MNRAS 000, 1–17 (2019) Preprint 11 December 2019 Compiled using MNRAS LATEX style file v3.

Galaxy interactions in IllustrisTNG-100, I: The power and

limitations of visual identification

Kelly A. Blumenthal1,2 , Jorge Moreno2,3,4 , Joshua E. Barnes1 , Lars Hernquist2 ,

Paul Torrey5,6 , Zachary Claytor1 , Vicente Rodriguez-Gomez7 , Federico Marinacci8 ,
and
1
Mark Vogelsberger5
Institute for Astronomy, University of Hawaii, 2680 Woodlawn Drive, Honolulu, HI 96822, USA
arXiv:1912.04371v1 [astro-ph.GA] 9 Dec 2019

2 Harvard-Smithsonian Center for Astrophysics, 60 Garden Street, Cambridge, MA 02138, USA

3 Department of Physics and Astronomy, Pomona College, 333 N. College Way, Claremont, CA 91711, USA
4 TAPIR, Mailcode 350-17, California Institute of Technology, Pasadena, CA 91125, USA
5 MIT Kavli Institute for Astrophysics & Space Research, Cambridge, MA 02139, USA
6 Department of Astronomy, University of Florida, 211 Bryant Space Sciences Center, Gainesville, FL, USA
7 Instituto de Radioastronomı́a y Astrofı́sica, Universidad Nacional Autónoma de México, Apdo. Postal 72-3, 58089 Morelia, Mexico
8 Department of Physics & Astronomy, University of Bologna, via Gobetti 93/2, 40129 Bologna, Italy

Accepted 2019 December 04. Received 2019 November 05; in original form 2019 May 06

ABSTRACT
We present a sample of 446 galaxy pairs constructed using the cosmological simulation
IllustrisTNG-100 at z = 0, with MFoF, dm = 1011 − 1013.5 M . We produce ideal mock
SDSS g-band images of all pairs to test the reliability of visual classification schema
employed to produce samples of interacting galaxies. We visually classify each image
as interacting or not based on the presence of a close neighbour, the presence of stellar
debris fields, disturbed discs, and/or tidal features. By inspecting the trajectories of the
pairs, we determine that these indicators correctly identify interacting galaxies ∼45%
of the time. We subsequently split the sample into the visually identified interacting
pairs (VIP; 38 pairs) and those which are interacting but are not visually identified
(nonVIP; 47 pairs). We find that VIP have undergone a close passage nearly twice as
recently as the nonVIP, and typically have higher stellar masses.
Further, the VIP sit in dark matter haloes that are approximately 2.5 times as
massive, in environments nearly 2 times as dense, and are almost a factor of 10 more
affected by the tidal forces of their surroundings than the nonVIP. These factors
conspire to increase the observability of tidal features and disturbed morphologies,
making the VIP more likely to be identified. Thus, merger rate calculations which
rely on stellar morphologies are likely to be significantly biased toward massive galaxy
pairs which have recently undergone a close passage.
Key words: galaxies: interactions – galaxies: evolution – galaxies: structure – meth-
ods: numerical – cosmology

1 INTRODUCTION may feed the central black hole, producing heightened ac-
tivity of the nucleus (e.g., Cutri & McAlary 1985; Dahari
Galaxy encounters have been used to explain the presence of 1985; Heckman et al. 1986a,b; Ellison et al. 2011; Hewlett
peculiar galaxies (e.g., Arp 1966), and facilitate our under- et al. 2017; Trakhtenbrot et al. 2017). Interacting and merg-
standing of galaxy evolution in a number of ways. Figure 1 ing galaxies have been shown to host heightened rates of
illustrates the typical merger sequence, from initial approach star formation (e.g., Joseph & Wright 1985; Kennicutt et al.
(far left) to final coalescence (far right). These encounters 1987; Whitmore & Schweizer 1995; Vigroux et al. 1996;
lead to significant changes in stellar and gas morphology Mirabel et al. 1998; Bridge et al. 2007; Scudder et al. 2012;
(e.g., Mihos 1995; Mihos et al. 1995; Malin & Hadley 1997; Moreno et al. 2015; Rich et al. 2015; Moreno et al. 2019). In
Côté et al. 1998; Knierman et al. 2003; Lotz et al. 2008; Wen the Local Universe, gas-rich mergers manifest themselves as
& Zheng 2016; Tapia et al. 2017), including the production (ultra)luminous infrared galaxies (e.g., Sanders et al. 1988;
of non-axisymmetric torques which enable gaseous inflows Hopkins et al. 2008). Until the James Webb Space Tele-
(e.g., Duc et al. 2004; Blumenthal & Barnes 2018), which

© 2019 The Authors

2 K. A. Blumenthal et al.

Figure 1. The typical view of the merger sequence includes, from left to right: (1) two galaxies coming in on their initial approach, (2)
just after the first pericentric passage, when tidal features are prominent, (3) near second pericentric passage, when there are significant
disruptions to the discs and still-visible tidal features, (4) prior to final coalescence when the galaxies’ nuclei are nearly completely
overlapping, and (5) a post-merger remnant, featuring clear tidal shells. These simulated three-color composite images are produced via
the same procedure described in §2.2.2 utilising SDSS g, r and i magnitudes.

scope begins operations, our knowledge of these objects at tions of binary galaxy pairs to derive a realistic observability
high redshifts will remain severely limited. However, indirect timescale. They find that applying this parameter to obser-
measurements such as the observation that early discs are vational data causes the widely divergent merger rates to
dominated by large clumps of gas and dust (e.g., Lotz et al. converge. Simons et al. (2019) use synthetic galaxy images
2006; Ravindranath et al. 2006; Whitaker et al. 2015) and from zoom simulations to determine how frequently galaxies
that many early ellipsoids are very compact (e.g., Buitrago are confused for discs in merger catalogues. These interlop-
et al. 2008; Williams et al. 2014; Barro et al. 2013), have led ers confuse the disc/spheroid ratio that is often used to de-
some to postulate that mergers were much more common in fine merger rates, a trend the authors found was dependent
the Early Universe (e.g., Conselice et al. 2004; Genzel et al. upon stellar mass. More recently there has been a push to
2008; Bezanson et al. 2009; Bournaud & Elmegreen 2009; apply deep learning (e.g., Bottrell et al. 2019; Pearson et al.
Dekel et al. 2009). These findings are consistent with ΛCDM 2019; Snyder et al. 2019) techniques to synthetic galaxy im-
cosmology, which dictates that the hierarchical structure of age catalogues to assess the completeness of observationally
the universe arises from sequential mergers throughout cos- derived catalogues.
mic time (e.g., White & Rees 1978). In this work we utilise the IllustrisTNG simulation with
A fundamental component of galaxy evolution, and by a volume of ∼1003 cMpc3 (hereafter TNG100-1), one of
extension hierarchical growth, is the galaxy merger rate. In the three main runs of the IllustrisTNG cosmological suite
its simplest form, the galaxy merger rate is calculated by (Marinacci et al. 2018; Naiman et al. 2018; Nelson et al.
dividing the fraction of galaxies undergoing a merger by the 2018; Pillepich et al. 2018b; Springel et al. 2018). The Il-
typical time a galaxy interaction will be observable. The lustrisTNG model (Weinberger et al. 2017; Pillepich et al.
merger fraction is often determined by counting the num- 2018a) employs state-of-the-art prescriptions for star forma-
ber of morphologically disturbed (both automatically or by tion, chemical evolution, and feedback due to active galactic
visual inspection, e.g., Lotz et al. 2008; Jogee et al. 2009; nuclei. Recent work has shown that the IllustrisTNG model
Shi et al. 2009; Lotz et al. 2010a; Bluck et al. 2012), or matches important observational benchmarks in the chemi-
the number of galaxies in close pairs (projected or 3D, e.g., cal and metallicty evolution of galaxies (e.g., Naiman et al.
Bundy et al. 2004; Kartaltepe et al. 2007; Bundy et al. 2018; Torrey et al. 2019), the quasar luminosity function and
2009; Robotham et al. 2014; Mundy et al. 2017; Snyder black hole mass relationships (e.g., Weinberger et al. 2018),
et al. 2017). The observability timescale is also variable, and and the overall morphologies of galaxies (e.g., Rodriguez-
may depend on the orbital parameters and initial conditions Gomez et al. 2019). Using a sample of galaxy pairs from
(e.g., Conselice 2006; Lotz et al. 2010b,a), the observational TNG100-1, we generate ideal mock SDSS images to identify
method used to characterize the merger (e.g., Lotz et al. what fraction of the interacting pairs are “observable.”
2008), and the redshift of the interaction (e.g., Snyder et al. This paper is structured as follows: in Section 2 we de-
2017). Due to the breadth of observational methods used scribe the cosmological simulation used, the methods asso-
to derive these quantities, the calculated merger rate varies ciated with its data products, and the pair sample prepara-
widely (e.g., Lotz et al. 2011). However, cosmological sim- tion; in Section 3 we present and discuss our results, which
ulations are providing insight into the limitations of purely compare the TNG100-1 sample of interacting pairs at two
observational studies (e.g., Rodriguez-Gomez et al. 2015). epochs; finally, in Section 4, we state our conclusions and
In this paper, we set out to answer the following ques- briefly describe our future and ongoing work.
tions that are fundamental to the calculation of the galaxy
merger rate across cosmic time:
(i) Does the stellar morphology of a merging pair reliably 2 METHODS
indicate its dynamical history?
(ii) What makes an interaction “visible”? 2.1 IllustrisTNG
(iii) Are merger catalogs derived solely from optical ob-
IllustrisTNG is a set of N-body/magnetohydrodynamic cos-
servations biased?
mological simulations with dark and baryonic matter. Grav-
Lotz et al. (2011) use small-scale hydrodynamic simula- ity is solved using a Tree-PM algorithm that implements

MNRAS 000, 1–17 (2019)

