Draft version August 6, 2024

Typeset using LATEX twocolumn style in AASTeX631

Stellar bars form dark matter counterparts in TNG50

Neil Ash ,1 Monica Valluri ,1 Yingtian Chen ,1 and Eric F. Bell 1

1 University of Michigan Department of Astronomy

1085 S. University
Ann Arbor, MI 48109, USA
arXiv:2408.01504v1 [astro-ph.GA] 2 Aug 2024

ABSTRACT
Dark matter (DM) bars that shadow stellar bars have been previously shown to form in idealized
simulations of isolated disk galaxies. Here, we show that DM bars commonly occur in barred disk
galaxies in the TNG50 cosmological simulation suite, but do not appear in unbarred disk galaxies.
Consistent with isolated simulations, DM bars are typically shorter than their stellar counterparts and
are 75% weaker as measured by the Fourier A2 moment. DM bars dominate the shape of the inner
halo potential and are easily identified in the time series of quadrupolar coefficients. We present two
novel methods for measuring the bar pattern speed using these coefficients, and use them to make
a measurement of the pattern speed and rotation axis orientation for one sample galaxy located in
one of the high-time resolution subboxes of TNG50. The stellar and dark bars in this galaxy remain
co-aligned throughout the last 8 Gyr and have identical pattern speeds throughout. Both the pattern
speed and rotation axis orientation of the bars evolve considerably over the last 8 Gyr, consistent with
torques on the bars due to dynamical friction and gas accretion. While the bar pattern speed generally
decreases over the time course, it is seen to increase after mergers. Rather than remaining static in
time, the rotation axis displays both precession and nutation possibly caused by torques outside the
plane of rotation. We find that the shape of the stellar and DM mass distributions are tightly correlated
with the bar pattern speed.

Keywords: Galaxy bars (2364), Barred spiral galaxies (136), Dark matter distribution (356), Galaxy
dark matter halos (1880), Milky Way dark matter halo (1049), Galaxy evolution (594),
Milky Way evolution (1052)

1. INTRODUCTION soula et al. 2013; Saha & Naab 2013; Petersen et al.
It has been well established that observed stellar bars 2016; Collier & Madigan 2021; Marostica et al. 2024).
are close to maximally rotating (rbar /rcorot ≡ R ≈ 1) Colin et al. (2006), Berentzen & Shlosman (2006), and
(e.g. Corsini 2010; Aguerri et al. 2015; Cuomo et al. Athanassoula (2007) claim that these DM bars lag be-
2019; Guo et al. 2019) while simulated bars are not, pre- hind their stellar bar by ≲ 10 degrees, creating a torque
senting a challenge to ΛCDM (e.g. Algorry et al. 2017; on the stellar bar which enhances the rate of pattern
Roshan et al. 2021; Ansar et al. 2023). One contributor speed slowdown. In contrast, Petersen et al. (2016) find
to this tension is the observed slowdown of simulated that the trapping of DM in bar-like orbits reduces the
bars caused by the transfer of angular momentum to rate of angular momentum transport to the outer halo
the dark matter (DM) halo through dynamical friction from the stellar bar, thereby reducing the rate at which
with material in resonance with the bar (e.g. Tremaine & the bar pattern speed decreases.
Weinberg 1984; Debattista & Sellwood 2000; Athanas- Because of the apparent tension between observed and
soula 2005). These interactions cause the bar to both simulated bars and the importance of bar-halo interac-
grow in strength and slow over time. Stellar bars in tions on the growth and evolution of stellar bars, it is
isolated simulations have also been found to trap DM both valuable and timely to pursue a detailed under-
into bar-like orbits, forming a DM “halo”, “shadow”, or standing of the response of the inner halo to the stel-
“ghost” bar (Athanassoula 2005; Berentzen & Shlosman lar bar within fully cosmological ΛCDM simulations,
2006; Colin et al. 2006; Athanassoula 2007; Athanas- where hierarchical galaxy assembly, gas physics, satel-
lite mergers, and baryonic feedback are each modelled.
2 Ash et al.

