGeant3 TGeant4 TFluka G3 geom G3 geom. ROOT geom. G4 geom. GEANT3transport GEANT4transport FLUKAtransport USERCODE VGM Flugg g3tog4 ROOT geom. VGM g2root Not yet
Additional geometry tasks in TFluka Creation of FLUKA materials corresponding to GEANT3 ones Not obvious – FLUKA does not support materials defined by effective Z Currently a patch is substituting some “averaged” materials with “close-to-real” accepted ones Considerable effort made by several people to do this – should be properly handled at detector level Assigning materials to FLUKA regions Handling low-energy neutron x-sec related indices Automatic PEMF file generation
TGeo developments related to TFluka Physical node ID management Equivalent to FLUKA “lattice” concept Int_t TGeoManager::GetCurrentNodeId() Void TGeoManager::CdNode(Int_t node_id) Avoids managing geometry states as objects (new/delete at each step) Adds just ~6 MB data in memory for full ALICE Boundary-tolerant distance computation algorithms Avoids potential numerical problems when crossing boundaries Leads to visible improvements – see next
Boundary tolerance The particle position (x,y,z) and the current physical volume supposed to contain this may become numerically inconsistent after crossing a boundary Symptoms depending on transport engine G3 transport careful not to “step” on boundaries. Even if something gets wrong (geometry error), “non- destructive” step recovery methods are built-in FLUKA transport systematically stepping on boundaries – numerically the new particle position may be as well in or out the new lattice; recovery methods are poor Hard to observe and debug – the systematic ranges from 1/10000 to 1/million with unpredictable effects
Boundary tolerance in TGeo Few strategies implemented in TFluka geometry interface to solve this “Push” strategy: the current point is pushed forward (~1e-10) to put it in the right geometry container 1e-10: quite empirical value – cannot work 100% Final solution implemented: boundary tolerance at geometrical shape level May the location assumption be wrong, give the transportation what it expects…
Boundary tolerance in TGeo (2) “I am inside – how long still ?” Red line “I am outside – how far next boundary ?” 0.0 !!! Implemented for all shapes used in ALICE ~1e-10 cm
Validation: TFluka vs. FLUKA Long way to current TFluka status Considerable effort – starting to pay-off Validation tests within AliRoot for TFluka vs. GEANT3 and ITS test beam (see talk from A.Morsch) Validation test against FLUKA native examples (geometry validation) – see next Geometry tests for simple examples Does the new geometry reproduce the boundary crossing sequence/positions ? - YES Do we get the same physics results in identical initial conditions ? – YES Can we reproduce the final random seeds with the 2 geometries – NOT YET, but working on understanding why
Setup 1: HCal Al+Pb with scintillators Compact (~5x10x5 cm) but complex geometry Boolean compositions for each layer (>100 components) 1 GeV protons in strong field (60 T !) Test case 1: Vacuum everywhere (no physics) Comparison of boundary crossings (w. or w.o. field) Test case 2: Materials on, physics on, same Input/pemf file Looking at Edep distribution
Setup 2: Al-Au-Al thin layers, low energy electron transport Trivial geometry, EM-cascade physics No field 1 MeV electrons along Z, all energy lost in material Comparison of shower profiles Identical run conditions Longitudinal and radial Edep distributions Z R
Results (1) Boundary crossings matching (no phys) Matching ratio: 100% with or w.o. field Difference between matched crossings at machine precision level No biasing observed
Results (2) Physics on 5000 protons 1GeV/c in 60T x-oriented field
Results (3) EM-cascade profiles (Ekin=1MeV) Low energy electron transport, primary electrons Problem when crossing boundaries observed – now fixed by FLUKA team
Conclusions Improvements in TFluka geometry interface Boundary tolerance implemented in TGeo Related fixes/adjustments in FLUKA/TFluka FLUKA VMC under extensive testing New geometry does not change/bias boundary sequence nor physics results Random seed reproducibility – next item to be followed Just preliminary not conclusive performance tests so far. We may expect however a non-negligible time penalty with respect to G3 simulation, depending on several factors