The Automatic Galaxy Collision Software

Abstract

The key to understanding the physical processes that occur during galaxy interactions is dynamical modeling, and especially the detailed matching of numerical models to specific systems. To make modeling interacting galaxies more efficient, we have constructed the `Automatic Galaxy Collision' (AGC) code, which requires less human intervention in finding good matches to data. We present some preliminary results from this code for the well-studied system Arp 284 (NGC 7714/5), and address questions of uniqueness of solutions. Comment: To appear in the Proceedings of the Galaxy Wars: Stellar Populations and Star Formation in Interacting Galaxies Conference

Full text

Page 1

arXiv:0908.3478v1 [astro-ph.CO] 24 Aug 2009
**FULL TITLE**
ASP Conference Series, Vol. **VOLUME**, c ? **YEAR OF PUBLICATION**
**NAMES OF EDITORS**
The Automatic Galaxy Collision Software
Beverly J. Smith1, Chris Carver1, Phillip Pfeiffer2, Sam Perkins2, Jason
Barkanic2, Steve Fritts2, Derek Southerland2, Dinikar Manchikalapudi2,
Matt Baker2, John Luckey2, Coral Franklin1, Amanda Moffett1,3, and
Curtis Struck4
1Dept of Physics and Astronomy, East Tennessee State University.
2Computer and Information Science Dept, East Tennessee State
University.3Now at University of North Carolina, Chapel Hill.
4Dept of Physics and Astronomy, Iowa State University.
Abstract.
galaxy interactions is dynamical modeling, and especially the detailed matching
of numerical models to specific systems. To make modeling interacting galaxies
more efficient, we have constructed the ‘Automatic Galaxy Collision’ (AGC)
code, which requires less human intervention in finding good matches to data.
We present some preliminary results from this code for the well-studied system
Arp 284 (NGC 7714/5), and address questions of uniqueness of solutions.
The key to understanding the physical processes that occur during
1 Introduction
Gravitational interactions and collisions between galaxies play an important
role in galaxy evolution. Computer simulations of such encounters can provide
detailed matches to observations of specific systems, providing information about
star formation triggering and mass transfer between galaxies (Struck & Smith
2003, hereafter SS03). However, the process of finding the best model for a
particular system is time-consuming, tedious, and subjective, requiring a series
of runs with the parameters being varied by hand, until the model matches the
observed appearance of the galaxy, with a good match being determined by eye.
Thus, only a handful of galaxy pairs have been modeled in detail. Furthermore,
little information is provided about the uncertainty in the parameters or the
uniqueness of the model.
To address these issues, we have created the ‘Automatic Galaxy Collisions’
(AGC) code. This code combines a galaxy interaction code with a genetic algo-
rithm and a pattern-matching routine to automatically find models that match a
particular galaxy. This modeling project is being done in parallel with a Spitzer
IR and GALEX UV imaging study of three dozen nearby interacting galaxies
(Smith et al. 2007, 2010a,b; Giroux et al. 2010). Our goal is to use the AGC
code to produce preliminary models for our full set of galaxies. These will then
be used as initial guesses for an N-body code, a method that was used in SS03.
Our approach parallels that of Theis & Kohle (2001) and Theis & Spinneker
(2003), who used a similar method to model the NGC 4449/DDO 125 and M51
encounters. Evolutionary codes have also been used to model the M81/M82/NGC
3077 system (Gomez et al. 2004) and the Milky Way/LMC/SMC interaction
1

Page 2

2
(Ruzicka et al. 2006). An alternative approach is used by Wallin et al. (2010),
who are harnessing the efforts of Galaxy Zoo participants to model galaxy inter-
actions. Another approach has been used by Barnes & Hibbard (2009), whose
“Identikit” interactive graphics program provides a way for users to find a model
for a particular system.
2 The Automatic Galaxy Collisions (AGC) Code
Our AGC code combines a standard restricted 3-body galaxy interaction code
(Wallin 1990) with a revised and updated version of the PIKAIA genetic algo-
rithm (Charbonneau 1995) and our own pattern-matching routine. The AGC
code runs a large number of galaxy interaction simulations with randomly-
selected interaction parameters. It then uses the pattern-matching routine to
determine how well each model fits the observations of the galaxy system. Based
on these fitnesses, it then runs a new generation of models, using parameters
closer to those of the real system. Over many generations, one can converge
upon a model which matches the real system very well. By using multiple runs
with different initial conditions, the question of uniqueness can be investigated.
The original PIKAIA program was converted to a parallel code by Metcalfe
(2001). We have re-engineered this software, repackaging and streamlining the
code, improving the genetic algorithm, making the code user-friendly, and cre-
ating a test suite to validate the code.
The AGC pattern matching routine begins with a sky-subtracted smoothed
image of the galaxy system being modeled. Next, for each timestep of the sim-
ulation, the model image is rotated to align the galaxy-galaxy axis with that of
the real system, scaled to match the radius of the primary galaxy, and regridded
to match the image. Then, a morphological chi-square is calculated for each
timestep, summing over each pixel, weighting the tail and sky pixels differently
from the disk regions, and scaling down the model disk points to account for dust
extinction. Finally, a weighted chi-square is computed to account for differences
between the model and the real kinematic line of nodes and the separation be-
tween the two model galaxies compared to that of the real galaxies. The fitness
is the reciprocal of this weighted chi-square.
3 Results on NGC 7714/5
The AGC code was first tested using the interacting galaxy pair NGC 7714/5
(Figure 1a). This system has been modeled ‘by hand’ using the Wallin (1990)
restricted 3-body code (Smith & Wallin 1992, hereafter SW92), and an N-body
code that includes gas hydrodynamics, star formation, and ISM heating (SS03).
Over 1.5 months, we did 36 AGC runs for NGC 7714/5 with different initial
random seed numbers. These runs were made on the ETSU ‘blackpearl’ com-
puter cluster, which consists of 60 Dell 1950 blade computers. In these runs,
only the 7 parameters that determine the positions, orientations, and orbits of
the two galaxies were allowed to vary. The 36 runs converged to four different
‘best-fit’ models. Models W and X have retrograde orbits, while Y and Z are
prograde. Models W and Z are near-head-on collisions, while X and Y are im-
pacts near the edge of the disk. All of these models approximately match the