In this paper, we demonstrate that DM bars are com- by the spherical harmonic BFE models. We compute
mon in the TNG50 cosmological simulation, matching these models using the AGAMA software package (Vasiliev
predictions from isolated, idealized N -body disk simu- 2019). Our BFE models use Ngrid = 25 logarithmically-
lations. The presence of these halo bars becomes evident spaced concentric shells centered on the potential mini-
in the time series quadrupolar Basis Function Expansion mum and lmax = mmax = 6 for our limiting expansion
(BFE) representations of either the DM density or po- order. We perform this expansion in the inertial simu-
tential. BFE models provide a rich representation of the lation box frame without any coordinate rotation.
bar structures and their evolution, and enable a mea- We use two novel methods to measure bar pattern
surement of both the pattern speed and rotation axis speed using the time series of the quadrupolar BFE co-
evolution provided sufficiently fine time resolution ≲ 10 efficients. These methods are described in full elsewhere
Myr. (Ash & Valluri 2024, in prep.). We provide a brief in-
troduction to each of these below.
2. SIMULATIONS
The simulation used in this work is TNG50 of the Illus- 3.1. Continuous Wavelet Transform
trisTNG simulation suite (Pillepich et al. 2019; Nelson The rotation of the bar leaves an oscillatory signal in
et al. 2019a,b). TNG50 is a magnetohydrodynamical the quadrupolar BFE coefficients which is related to the
N −body simulation run using the moving mesh AREPO pattern speed. To determine the time dependence of
code (Weinberger et al. 2020). It evolves a box of side frequencies represented in our BFE coefficients, we uti-
length ∼ 50 cMpc from z = 127 to z = 0, with a DM lize a Continuous Wavelet Transform (CWT). A CWT
mass resolution of ∼ 3.1 × 105 h−1 M⊙ and mean bary- is the convolution of an input signal x(t) with a wave
onic mass resolution of 5.7 × 104 h−1 M⊙ . The simula- function ψ(t):
tion adopts the Planck 2015 cosmology with h = 0.6774 Z ∞  
(Planck Collaboration et al. 2016). We make use of 1 t−b
X(a, b) = 1/2 x(t)ψ̄ , (1)
Subbox-0, a sub-volume within TNG50 of side length |a| −∞ a
∼ 7.5h−1 cMpc, and the suplementary data catalog in-
troduced in Nelson et al. (2019a). Subbox-0 is a rel- where X(a, b) is the CWT of the input signal, a specifies
atively dense environment containing ∼ 6 Milky Way- a “scale” value, and b specifies the center for the convo-
mass halos. It provides ∼ 3600 snapshots, offering a lution. In contrast to a Fourier transform, in CWT ψ(t)
time resolution of ∼ 8 Myr near z = 0 allowing bar ro- represents a wavelet for which lim|t|→∞ ψ(t) = 0. We
tation to be fully resolved. Finally, we make use of the choose a complex Morlet wavelet (Grossmann & Morlet
catalogue of TNG50 barred galaxies and their proper- 1984), which defines a wavelet using a complex expo-
ties produced by Rosas-Guevara et al. (2022) (hereafter nential with a Gaussian window function:
 2
RG22). 1 −t
Ψ(t) = √ exp exp(i2πCt). (2)
3. METHODS πB B
We use several methods in this work which have been The value B representing the width of the Gaussian is
well established. Determination of principal axis lengths equivalent to twice the variance, and the value C gives
a, b, c and orientation is performed using the iterative the central frequency of the complex exponential. We
shape-tensor method (e.g. Emami et al. 2021; Ash & find that choosing B = 10 and C = 1 gives a good trade-
Valluri 2023, give our exact conditions to terminate it- off between precision in the time and frequency domains
eration). and adopt these values for this work.
To assess whether DM halo bars are common in CWT requires uniform time sampling. We achieve
TNG50 barred galaxies, we take the catalog of 349 this by taking BFE quadrupolar coefficients generated
galaxies at z = 0 presented in RG22. This cata- from the subbox snapshots (which do not have a uni-
log contains massive disk galaxies with stellar masses form time spacing) and linearly interpolating between
M ∗ > 1010 M⊙ , of which ∼ 30% are barred. Using this snapshots on a grid with spacing ∆t ≈ 4.8 Myr. We use
galaxy set, we compute the in-plane Fourier m = 2 am- a grid of 200 scale values logarithmically spaced between
plitude A2 (r) at z = 0 for both the stellar and DM par- the Nyquist frequency (a = 2, f ∼ 105 Gyr−1 ) and the
ticle distribution, using the definitions defined by RG22 frequency corresponding to one full period in our full 8
and cylindrical bins of width 0.1 kpc and height 1 kpc Gyr timecourse (a = 1679, f ∼ 1/8 Gyr−1 ). We note
centered on the mid-plane and potential minimum. that, since the value of the quadrupolar spherical har-
The presence of a stellar bar leaves an aspherical im- monics goes through two full periods for one complete
print on the potential and density which is well encoded rotation in either azimuth or polar angle, the physical
 AASTeX v6.3.1 Sample article 3

