Geantino Maps for validation of geometry description

Introduction

This page is the central repository for geantino maps of ATLAS detector components, as generated by the core simulations team for geometry validation studies. Please feel free to add your plots!

In very simple terms, geantino maps are obtained by shooting neutral geantinos and histogramming the integrated radiation lengths traversed inside the detector material. Quantitatively, for each distinct material i, we compute its thickness (Thick_i) for a step i inside the material, as Thick_i = stepLength_i / radLength_i, where the Geant4 stepLength_i spans the whole step traversed by the geantino within that material (actually a Geant4 volume).

The thickness gets integrated over several steps, according to the name of the subdetector component (say Muon). Therefore, in case a sub-component needs to be mapped (eg TGC), all other sub-components (MDT, CSC, etc) will have to be turned off in order to prevent their thicknesses to be combined (as Muon). That's because the volume names start with the system name, as Muon::TGC... Please check here on how to turn some AGDD (muon) sub-components off. Note that it is ok to leave eg Calorimeter active, as the different components are integrated separately.

Generating geantino maps

There are currently two distinct methods for generating the maps: using random sampling or single central sampling.

Random sampling

In random samplings (the original algorithm), geantinos are shot randomly, within fixed intervals (say of eta,phi), and the results are displayed using 2-D =TProfile=s, which means that the geantino map contains the average thickness traversed by all geantinos within a given profile bin. Ideally, with high statistics, the bin contains the average thickness within the area covered by the bin. The disadvantage of this method is that it is CPU-time consuming, as it may require a very large number of geantinos to reduce the uncertainties in the bin averages.

Single central sampling

A significant improvement in the generation time can be obtained by just shooting one geantino at the center of each histogram bin (no averaging is necessary, hence a common TH2F is good enough as long as a single measurement is done per bin). The drawback here is that the ranges and loops over the independent variables (e.g. x,y) have to be chosen carefully: as the loops don't nest, numbers of bins with a common factor (like nbinsX=100 and nbinsY=150) will cause the "one shot per bin" rule to be false. That's a feature of the definition of loops in the particle gun. A simple rule of thumb is to make e.g. nbinsY = nbinsX+1.

Comparing the two methods

The two algorithms described above were used for the same geometry: TGC supports only, everything else turned off. Also the same histogram limits were used: eta in the range (1.1; 2.6) and phi in the range (0; π/2). The resulting histograms are compared in the table below.

Comments	Random sampling	Single central sampling
Statistics and histograms used	(200 eta bins) x (200 phi bins), filled with 1 million events	(200 eta bins) x (201 phi bins), filled with 40200 events
At first no significant differences, although the single sampling method seems to resolve better what seem to be spaces between close bars. This view offers no information on the radiation length itself at each point, we need to look at 3D plots, see next rows.
Projecting both histograms along the eta axis, we observe excellent agreement for high eta values (note slightly different scales in y-axis). For low eta, there are clear fluctuations from random samplings, while the central sampling seems to be pretty continuous. The increased thickness at low etas seems to agree with expectations (due to longer trajectories inside the material) for single sampling only, while this trend is unclear from random sampling. The spikes are probably due to justaposed bars at those points.
This view seems to have very good agreement everywhere, except for the highest spikes, where the methods disagree on how high the spikes are. With a careful view, the single sampling seems to be more symmetric with respect to the spikes.
The perspective suggests that this quadrant contains three bars twice as thick as the remaining bars, maybe due to structural reasons: the thicker bars are probably the ones which support the TGC weight, while the thinner ones just provide support to individual chambers.

Maps of some components

The plots shown here were obtained using Geant4 user actions. The code will be posted here later.

TGC supports

MDT supports

	Geantino map of MDT support (eta-phi view), using 20000 events in random sampling.

The next geantino maps were produced using single central sampling:

There seems to be a lot of fluctuations here in these plots. Are the fluctuations real or just binning effect (sampling at center of bin sometimes may miss objects, when those are too thin to be resolved by the histogram binning resolution)?
Zooming in an interesting region, we can see that some bars are justaposed, according to the Geant4 description. At this zoom level (10mm x 10mm bins), the fluctuations seem to be due to some "dashed" pattern in the support. Dashed support structures don't seem much useful...
Further zooming (2mm x 2mm bins) confirms that the fluctuating thicknesses are not real, but due to binning resolution.

Source code used to make the plots

All the geantino map plots in this page were produced by the classes RadLengthSampler (eta vs. phi maps) and DetThicknessSamplerXY (x vs. y maps). The source code used has never been commited to CVS or SVN (maybe it should?). For now, it is available from the attachments table below. The code was used, as it is, with release 14, but in principle it should work just fine with release 15 (not tested though).

Page history

• 31 July 2008 - created
• 05 June 2009 - updated: attached source code and its section.

-- GuilhermeLima -