Monte Carlo simulations and generation of the SPI response

Full text

Page 1

A&A 411, L81–L84 (2003)
DOI: 10.1051/0004-6361:20031171
c ? ESO 2003
Astronomy
&
Astrophysics
Monte Carlo simulations and generation of the SPI response?
S. J. Sturner1,2, C. R. Shrader1,2, G. Weidenspointner1,2,3, B. J. Teegarden1, D. Atti´ e4, B. Cordier4,
R. Diehl5, C. Ferguson6, P. Jean3, A. von Kienlin5, Ph. Paul3, F. S´ anchez7, S. Schanne4,
P. Sizun4, G. Skinner3, and C. B. Wunderer5
1Code 661, NASA/Goddard Space Flight Center, Greenbelt, MD 20771, USA
2Universities Space Research Association, 7501 Forbes Blvd. #206, Seabrook, MD 20706, USA
3Centre d’´Etude Spatiale des Rayonnements, BP 4346, 31028 Toulouse Cedex 4, France
4DSM/DAPNIA/Service d’Astrophysique, CEA Saclay, 91191 Gif-sur-Yvette, France
5Max-Planck-Institut f¨ ur extraterrestrische Physik, Giessenbachstr. 1, 85748 Garching, Germany
6Dept. of Physics and Astronomy, University of Southampton, Southampton, S017 1BJ, UK
7CSIC-UV, Edificio Institutos de Paterna, Instituto de F´ ısica Corpuscular, PO Box 22085, 46071, Valencia, Spain
Received 10 July 2003 / Accepted 31 July 2003
Abstract. In this paper we discuss the methods developed for the production of the INTEGRAL/SPI instrument response.
The response files were produced using a suite of Monte Carlo simulation software developed at NASA/GSFC based on the
GEANT-3 package available from CERN. The production of the INTEGRAL/SPI instrument response also required the devel-
opment of a detailed computer mass model for SPI. We discuss our extensive investigations into methods to reduce both the
computation time and storage requirements for the SPI response. We also discuss corrections to the simulated response based
on our comparison of ground and inflight calibration data with MGEANT simulations.
Key words. INTEGRAL – SPI – gamma rays – methods: data analysis – methods: numerical
1. Introduction
In this paper we discuss the work that has gone into the pro-
duction of the Spectrometer for INTEGRAL (SPI) instrument
response matrices. We discuss the use of the SPI instrumental
calibration to evaluate the accuracy of the simulation software
and the SPI computer mass model developed at NASA/GSFC.
The methods that we have tested and implemented to produce
response matrices with sufficient angular and energyresolution
within reasonable limitations of computation and storage re-
sources are then described. This is followed by a discussion
of the use of SPI ground and flight calibration results to pro-
duce correction tables in order to correct inaccuracies inherent
to simulated responses.
2. Simulations and the SPI mass model
SPI went through an extensive ground calibration cam-
paign during April 2001 at the Commissariat ` a l’´Energie
Atomique(CEA)/Directiondes ApplicationsMilitaires (DAM)
Send offprint requests to: S. J. Sturner,
e-mail: sturner@swati.gsfc.nasa.gov
?Based on observations with INTEGRAL, an ESA project with
instruments and science data center funded by ESA member states
(especially the PI countries: Denmark, France, Germany, Italy,
Switzerland, Spain), Czech Republic and Poland, and with the par-
ticipation of Russia and the USA.
site in Bruy` eres-le-Chˆ atel, France (Atti´ e et al. 2003). We
used our MGEANT software (Sturner et al. 2000) to pro-
duce high-fidelity Monte Carlo simulations of these cali-
bration runs. MGEANT is a gamma-ray instrument simu-
lation package developed at NASA/GSFC that is based on
GEANT 3. It has several benefits over standard GEANT 3
(see wwwasd.web.cern.ch/wwwasd/geant).The MGEANT
source code contains several beam geometry and spectrum
generators that allow users to customize their executable
at compile time. The MGEANT releases are available at
our website lheawww.gsfc.nasa.gov/docs/gamcosray/
legr/integral.
Comparisons between the ground calibration and simu-
lated spectra led to improvements in the MGEANT software
as well as in the modeling of SPI and the calibration site.
This includes improvements to the modeling of the SPI Anti-
Coincidence Shield (ACS) and the inclusion of Doppler broad-
ening when simulating Compton scattering by bound elec-
trons through the use of the GLECS add-on package (see
nis-www.lanl.gov/∼mkippen/actsim/glecs for details).
A highlydetailed computermass model for SPI and the ground
calibration site was developed with particular emphasis on the
SPI cryostat including detector modules, preamplifiers, and
support structures (see Fig. 1). Much of the SPI mass model
was derived from technical drawings provided by various SPI
team institutions. A notable exception is the cryostat and its
support structure which was translated directly from CAD
Letter to the Editor