frequency of a rotating object is half the frequency ob- which is expressed in terms of a fixed pattern speed as
served in its quadrupole coefficients. We therefore divide
α(tj ) = Ωp · (tj − tj−1 ). (8)
the frequencies obtained in the CWT analysis by 2. We
make use of the N -dimensional CWT implemented in The summation in equation 5 may be performed with
the pywt software package (Lee et al. 2019) to simulta- an arbitrary range of snapshots and radial grid nodes.
neously perform a CWT over all radii and quadrupolar In the limit that a majority of the time evolution of
orders l = 2, −2 ≤ m ≤ 2. the BFE coefficients is explained by a 3D rotation (as
is the case for a stellar bar rotating as a solid body),
3.2. Rotation axis and pattern speed fitting
our function χ2 is minimized when R(tj ) matches the
The rotation of a spherical harmonic function Yml can system rotation and each of the coefficients are rotated
be expressed as a linear combination of up to 2l+1 other into the co-rotating frame. Determining the rotation of
spherical harmonic functions of the same order l. The a system is therefore reduced to a minimization problem
weighting coefficients to perform the rotation are taken of three variables; θ and ϕ describing the orientation of
from the Wigner D-matrices: the rotation axis, and Ωp describing the pattern speed/
l rotational velocity, and can be performed using standard
(l) χ2 -minimization techniques.
X
Yml (R−1 n) = Dmm′ [R]Yml ′ (n), (3)
m′ =−l To allow for a time-varying pattern speed, we perform
a fit for the rotation properties at each snapshot with a
Where n is a unit vector within the non-rotated basis window filter 30 snapshots across and centered on the
and R is either a rotation matrix or unit quaternion current snapshot. We further adopt Gaussian weights
(l)
describing the rotation. The matrix elements Dmm′ [R] wj centered on the current snapshot and with a stan-
are defined as: dard deviation of ∼ 42 snapshots (∼ 290 Myr for the
(l) median snapshot spacing, ∼ 360 Myr near z = 0 and
Dmm′ [R] ≡ ⟨lm|R|lm′ ⟩ . (4)
∼ 200 Myr near z = 2). We choose these weights to ap-
We use this property to define a method to fit si- proximately correspond to the Gaussian kernel used for
multaneously the axis of rotation and pattern speed our CWT analysis, near the typical bar pattern speed
for the stellar bar directly from its quadrupolar BFE of 30 − 60 km s−1 kpc−1 . We restrict the radial extent
coefficients, described in full by (Ash & Valluri 2024, of our fitting to the bar region, from 0.2 ≲ r ≲ 1 comov-
in prep.). In brief, this method attempts to determine ing h−1 kpc. By performing our fitting routine in this
the co-rotating frame in which the time variation of the manner, we are able to take advantage of the extra in-
quadrupolar l = 2 coefficients is minimized. We measure formation provided by many snapshots’ BFE coefficients
this time variation as: while also approximating the time-variation of the rota-
2
tion axis and pattern speed.
1 X wj {ρm m
l (tj−1 , ri ) − Rρl (tj , ri )} Our fit results do not depend on whether we use the
χ2 = ,
W i,j,m σ 2 [ρm
l (tj , ri )] l=2 coefficients from the expansion of the density or the
(5) potential. The width of the Gaussian kernel and win-
P
where wj are weighting coefficients for which j wj = dow function additionally do not affect the mean value
W , tj and tj−1 are subsequent simulation snapshots, ri of our fits but instead act to suppress the influence of
is a radial grid node, σ 2 [ρm
l (tj , ri )] is an estimate of the
noisy snapshots. Finally, the exact value we assume for
squared coefficient uncertainty, and σ 2 [ρm
l (tj , ri )] does not impact our fit results (since we
assume it to be constant), but rather helps the mini-
l
X ′ (l) mization routine to converge. We assume σ[ρm l (tj , ri )]
Rρm
l (tj , ri ) = ρm
l (tj , ri )Dmm′ [R(tj )] (6) th
to be 10% of the 99 percentile of the absolute value
m′ =−l
of the quadrupolar coefficient values. For a more in-
represents the new BFE coefficient value ρm l at snapshot depth investigation of the BFE coefficient uncertainties,
tj after the rotation R(tj ) is applied. R(tj ) is a function see (Ash & Valluri 2024, , in prep.). To calculate the
of three parameters: rotated coefficient modes, we make use of the python
software packages quaternionic and spherical devel-
R(tj ) = R[θ, ϕ, α(tj )], (7) oped by Michael Boyle.
Where θ and ϕ specify the polar and azimuthal angle of 4. RESULTS
the axis of rotation (respectively, in the inertial simu-
4.1. Prevalence of DM bars
lation box frame) and α(tj ) gives the angle of rotation,
4 Ash et al.

Figure 1. Bar metrics applied to the RG22 disk galaxy catalog for DM (left) and stellar (right) particle distributions. Top row:
A2 (r), with the median (dashed line) and interquartile ranges (shaded region) shown for both the barred (red) and unbarred
(blue) populations as defined by RG22. The presence of a stellar bar is easily identified by the peak in A2 (r). In the DM,
the median A2 (r) curves show that the DM in barred galaxies is typically elongated within ∼ 2 kpc more strongly than in
unbarred galaxies. Middle row: histograms for A2,max . Medians (dashed lines) and the interquartile region (dotted lines) of
the distributions are shown. The ratio of stellar bar to DM bar strength is approximately 4 : 1 for the median barred galaxy
in this catalog, as measured by A2,max . Bottom row: cumulative histogram of the difference in phase angles φ(r) between the
stellar and DM mass distributions, measured where the bar amplitudes are typically highest. Typical measurement uncertainty
is indicated by the black line. The distribution of |φ∗ − φDM | is consistent with no net offset between stellar bars and their DM
counterparts.

In figure 1, we show the median and interquartile re- the ratio of DM to stellar bar strength is ∼ 0.26 for the
gion for the amplitude of A2 (r) for both stars and DM median barred galaxy in our sample.
in the barred and unbarred populations. The presence The elongation in the DM distribution is typically
of a bar is clear in the stellar population, with a median aligned with the major axis of the stellar bar, as re-
peak amplitude (defined here as the amplitude at the vealed by the phase angles of the DM and stellar m = 2
first maximum determined by a cubic spline outside 0.3 moment φ(r). We measure the difference in phase an-
kpc) of A2,max ∼ 0.36, as compared to the median peak gles |φ∗ − φDM | using all particles with |z| ≤ 0.5 and
in the unbarred population of A2,max ∼ 0.08. While within the the radial range 0.75 ≤ r ≤ 1.25 kpc, where
the DM of the barred galaxies features a lower A2,max both DM and stellar bar strengths peak for the median
of ∼ 0.09, there is a systematic offset in the DM A2 (r) barred galaxy. We find no evidence for a net offset be-
between the barred and unbarred galaxies. By taking tween the DM and stellar bars at a population level:
a ratio of A2,max for DM and stellar bars, we find that In barred galaxies, |φ∗ − φDM | is less than our typical
measurement uncertainty (∼ 4◦ , estimated by 100 boot-
 AASTeX v6.3.1 Sample article 5

