Draft version November 10, 2018

Preprint typeset using LATEX style emulateapj v. 12/14/05

NONLINEAR DYNAMICAL FRICTION OF A CIRCULAR-ORBIT PERTURBER IN A GASEOUS MEDIUM

Woong-Tae Kim1,2
1 Center for the Exploration of the Origin of the Universe (CEOU), Astronomy Program, Department of Physics & Astronomy, Seoul
National University, Shillim-Dong, Kwanak-Gu, Seoul 151-742, Republic of Korea
2 Astronomy Program, FPRD, Department of Physics and Astronomy, Seoul National University, Shillim-Dong, Kwanak-Gu, Seoul
151-742, Republic of Korea
Draft version November 10, 2018
arXiv:1010.1995v1 [astro-ph.GA] 11 Oct 2010

ABSTRACT
We use three-dimensional hydrodynamic simulations to investigate the nonlinear gravitational re-
sponses of gas to, and the resulting drag forces on, very massive perturbers moving on circular orbits.
This work extends our previous studies that explored the cases of low-mass perturbers on circular
orbits and massive perturbers on straight-line trajectories. The background medium is assumed to be
non-rotating, adiabatic with index 5/3, and uniform with density ρ0 and sound speed a0 . We model
the gravitating perturber using a Plummer sphere with mass Mp and softening radius rs in a uniform
circular motion at speed Vp and orbital radius Rp , and run various models with differing R ≡ rs /Rp ,
M ≡ Vp /a0 , and B ≡ GMp /(a20 Rp ). A quasi-steady density wake of a supersonic model consists of a
hydrostatic envelope surrounding the perturber, an upstream bow shock, and a trailing low-density
region. The continuous change in the direction of the perturber motion makes the detached shock
distance reduced compared to the linear-trajectory cases, while the orbit-averaged gravity of the per-
turber gathers the gas toward the center of the orbit, modifying the background preshock density to
ρ1 ≈ (1 + 0.46B 1.1)ρ0 depending weakly on M. For sufficiently massive perturbers, the presence of
a hydrostatic envelope makes the drag force smaller than the prediction of the linear perturbation
theory, resulting in F = 4πρ1 (GMp )2 /Vp2 × (0.7ηB−1 ) for ηB ≡ B/(M2 − 1) > 0.1; the drag force for
low-mass perturbers with ηB < 0.1 agrees well with the linear prediction. The nonlinear drag force
becomes independent of R as long as R < ηB /2, which places an upper limit on the perturber size for
accurate evaluation of the drag force in numerical simulations.
Subject headings: hydrodynamics — ISM: general — shock waves

1. INTRODUCTION galactic nuclei (e.g., Begelman et al. 1980; Menou et al.

Dynamical friction (DF) arising from the gravitational 2001; Ferrarese & Ford 2005).
interaction of a massive body with a background medium Using a time-dependent linear perturbation theory,
is considered to be one of the most powerful mechanisms Ostriker (1999) showed that the drag force on a perturber
responsible for the orbital decay of astronomical objects with mass Mp moving at speed Vp on a rectilinear tra-
(e.g., Binney & Tremaine 2008 and references therein). jectory through a uniform gaseous medium with density
DF occurs not only in a collisionless background in which ρ0 and sound speed a0 is given by
long-range, two-body interactions transfer momentum
  