Page 2

L82S. J. Sturner et al.: Monte Carlo simulations and generation of the SPI response
Fig.1. Shown are two detailed views of the NASA/GSFC SPI com-
puter mass model. On the left is a cut-away view of SPI. On the right
is a rendered view of SPI in which the outer support tube has been
removed for clarity.
files at CEA/Saclay. Once these improvements were made, we
achieved a remarkably good agreement between data and sim-
ulation (Fig. 2).
We subsequently determined that it was necessary to in-
clude mass models for the rest of the INTEGRAL spacecraft,
including IBIS and JEM-X, when generating the SPI inflight
response. The most important reason for this is that the JEM-X
masks are within the partially coded field of view of SPI. Thus
the mass model used for generating the SPI responses contains
the SPI mass model combined with The INTEGRAL Mass
Model (TIMM) developed at the University of Southampton
(Ferguson et al. 2003).
3. Method for response generation
We have conducted investigations into methods to reduce both
the computation time and storage requirements for the SPI in-
strument response which is separated into two parts: the imag-
ing response function or IRF and the redistribution matrix file
or RMF. The IRFs are a set of FITS files which contain the de-
tector effective area as a function of the detector number, the
incident photon direction, and the incident photon energy. The
RMFs contain information about the energy distribution of de-
tector counts for photons of a given incident energy.
A full Monte Carlo calculation of the SPI response includ-
ing the energy-redistributionportion, was determined to be un-
realistically CPU and storage intensive. Accurately calculating
the energy redistribution portion of the response using such
an approach would require good statistics in every count bin
for every detector, incident photon direction, and input photon
energy. Given the energy and angular resolution of SPI, this
would require the simulation of an enormous number of inci-
dent photons.
We were able to reduce the number of simulated photons
that were necessary by separating the response generation pro-
cess into 2 parts: a ray tracing part that takes into account
0200400600
800
Energy (keV)
1.5
1.0
0.5
0.0
Counts/s/keV
Fig.2. Comparison of SPI Bruy` eres-le-Chˆ atel ground calibration data
for a137Cs source (black) and the MGEANT simulation (blue), after
MGEANT and mass model improvements, for single detector events
summed over the entire camera. Note that the low-energy line seen
in the data was not included in the simulation and that the vertical
scale has been greatly expanded to exagerate the small differences be-
tween the curves. For example, the count rate in the 661−662 keV
photopeak bin of the calibration data is ∼31 counts/s. The total counts
above 50 keV in the two curves differ by only ∼3%.
radiation processes in the mask and shield, and a Monte Carlo
part that takes into account radiation processes in the cryo-
stat and detector modules (Fig. 3). The ray tracing portion
reproduces the efficiency modulation on small angular scales
produced by the mask and shield with the limitation that an
interactionis treatedasabsorption.Thisis a reasonableapprox-
imation since any scattering that takes place in this portion of
the mass model is generally occurring far from the detector ar-
ray. Thus a photon that scatters there is unlikely to interact in
a detector. We then use full Monte Carlo techniques, including
an accurate treatment of scattering, to model the interaction of
photons once they reach the top of the cryostat. Since Monte
Carlo techniques are used for interactions within the detectors,
this method is used for the calculation of responses for both
single and multiple detector events.
This separation into ray tracing and Monte Carlo portions
significantly reduces the overall calculation time because the
number of directions for which the Monte Carlo portion is run
is greatlyreduced.The raytracingpart containsthe modulation
of the incoming photon beam due to the mask. This requires a
fine grid of incidence angles. 0.5◦pitch was chosen based on
the expected 2−3◦angular resolution of SPI. The Monte Carlo
portion varies only over larger angular scales and thus need
only be calculated on a much coarser angular grid, 5◦pitch.
We have further reducedthe numberof Monte Carlo events
required by utilizing symmetries in the detector array and by
implementing a decomposition of the IRFs and RMFs. We
have found that for a given input photon energy, the shape of
the RMF can be well characterized by a linear combination
of 3 template shapes whose normalizations are stored in the
IRF. These 3 components correspond to the detector response
to (1) photopeak events, (2) non-photopeakevents that interact
first in a detector, and (3) photons that interact first in passive
Letter to the Editor