strap iterations) in 71% of barred galaxies and less than the bar region from larger radii. This could be indica-
twice our uncertainty in 94% of barred galaxies. This tive of gas accretion into the bar region. In the bottom
alignment of the m = 2 Fourier modes is only weakly ob- row of figure 2, we show contour plots of the 3D density
served in unbarred galaxies, which instead show a nearly contributions in the x − y plane from each of the three
uniform distribution in phase angle. We note that the particle types at present day, rotated to align the bar
maximum possible separation of phase angles is 90◦ , be- major axis to x and rotation axis to z. Density con-
cause the m = 2 mode is symmetric for reflections about tours from our BFE models are overplotted. The DM
the origin. density is elongated in the same direction as the stel-
Taken together, we can observe that the DM distri- lar bar. This is reflected by the peak amplitudes of the
bution in the inner regions of barred galaxies is system- l, m = 2 potential coefficients, which act as a correction
atically more elongated than in unbarred galaxies, and to a spherical potential and are a factor ∼ 9 lower in the
that this elongation is aligned to the stellar bar ma- DM compared to the stellar coefficients. This is con-
jor axis. The presence of DM bars in the TNG50 cos- sistent with the lower A2,max amplitudes for DM bars
mological ΛCDM simulation is consistent with previous compared to stellar bars shown in figure 1, nonetheless
results from a large body of idealized N -body simula- the presence of the DM bar is easily identified by the
tions of isolated galaxies (Athanassoula 2005; Berentzen oscillation in the quadrupolar coefficients. The contour
& Shlosman 2006; Colin et al. 2006; Athanassoula 2007; plots of the gas density show a trailing-arm like feature
Saha & Naab 2013; Petersen et al. 2016; Collier & Madi- beginning at the outer edge of the bar and extending for
gan 2021; Marostica et al. 2024), however the degree of several kpc.
alignment between the DM and stellar bars is inconsis- We see evidence for DM “wake” structures in the BFE
tent with Athanassoula (2007) who finds the two bars coefficients. These structures emerge just beyond the
to be misaligned by ≲ 10◦ in her simulations. bar region as faint filaments which extend to larger radii
and are sloped generally downwards. This downward
4.2. Detailed investigation of SubhaloID 574286 in slope implies that they reach a given phase angle after
Subbox-0 the DM/stellar bar, and that the time at which they ar-
rive at this phase angle is radius dependent. This picture
The ratio of A2 (r) Fourier amplitude between barred
is consistent with that of e.g. Athanassoula (2005) and
and unbarred DM distributions is considerably smaller
Petersen et al. (2016), who show that DM wakes may
relative to the same ratio for stellar distributions. At-
trail behind stellar bars and exert a negative torque on
tempting to detect the presence of a DM bar using A2 (r)
them.
is therefore challenging. In contrast, we find DM bars
are easily detectable in the time series of the quadrupo-
4.2.1. Pattern speed and rotation axis evolution
lar BFE coefficients. In these coefficients, the DM bar
produces a semi-sinusoidal oscillation in the inner re- To examine the evolution of the bar pattern speed
gions as the bar structure comes into and out of phase over the ∼ 8 Gyr duration, we use both a CWT analy-
with a given spherical harmonic basis function. We sis on the BFE coefficients and our Wigner D-matrix
chose one galaxy (SubhaloID 574286), which belongs to method, introduced in section 3.2. We perform our
the RG22 catalogue and lies within Subbox-0 for the CWT simultaneously over time and radius for each of
last ∼ 8 Gyr, for closer study. In figure 2, we show the the quadrupolar BFE terms for both stellar and DM
l, m = 2 coefficient values across time and radius for BFE coefficients, and average over the quadrupolar m
the gas, star, DM contributions to the potential of this moments to account for the arbitrary orientation of our
galaxy. The coefficients are rotated to approximately reference frame. CWT results for the stellar and DM
align the rotation axis to the z-axis, which places the BFE coefficients are shown in the top and bottom of
azimuthal oscillations of the m = 2 spherical harmonic the left column of figure 3, respectively. The CWT am-
within the plane of rotation. plitude shows remarkable agreement with the pattern
The presence of the bar is apparent in the inner regions speed fit performed using our Wigner D-matrix method,
of the galaxy as unresolved oscillations in the coefficient which falls nearly exactly along the peak amplitude of
amplitudes. These oscillations are fully resolved in the the CWT at each point in time. This is not a trivial
Subbox-0 snapshots (shown in the inset plots in fig 2). result; the CWT measures frequencies only in the coef-
Notably, the coefficients of the stellar and DM poten- ficients of the quadrupolar spherical harmonic functions,
tials share a common frequency and are in phase with which have differing orientations and are not necessarily
one another. The gas coefficients do not follow this fre- aligned to the plane of the bar’s rotation. In contrast,
quency, but instead appear to periodically extend into the Wigner D-matrix fit takes advantage of all 5 coeffi-
6 Ash et al.