1 1+M
from an object in motion to the field particles, but also 4πρ0 (GMp )2  2 ln 1−M − M,  M < 1,
Flin = 2 V t

in a continuous gaseous medium where a perturber loses Vp 1 1

p
 ln 1 − 2 + ln
2 M rmin , M > 1,
its momentum due to the gravitational drag force ex-
(1)
erted by its own induced wake. The DF in a gaseous
where M ≡ Vp /a0 is the Mach number, and rmin is the
background is likely to play an important role in the
characteristic size of the perturber. The key features
orbital decay of companions in common-envelope bina-
of equation (1) are that (i) the gaseous DF force be-
ries, supermassive black holes (SMBHs) at galaxy cen-
comes identical to the classical formula of Chandrasekhar
ters, massive galaxies in galaxy clusters, etc. For in-
(1943) for the collisionless drag when M ≫ 1, (ii) the
stance, in a binary system, DF causes a low-mass com-
gaseous DF force is more efficient than the collisionless
panion engulfed by an extended primary to spiral in to-
drag for M ∼ 1, and (iii) the gaseous DF force is non-
ward the core of the primary, which in turn spins up
vanishing even for subsonic perturbers with M < 1 (see
the primary envelope, greatly affecting the subsequent
also Just & Kegel 1990). The last point improves the
evolution of the system (e.g., Taam & Sandquist 2000;
previous notion that the gaseous drag is absent in sub-
Edgar 2004; Nordhaus & Blackman 2006; Maxted et al.
sonic cases due to the front-back symmetry in the steady-
2009 and references therein). Also, numerical simulations
state wakes (e..g, Dokuchaev 1964; Ruderman & Spiegel
show that the DF due to a gaseous medium expedites
1971; Rephaeli & Salpeter 1980). Several recent studies
the growth of SMBHs by mergers in colliding galaxies
showed that equation (1) can be applicable, with some
(e.g., Milosavljević & Merritt 2003; Escala et al. 2004,
modifications, to more general cases such as in a radi-
2005; Dotti et al. 2006; Mayer et al. 2007; Colpi & Dotti
ally stratified medium (Sánchez-Salcedo & Brandenburg
2009), potentially explaining the ubiquity of SMBHs at
2001), for circular-orbit perturbers (Kim & Kim 2007;
Electronic address: [email protected] Kim et al. 2008), for perturbers with relativistic speed
(Barausse 2007) or in accelerating motion (Namouni
2 W.-T. Kim