 Galaxy Mergers in TNG100-1 3
a particle mesh on large scales and a tree code on small has been set to 8 M , and the yield tables have been up-
scales. Gas is treated as an ideal fluid on an unstructured dated (Naiman et al. 2018); (5) it includes an ideal treatment
mesh (AREPO; Springel 2010) that incorporates an ideal of magnetohydrodynamics. Further, the IllustrisTNG model
treatment of magnetohydrodynamics (Pakmor et al. 2011, was run at three different volumes to generate a simulation
2016). Gas is allowed to cool via metal-lines and radiation, series that spans a wide dynamical range: TNG50 (Nelson
and can also heat radiatively by exposure to a redshift- et al. 2019; Pillepich et al. 2019), TNG100 and TNG300.
dependent radiation field (e.g., Katz et al. 1996; Faucher- Each of these runs was initialised with three (TNG100 and
Giguère et al. 2009). High density gas can self-shield (Vo- TNG300) or four (TNG50) sets of initial conditions, often
gelsberger et al. 2013), under the appropriate optical depth indicated as e.g., TNG100-1. For more information on the
conditions. The interstellar medium (ISM) is modeled with simulation series structure, see Appendix A1. In this work,
an effective two phase model, following Springel & Hern- we utilise the run TNG100-1 for several reasons: (1) it has
quist (2003): cool clouds are in pressure equilibrium with the same set of initial conditions as the original Illustris run;
the hot diffuse medium. These simulations cannot describe (2) it has the largest number of resolution elements for its
the ISM structure in detail, but do include mass exchange volume; and (3) the volume is large enough to contain many
via cooling, star formation, and the evaporation of clouds examples of interacting galaxies (c.f., TNG50), but not too
by supernovae. This acts to harden the equation of state of large that these galaxies are poorly resolved (c.f., TNG300,
the star forming gas, and also stabilizes gas against instabil- which typically has a baryon mass resolution on the order
ity. The ISM prescription does not reach low (high) enough of 107 M ).
temperatures (densities) to properly describe the molecular Many parameters and model choices of the IllustrisTNG
gas component. These simulations do not include modeling model were calibrated using observational scaling relations
of cosmic rays nor explicit radiative transfer. and galaxy properties (Pillepich et al. 2018a). Several works
Each star particle represents a stellar population, not outline the successes of this model. Nelson et al. (2018)
an individual star, based on empirical models that include shows that the colour bimodality, which was absent in the
stellar evolution, enrichment, mass and metal returns and original Illustris, possibly due to the previous implementa-
supernova rates (Pillepich et al. 2018a). Star formation and tion of black hole feedback, was present in both TNG100
supernovae drive outflows in galaxies. Gas mass is ejected and TNG300. Rodriguez-Gomez et al. (2019) compares syn-
from star forming regions such that the wind velocity is thetic images from TNG100 to an analogous sample from
proportional to the dark matter velocity dispersion. Due Pan-STARRS. They find TNG100 to be a significant im-
to resolution limitations, outflowing material is initially hy- provement over the original Illustris suite, particularly with
drodynamically decoupled, and is re-coupled at a density respect to the galaxy morphologies. Additionally, chemical
threshold. The winds carry a sufficient metal content out of evolution (Naiman et al. 2018), galaxy mass-metallicity rela-
the galaxy, to approximately match the mass-metallicity (or, tions (Torrey et al. 2019), and the present day quasar lumi-
M-Z) relation (for further details, see Pillepich et al. 2018a). nosity function (Weinberger et al. 2018) are broadly consis-
Black holes (BHs) – and the feedback due to active tent with observations. Despite its relative success, there are
galactic nuclei (AGN) – are a key part of this simulation, still areas of contention between the IllustrisTNG model and
in particular the production of quiescent galaxies. Given the the observed universe. For example, TNG100 may underpro-
resolution of the simulation, black hole formation cannot be duce bulge-dominated galaxies, and may overproduce red
self-consistently modeled, so once a galactic halo reaches a discs and blue spheroids (e.g., Huertas-Company et al. 2019;
certain mass, a seed black hole particle is inserted at its Rodriguez-Gomez et al. 2019). The high-redshift quasar lu-
centre, which then acts as a sink particle. The black hole minosity function, driven by the feedback mechanisms em-
is thus tied to the potential minimum and grows by sub- ployed by supermassive black holes, may be in tension with
sequent mass accretion via Eddington-limited Bondi-Hoyle observations (Weinberger et al. 2018; Habouzit et al. 2019).
accretion (Springel et al. 2005). The channel of AGN feed- Additionally, it has been suggested that there is contention
back (Sijacki et al. 2007; Weinberger et al. 2017) depends between the observed and simulation H2 content in high red-
upon the accretion rate. At low accretion rates, the galaxy shift galaxies (Popping et al. 2019).
experiences a wind (or, kinetic mode), wherein kinetic en-
ergy is deposited into the gas around the black hole. The
duty cycle then ensures star formation remains suppressed. 2.1.2 Friends-of-Friends Groups and Subhaloes
At high accretion rates, the galaxy enters the thermal (or, The Friends-of-Friends (hereafter FoF) algorithm utilizes
quasar) mode; the strength of this feedback mode is a func- percolation to construct associated groups of particles
tion of the black hole mass. Full details of the IllustrisTNG (Davis et al. 1985). Dark matter particles (or chains of par-
BH feedback model are available in Weinberger et al. (2017). ¯ where
ticles) are said to be linked if they are closer than bl,
l¯ is the mean interparticle distance and is related to the
simulation’s mean number density, l¯ = n̄ −1/3 . The free pa-
2.1.1 IllustrisTNG vs. Illustris
rameter b is the ratio between the maximum linking distance
The IllustrisTNG model differs from its earlier counterpart, and the interparticle separation for a homogeneous system;
Illustris (Vogelsberger et al. 2014b,a; Genel et al. 2014; Si- in IllustrisTNG b = 0.2. Taken together, this all represents
jacki et al. 2015), in several ways: (1) it includes isotropic an approximate density threshold below which particles are
winds with velocities that scale according to the halo virial not considered associated. The baryonic (gas and stars) ma-
mass; (2) the supernova energy has two components (ther- terial is assigned to a particular FoF group based on the
mal and kinetic) which are applied to winds; (3) the wind en- membership of the nearest dark matter particle. Subhaloes,
ergy is metallicity-dependent; (4) the supernova mass limit on the other hand, are identified via the subfind algorithm

MNRAS 000, 1–17 (2019)

4 K. A. Blumenthal et al.
(Springel et al. 2001). This iteratively strips away particles using Bruzual & Charlot (2003) (assuming no dust). These
that are unbound from the central structure, until a bound include SDSS g, r, i, z, Buser U, B, V, and Palomar K. Here,
system above a certain size remains – in the case of Illus- we generate ideal mock SDSS g-band images using all star
trisTNG, this is 20 particles. In many cases, as in this work, particles bound to a FoF group. This band was chosen to
subhaloes are considered galaxies, while FoF groups may facilitate future comparisons with wide-field observational
contain pairs or groups of galaxies. surveys. To calculate the luminosity of each star particle,
we determine the true SDSS g magnitude

2.2 Galaxy Pair Samples Mtrue = Mobs − χk filter (1)

2.2.1 Parent Sample where Mobs is the TNG100 SDSS g-band absolute magni-
tude, kfilter is the filter-dependent first order extinction cor-
We select FoF haloes in the most recent snapshot (i.e., z = 0) rection, and χ is the airmass, assumed to be 1.3 for all SDSS
with a FoF group total dark matter mass between 1011 and bands in TNG100. The apparent magnitudes are needed to
1013.5 M . Additionally, subhaloes are required to have a derive the flux:
total dark matter mass between 1010.5 and 1013 M . These
mass cuts ensure that we limit ourselves to well-resolved mtrue = Mtrue + µ (2)
galaxies and haloes; that we avoid systems in which visual The distance modulus, µ, is calculated for each set of galaxy
features are driven by environmental, non-merger related pairs using a set distance of 35 Mpc for every system. The
processes. Subhaloes with a total dark halo mass less than flux is then
1010.5 M are likely to be poorly resolved in both the dark
and baryonic material. To ensure the proper mass resolution f = 100.4(mtrue −mzp ) (3)
of the stars, we place a final limit on the subhalo total stellar where mzp is the zero-point of the desired filter. For our
mass such that both subhaloes in the pair have a total stellar images, we use kfilter = 0.15, χ = 1.3, and mzp = 25.11
mass above 109 M . We place this restriction on the stellar (Stoughton et al. 2002). We project the three-dimensional
mass primarily because a preliminary inspection of the im- distribution of particles onto a flat two-dimensional plane,
ages described in Section 2.2.2 shows that it is very difficult and apply a 2D Gaussian smoothing function with FWHM
to identify tidal features in systems with stellar mass below equal to the radius of a sphere enclosing the 32 nearest star
109 M (see Section 2.2.2. However, abundance matching particles, following Torrey et al. (2015). For simplicity, we
(e.g. Sawala et al. 2015) indicates that this might lead to a use the x and y coordinates to define this plane, and do not
sample of galaxies with systematically high stellar masses. assume a location or viewing angle for an observer. Thus
No limit is placed on the distance between the subhaloes, the sample represents a random set of orientations with no
although they are required to belong to the same FoF halo. preferred observing direction. Further the images include no
We do not consider pairs that straddle two FoF haloes (as treatment of dust attenuation nor a convolution with the
in, e.g., Moreno 2012; Moreno et al. 2013), and note that SDSS resolution. This affords us optimal conditions to “ob-
these systems are not only relatively rare, but are likely to serve” any tidal features in the mocks. Figure 1 contains five
be unbound (and as such, not orbiting one another). We (rgb) examples of our ideal mock observations. For the full
consider only pairs of galaxies with a stellar mass ratio be- postage stamp collection of the interacting pairs, refer to
tween unity and 1:4 (“major merger”) at the present day. Appendix B.
Lastly, the majority of observations (e.g., Bridge et al. 2010;
Ellison et al. 2010; Larson et al. 2016; Ventou et al. 2017;
Mantha et al. 2018) and idealised simulations (e.g., Toomre 2.2.3 Visual Classification Scheme
& Toomre 1972; Barnes & Hernquist 1991, 1996; Di Matteo The merger sequence is defined by the presence (or ab-
et al. 2008; Lotz et al. 2008; Rupke et al. 2010; Bournaud sence) of tidal features. Larson et al. (2016) devised a merger
et al. 2011; Hopkins et al. 2013; Moreno et al. 2015, 2019) stage classification scheme that includes non-interacting sin-
of galaxy mergers typically assume the system is composed gle galaxies (s), minor mergers (m) and major mergers,
of only two galaxies. In order to approximate this assump- ranging from before first pericentric passage through final
tion, we required that any tertiary subhalo in the FoF group coalescence and post-merger remnant (M1−M5) of Ultra-
be at most 1/16 the stellar mass of the primary (or, most Luminous Infrared Galaxies (ULIRGs). The major merger
massive) halo. It should be noted that this restriction will sequence is as follows (Figure 1):
not exclude all recent minor mergers. There may be sys- M1 - Galaxies are well separated and appear to be on their
tems with strong tidal features at the present day due to initial approach.
low mass ratio interactions in the past. However, observers M2 - Tidal features (bridges and tails) are clearly visible,
do not have unlimited knowledge about their targets. By not and likely just after the first close passage.
removing these objects, we remain more closely connected M3 - Two individual nuclei are visible in highly disturbed
to observational surveys of interacting galaxies. Our final set overlapping discs. The tidal tails are still well defined.
of galaxies contains 446 binary galaxy pairs at z = 0. M4 - The two nuclei have now coalesced, but the tidal de-
bris are still visible.
M5 - A post-merger remnant, with a diffuse outer shell, and
2.2.2 Ideal Mock SDSS Images
little-to-no evidence of tidal features.
We generate ideal mock SDSS images for each of the 446 Using this merger stage classification as a guide, three of
galaxy pairs in our sample. TNG100 provides magnitudes the authors independently classified the pairs as either in-
in eight bands for each star particle, which are calculated teracting (roughly, stage M2-M5) or not interacting (s-M1).