Page 3

S. J. Sturner et al.: Monte Carlo simulations and generation of the SPI responseL83
Ray Tracing
Monte Carlo
Full Monte Carlo
Cryostat
Shield
Mask
Detector Array
Fig.3. Schematic illustration of the decomposition of the response
generation process into ray tracing and Monte Carlo parts.
material such as the cryostat. These templates, to a good ap-
proximation, do not vary with detector or direction, only their
normalizationchanges.Thusoncethetemplatesaredetermined
for a given photon energy, it is only the normalization which
need be calculated for every direction and detector. This pa-
rameterization requires large numbers of events only in each of
the 3 components, thus greatly reducing the computation time
and storage requirements for the simulated event data. In ad-
dition, the RMF decomposition into static components can be
exploited in the analysis software leading to a significant re-
duction of memory use.
The Monte Carlo and ray tracing data sets were created us-
ing the MGEANT simulation software. The response genera-
tion software then combines this information into a set of IRFs
forarangeofincidentphotonenergies.TheresultingIRFs each
contain a four dimensional matrix where the four dimensions
are: two angular dimensions (a parameterization of θ and φ), a
third dimension for the set of detectors, and a fourth dimension
for the 3 components. An example of one plane of an IRF is
shown in Fig. 4.
4. Correction factors
Once the initial set of IRFs have been generated, a series of
multiplicative correction factors are applied. These factors are
organized into a set of tables that are used to determine a total
correction factor for each element of each IRF.
4.1. Correcting for variations in the Ge crystal mass
The first set of correction factors is to compensate for minor
mass differences among the 19 detectors. Given the way in
which the detector numbers are referenced within MGEANT,
it was necessary to make the detectors identical in the SPI mass
model. The masses of the 19 Ge crystals in the actual SPI de-
tectors are known to vary by up to 3.7% about their mean mass
of 950.91g which is used in the mass model. Atti´ e et al. (2003)
Fig.4. This illustratesthe SPI IRF at 508.33 keV for the photopeak ef-
fectiveareaasafunctionof direction for detector 0. Notethat themask
pattern is clearly seen. The hexagonal boundary is a consequence of
the shape of SPI˜Os BGO collimator.
has shown,using analysis of the Bruy` eres-le-Chˆ atelcalibration
data, that variations in individual detector efficiencies about
their mean closely follow their variation in mass.
Given the good agreement between detector mass and ef-
ficiency variations, the first set of correction factors is merely
the ratio of the actual detector Ge crystal mass to their average
mass. For cases in which a photon interacts in multiple detec-
tors, the correction factor is the ratio of the average mass for
those detectors to the mass used in the model.
4.2. Correcting for simplification of the mask support
structure in the mass model
The SPI mask support in the computer mass model is a simpli-
fication of the actual support structure. The main part of actual
support structure is a honeycomb panel with 1/8 inch diame-
ter hexagonal cells made of NOMEX1aramid fiber. In the SPI
mass model, this honeycomb panel is replaced by a uniform
panel of NOMEX with the same nominal density as the hon-
eycomb. This simplified structure has an average transparency
very similar to that of the actual flight model (FM) support
panel but the model transparency shows less variation with an-
gle at low energies than does the FM panel (see Fig. 5). To
correct for this, we created a table of multiplicative correction
factors using measured transparency values and values calcu-
lated from the GEANT cross sections and the mass model.
A set of transparency measurements were made on the FM
support panel at 17, 21, 31, 35, 59.5, 81, 356, and 511 keV for
incidence angles of 0◦, 1◦, 2◦, and 10◦from normal (S´ anchez
et al. 2001). Measurements were performed for a set of 64 dif-
ferent locations on the panel (locations with no tungsten mask
elements present) from which a mean transparency was de-
termined for each photon energy and incidence angle. To ex-
tend the table to higher photon energies, measurements were
also made made using the STM/QM mask support at 898 and
1836.1 keV. We then calculated the transmission of the mass
model mask support at these energies and angles. The table
of correction factors was then constructed by taking the ratio
1NOMEX isa registered trademark of the E. I.du Pont deNemours
and Company.
Letter to the Editor