Figure 2. Contributions to the potential from stellar (left columns), DM (center columns), and gas (right columns) particles
for TNG50 subhaloID 574286. Top row: rotated l, m = 2 quadrupolar coefficients for the ’coarse’ time resolution main TNG50
box (left hand side of each column) next to the high time resolution Subbox-0 snapshots between 0.2 ≤ tlb ≤ 1.2 Gyr (right hand
side of each column, tick intervals are 250 Myr) for the same halo. The location of the ’coarse’ time resolution snapshots are
shown here as dashed horizontal lines. Aliased oscillations caused by the bar’s rotation are visible in the ’coarse’ time resolution
snapshots, but can only be resolved using the Subbox-0 snapshots. Bottom row: contours for the BFE density model of each
particle type in the region near the bar (r ≲ 4 kpc) are shown overplotted on the mid-plane mass density at z ∼ 0. Each of the
plots are taken in primed coordinates rotated such that the bar major axis lies along x′ , and minor axis along z ′ (out of the
page). The DM density is aspherical and elongated in the direction of the stellar bar.

cients to determine the pattern speed within the plane Throughout much of its evolution, we find that the
of rotation at each time. DM and stellar bars show a decreasing pattern speed.
The DM and stellar bars share a common pattern The bar is impacted by several mergers, with 5 larger
speed throughout the duration of the 8 Gyr time course, mergers whose pericenter passages are around 7.5, 5.2,
as shown by our CWT analysis. While the amplitude of 4.8, 1.5, and 0.1 Gyr ago (see the bottom panel of figure
the CWT is a factor ∼ 6 higher on average for the stel- 3). The merger 1.5 Gyr ago appears to deliver addi-
lar coefficients, the DM and stellar CWT show a nearly tional angular momentum to the bars, causing the pat-
identical pattern speed evolution, and the pattern speed tern speed to increase from ∼ 35 to ∼ 40 km s−1 kpc−1
fit using our D-matrix method with the stellar particles over roughly 1.2 Gyr. The earlier mergers do not ap-
precisely matches the central frequency for both the DM pear to cause a significant changes to the pattern speed,
and stellar bars. though oscillations in the Wigner D-matrix fit are visible
between 7-8 Gyr ago.
 AASTeX v6.3.1 Sample article 7

Figure 3. Evolution of the bar pattern speed (left column) and rotation axis orientation (right column) for galaxy 574286
located in Subbox-0 of TNG50 over the last 8 Gyr. The bar pattern speed was measured using both the Wigner D-matrix
method using the star particles (cyan line) and by a Continuous Wavelet Transform using the stellar (top) and DM (bottom)
particles. The CWT slices shown are taken at ∼ 2.1 comoving kpc, and show the average CWT amplitude over all 5 quadrupolar
BFE terms. The pattern speed of the DM and stellar bars evolve synchronously throughout the last 8 Gyr, and slow down
for most of the evolution until they are spun up by a merger beginning around ∼ 2 Gyr ago. This merger is evident in the
power in aspherical terms (l > 0) of the density BFE in DM (bottom panel), along with several others (approximate pericenter
passages marked by the magenta dashed lines). The rotation axis of the bar was determined from the stellar particles using the
Wigner D-matrix method and shows both steady precession and nutation throughout the time course, roughly evolving along
an angular circle of diameter ∼ 45◦ . The precession frequency varies over the time course, at times rising suddenly (see e.g.,
the blow-up of a loop structure on the bottom right) which may correspond to merger events (see bottom panel) where the bar
could be subjected to sudden torquing from a massive body outside of the bar rotation plane. We mark the times of satellite
pericenter passages in this plot as magenta stars. Nutations are visible as the higher frequency arcs which occur repeatedly
throughout the loop structure and continuously during the time course. The rotation axis precession is shown in an arbitrary
inertial reference frame in which ϕ′′ and θ′′ refer to the azimuthal and polar angles respectively.
8 Ash et al.