MNRAS 000, 1–17 (2019)

 Galaxy Mergers in TNG100-1 5
Visual cues including the projected distance, tidal features, Does the pair appear
and stellar debris were used. Following Galaxy Zoo (Lintott to be interacting?
et al. 2008), we adopt the group consensus (2 out of 3) as Yes No
Is the pair Yes 38 47
the morphological classification for any given pair.
actually
Identifying merger stage in this way may be subject to No 18 214
interacting?
certain pitfalls. For example, the production of tidal features
Recently entered the same group 97
is dependent upon encounter geometry, and can, in some
Group 22
retrograde encounters, be completely absent (e.g., Toomre
Not Orbiting 10
& Toomre 1972; Di Matteo et al. 2007). Further, Nevin
Total Pairs 446
et al. (2019) study the effects of initial conditions on various
quantitative morphology measures which are often linked to Table 1. A complete description of the classifications for all pairs
merger activity. They find that the initial conditions, in- in the parent sample. In the top section, we report the results of
cluding viewing angle, may affect the derived quantitative the combined visual (“Does the pair appear to be interacting?”)
morphologies, thus impacting the perceived merger stage. and trajectory schema (“Is the pair actually interacting?”), in-
Dubinski et al. (1996); Mihos et al. (1998); Springel & White cluding the VIP and nonVIP which are the subsamples used in
this paper. Note that the report from the trajectory is taken to
(1999); Barnes (2016) perform systematic theoretical stud-
be “truth.” The middle section reports the number of pairs which
ies on the tidal response of interacting galaxies, spanning a were manually eliminated from the sample.
wide range of galaxy structures. They find that the visibility
of tidal features additionally depends on the galaxy’s inter-
nal structure (e.g., the dark halo mass and concentration), affect the presence of visual features. Additionally, we can
and that under certain circumstances, galaxy pairs may not expect a typical TNG100-1 tidal feature to have about 200
show any obvious signs of interaction. star particles, corresponding to less than 1% relative error
Lotz et al. (2008) use quantitative morphological met- in the feature’s resolution.
rics (e.g., Gini and M20 of Lotz et al. 2004) of simulated
galaxy mergers to dive deeper into the idea of time-scales
for tidal features. They determine when, over the course of 2.2.4 Trajectory Classification
an interaction, the tidal response and subsequent morpho-
logical disruption are greatest, and find that galaxies tend Using the Sublink merger tree (Rodriguez-Gomez et al.
to exhibit strong tidal features at first pericentre and near 2015), we extract the 3D orbital motion of the secondary
final coalescence. In contrast, the morphologies at interme- with respect to the primary. A pair is considered “interact-
diate passes are largely consistent with a control sample of ing” if it has had at least one close passage, the pair is at or
isolated galaxies. nearing an apocentre (i.e., there is an apparent turnover in
In addition to the object’s look-back time, internal the relative separation), and there is apparent orbit decay
structure, and encounter geometry, the observer viewing an- (i.e., sequentially deeper pericentric passages). For interact-
gle can also drastically alter the prominence of tidal features. ing pairs which have had only one close passage, we require
For example, Pop et al. (2018) studied shell stellar debris the pericentric distance be less than ∼150 kpc. In considering
fields (e.g., their Figure 1, and the last panel of our Figure the full trajectories, we more reliably remove those galaxies
1) which they found to be present predominantly in merger which have merely flown past one another. Whilst it could
remnants. The authors show that whether or not a shell is be argued that such systems have interacted, they are not
visible in a particular projection depends on the orbital tra- currently interacting, and are thus not part of our sample.
jectory of the progenitor system.
Though in this work we do not directly account for the
2.2.5 Sample Selection Summary
viewing angle, we follow Rodriguez-Gomez et al. (2019) and
adopt a fixed projection. This is directly analogous to ob- In Table 1, we provide a full account of the results of our
servational studies; one of the main goals of this paper is various classification schema. Some pairs were manually re-
to test how well observationally-derived and morphology- moved from the sample. This includes systems with multiple
based classifications identify interacting systems. Projection prominent subhaloes, which comprised only ∼5 per cent of
will be important for the observability of individual systems’ the parent pairs sample. Pairs were also discounted if they
tidal features. However in this work, we look at the prop- have only been in the same group for less than 1 Gyr (∼22
erties of the interacting pairs as a whole. In doing so, we per cent). These are exclusively subhaloes determined not to
sample over a random set of viewing angles, thereby min- be interacting, based on the aforementioned criteria. There
imizing the biases of individual interacting pairs. Clearly, are a small number of subhaloes (∼2 per cent) which appear
schema which depend entirely on tidal features in the stel- to be interacting based on their morphologies in the mock
lar material might be biased in a variety of ways, thereby images and/or their trajectories, but were not orbiting one
affecting the kinds of interactions captured. another. Namely, their orbits appear to be dominated by
Several studies have shown that the internal structure structures outside the FoF group. The majority (∼48 per
of tidal features is highly dependent upon spatial resolution cent) of our parent sample are not interacting and are not
(e.g., Wetzstein et al. 2007). However, the broader struc- visually identified as mergers. However, there is a small frac-
ture of tidal features is less sensitive to this parameter. For tion (∼4 per cent) of the non-interacting sample which were
example, Moreno et al. (2015) and Moreno et al. (2019) per- misidentified as mergers. Pairs which were visually identified
form similar simulations of galaxy interactions: improving as mergers (§2.2.3) and were found to be interacting (§2.2.4)
the spatial resolution by two orders of magnitude does not are hereafter referred to as Visually Identified Pairs, or VIP.

MNRAS 000, 1–17 (2019)

6 K. A. Blumenthal et al.

Figure 2. Galaxy-pair property distributions for the Visually Identified Pairs (VIP; salmon) and the non-Visually Identified Pairs
(nonVIP; purple). The panels show: (a) the current (z = 0) stellar mass ratio, (b) z = 0 3D separations, (c) relative velocity, and (d) FoF
group dark matter mass. The vertical coloured lines correspond to the medians of each sample, and colored rectangles indicate the range
within ± one median absolute deviation from the median.

Figure 3. Individual galaxy distributions for the Visually Identified Pairs (VIP; salmon) and the non-visually identified pairs (nonVIP;
purple). The panels show: (a) stellar mass, (b) star formation rate, (c) total gas mass, and (d) the star forming gas. The vertical coloured
lines correspond to the medians of each sample, and colored rectangles indicate the range within ± one median absolute deviation from
the median. MNRAS 000, 1–17 (2019)
 Galaxy Mergers in TNG100-1 7
The nonVIP, then, are those pairs which are interacting,
but were not selected visually. The majority of interacting
pairs are nonVIP. This is because many of them lack the
prominent tidal features or disrupted morphologies that are
typically used to identify mergers (§2.2.3; see also Appendix
B for visual examples).

3 RESULTS AND DISCUSSION

3.1 The VIP and nonVIP Samples
With the sample of pairs defined, and delineated into VIP
and nonVIP, we investigate the bulk and individual proper- Figure 4. Here we show a (nonVIP) pair at the present day (left)
and at its last pericenter (right), which occurred about 1.3 Gyr
ties of the interacting pairs at the present day, and at the
prior. The stellar mass ratio has stayed at ∼3:1 over this period.
time of their last pericentres. Among other properties dis- Bars in the bottom left corner of each image indicate 50 kpc. The
cussed here, we will show that the VIP are not only more inset in the left panel shows an enlarged image of the primary
massive, but they have undergone a close passage more re- galaxy to highlight its visible structures.
cently than the nonVIP.

2.0
3.1.1 Present-day (z = 0) Properties VIP
nonVIP
Panel (a) of Figure 2 shows that the distribution in stellar
mass ratios, which peak at unity and peter out toward larger
1.5
mass ratios for both samples, with the VIP having slightly
larger median stellar mass ratio. Though we do see this trend
PDF

with the median values, the VIP and nonVIP stellar mass 1.0
ratios are not distinct distributions: a two-sided KS test in-
dicates (p ≈ 0.3) that these are drawn from the same sam-
ple. Panel (b) shows the present-day (z = 0) 3D separation, 0.5
with the VIP having separations shifted to smaller values
(here, the two-sided KS test indicates that VIP and nonVIP
relative separations are drawn from distinct distributions: 0.0
p ≈ 2 × 10−3 ). Interacting pairs at wider projected separa- 1.0 0.5 0.0 0.5
tions may be overlooked in preparing samples of merging log(t0 - tLP) [Gyr]
galaxies. There has, however, been some work which indi-
cates that interacting galaxies may exhibit heightened rates
of star formation, even with separations as large as 150 kpc Figure 5. Time since last pericenter (t0 - t L P ) distribution and
(e.g. Patton et al. 2013). In panel (c), we present the relative median values (vertical lines) for the VIP (salmon) and the non-
velocity distributions (i.e. the difference between the subhalo VIP (purple). The VIP have more recently undergone a close pas-
sage. Vertical coloured lines show the position of the median, and
velocities); the VIP and nonVIP samples attain low veloci-
the coloured rectangles indicate the range within ± one median
ties, facilitating their interaction and eventual merging. The absolute deviation from the median.
VIP do appear to be moving faster on average than the non-
VIP (p ≈ 0.05), which hints at their local dynamics. That
is, the relative velocities of these galaxies may be affected The present-day contribution due to cold gas appears to be
by their environment (§3.4), despite our efforts to avoid this nearly the same for both samples (p ≈ 0.23). This suggests
using our FoF group mass cuts. Finally, panel (d) displays that the gas reservoir available for star formation is not sig-
the VIP and nonVIP FoF group dark matter mass distri- nificantly different for the VIP or nonVIP, consistent with
butions; the VIP inhabit slightly more massive FoF haloes the findings of panel (b).
(p ≈ 4 × 10−4 ).
In Figure 3, panel (a) shows that VIP galaxies tend to
3.1.2 A Comparative Epoch: Last pericentric (LP) passage
have higher stellar masses than the nonVIP (p ≈ 1.3 × 10−6 ).
Because visual classification is based on tidal disruptions in We additionally utilise merger trees to study the interacting
the stellar material, we might predict that VIP galaxies to pairs at the time of their last pericentric passage (LP) – a
have a higher stellar mass (i.e., more stars to disrupt) on av- local maximum in the strength of their interaction. We note
erage. In fact, we do find that the VIP median stellar mass that the interacting pairs do not all reach their respective
is about one-half dex greater than the nonVIP, consistent LP events at the same time, but are at a dynamically sim-
with the findings in panel (d) of Figure 2. Panel (b) shows ilar moment in their histories. In this way, we analyse all
that the present-day star formation rate (SFR) is relatively interacting pairs at a point in time when the effects of their
consistent for both samples (this is confirmed by a two-sided interaction are at a near a peak. Figure 4 shows a nonVIP
KS test with p ≈ 0.33). Panels (c) and (d) show the total galaxy image at the present-day (left), and at its last peri-
gas mass and the cool (star forming) gas mass, respectively. centre (right). At this pair’s LP, there is a clear tidal debris