Page 3

3
-10 -10 -5 -50055
-10-10
-5-5
00
55
X
RUN 16
Figure 1.
North is up while east is to the left. NGC 7714 is the larger galaxy to the
west. Right: The best-fit simulation for Run 16 - a Z model.
Left: Arp (1966) Atlas photograph of NGC 7714/7715 (Arp 284).
pair separation and basic kinematics, and all do an adequate job of reproducing
the bridge. To varying degrees, each model type also reproduces the tidal tails,
except for the X model, which does not match the western NGC 7714 tails well.
The fact that different runs produced different ‘best’ models shows that multiple
solutions are possible when the input is limited to an optical image with line of
node information, and a restricted 3-body code is used. However, the quality of
the best-fit models and the limited number of solutions shows that within these
limitations the code works adequately, and limits the set of candidate solutions
to explore further.
Of the four model types, model Z has the highest fitness and the best match
to the system by eye. The results of run 16, a Z model, is shown in Figure 1b.
This does a good job of producing the bridge and the long western tail. In
3-dimensional views, a long tail extends from the eastern side of NGC 7715
and curves behind NGC 7714 away from the observer. This is not visible ‘from
Earth’ since it is in the same plane as the bridge and both galaxies.
While the parameters of the earlier ‘by hand’ SW92 model do not exactly
match any of our four model types, it exhibits similarities to our X model,
being a retrograde encounter with an off-center collision. In contrast, the ‘by
hand’ SS03 hydrodynamical model most closely resembles our Z model, being a
prograde near-head-on collision. In particular, both the SS03 model and our Z
model produce a long eastern tail that curves back behind the bridge.
Figure 2a shows the corresponding fitness versus generation plot for run
16, showing an improvement in the fitness with increasing generation. After 61
generations, the best fit model has a large impact parameter and no tails. By
generation 97, the best-fit model had changed to an almost head-on collision.
At this point, the western tail of NGC 7714 is reasonably well-matched, but
the NGC 7715 tails are incorrect (Figure 2b). By the 259th generation, this
discrepancy has been fixed, and the tails are aligned with the bridge.

Page 4

4
-15 -15-10-10 -5-5
X
0055
-15 -15
-10-10
-5-5
00
55
RUN 16, GENERATION 97
Figure 2.
with generation. Right: The best-fit simulation in Generation 97 for Run 16.
Left: Fitness plot for Run 16, showing the evolution of fitness
To further test our code, we took the best-fit model from Run 16 (Figure
1b) as the ‘real’ galaxy. We then ran eight ‘rerun’ runs, with this as our ‘input
image’. All 8 models converged to the same basic four model types. This shows
that, within the limitations of the code, we can reproduce the basic structure
of the galaxy. Rerun 6 gave us the closest match, producing a Z model that
matches the Run 16 model in four of the seven parameters. The main difference
is the appearance of the NGC 7715 tails, which are not lined up with the bridge.
Acknowledgments.
Term Space Astrophysics (LTSA) grant NAG5-13079 and East Tennessee State
University Research Development grant RD0094.
This research has been supported by NASA Long
References
Arp, H. C. 1966, Atlas of Peculiar Galaxies
Barnes, J. E. & Hibbard, J. E. 2009, AJ, 137, 3071.
Charbonneau, P. 1995, ApJS, 101, 309
Giroux, M. L., et al. 2010, this proceedings.
Gomez, J. C., et al. 2004, Proceedings of the XIII Astronomical Data Analysis Software
and Systems conference, ASP Conference Proceedings
Metcalfe, T. PhD Dissertation, University of Texas, Department of Astronomy, 2001.
Ruzicka, A., et al. 2006, IAU Symp. 235, Galaxy Evolution Across the Hubble Time.
Smith, B. J., et al. 2010a, in preparation.
Smith, B. J., et al. 2010b, this proceedings.
Smith, B. J., et al. 2007, AJ, 133, 791 2007
Smith, B. J. & Wallin, J. F. 1992, ApJ, 393, 544.
Struck, C. & Smith, B. J. 2003, ApJ, 589, 157.
Theis, C. 2004, IAU Symp. 220, p. 461
Theis, C. & Kohle, S. 2001, A&A, 370, 365
Theis, C. & Spinneker, C. 2003, Ap&SS, 284, 495
Wallin, J. F., et al. 2010, this proceedings
Wallin, J. F. 1990, AJ, 100, 1477.