Page 4

L84S. J. Sturner et al.: Monte Carlo simulations and generation of the SPI response
20.0
30.0
40.0
50.0
60.0
70.0
80.0
90.0
0.0 20.040.060.0 80.0100.0
FM - 0o
FM - 1o
FM - 2o
FM - 10o
MGEANT - 0o
MGEANT - 1o
MGEANT - 2o
MGEANT - 10o
Transparency (%)
Energy (keV)
(a)
20.0
30.0
40.0
50.0
60.0
70.0
80.0
90.0
0.0 20.040.0 60.0 80.0 100.0
FM - 0o
FM - 1o
FM - 2o
FM - 10o
MGEANT - 0o
MGEANT - 1o
MGEANT - 2o
MGEANT - 10o
Transparency (%)
Energy (keV)
(b)
Fig.5. Comparison of the transparency measurements of the FM hon-
eycomb mask support panel and the transparency calculated for the
mass model panel before a) and after b) corrections were applied.
Since the correction factors are merely the ratio of the measured tran-
parencies to the model transparencies, the corrected model data are,
by definition, equal to the measured FM data.
of the measured transparency to the calculated mass model
transparency.
4.3. Efficiency correction factors
We have found that the photopeak effective areas in simula-
tions were ∼10% larger than found in the ground calibration
data. Thus a table of correction factors was needed to compen-
sate for this difference. The 8 meter source data from the SPI
ground calibration were analyzed using the ISDC analysis sys-
tem including single, double, and triple detector events (Diehl
et al. 2003). An effective area was calculated for each of the
strong lines from the calibration sources. The same analysis
was performed on the simulated data. Ratios of the calibration
to simulated effective areas for single, double, and triple detec-
tor events were calculated as a function of line energy. These
values were averaged overevent type and then used as an inter-
polation grid to calculate the efficiency correction at each IRF
energy.
Initial analysis of the ground calibration data was per-
formedonlyonlineswithenergiesatorabovethe59.5keVline
10
101
1
10
102
2
10
103
3
10
104
4
10
101
1
10
102
2
Energy (keV)
Energy (keV)
Effective area (cm )

Calibration

IRF - Release 3

IRF - Release 1
Effective Area (cm2)
Fig.6. The SPI camera photopeak effective areas, as contained in the
SPI IRFs, for all event multiplicities before (blue) and after (black)
the low-energy correction was applied. Releases 1 (Nov. 2002) and 3
(July 2003) of the IRFs correspond to IRF groups spi irf grp 0013
and spi irf grp 0015 at the ISDC, respectively. Other correction fac-
tors discussed in the text have been applied to both IRF data sets. Also
shown are the Bruy` eres-le-Chˆ atel data points of Atti´ e et al. (2003)
which have been corrected for absorption by the mask.
from241Am. The efficiency correction at energies <59.5 keV
were initially assumed to be constant at the value for 59.5 keV.
We found through analysis of early flight data from the Crab
Nebula and pulsar that this assumption was deficient. The SPI
efficiency decreases faster than predicted by the simulated re-
sponse at low energies and thus revised correction factors were
needed in the energy range 20−59.5 keV. Additional analy-
ses of weaker, low-energy lines from the Bruy` eres-le-Chˆ atel
groundcalibration data were performed(Atti´ e et al. 2003). The
new analyses included the 20.80 and 26.35 keV lines from
241Am and the 30.80 and 35.07 keV lines from133Ba. In Fig. 6
we show the SPI effective area summed over all detectors be-
fore and after the revision at low energies.
Periodically throughoutthe mission, INTEGRAL will con-
tinue to make observations of calibration sources such as the
Crab Nebulaand pulsarin orderto performinflightcalibrations
of the instruments. We will continue to use these observations
to monitor the accuracy of the SPI response and perform up-
dates of the IRFs as needed.
Acknowledgements. The SPI project has been completed under the
responsibility and leadership of CNES. We are grateful to ASI, CEA,
CNES, DLR, ESA, INTA, NASA, and UCL for their support.
References
Atti´ e, D., Cordier, B., Gros, M., et al. 2003, A&A, 411, L71
Diehl, R., Baby, N., Beckmann, V., et al. 2003, A&A, 411, L117
Ferguson, C., Barlow, E., Bird, A., et al. 2003, A&A, 411, L19
S´ anchez, F., Reglero, V., Chato, R., et al. 2001, in Proc. of the 4th
INTEGRAL Workshop: Exploring the Gamma-Ray Universe, ed.
A. Gimenez, V. Reglero, & C. Winkler (ESA, Noordwijk) 661
Sturner, S. J., Seifert, H., Shrader, C., & Teegarden, B. J. 2000, in
Proc. of the Fifth Compton Symp., ed. M. L. McConnell, & J. M.
Ryan (AIP, New York), 814
Letter to the Editor