MNRAS 000, 1–17 (2019)

8 K. A. Blumenthal et al.
q
field with several star forming regions in the primary galaxy. virial velocity, and the orbits are parabolic: V = 2GMdyn /R
The LP’s span a range of ages relative to the present day (e.g., Moreno et al. 2013), where the dynamical mass Mdyn
of 70 Myr to 5.66 Gyr (Figure 5). The VIP have more re- is defined as the sum of the galaxies’ dark matter masses.
cently undergone a close passage than the nonVIP by nearly The ideal trend (i.e., parabolic) is shown for both the VIP
a factor of two (p ≈ 0.01). (dashed salmon) and nonVIP (dashed purple) samples, us-
ing their corresponding median separations at the present
day (RVIP = 88.8 kpc; RnonVIP = 141.4 kpc). The solid lines
3.1.3 The Failures of Morphological Identification
indicate lines of best fit for each subsample (mnonVIP = 0.36
In ∼55 per cent of the interacting pairs, it was unclear if an and mVIP = 0.17, with standard errors σnonVIP = 0.067 and
interaction was underway (that is, the nonVIP). There are a σVIP = 0.068). Outliers have a variable impact on the fitted
number of reasons why morphological identification schema slopes for both the VIP and the nonVIP, ranging from 0.08
may fail: to 0.47 dex. Note that their slopes differ from the parabolic
case. There is a substantial amount of scatter in these sam-
(i) Due to the finite resolution of the simulations, star ples, particularly at the high mass end, where a subset of
particles have relatively large masses. This may inhibit our the VIP dip to lower relative velocities. That these galaxies
ability to resolve the fine-grained structures within tidal in- have lower relative velocities than what might be expected
teractions (e.g., Wetzstein et al. 2007). based on their dynamical mass may be indicative of their
(ii) The stellar material may not be the best indicator visual identification.
of a tidal interaction (discussed as the “internal properties” Figure 7 shows the VIP and nonVIP at last pericentre
in e.g., Darg et al. 2010). The gas disc has been shown to (coloured points), compared with their positions at z = 0
be as large, if not larger than the stellar disc (e.g., Broeils (gray points). The interacting pairs at LP are significantly
& Rhee 1997). Thus, gas discs are much more likely to be closer together, and are moving much faster (left panel) than
perturbed by one another, even in the case of wide pericen- they are at z = 0. This is expected: an interacting pair should
tric distances. Integral Field Unit surveys (e.g., Croom et al. reach a local maximum in its relative velocity at each peri-
2012; Sánchez et al. 2012; Bundy et al. 2015) of interacting centric passage (or conversely, should reach a local minimum
galaxies may be necessary to get a realistic measurement of in its velocity at each apocentre, e.g., Figure 4 of Moreno
the local merger rate. et al. 2019). Similarly, the right panel of Figure 7 shows
(iii) The present-day separations (Figure 2) are larger that the the Mdyn − Vrel relationship is much tighter at LP
than expected from observationally motivated merger cat- than at z = 0 (mnonVIP = 0.21 and mVIP = 0.33, with stan-
alogues. What observers assume to be the first passage may, dard errors σnonVIP = 0.05 and σVIP = 0.04). Contrary to the
in many cases, be the second (e.g., Patton et al. 2013). z = 0 behaviour, the nonVIP appear to have greater scat-
(iv) If encounters are sufficiently wide, tidal forces may ter, particularly at the low mass end, where a subset achieve
not be strong enough to produce visible (i.e. observable) higher velocities than their dynamical mass might suggest.
bridges and tails. The effects of these outliers on the slope is similar to that
(v) If an encounter has occurred within the last Gyr, it of the present day population, with a range of 0.02 to 0.33
is more likely to host obvious tidal features. As time passes, dex. Moreover, the best fit lines to the data (solid) and the
material from the bridge and tails settles back into the discs, parabolic fiducial curves (dashed) do not agree at last peri-
and is able to phase-mix with the surrounding material (e.g., center. This is expected at this epoch, as the orbital elements
Lotz et al. 2008, 2010a). of the interaction will change rapidly near a close passage.
The parabolic trends are elevated from the z = 0 case, as
the median separation values used are RVIP = 57.63 kpc and
3.2 Galaxy pair dynamics RnonVIP = 76.73 kpc. These curves would be translated down
with larger separations. Although the best-fitting lines are
To better understand the distinctions between the VIP and still notably different, the VIP slope is now more consistent
nonVIP, we discussed the fundamental physical properties of with the parabolic slopes.
the pairs in the previous section. Here, we will show that the Deviations from the fiducial parabolic slope hint at
VIP are closer together and move faster than the nonVIP at the shortcomings of our assumptions regarding galaxy
both the present day and LP. interactions. In particular, this implies that the constant
Figure 6 focuses on the dynamical properties of the weak gravitational encounters that galaxies experience
VIP (salmon diamonds) and nonVIP (purple circles) at the throughout their evolution impact the orbits in measurable
present day. Consistent with panels (b) and (c) of Figure 2, ways. This is made most evident at infall, an epoch dis-
the left panel of Figure 6 shows the VIP and nonVIP are cussed in our forthcoming paper (Blumenthal et al. in prep.)
offset from one another: the VIP have smaller 3-dimensional
separations, and move with slightly faster relative velocities
than the nonVIP. Figure 5 shows that the VIP have more
recently undergone a pericentric passage. Thus, their sepa-
3.3 Star formation main sequence
rations will naturally be smaller at the present day, and as
they are closer to a pericentre, the VIP should have higher In Section 3.1, we presented physical properties such as the
velocities than the nonVIP, which are typically closer to an stellar mass and star formation rate for the VIP and non-
apocentre. The right panel of Figure 6 shows a moderate VIP. It is known that these two parameters often trace one
linear trend between the dynamical mass and relative ve- another, forming the star formation main sequence.
locity. This is expected if the relative velocity traces the The star formation main sequence (e.g., Noeske et al.

MNRAS 000, 1–17 (2019)

 Galaxy Mergers in TNG100-1 9

3
10
−1 ]
Vrel [km s

2
10

nonVIP
VIP

0 1 2
10 10 10 11.0 11.5 12.0 12.5 13.0

Rsep [kpc] log(Mdyn ) [M⊙ ]

Figure 6. The dynamics of the interacting pairs at z = 0: relative velocity as a function of pair separation (left) and of dynamical mass
(bottom) for the VIP (salmon diamonds) and the nonVIP (purple circles) sample. The dashed lines in the right panel indicate the trend
expected from a parabolic trajectory, whilst the solid lines are a fit to the data.

3
10
−1 ]
Vrel [km s

2
10

(non)VIP, z=0
nonVIP, LP
VIP, LP

0 1 2
10 10 10 11.0 11.5 12.0 12.5 13.0

Rsep [kpc] log(Mdyn ) [M⊙ ]

Figure 7. Analogous to Figure 6, but at last pericenter (LP). Values at LP are reported in colour, and in gray points for comparison
with z = 0 values form Figure 6.

2007; Daddi et al. 2007; Elbaz et al. 2007; Salim et al. 2007; ley” of transitioning galaxies between the two. Merging and
Rodighiero et al. 2011; Bluck et al. 2016, 2019; Donnari et al. interacting galaxies, which themselves are examples of star-
2019, hereafter SFMS) defines a general trend of all star bursting systems, have been shown to lie above the SFMS
forming galaxies: the star formation rate is tightly correlated (e.g., Puech et al. 2014; Willett et al. 2015). In particular,
with the stellar mass. That this relationship holds for a wide Hung et al. (2013) show that for z ∼ 0.4 galaxies, distance
range of redshifts (e.g., Noeske et al. 2007; Lee et al. 2015), above the SFMS is correlated with disturbed morphologies.
several orders of magnitude in stellar mass, and a relatively However, other studies (Willett et al. 2015; Brennan et al.
small spread in star formation rate implies that star forming 2017) are unable to to confirm this morphological depen-
galaxies behave in a self-regulatory manner with a fairly con- dence.
sistent star formation history throughout cosmic time (e.g., Figure 8 shows the star formation main sequence for
Bouché et al. 2010; Lilly et al. 2013). Outliers above the all galaxies (that is, each point is an individual galaxy) at
SFMS (starbursts) are thought to represent an important z = 0 which meet the same mass criteria as the interacting
stage (that is, mergers) in galaxy evolution, though their pairs (grayscale hexagons; the black dashed line shows our
relative contribution to the star formation density is still fiducial SFMS fit), the VIP (diamonds, outlined in black),
debated (e.g., Cox et al. 2008; Rodighiero et al. 2011; Hung and the nonVIP (circles). The interacting pairs are coloured
et al. 2013; Brennan et al. 2015; Willett et al. 2015; Bren- by the change in the log of their star formation rates from
nan et al. 2017; Ellison et al. 2018). Quiescent galaxies lie in the present day to LP. These colours enable mapping from
a so-called “red cloud” below the SFMS, with a “green val- z = 0 to LP, and show how the galaxies have evolved since

MNRAS 000, 1–17 (2019)