We find that the bar rotation axis is not static in time, beyond the bar region for both DM and stars, whereas
rather it undergoes both precession and nutation over they are defined to occur where A2 (r) is at a minimum
the full 8 Gyr time course. We show this behavior in the (if the minimum exists). Because of the lower A2 (r)
polar plot of figure 3. Neither the precession or nutation amplitudes in DM bars and the apparent discrepancy
frequencies are constant in time, as demonstrated by the between C2 /C0 and A2 (r), we do not attempt to define
change in curvature of the track describing axis orien- a metric to determine the length of a DM bar in this
tation evolution. At a few different points in the time work.
course, the precession frequency becomes quite large, Instead of attempting to measure the time evolution
forming closed “loops” in the track (see the cutout at the of the DM and stellar bar lengths, we track the evolution
bottom of the right-hand column). The disk short axis of the bar shape in terms of its intermediate:major b/a
outside the bar region undergoes a similar amount of and minor:major c/a axial ratios, determined using the
precession to the bar rotation axis over the time course. iterative shape tensor method with an enclosed volume
However, the disk axis and bar rotation axis remain mis- of radius 1 kpc, corresponding roughly to where the bar
aligned over the full 8 Gyr by 5-10 degrees. Therefore, A2 (r) amplitude is at its maximum. The time series of
some but not all of the rotation axis evolution can be the three principal axis ratios for both the stellar and
accounted for by global tumbling of the disk. DM bars are shown in the upper right of figure 4. We
Determining the exact cause of the rotation axis evolu- note that this is not itself a measure of the axis ratios
tion is beyond the scope of this work. Assuming that the of the bars, but rather a measure of the axis ratios of
bar’s rotation is well approximated by a solid body, there the mass distributions within roughly 1 kpc. For this
are likely contributions from both torque-free precession reason, at early times the b/a axis ratios may be higher
and torques which are misaligned with the bar’s angular if this volume extends beyond the bar into the disk/halo
momentum axis. This second scenario may especially be region, where the mass distribution is more axisymmet-
relevant during mergers, when perturbations from mas- ric.
sive bodies outside of the disk plane are likely, however We find that the axis ratios for both the DM and stel-
the bar may be torqued by the disk itself when the bar’s lar distributions become more prolate (b/a is lowered)
rotational plane is misaligned with the disk plane. No- over much of their evolution, while their c/a ratio is
tably, many or possibly all of the “loop” structures seen largely unchanged. Impacts of mergers are visible in the
in the rotation axis track coincide with satellite mergers b/a ratios. In the stellar distribution, oscillations in the
(see fig 3). b/a ratio are visible between 6 − 4.5 Gyr ago, which may
indicate a “ringing” response to a merger which occurs
4.2.2. Bar shape evolution
during this time (We show the influence of this merger
It is not possible for us to follow conventional bar in the bottom panel of 3). Beginning around tlb = 3
length measurements based on A2 (r) to measure the Gyr, the steady decline in the b/a axis ratios is halted,
length of a DM bar. RG22 calibrate their definitions for and b/a begins to increase around tlb = 2 Gyr in both
the existence and radius of a bar based on the A2 (r) val- DM and stellar distributions.
ues for stellar bars, but because DM bars are far more The evolution of the b/a ratio is evidently coupled to
spherical and their A2 (r) are far lower than those for the pattern speed evolution, illustrated in the bottom
stellar bars, these criteria are not appropriate for DM right panel of figure 4. We find that the bar evolves
bars. The analog to the A2 Fourier moment is the nor- on a nearly linear track in the b/a − Ωp plane, with
malized quadrupole power, C2 /C0 , which is defined as lower pattern speeds corresponding to a more prolate
l bar shape in the plane of rotation. This picture is con-
1 X (ρm
l )
2
sistent with the classical picture of the bar growing in
C2 /C0 = 0 2
, (9)
2l + 1 (ρ0 ) l=2 strength by transporting angular momentum into the
m=−l
outer halo (Debattista & Sellwood 2000; Athanassoula
is correlated with A2 but still does not allow a simple
2003). This evolution is evidently disrupted by the late
measurement of bar length. We show its behavior as
merger between 1−3 Gyr ago, which raises the b/a ratio
a function of radius and time in the left two panels of
in both DM and stellar bars, and increases the pattern
figure 4. Notably, C2 /C0 shows a large amplitude out to
speed. It is likely that this merger is responsible for the
roughly ∼ 2 kpc due to the bars in both the stellar and
increase in bar pattern speed, possibly by delivering gas
DM populations, in rough analogy to A2 (see the left side
to the disk which can contribute angular momentum to
of figure 4. The bar lengths measured by RG22 (over-
the bar (Beane et al. 2023).
plotted in cyan) at tlb = 0, 5.2, and 7.9 Gyr correspond
roughly to where C2 /C0 approaches its minimum just
 AASTeX v6.3.1 Sample article 9

Figure 4. Evolution of bar shapes. The normalized spherical harmonic quadrupole power C2 /C0 (left columns) demonstrate
the presence of the bar in both DM and stars by the relatively large amplitude at small radii, which falls to some minimum
value around ∼ 3 kpc before rising again due to the disk/halo component. C2 /C0 does not share a common minimum with the
Fourier A2 (r) (bar length as measured by RG22, cyan squares at tlb = 0, 5.2, and 7.9 Gyr) The corotation radius of the bars is
overplotted (pink dotted line), demonstrating that the stellar bar is always a fast bar (rbar ≤ rcorot ≤ 1.4 · rbar , shaded cyan
region shows 1 − 1.4 · rbar ). We track the evolution of the triaxial intermediate:major (b/a) and minor:major (c/a) axis ratios
of the DM and stellar mass distributions measured at 1 kpc (top right). While the DM axis ratios are higher than the those of
the stellar bar at all times, in both cases the b/a ratio decreases for much of the evolution while the c/a ratio remains relatively
constant. The evolution of the b/a axis ratio is coupled to the bar pattern speed evolution, as shown by the panel in the bottom
right. In the b/a − Ωp plane, the stellar bar is seen to evolve steadily along a ∼linear track, with its pattern speed decreasing as
the bars become more prolate in the plane of rotation. The impact of mergers are visible is this plot as vertical strips around
tlb ∼ 8, 5.5, and 1.5 Gyr where there are sudden changes in the bar shape without much change in pattern speed. As in figure
3, we mark the times of satellite pericenter passages with vertical dashed magenta lines and magenta stars.

