STAR  Computing  Simulation  
Track by track comparison of the Geant 3 and TGeo propagation engines


We have conducted a test aimed at validating the TGeo propagation engine, as compared to the standard Geant 3.21.

In the course of this test, an identical set of kinematic tracks was injected into respective models of the STAR detector, produced from the same source (based on the standard STAR description within Geant 3.21; the TGeo model was obtained by using a converter now incorporated in the St_geant_Maker, and/or by running a standalone utility g2root (such results are concidered separately).

Consider a particular volume. Tracks entering this volume will travel a certain distance inside it before exiting. The following metrics were considered in this test:

  1. The average length of tracks having traversed the volume
  2. The width of the distribution of the above
The rationale behind it was of course that on the one hand it allows to cover a large number of volumes without specific coding, and that on the other hand close numerical results in these metrics would signal the consistency of track propagation across both engines.


Let's enumerate the two cases as 1 and 2. The following convention was used:

  • Nk1,2 -- numbers of entries into the volume number "k"
  • Ak1,2 -- average track lengths in the volume number "k"
  • Dk1,2 -- widths of the above, in the volume number "k"
The following "goodness" vlaues were then defined:
    dN = (N1-N2)/sqrt(N1+N2)
    dA = (A1-A2)/sqrt(D1*D1/N1 + D2*D2/N2)

Test cases included a "TPC ONLY" geometry, as well as the full STAR with the exception of the SVT, which due its complexity merits its own study. We also used different conversion schemes as mentioned above. In the end, one has the following sets of data:

  1. tpc_only + GSTAR
  2. tpc_only + STAR conversion to TGeo
  3. tpc_only + G2ROOT conversion to TGeo
  4. full-svt + GSTAR
  5. full-svt + STAR conversion to TGeo
  6. full-svt + G2ROOT conversion to TGeo


Below is a condensed summary of the tests that have been done:

  • 1 vs 2: discrepancies were found which can be explained by (a) handling of the divisions in Geant 3 vs emulation of the divisions with volume copies in TGeo, which may lead to tiny cracks between volumes due to finite numerical accuracy (b) Imprefection in the STAR conversion scheme for volumes of zero size, whereby there is a duplicate name rejected by TGeo
  • 1 vs 3: match
  • 4 vs 5: discrepancies of the kind described above
  • 4 vs 6: g2root failure

Last updated by Maxim Potekhin on 06/07/2004