 10 K. A. Blumenthal et al.
1.5 1.5
1.0 1.0
0.5 0.5
0.0 0 1 0.0 0 1
2 2

1 1
log(SFR) [M yr 1]

log(SFR) [M yr 1]
0 0

1 1

2 2

3 3

2 2

1 1
log(SFR) [M yr 1]

log(SFR) [M yr 1]

0 0

1
1

2
2
3
3
4
4
9.0 9.5 10.0 10.5 11.0 11.5 9.0 9.5 10.0 10.5 11.0 11.5
log(M ) [M ] log(M ) [M ]

1.6 0.8 0.0 1.6 0.8 0.0

log(SFRz = 0) - log(SFRLP) log(SFRz = 0) - log(SFRLP)

Figure 8. Top: The SFMS at the present day. The positions of Figure 9. The axes and background galaxies are the same as
all TNG100-1 galaxies at z = 0 which meet the same mass crite- Figure 8, but shown here are the VIP and nonVIP at their last
ria as the interacting pairs are shown in the grayscale hexagons, close passage. Compared to the present day values, the interact-
with the fiducial fit to that SFMS indicated by the black dashed ing pairs sit slightly higher on the MS, and fewer of them lie in
line. The VIP (diamonds, outlined in black) and nonVIP (cir- the green valley.
cles) are coloured by the log in the change of their SFR. Bottom:
Distance from the SFMS fiducial line, ∆log(SFR) for the same
samples as above. Despite the fact that the (non)VIP are inter-
acting pairs, there is no apparent offset above the star formation
main sequence, though they are offset from the median value
of the total TNG100-1 sample (black solid line). The VIP have
more scatter in ∆log(SFR), perhaps indicating that their inter-
action has triggered a dramatic change in morphology, toward
compact quiescent spheroids.

MNRAS 000, 1–17 (2019)

 Galaxy Mergers in TNG100-1 11
their last close passage. Blue colours indicate that a galaxy – partially as measured by the total FoF group mass – is
has increased its rate of forming stars since LP, whilst red marginally more dense than that of the nonVIP.
colours indicate a decrease in SFR since LP. The bottom
panel of Figure 8 shows the VIP and nonVIP distances,
defined as ∆log(SFR) = log(SFR) - log(SFR|MS ), from the 3.4 Environment
fiducial SFMS line as a function of stellar mass. First, we In the previous sections, we have detailed the intra- and
note that the VIP and nonVIP are consistent with the en- inter-galaxy properties of the interacting pairs. In this sec-
tire set of local TNG100-1 galaxies. The nonVIP appear to tion, we describe the external forces acting on these systems
exhibit a tight scatter around the SFMS fiducial line, whilst through three environmental metrics: the nearest neighbour,
the VIP display a larger spread. The stellar mass appears the interaction strength, and the FoF group mass.
to increase with increasing FoF group mass, as is expected There is no universal definition of galactic environment
from abundance matching (e.g., Colı́n et al. 1999; Kravtsov (e.g., Muldrew et al. 2012, and references therein). Many
& Klypin 1999; Kravtsov et al. 2004; Vale & Ostriker 2004; studies attempt to compare the various definitions of this
Conroy & Wechsler 2009; Behroozi et al. 2010; Guo et al. fundamental property (e.g., Cooper et al. 2005; Gallazzi
2010; Moster et al. 2010). That there are more VIP at higher et al. 2009; Wilman et al. 2010; Haas et al. 2012; Shattow
stellar and halo masses (see also Figures 2 and 3) may indi- et al. 2013; Fossati et al. 2017). Parameterisations used to
cate the VIP experience a dramatic change in morphology, characterise the environment include the local number den-
perhaps toward compact quiescent spheroids (e.g., Ellison sity (e.g., Dressler 1980; Lewis et al. 2002; Cooper et al.
et al. 2018). 2005; Shattow et al. 2013), measurements of galaxy clus-
Regardless of the orbital geometry, the tidal interac- tering (e.g., Skibba et al. 2013, 2015; Gunawardhana et al.
tion at a pericentre will invariably draw material from the 2018), and placement within cosmic structures (e.g., Yang
outskirts of each galaxy toward the center (e.g., Barnes & et al. 2007; Darvish et al. 2014; Kuutma et al. 2017; Liao &
Hernquist 1996; Mihos & Hernquist 1996; Rupke et al. 2010; Gao 2019).
Moreno et al. 2015; Blumenthal & Barnes 2018). In the case The top panels of Figure 10 show the placement of the
of prograde interactions, a significant amount of gas can VIP (salmon diamonds) and nonVIP (purple circles) sam-
be funneled toward the galaxy’s nucleus, sparking a burst ples in the TNG100-1 cosmic web, as traced by all subhaloes
of star formation (e.g., Alonso-Herrero et al. 2000; Barnes in the z = 0 slice. Qualitatively, the VIP typically lie in
2004; Evans et al. 2008; Chien & Barnes 2010; Moreno et al. denser regions than the nonVIP (which seem to mostly oc-
2015; Larson et al. 2016). Thus, “observing” the VIP and cupy voids). The marker sizes in each of the three panels
nonVIP at the time of their most recent pericentres would scale linearly with the log of three independent measures of
naturally push the points in Figure 8 up to higher star forma- environment: the total FoF group mass (left, §3.4.1), the nth
tion rates. Figure 9 shows the star formation main sequence nearest neighbour, Σn (e.g., Dressler 1980; Lewis et al. 2002,
(top) for the interacting pairs at their last pericentres, in ad- middle, §3.4.2), and the interaction strength, Qint (e.g., Ver-
dition to the distance from the main sequence fiducial line ley et al. 2007, right, §3.4.3). The structure of the cosmic
(∆log(SFR); bottom). Note that the galaxies which have met web is defined by the location of haloes containing individ-
the stellar mass threshold at z = 0 might not achieve this ual, groups and clusters of galaxies. Thus, the mass of a
limit at LP. There are only a few galaxies whose masses halo is indicative of its placement within this structure. The
become unreliable; they are removed from this LP analysis. nearest neighbour statistic measures environment based on
The colours in both of these panels are the same as in Figure the number density of nearby galaxies, regardless of mass.
8. Using these colours, we note that there are some galax- On the other hand, the interaction strength measures the
ies which appear to move out of the bottom right part of balance of external tidal forces from all galaxies within an
the SFMS (the so-called “red and dead” galaxies) between aperture with the binding force of the galaxy. Whilst it can
LP and the present day. This may imply that membership be difficult to discern any trends from the cosmic web pan-
in the various regions of the SFMS is fluid: galaxies might els, the bottom panels of Figure 10 show the subsequent
undergo periods of starbursts and relative quiescence (e.g., distributions for each of these environmental metrics.
Forbes et al. 2014a,b).
The interacting pairs’ shift above the main sequence 3.4.1 FoF Group Mass
from the present day to LP cannot be explained by the ver-
tical translation of the SFMS with increasing redshift (e.g., The bottom right panel of Figure 10 shows the distribution
Noeske et al. 2007; Lee et al. 2015) alone. If that were true, of FoF group total masses for the VIP (salmon) and nonVIP
the VIP should have systematically lower ∆log(SFR) than (purple). This indicates that the VIP sit in preferentially
the nonVIP in Figure 9. That this is not the case implies more massive haloes (p ≈ 4 × 10−4 ), surpassing the nonVIP
the difference in the merger-driven starbursts is mediated by nearly half an order of magnitude. This is consistent with
by the strength of the interaction, which we have shown is the fact that most massive haloes are likely to sit in nodes or
significantly boosted in the VIP. This is supported by the at intersections of filamentary structures (e.g., Bond et al.
observation that the VIP have higher stellar masses than 1996; Joachimi et al. 2015, and sources therein).
the nonVIP (Figure 3), and inhabit FoF groups with more
massive dark matter haloes (Figure 2). Studies have shown
3.4.2 nth Nearest Neighbour
(e.g., Sobral et al. 2011) a connection between the stellar
mass, star formation rate, and density of environment. In The nth nearest neighbour statistic is a number density mea-
later sections (§3.4), we will show that the VIP environment surement that uses the distance to the nth nearest neighbor,

MNRAS 000, 1–17 (2019)

12 K. A. Blumenthal et al.

2.0
VIP 0.5
nonVIP
1.5 1.0 0.4

0.3
PDF
PDF

PDF
1.0
0.5 0.2
0.5
0.1

0.0 0.0 0.0

11.25 11.50 11.75 12.00 12.25 12.50 12.75 13.00 2.0 2.5 3.0 3.5 4.0 4.5 4 3 2 1 0 1
log(MFoF group) [M ] log( n) [Mpc 3] log(Qint)
Figure 10. Top: Placement of the nonVIP (purple circles) and VIP (salmon diamonds) in the TNG100-1 cosmic web. The point sizes
are scaled by the total group mass (MFoF group , left), nearest neighbour statistic (Σn , centre), and the interaction strength (Qint , right).
Bottom: We also show the distributions of the three environmental measures for the VIP (salmon) and nonVIP (purple), as before.
Vertical coloured lines show the median values, and the corresponding coloured rectangles indicate the range within ± one median
absolute deviation of the median. The VIP sit in significantly more massive haloes, in denser environments and are more affected by
their surroundings than the nonVIP.

rn , to define the volume. That is, we take all masses to be the total mass within twice the stel-
n−1 lar half-mass radius, and the diameter of the central galaxy
Σn = 4 (4) which corresponds to that mass (that is, four times the stel-
3 πrn
3
lar half-mass radius). This value is calculated in a number
where the numerator n−1 is used to discount the central of different ways in Verley et al. (2007), including using a
or, primary galaxy. (Note that here we employ a three- fixed and infinite aperture (that is, all galaxies within a fixed
dimensional version of what is typically used by observers.) volume). They find that there was very little difference be-
Thus, centrals with larger Σn sit in denser environments. For tween the two, as distant galaxies will contribute only a small
the purposes of this paper, we adopt n= 5. The bottom left amount to the tidal field of the central. To accommodate
panel of Figure 10 shows that the VIP lie in preferentially the large present-day separations of some of our interacting
denser environments than the nonVIP (p ≈ 0.07). pairs, we use an aperture of 5 Mpc (Figure 10, bottom right).
The VIP are affected by the tidal effects of their neighbours
nearly ten times as much as the nonVIP (p ≈ 7 × 10−4 ).
3.4.3 Interaction Strength
One major drawback of the Σn measure is that it does not
3.4.4 The Effects of Environment
account for the mass of neighbouring galaxies. The interac-
tion strength, Qint , thus serves a useful counterpoint to Σn in In the previous subsections, we demonstrated that the VIP
its careful accounting of the tidal effect of nearby galaxies. belong to more massive FoF haloes, sit in denser environ-
Verley et al. (2007) defined the interaction strength as the ments, and are more affected by interactions with their
ratio of the cumulative tidal forces tugging on the galaxy neighbours than the nonVIP. Here we disentangle the ef-
from all neighbouring galaxies within a set aperture, and fects of mass and environment and show that although the
the binding force keeping the central together: (more massive) VIP are in systematically more dense envi-
F Mn Dc Mc ronments, there is no statistically significant difference be-
Qint ≡ tidal , Ftidal = 3
, Fbind = 2 (5) tween the ∆log(SFR) of the nonVIP and VIP, when control-
Fbind Rnc Dc
ling for stellar mass.
where Mn is the mass of the neighbor, Rnc is the distance Figure 11 shows the ∆log(SFR) as a function of stel-
from the central galaxy to that neighbor, Mc is the mass of lar mass. The stellar mass distribution is split into three
the central, and Dc is the diameter of the central. Following bins: 9.0 ≤ log(M?) < 9.75, 9.75 ≤ log(M?) < 10.5, and
observational studies (for which this metric was developed), log(M?) ≥ 10.5. These mass increments were chosen to sep-