5. DISCUSSION AND CONCLUSIONS (2024) form DM bars in the presence of hydrodynam-

We have shown using a catalogue of disk galaxies in ically modelled gas, but to the best of our knowledge
TNG50 that DM “halo” bars are common in the pres- DM bars have not been previously noted in cosmolog-
ence of stellar bars, and that these DM bars are aligned ical simulations where mergers, substructure, and gas
with their stellar counterparts. DM bars have been physics are each modelled. The presence of these bars
previously observed to form in isolated N −body sim- is readily identified in BFE models of the DM potential
ulations of disk galaxies with live halos. Berentzen & as semi-sinusoidal oscillations of the quadrupole terms
Shlosman (2006) demonstrate the formation of DM bars in the inner ∼ 3 kpc. Using these BFE components,
in cosmologically motivated halos and Marostica et al. we have shown that it is possible to measure bar pat-
10 Ash et al.

tern speed and the orientation of its rotation axis, pro- ously suggested Athanassoula (2007). Indeed, Petersen
vided a snapshot spacing of order ∼ 10 Myr. The power et al. (2016) find that prohibiting the formation of the
in the quadrupolar BFE coefficients may enable a mea- DM bar in their simulations enhances the rate of bar
surement of bar lengths, in analogy to the m = 2 Fourier pattern speed slow-down. Negative torques on the bar
amplitude. We leave this determination for later work. are most likely caused by near-resonant material in the
In one sample galaxy taken from Subbox-0, the shape trailing DM wake. Some of this material can become
of the stellar mass distribution in the inner ∼ 1 kpc ap- trapped in bar-like orbits, prohibiting it from negatively
pears coupled to the pattern speed evolution, with the torquing the bar.
b/a axis ratio shrinking as the pattern speed declines. We trace the detailed evolution of the rotation axis
This picture is consistent with a large body of simula- orientation of the stellar bar in our sample galaxy and
tions and theoretical work, showing that bars can be- observe both precession and nutation which occur con-
come longer and slower due to dynamical friction with tinuously and evolve over the full 8 Gyr time course,
a trailing DM wake (e.g. Tremaine & Weinberg 1984; with remarkable similarity to rigid-body rotation dy-
Debattista & Sellwood 2000; Athanassoula 2005). In- namics. This precession is long-lived and does not decay,
deed, we observe evidence for a trailing wake structure nor does the rotation axis tend to a fixed position. We
extending beyond the DM bar in the quadrupolar BFE see several distinct points at which the frequency of pre-
coefficients of the DM potential (see figure 2). Both cession visibly rises, in some cases forming closed “loop”
the DM and stellar bars rotate very fast (∼ 60 km s−1 structures in the axis orientation track. We suggest that
kpc−1 ) at early times (tlb = 8 Gyr), and show a steady these loops likely correspond to times during which the
decline to ∼ 35 km s−1 kpc−1 around 2 Gyr ago. At inner halo is impacted by mergers, when the bars may
this stage, the host halo is impacted by a merger, and experience significant off-axis torques. We save further
both the DM and stellar bars coherently spin up some investigation of these torques, rotation axis evolution,
∼ 5 km s−1 kpc−1 , to roughly 40 km s−1 kpc−1 . The and the DM wake structure for future work.
merging satellite may have delivered gas into the bar re-
gion, which could provide a positive torque on the bars We thank the IllustrisTNG collaboration for providing
and possibly cause the observed spin up (Sellwood & access to the TNG50 simulation data and their virtual
Debattista 2006; Beane et al. 2023). The pattern speed jupyterlab workspace. We additionally would like to
does not reach a steady-state after being spun up, but thank Behzad Tahmasebzadeh, Eugene Vasiliev, and Le-
instead begins to slow again roughly 800 Myr ago at a andro Beraldo e Silva, for numerous suggestions and illu-
rate nearly identical to the pre-merger slowdown rate. minating discussions which helped make this work possi-
At no point in the time course does the pattern speed ble. MV & NA gratefully acknowledge financial support
dip below 34 km s−1 kpc−1 . from NASA-ATP award 80NSSC20K0509. MV also ac-
Because DM bars appear to form in the presence of knowledges support from the National Science Founda-
stellar bars when a live N -body halo is modelled in tion grant AST-2009122.
both idealized and cosmological simulations, we cannot
gauge the impact of the DM bar on angular momentum Software: astropy (The Astropy Collaboration
transport in this work. However, we find no evidence et al. 2018), AGAMA (Vasiliev 2019) numpy (Har-
in TNG50 that DM bars lag behind their stellar coun- ris et al. 2020), scipy (Virtanen et al. 2020),
terparts. This suggests that in ΛCDM DM bars are quaternionic https://quaternionic.readthedocs.io/en/
unable to exert a negative torque on the stellar bar to latest/, spherical https://pypi.org/project/spherical/,
contribute to pattern speed slowdown as has been previ- PyWavelets (Lee et al. 2019).

