GISMO vs. GEANT Simulation of SLD


This study extends a recent attempt to verify the GISMO EC and barrel algorithms, where GISMO was configured to SLD LAC (barrel) and luminosity monitor (EC) specifications.  Data from that previous study indicates a very good hadronic resolution for the LAC, too good when compared to accepted SLD data.  GISMO seems to produce 0.40E hadronic resolution independent of detector specs.  GEANT and GISMO seem to disagree on the gross energy deposition of an ionizing hadron.

Hadronic showers can be very complex leaving many factors to be investigated.   This study has three objectives:

i.) To isolate GHEISHA production in GISMO.  GISMO uses GHEISHA to interact hadrons and produce shower particle ids and momenta.  If there were an over population of certain particle types (pi-0 for example) HAD resolution might be affected.

ii) To study GEANT calorimetry in more detail and to provide a point-by-point comparison to GISMO.  Such comparison will isolate aspects of both shower simulations and highlight their differences.

iii) To uncover any relevant coding errors.  If coding errors exist in GISMO's shower simulation, symptoms might impact calorimetry and resolution in some unanticipated way.


ROOT and other FLCD tools are listed in the previous study "Methods".  GEANT data was analyzed using the SLD packages MIDAS and MFIT.  MFIT and ROOT both use the CERN MINUIT Minimization Package. 

Data and Results

GEANT testbeam pions (SLD type: 17) were generated through SLD MCDiag, and given constant momentum and theta values.  The SLD subsystem processing and geometry were removed via Ducskal:konly.ida.

The geometry of GISMO's barrel LAC is described in the previous study.  The layering here includes Al components:

which represent the cryostat and housing of each LAC component. 

The 5k pion events range in energy from 4 to 18 GeV in 2 GeV increments.  Incident energies above 6 GeV were fitted with a 3rd order polynomial approximating the non-Gaussian tail.  The resolutions and (s/<E>)*E are tabulated below in order of increasing E where <E> is detector response and E is incident energy.  Table error entries refer error on (s/<E>)*E values.

- GEANT - - - GISMO -
s/<E> (s/<E>)*E error - s/<E> (s/<E>)*E error
- - - - - - -
0.32 0.63 0.10 - 0.27 0.53 0.02
0.24 0.60 0.07 - 0.22 0.55 0.03
0.20 0.57 0.16 - 0.21 0.61 0.01
0.20 0.64 0.14 - 0.19 0.61 0.02
0.18 0.63 0.14 - 0.18 0.62 0.03
0.18 0.68 0.15 - 0.17 0.65 0.03
0.17 0.69 0.13 - 0.16 0.62 0.02
0.19 0.79 0.28 - 0.18 0.76 0.04

When a shower is broken into SLD established sub-sections, energy deposition readout by GISMO is ~10% of GEANT values.  The absolute dE/ dx differs by ~7%.  GISMO and GEANT differ in energy loss mechanisms and calculation of dE/ dx.  SLD parameterization was disabled allowing for full shower simulation, and resulted in a 10% improvement in GEANT resolution.

GHEISHA interaction of the primary particle is simulation independent.  It produced nearly the same shower particles whether conducted by GISMO or by GEANT.   The shower particle momenta, and track ids were nearly identical.

An error was discovered in the code responsible for GISMO barrel layering.  The problem was addressed and it's related symptoms do not appear in the data listed above.


There are 10X the number of events at each energy level compared with the last study (5000 events vs. 500).   The increased reliability of more statistics, our discovery of a coding error, and the validation of GHEISHA production in GISMO make the results here much more convincing.

SLD hybridization of hadronic showers has a significant impact on GEANT resolution for higher energy primaries (E>10 GeV).   Allowing full showers in GEANT greatly improves it's resolution, and the agreement between GISMO and GEANT data.

The differences between the two simulations seem reconcilable, including their resolutions which are in much more satisfactory agreement.  From this we conclude that GISMO is demonstrating valid calorimetry.

C.Colgur and R.Dubois

Last Modified: 01/13/04 12:59 PM