MNRAS 000, 1–17 (2019)

 Galaxy Mergers in TNG100-1 13
arately analyse “normal” star forming galaxies (log(M?) < 2
log(MFoF grp) 11.71 log(MFoF grp) 12.12 log(MFoF grp) 12.69
10.5) from those which have begun to dip below the main 1 log(MFoF grp) < 11.71 log(MFoF grp) < 12.12 log(MFoF grp) < 12.69
sequence (log(M?) ≥ 10.5). Environment is considered inde-
pendently within each of these bins. Different colours indi- 0
cate the galaxies which sit in a relatively low (light colours)
or high (dark colours) density environment. The median en- 1
vironmental measure in each mass bin is used to delineate 2
between low and high densities. We present this in Figure 11
for the MFoF group (top), Σn (middle) and Qint (bottom). As 3
before, the VIP and nonVIP are distinguished by diamond
and circle markers, respectively. Centrals and satellites are 4
2
indicated by the marker size (large and small, respectively). log( n) 2.78 log( n) 3.0 log( n) 3.21
The median ∆log(SFR) values for both environment bins 1 log( n) < 2.78 log( n) < 3.0 log( n) < 3.21
within each mass bin are displayed as stars with error bars
0

log(SFR)
indicating the first and third quartiles (that is, the width
of the distribution). For comparison, the median value of
1
∆log(SFR) of the underlying TNG100-1 distribution is dis-
played by the X’s. 2
3
The first two mass bins of all three environmental met-
rics shown in Figure 11 show no clear trend with environ- 4
2
ment. That is, not only do they show no distinction between log(Qint) -2.63 log(Qint) -2.56 log(Qint) -1.88
high and low density environments, but they are consistent 1 log(Qint) < -2.63 log(Qint) < -2.56 log(Qint) < -1.88
with the background distribution of all TNG100-1 galaxies
(coloured X’s). Only in the largest mass bin do we see any 0
significant difference between the low and high density en-
1 All TNG
vironments across all samples. In the largest mass bin of
Interacting Pairs
the top panel (MFoF group ), the less massive FoF groups have 2 Centrals
systematically higher ∆log(SFR) than the high mass FoF Satellites
groups, as these are likely quenched or are in the process of 3 VIP
nonVIP
quenching. It should be noted however that in this panel,
4
the interacting pair sample (and its individual components) 9.0 - 9.75 9.75 - 10.5 10.5 - 11.25
are consistent with the background.
log(M ) [M ]
The middle and bottom panels of Figure 11 indicate Figure 11. We separate the stellar mass into three bins, and fur-
that denser environments foster higher star formation rates ther split the subsample into two environmental bins based on the
only within the highest mass bin. Whereas before, the inter- median environmental measure in that mass bin. Stars indicate
the median value of ∆log(SFR) for all interacting pairs within
acting pairs behaved similarly to the background TNG100-1
each mass bin with error bars that correspond to the median ab-
galaxies, in the Σn and Qint panels, the interacting pairs di-
solute deviation. Markers represent the the median ∆log(SFR)
verge significantly from the background TNG100-1 galaxies. for the galaxies within that mass bin. This is shown for the FoF
Further, that the environmental dependence of ∆log(SFR) group mass (top), Σn (middle), and Qint (bottom). X’s represent
only becomes appreciable at higher masses – when AGN the background distribution of all TNG100-1 galaxies. Stars indi-
activity and quenching begin to dominate a galaxy’s evolu- cate the sample of interacting pairs (that is, the VIP and nonVIP
tion – implies environment plays a larger role in suppressing together). As before, the nonVIP and VIP are indicated by cir-
quenching than it does in boosting star formation. cles and outlined diamonds, respectively. Satellites and centrals
are also shown, and are distinguished by symbol size: small and
large, respectively.
Though there is no clear distinction between the VIP
and nonVIP at any mass bin (except for the highest mass bin
of Qint ), the satellites and centrals appear to have divergent
4 CONCLUSIONS
evolutionary pathways. The centrals dip low in ∆log(SFR)
at high masses whereas the satellites are only moderately In this paper, we identify a set of paired galaxies from the
affected. This implies that centrals are likely to quench be- z = 0 snapshot of the TNG100-1 simulation of IllustrisTNG
fore their satellites. It may be that the evolution of satel- (Marinacci et al. 2018; Naiman et al. 2018; Nelson et al.
lites is more sensitive to environment, whilst the evolution 2018; Pillepich et al. 2018b; Springel et al. 2018). We
of centrals is depends more strongly upon mass; perhaps an generate ideal mock SDSS g-band images of all pairs and
example of the interplay between “environment quenching” visually classify each as interacting or not interacting. We
and “mass quenching” (e.g., Peng et al. 2010; Bluck et al. then confirm using the information from the Sublink merger
2016, 2019). Thus, the relative importance of environment tree, and find that of the interacting pairs, we correctly
and stellar mass depends upon which component of the in- identify 38 (the Visually Identified Pairs, or VIP) and miss
teraction is the subject of inquiry. 47 (the Non-Visually Identified Pairs, or nonVIP). Our

MNRAS 000, 1–17 (2019)

14 K. A. Blumenthal et al.
analysis includes a detailed study of the interacting pairs’ the NSF (AST Award Number 1516374), by the Australian
present day properties, as well as their properties at their Research Council (ARC), and by the Harvard Institute for
respective last pericentres. Theory and Computation, through their Visiting Scholars
Program. FM acknowledges support through the Program
Our primary findings are as follows: “Rita Levi Montalcini” of the Italian MIUR. MV acknowl-
edges support through an MIT RSC award, a Kavli Research
(i) Stellar morphologies are not ideal for identifying in-
Investment Fund, NASA ATP grant NNX17AG29G, and
teractions as the visibility of stellar tidal features, which is
NSF grants AST-1814053 and AST-1814259. The authors
in part dependent upon the environment and the time since
would like to thank the anonymous reviewer, and the Scien-
the last close passage.
tific Editor Prof. Joop Schaye, both of whom provided in-
(ii) Using the merger trees, we trace the interacting pairs
valuable feedback on the manuscript. In addition, we thank
back to their time of last pericentric (LP) passage.
Dylan Nelson and Annalisa Pillepich for their thoughtful
(iii) The VIP have more recently undergone a close pas-
comments. The simulation used in this work, TNG100-1, is
sage than the nonVIP by about a factor of two. As a result,
one of the flagship runs of the IllustrisTNG project, and
their tidal features are easier to observe. Merger classifica-
was run on the HazelHen Cray XC40-system at the High
tions are thus biased toward recent interactions.
Performance Computing Center Stuttgart as part of project
(iv) Compared with the nonVIP, the VIP sit in very
GCS-ILLU of the Gauss Centre for Supercomputing.
different environments. The VIP are: in groups which are
nearly 2.5 times as massive; in nearly twice as dense sur-
roundings; and are affected by interactions with their neigh-
bours by nearly an order of magnitude more than the non-
VIP. Classification schema based on stellar morphologies are REFERENCES
biased toward dense environments. Alonso-Herrero A., Rieke G. H., Rieke M. J., Scoville N. Z., 2000,
(v) Though the VIP sit in distinct environments from the ApJ, 532, 845
nonVIP, the visibility of a pair does not depend strongly on Arp H., 1966, The Astrophysical Journal Supplement Series, 14,
environment, when correcting for stellar mass. 1
Barnes J. E., 2004, MNRAS, 350, 798
Care should be taken when interpreting these results Barnes J. E., 2016, MNRAS, 455, 1957
within the context of large observational catalogues of Barnes J. E., Hernquist L. E., 1991, ApJ, 370, L65
galaxy pairs (e.g., Ellison et al. 2008, 2010; Patton et al. Barnes J. E., Hernquist L., 1996, ApJ, 471, 115
2016), as the mass range covered in this work is relatively Barro G., et al., 2013, ApJ, 765, 104
limited. The roughly 45% merger recovery rate that we Behroozi P. S., Conroy C., Wechsler R. H., 2010, ApJ, 717, 379
present here should not be taken as a completeness correc- Bezanson R., van Dokkum P. G., Tal T., Marchesini D., Kriek
tion. This work can only offer a critique of the observational M., Franx M., Coppi P., 2009, ApJ, 697, 1290
Bluck A. F. L., Conselice C. J., Buitrago F., Grützbauch R.,
surveys used to derive merger rates (e.g. Kartaltepe et al.
Hoyos C., Mortlock A., Bauer A. E., 2012, ApJ, 747, 34
2007; Kitzbichler & White 2008; Robotham et al. 2014). To
Bluck A. F. L., et al., 2016, MNRAS, 462, 2559
fully answer this question would require a realistic mock Bluck A. F. L., et al., 2019, MNRAS, p. 363
(that is, not ideal mock as described above) survey of all Blumenthal K. A., Barnes J. E., 2018, MNRAS, 479, 3952
galaxy pairs, though this is beyond the scope of this work. Bond J. R., Kofman L., Pogosyan D., 1996, Nature, 380, 603
However, we have illuminated distinct biases inherent in ob- Bottrell C., et al., 2019, arXiv e-prints, p. arXiv:1910.07031
servational galaxy pair catalogs. If these are used to deter- Bouché N., et al., 2010, ApJ, 718, 1001
mine the merger rate, the result is likely to be biased toward Bournaud F., Elmegreen B. G., 2009, ApJ, 694, L158
close pairs (§3.2), high stellar masses (§3.1.1) that may result Bournaud F., et al., 2011, ApJ, 730, 4
in particularly prominent tidal features, and recent pericen- Brennan R., et al., 2015, MNRAS, 451, 2933
tric passages (§3.1.2). Brennan R., et al., 2017, MNRAS, 465, 619
Bridge C. R., et al., 2007, ApJ, 659, 931
Bridge C. R., Carlberg R. G., Sullivan M., 2010, ApJ, 709, 1067
In addition to the intrinsic and dynamic properties of
Broeils A. H., Rhee M.-H., 1997, A&A, 324, 877
galaxies, the production of tidal features depends on the in- Bruzual G., Charlot S., 2003, MNRAS, 344, 1000
teraction geometry. In a forthcoming paper (Blumenthal et Buitrago F., Trujillo I., Conselice C. J., Bouwens R. J., Dickinson
al. in prep), we investigate the orbital characteristics of the M., Yan H., 2008, ApJ, 687, L61
interacting pairs sample. This work will provide a realistic Bundy K., Fukugita M., Ellis R. S., Kodama T., Conselice C. J.,
set of parameters from which to produce the initial condi- 2004, ApJ, 601, L123
tions of future idealized simulations, including the eccen- Bundy K., Fukugita M., Ellis R. S., Targett T. A., Belli S., Ko-
tricities, inclinations, and first pericentric separations. Ad- dama T., 2009, ApJ, 697, 1369
ditionally, we will assess orbital stability and the validity of Bundy K., et al., 2015, ApJ, 798, 7
the Keplerian approximation. Chien L.-H., Barnes J. E., 2010, MNRAS, 407, 43
Colı́n P., Klypin A. A., Kravtsov A. V., Khokhlov A. M., 1999,
ApJ, 523, 32
Conroy C., Wechsler R. H., 2009, ApJ, 696, 620
ACKNOWLEDGEMENTS Conselice C. J., 2006, ApJ, 638, 686
Conselice C. J., et al., 2004, ApJ, 600, L139
KB acknowledges partial support from the National Sci- Cooper M. C., Newman J. A., Madgwick D. S., Gerke B. F., Yan
ence Foundation (NSF) Graduate Research Fellowship un- R., Davis M., 2005, ApJ, 634, 833
der Grant No. DGE-1329626. Support for JM is provided by Côté P., Marzke R. O., West M. J., 1998, ApJ, 501, 554