REFERENCES
Aguerri, J. A. L., Méndez-Abreu, J., Falcón-Barroso, J., Ansar, S., Pearson, S., Sanderson, R. E., et al. 2023, Bar
et al. 2015, Astronomy & Astrophysics, 576, A102, formation and destruction in the FIRE-2 simulations,
arXiv, doi: 10.48550/ARXIV.2309.16811
doi: 10.1051/0004-6361/201423383
Ash, N., & Valluri, M. 2023, The Astrophysical Journal,
Algorry, D. G., Navarro, J. F., Abadi, M. G., et al. 2017, 955, 111, doi: 10.3847/1538-4357/acf30c
Ash, N. F., & Valluri, M. 2024
Monthly Notices of the Royal Astronomical Society, 469,
Athanassoula, E. 2003, 17, 28. https:
1054, doi: 10.1093/mnras/stx1008 //ui.adsabs.harvard.edu/abs/2003RMxAC..17...28A
 AASTeX v6.3.1 Sample article 11

—. 2005, Annals of the New York Academy of Sciences, Nelson, D., Springel, V., Pillepich, A., et al. 2019b,
1045, 168, doi: 10.1196/annals.1350.013 Computational Astrophysics and Cosmology, 6, 2,
—. 2007, Monthly Notices of the Royal Astronomical doi: 10.1186/s40668-019-0028-x
Society, 377, 1569, doi: 10.1111/j.1365-2966.2007.11711.x
Petersen, M. S., Weinberg, M. D., & Katz, N. 2016,
Athanassoula, E., Machado, R. E. G., & Rodionov, S. A.
Monthly Notices of the Royal Astronomical Society, 463,
2013, Monthly Notices of the Royal Astronomical
1952, doi: 10.1093/mnras/stw2141
Society, 429, 1949, doi: 10.1093/mnras/sts452
Beane, A., Hernquist, L., D’Onghia, E., et al. 2023, The Pillepich, A., Nelson, D., Springel, V., et al. 2019, Monthly
Astrophysical Journal, 953, 173, Notices of the Royal Astronomical Society, 490, 3196,
doi: 10.3847/1538-4357/ace2b9 doi: 10.1093/mnras/stz2338
Berentzen, I., & Shlosman, I. 2006, The Astrophysical Planck Collaboration, Ade, P. A. R., Aghanim, N., et al.
Journal, 648, 807, doi: 10.1086/506016 2016, Astronomy & Astrophysics, 594, A13,
Colin, P., Valenzuela, O., & Klypin, A. 2006, The doi: 10.1051/0004-6361/201525830
Astrophysical Journal, 644, 687, doi: 10.1086/503791
Rosas-Guevara, Y., Bonoli, S., Dotti, M., et al. 2022,
Collier, A., & Madigan, A.-M. 2021, The Astrophysical
Monthly Notices of the Royal Astronomical Society, 512,
Journal, 915, 23, doi: 10.3847/1538-4357/ac004d
5339, doi: 10.1093/mnras/stac816
Corsini, E. M. 2010, Direct measurements of bar pattern
speeds, arXiv, doi: 10.48550/ARXIV.1002.1245 Roshan, M., Ghafourian, N., Kashfi, T., et al. 2021,
Cuomo, V., Lopez Aguerri, J. A., Corsini, E. M., et al. Monthly Notices of the Royal Astronomical Society, 508,
2019, Astronomy & Astrophysics, 632, A51, 926, doi: 10.1093/mnras/stab2553
doi: 10.1051/0004-6361/201936415 Saha, K., & Naab, T. 2013, Monthly Notices of the Royal
Debattista, V. P., & Sellwood, J. A. 2000, The Astronomical Society, 434, 1287,
Astrophysical Journal, 543, 704, doi: 10.1086/317148 doi: 10.1093/mnras/stt1088
Emami, R., Genel, S., Hernquist, L., et al. 2021, The
Sellwood, J. A., & Debattista, V. P. 2006, The
Astrophysical Journal, 913, 36,
Astrophysical Journal, 639, 868, doi: 10.1086/499482
doi: 10.3847/1538-4357/abf147
Grossmann, A., & Morlet, J. 1984, SIAM Journal on The Astropy Collaboration, Price-Whelan, A. M., Sipőcz,
Mathematical Analysis, 15, 723, doi: 10.1137/0515056 B. M., et al. 2018, The Astronomical Journal, 156, 123,
Guo, R., Mao, S., Athanassoula, E., et al. 2019, Monthly doi: 10.3847/1538-3881/aabc4f
Notices of the Royal Astronomical Society, 482, 1733, Tremaine, S., & Weinberg, M. D. 1984, Monthly Notices of
doi: 10.1093/mnras/sty2715 the Royal Astronomical Society, 209, 729,
Harris, C. R., Millman, K. J., van der Walt, S. J., et al. doi: 10.1093/mnras/209.4.729
2020, Nature, 585, 357, doi: 10.1038/s41586-020-2649-2
Vasiliev, E. 2019, Monthly Notices of the Royal
Lee, G., Gommers, R., Waselewski, F., Wohlfahrt, K., &
Astronomical Society, 482, 1525,
O’Leary, A. 2019, Journal of Open Source Software, 4,
1237, doi: 10.21105/joss.01237 doi: 10.1093/mnras/sty2672
Marostica, D. A., Machado, R. E. G., Athanassoula, E., & Virtanen, P., Gommers, R., Oliphant, T. E., et al. 2020,
Manos, T. 2024, Galaxies, 12, 27, Nature Methods, 17, 261, doi: 10.1038/s41592-019-0686-2
doi: 10.3390/galaxies12030027 Weinberger, R., Springel, V., & Pakmor, R. 2020, The
Nelson, D., Pillepich, A., Springel, V., et al. 2019a, Astrophysical Journal Supplement Series, 248, 32,
Monthly Notices of the Royal Astronomical Society, 490, doi: 10.3847/1538-4365/ab908c
3234, doi: 10.1093/mnras/stz2306