MNRAS 000, 1–17 (2019)

 Galaxy Mergers in TNG100-1 15
Cox T. J., Jonsson P., Somerville R. S., Primack J. R., Dekel A., Kennicutt Robert C. J., Keel W. C., van der Hulst J. M., Hummel
2008, MNRAS, 384, 386 E., Roettiger K. A., 1987, AJ, 93, 1011
Croom S. M., et al., 2012, MNRAS, 421, 872 Kitzbichler M. G., White S. D. M., 2008, MNRAS, 391, 1489
Cutri R. M., McAlary C. W., 1985, ApJ, 296, 90 Knierman K. A., Gallagher S. C., Charlton J. C., Hunsberger
Daddi E., et al., 2007, ApJ, 670, 156 S. D., Whitmore B., Kundu A., Hibbard J. E., Zaritsky D.,
Dahari O., 1985, The Astrophysical Journal Supplement Series, 2003, AJ, 126, 1227
57, 643 Kravtsov A. V., Klypin A. A., 1999, ApJ, 520, 437
Darg D. W., et al., 2010, MNRAS, 401, 1552 Kravtsov A. V., Berlind A. A., Wechsler R. H., Klypin A. A.,
Darvish B., Sobral D., Mobasher B., Scoville N. Z., Best P., Sales Gottlöber S., Allgood B., Primack J. R., 2004, ApJ, 609, 35
L. V., Smail I., 2014, ApJ, 796, 51 Kuutma T., Tamm A., Tempel E., 2017, A&A, 600, L6
Davis M., Efstathiou G., Frenk C. S., White S. D. M., 1985, ApJ, Larson K. L., et al., 2016, ApJ, 825, 128
292, 371 Lee N., et al., 2015, ApJ, 801, 80
Dekel A., Sari R., Ceverino D., 2009, ApJ, 703, 785 Lewis I., et al., 2002, MNRAS, 334, 673
Di Matteo P., Combes F., Melchior A.-L., Semelin B., 2007, A&A, Liao S., Gao L., 2019, MNRAS, 485, 464
468, 61 Lilly S. J., Carollo C. M., Pipino A., Renzini A., Peng Y., 2013,
Di Matteo P., Bournaud F., Martig M., Combes F., Melchior ApJ, 772, 119
A. L., Semelin B., 2008, A&A, 492, 31 Lintott C. J., et al., 2008, MNRAS, 389, 1179
Donnari M., et al., 2019, MNRAS, 485, 4817 Lotz J. M., Primack J., Madau P., 2004, AJ, 128, 163
Dressler A., 1980, ApJ, 236, 351 Lotz J. M., Madau P., Giavalisco M., Primack J., Ferguson H. C.,
Dubinski J., Mihos J. C., Hernquist L., 1996, ApJ, 462, 576 2006, ApJ, 636, 592
Duc P.-A., Bournaud F., Masset F., 2004, A&A, 427, 803 Lotz J. M., Jonsson P., Cox T. J., Primack J. R., 2008, MNRAS,
Elbaz D., et al., 2007, A&A, 468, 33 391, 1137
Ellison S. L., Patton D. R., Simard L., McConnachie A. W., 2008, Lotz J. M., Jonsson P., Cox T. J., Primack J. R., 2010a, MNRAS,
AJ, 135, 1877 404, 575
Ellison S. L., Patton D. R., Simard L., McConnachie A. W., Lotz J. M., Jonsson P., Cox T. J., Primack J. R., 2010b, MNRAS,
Baldry I. K., Mendel J. T., 2010, MNRAS, 407, 1514 404, 590
Ellison S. L., Patton D. R., Mendel J. T., Scudder J. M., 2011, Lotz J. M., Jonsson P., Cox T. J., Croton D., Primack J. R.,
MNRAS, 418, 2043 Somerville R. S., Stewart K., 2011, ApJ, 742, 103
Ellison S. L., Sánchez S. F., Ibarra-Medel H., Antonio B., Mendel Malin D., Hadley B., 1997, Publications of the Astronomical So-
J. T., Barrera-Ballesteros J., 2018, MNRAS, 474, 2039 ciety of Australia, 14, 52
Evans A. S., et al., 2008, ApJ, 675, L69 Mantha K. B., et al., 2018, MNRAS, 475, 1549
Faucher-Giguère C.-A., Lidz A., Zaldarriaga M., Hernquist L., Marinacci F., et al., 2018, MNRAS, 480, 5113
2009, ApJ, 703, 1416 Mihos J. C., 1995, ApJ, 438, L75
Forbes J. C., Krumholz M. R., Burkert A., Dekel A., 2014a, MN- Mihos J. C., Hernquist L., 1996, ApJ, 464, 641
RAS, 438, 1552 Mihos J. C., Walker I. R., Hernquist L., Mendes de Oliveira C.,
Forbes J. C., Krumholz M. R., Burkert A., Dekel A., 2014b, MN- Bolte M., 1995, ApJ, 447, L87
RAS, 443, 168 Mihos J. C., Dubinski J., Hernquist L., 1998, ApJ, 494, 183
Fossati M., et al., 2017, ApJ, 835, 153 Mirabel I. F., et al., 1998, A&A, 333, L1
Gallazzi A., et al., 2009, ApJ, 690, 1883 Moreno J., 2012, MNRAS, 419, 411
Genel S., et al., 2014, MNRAS, 445, 175 Moreno J., Bluck A. F. L., Ellison S. L., Patton D. R., Torrey P.,
Genzel R., et al., 2008, ApJ, 687, 59 Moster B. P., 2013, MNRAS, 436, 1765
Gunawardhana M. L. P., et al., 2018, MNRAS, 479, 1433 Moreno J., Torrey P., Ellison S. L., Patton D. R., Bluck A. F. L.,
Guo Q., White S., Li C., Boylan-Kolchin M., 2010, MNRAS, 404, Bansal G., Hernquist L., 2015, MNRAS, 448, 1107
1111 Moreno J., et al., 2019, MNRAS, 485, 1320
Haas M. R., Schaye J., Jeeson-Daniel A., 2012, MNRAS, 419, Moster B. P., Somerville R. S., Maulbetsch C., van den Bosch
2133 F. C., Macciò A. V., Naab T., Oser L., 2010, ApJ, 710, 903
Habouzit M., Volonteri M., Somerville R. S., Dubois Y., Peirani Muldrew S. I., et al., 2012, MNRAS, 419, 2670
S., Pichon C., Devriendt J., 2019, MNRAS, 489, 1206 Mundy C. J., Conselice C. J., Duncan K. J., Almaini O., Häußler
Heckman T. M., Beckwith S., Blitz L., Skrutskie M., Wilson A. S., B., Hartley W. G., 2017, MNRAS, 470, 3507
1986a, ApJ, 305, 157 Naiman J. P., et al., 2018, MNRAS, 477, 1206
Heckman T. M., Smith E. P., Baum S. A., van Breugel W. J. M., Nelson D., et al., 2015, Astronomy and Computing, 13, 12
Miley G. K., Illingworth G. D., Bothun G. D., Balick B., Nelson D., et al., 2018, MNRAS, 475, 624
1986b, ApJ, 311, 526 Nelson D., et al., 2019, MNRAS, p. 2010
Hewlett T., Villforth C., Wild V., Mendez-Abreu J., Pawlik M., Nevin R., Blecha L., Comerford J., Greene J., 2019, ApJ, 872, 76
Rowlands K., 2017, MNRAS, 470, 755 Noeske K. G., et al., 2007, ApJ, 660, L43
Hopkins P. F., Hernquist L., Cox T. J., Kereš D., 2008, The As- Pakmor R., Bauer A., Springel V., 2011, MNRAS, 418, 1392
trophysical Journal Supplement Series, 175, 356 Pakmor R., Springel V., Bauer A., Mocz P., Munoz D. J.,
Hopkins P. F., Cox T. J., Hernquist L., Narayanan D., Hayward Ohlmann S. T., Schaal K., Zhu C., 2016, MNRAS, 455, 1134
C. C., Murray N., 2013, MNRAS, 430, 1901 Patton D. R., Torrey P., Ellison S. L., Mendel J. T., Scudder
Huertas-Company M., et al., 2019, arXiv e-prints, p. J. M., 2013, MNRAS, 433, L59
arXiv:1903.07625 Patton D. R., Qamar F. D., Ellison S. L., Bluck A. F. L., Simard
Hung C.-L., et al., 2013, ApJ, 778, 129 L., Mendel J. T., Moreno J., Torrey P., 2016, MNRAS, 461,
Joachimi B., et al., 2015, Space Sci. Rev., 193, 1 2589
Jogee S., et al., 2009, ApJ, 697, 1971 Pearson W. J., Wang L., Trayford J. W., Petrillo C. E., van der
Joseph R. D., Wright G. S., 1985, MNRAS, 214, 87 Tak F. F. S., 2019, A&A, 626, A49
Kartaltepe J. S., et al., 2007, The Astrophysical Journal Supple- Peng Y.-j., et al., 2010, ApJ, 721, 193
ment Series, 172, 320 Pillepich A., et al., 2018a, MNRAS, 473, 4077
Katz N., Weinberg D. H., Hernquist L., 1996, ApJS, 105, 19 Pillepich A., et al., 2018b, MNRAS, 475, 648

MNRAS 000, 1–17 (2019)

16 K. A. Blumenthal et al.
Pillepich A., et al., 2019, MNRAS, p. 2024 Numerical Parameters Cosmological Parameters
Pop A.-R., Pillepich A., Amorisco N. C., Hernquist L., 2018, MN-
Volume 110.73 Mpc3 Ωdm 0.3089
RAS, 480, 1715
Popping G., et al., 2019, ApJ, 882, 137 Ngas 18203 Ωbaryon 0.0486
Puech M., Hammer F., Rodrigues M., Fouquet S., Flores H., Dis- Ndm 10203 ΩΛ 0.6911
seau K., 2014, MNRAS, 443, L49 mbaryon 1.4 × 106 M h 0.6744
Ravindranath S., et al., 2006, ApJ, 652, 963 mdm 7.6 × 106 M
Rich J. A., Kewley L. J., Dopita M. A., 2015, ApJS, 221, 28 baryon 0.7 kpc
Robotham A. S. G., et al., 2014, MNRAS, 444, 3986
dm 1.4 kpc
Rodighiero G., et al., 2011, ApJ, 739, L40
Rodriguez-Gomez V., et al., 2015, MNRAS, 449, 49
Table A1. Left: Numerical specifications of TNG100-1. Right:
Rodriguez-Gomez V., et al., 2019, MNRAS, 483, 4140
Cosmological parameters used in the IllustrisTNG model.
Rupke D. S. N., Kewley L. J., Barnes J. E., 2010, ApJ, 710, L156
Salim S., et al., 2007, The Astrophysical Journal Supplement Se-
ries, 173, 267 White S. D. M., Rees M. J., 1978, MNRAS, 183, 341
Sánchez S. F., et al., 2012, A&A, 538, A8 Whitmore B. C., Schweizer F., 1995, AJ, 109, 960
Sanders D. B., Soifer B. T., Elias J. H., Madore B. F., Matthews Willett K. W., et al., 2015, MNRAS, 449, 820
K., Neugebauer G., Scoville N. Z., 1988, ApJ, 325, 74 Williams C. C., et al., 2014, ApJ, 780, 1
Sawala T., et al., 2015, MNRAS, 448, 2941 Wilman D. J., Zibetti S., Budavári T., 2010, MNRAS, 406, 1701
Scudder J. M., Ellison S. L., Mendel J. T., 2012, MNRAS, 423, Yang X., Mo H. J., van den Bosch F. C., Pasquali A., Li C.,
2690 Barden M., 2007, ApJ, 671, 153
Shattow G. M., Croton D. J., Skibba R. A., Muldrew S. I., Pearce
F. R., Abbas U., 2013, MNRAS, 433, 3314
Shi Y., Rieke G., Lotz J., Perez-Gonzalez P. G., 2009, ApJ, 697,
1764 APPENDIX A: TECHNICAL DETAILS
Sijacki D., Springel V., Di Matteo T., Hernquist L., 2007, MN-
RAS, 380, 877 A1 Data Structure
Sijacki D., Vogelsberger M., Genel S., Springel V., Torrey P., Sny-
The IllustrisTNG model is run at three different volumes
der G. F., Nelson D., Hernquist L., 2015, MNRAS, 452, 575
(TNG50, TNG100 and TNG5300), each with a dark matter
Simons R. C., et al., 2019, ApJ, 874, 59
Skibba R. A., Sheth R. K., Croton D. J., Muldrew S. I., Abbas
only run, and a dark plus baryonic matter run. Additionally,
U., Pearce F. R., Shattow G. M., 2013, MNRAS, 429, 458 there are three (TNG100 and TNG300) or four (TNG50)
Skibba R. A., et al., 2015, ApJ, 807, 152 iterations for each simulation that correspond to different
Snyder G. F., Lotz J. M., Rodriguez-Gomez V., Guimarães initial conditions and resolutions. All simulations contain
R. d. S., Torrey P., Hernquist L., 2017, MNRAS, 468, 207 100 nearly logarithmically spaced snapshots that span a red-
Snyder G. F., Rodriguez-Gomez V., Lotz J. M., Torrey P., Quirk shift range of z = [0 − 20]. Particle data is available for all
A. C. N., Hernquist L., Vogelsberger M., Freeman P. E., 2019, snapshots, and is organised based on three criteria: bind-
MNRAS, 486, 3702 ing energy, subfind halo membership, and friend-of-friend
Sobral D., Best P. N., Smail I., Geach J. E., Cirasuolo M., Garn (FoF) group membership. Subfind haloes (or, ‘subhaloes’)
T., Dalton G. B., 2011, MNRAS, 411, 675
are defined based on the subfind algorithm (Springel et al.
Springel V., 2010, MNRAS, 401, 791
Springel V., Hernquist L., 2003, MNRAS, 339, 289
2001), which links together baryonic and dark matter par-
Springel V., White S. D. M., 1999, MNRAS, 307, 162 ticles into locally over-dense and bound groups. The FoF
Springel V., White S. D. M., Tormen G., Kauffmann G., 2001, haloes (or simply, ‘groups’) are explicitly defined only for
MNRAS, 328, 726 the dark matter particles using the FoF algorithm (Davis
Springel V., Di Matteo T., Hernquist L., 2005, MNRAS, 361, 776 et al. 1985) with linking length b = 0.2, however baryonic
Springel V., et al., 2018, MNRAS, 475, 676 particles’ membership to a FoF group is based on the mem-
Stoughton C., et al., 2002, AJ, 123, 485 bership of the closest dark matter particle.
Tapia T., Eliche-Moral M. C., Aceves H., Rodrı́guez-Pérez C.,
Borlaff A., Querejeta M., 2017, A&A, 604, A105
Toomre A., Toomre J., 1972, ApJ, 178, 623 A2 Merger Trees
Torrey P., et al., 2015, MNRAS, 447, 2753
Torrey P., et al., 2019, MNRAS, 484, 5587 Merger trees (Rodriguez-Gomez et al. 2015) use baryonic in-
Trakhtenbrot B., Lira P., Netzer H., Cicone C., Maiolino R., formation within subfind haloes to trace mergers as a func-
Shemmer O., 2017, ApJ, 836, 8 tion of time. Such a merger tree is constructed by using
Vale A., Ostriker J. P., 2004, MNRAS, 353, 189 three fundamental links: the Descendant, First Progenitor
Ventou E., et al., 2017, A&A, 608, A9 (FP), and Next Progenitor (NP). For a schematic of this
Verley S., et al., 2007, A&A, 472, 121 network, refer to Figure 4 from Nelson et al. (2015). The de-
Vigroux L., et al., 1996, A&A, 315, L93 scendant link tracks subfind haloes through time. The FP is
Vogelsberger M., Genel S., Sijacki D., Torrey P., Springel V., the subfind halo with the largest mass history (i.e. the sum
Hernquist L., 2013, MNRAS, 436, 3031
of this subhalo’s progenitor masses along the main branch)
Vogelsberger M., et al., 2014a, MNRAS, 444, 1518
of a given Descendant. The NP has the next largest mass
Vogelsberger M., et al., 2014b, Nature, 509, 177
Weinberger R., et al., 2017, MNRAS, 465, 3291 history subfind halo of a that Descendant. A merger occurs
Weinberger R., et al., 2018, MNRAS, 479, 4056 when two subfind haloes share a Descendant. Put another
Wen Z. Z., Zheng X. Z., 2016, ApJ, 832, 90 way, a merger occurs when a Descendant has both a First
Wetzstein M., Naab T., Burkert A., 2007, MNRAS, 375, 805 and Next Progenitor. When parsing a merger tree, it is often
Whitaker K. E., et al., 2015, ApJ, 811, L12 useful to consider only the Main Progenitor Branch (MPB),

MNRAS 000, 1–17 (2019)

 Galaxy Mergers in TNG100-1 17
which can be considered the “trunk” of the tree. This pro-
vides information only directly linked to the MPB. Parsing
a merger tree requires at least two identifiers: the identi-
fication (ID) numbers of the First and Next Progenitors.
Walking back along the MPB, each First Progenitor (FP) is
defined by its index in the subhalo catalog at that snapshot
until a FP can no longer be defined. For each FP, there is
a network of Next Progenitors (NP) which were involved in
a merger. Similarly, we terminate the merger tree traversal
when there are no more Next Progenitors for a given FP.

APPENDIX B: VISUAL EXAMPLES

In this Appendix, we present our ideal mock SDSS g-band
images for each of the 85 interacting pairs. These are or-
ganized roughly by their FoF group mass, with the most
massive haloes at the top of the figure.

This paper has been typeset from a TEX/LATEX file prepared by

the author.

MNRAS 000, 1–17 (2019)

18 K. A. Blumenthal et al.

MNRAS 000, 1–17 (2019)

 Galaxy Mergers in TNG100-1 19

MNRAS 000, 1–17 (2019)

20 K. A. Blumenthal et al.

MNRAS 000, 1–17 (2019)

 Galaxy Mergers in TNG100-1 21

MNRAS 000, 1–17 (2019)

22 K. A. Blumenthal et al.

Figure B1. Ideal mock SDSS images of the TNG100-1 interacting pairs at z = 0. Galaxies are ordered roughly by their FoF group mass,
with the most massive haloes at the beginning, and the less massive haloes toward the end.

MNRAS 000, 1–17 (2019)