2010), etc. In particular, Kim & Kim (2007, hereafter This certainly introduces a few important caveats that
KK07) used a semi-analytic method to show that equa- should be noted from the outset. If the gaseous medium
tion (1) is a good approximation to the DF force on is supported primarily by rotation, as in protoplane-
circular-orbit perturbers provided Vp t = 2Rp , where Rp tary disks, the perturber launches spiral waves at Lind-
is the orbital radius. blad resonances that propagate away from it (e.g., Ward
While all the theoretical studies mentioned above con- 1997; Tanaka et al. 2002; Chambers 2009; Lubow & Ida
sider low-mass perturbers and find various astrophysi- 2010 and references therein), which is quite different
cal applications (e.g., Narayan 2000; El-Zant et al. 2004; from quasi-steady density wakes with long trailing tails
Kim et al. 2005; Kim 2007; Conroy & Ostriker 2008; produced in a non-rotating medium (e.g., KK07). If
Villaver & Livio 2009), there are some situations such as the medium is instead supported by thermal pressure,
in orbital decay of SMBHs or companions in common- it should have a stratified density in the radial direc-
envelope binaries, where perturbers have so large masses tion, as in intracluster media or common-envelope bi-
that the induced density wakes are in the nonlinear naries. In this case, the gradient in the background
regime. Using hydrodynamic simulations, Kim & Kim density profile can be ignored if the Bondi radius is
(2009, hereafter KK09) extended the work of Ostriker much smaller than both the pressure scale length and
(1999) to study nonlinear DF force for a very massive the orbital radius, that is, if M ≪ (M (Rp )/Mp )1/4
perturber on a straight-line trajectory. By modeling a and B ≪ 1, where M (Rp ) is the dynamical mass en-
perturber using a Plummer sphere with softening radius closed within Rp and B is the dimensionless perturber
rs , KK09 found that a nonlinear supersonic wake is char- mass defined in §2 below. Although the first condition
acterized by a detached bow shock and a hydrostatic en- is readily met, for example, in common-envelope bina-
velope near the perturber. The resulting drag force de- ries with planets or brown dwarfs as low-mass compan-
pends solely on the dimensionless parameter ηA defined ions (e.g., Nordhaus & Blackman 2006), the second con-
as dition is not well satisfied especially for very massive per-
A GMp turbers with B >
ηA ≡ , with A ≡ 2 , (2) ∼ 1 considered in this paper. In the latter
M2 − 1 a0 rs case, the wakes and the associated DF forces are likely
to be affected by the density gradient of the background
and is given by medium (e.g., see Just & Peñarrubia 2005 for collision-
F  η −0.45
A
less cases). Neglecting the density stratification in the
= , for ηA > 2, (3) background medium also suppresses gas buoyancy, pre-
Flin 2
cluding the potential effects of convective motions and
while F/Flin ≈ 1 for ηA < 0.7. The reduction of the gravity modes in heating the medium by resonant ex-
nonlinear DF force compared to the linear estimate is citations (e.g., Balbus & Soker 1990; Lufkin et al. 1995;
due to the presence of a hydrostatic envelope that makes Kim 2007) and other processes.
the density distribution spherically symmetric near the In addition to the above assumptions, we ignore the
perturber. This clearly demonstrates that the nonlinear orbital motion of the background gas with respect to
effect can be significant if a perturber is very massive. the center of mass of the whole system, which can be
Since astronomical objects usually follow curvilinear a valid approximation only when B/M2 ≪ 1. If this
rather than straight-line trajectories, it is interesting to condition is not satisfied, the neglect of the centrifugal
see whether equation (3) remains valid for perturbers on and Coriolis forces arising from the orbital motion of the
circular orbits. For this purpose, we in this paper take background may affect the density wakes and the drag
one step further from KK09 to study the nonlinear gravi- forces (e.g., Adams et al. 1989; Ostriker et al. 1992). We
tational responses of gas to, and the associated drag force also treat the gas using an adiabatic equation of state,
on, a very massive perturber moving on a circular orbit. which implicitly assumes that the orbital energy of the
This work also extends the linear perturbation analyses perturber is converted to heat as it spirals inward, and
of KK07 by considering the nonlinear effect on the den- thus is valid only if radiative loss is negligible. By ignor-
sity wakes. The intensity of gravitational influence of a ing self-gravity, we do not consider any back reaction of
perturber to the background gas can be measured by the the gas on the perturber.
Bondi radius rB = GMp /a20 . For perturbers on straight- Given these limitations and constraints, we by no
line trajectories, the softening radius (or the perturber means attempt to apply the results of this work to real
size) is the lone length scale with which the Bondi ra- astronomical systems. Nevertheless, the idealized mod-
dius can be compared. The drag force, correspondingly, els considered in this paper help to isolate the effect
depends on rs and Mp only through the dimensionless of the perturber size or orbital radius on the nonlin-
parameter A: increasing rs is equivalent to decreasing ear DF force. The results of this work will be partic-
Mp . Equation (3) then predicts that the drag force on a ularly useful to justify large perturber sizes employed in
highly nonlinear object with A ≫ 1 would be negligible, recent hydrodynamic simulations, such as for SMBHs at
which results simply from the fact that rs is inseparable galactic nuclei (e.g., Escala et al. 2004, 2005; Dotti et al.
from Mp . On the other hand, circular orbits with orbital 2006, 2007; Mayer et al. 2007; Cuadra et al. 2009) and
radius Rp naturally introduce an additional length scale, companions in common-envelope binaries (e.g., Ruffert
so that the dependence of the drag force on rs can be 1993; Sandquist et al. 1998; Ricker & Taam 2008), etc.
explored independently of the perturber mass. These simulations usually treat the perturber using a
While a perturber in real astronomical situations is softened point mass, with its size inevitably limited by
likely to move through a background medium that is numerical resolution. For instance, N -body/SPH simu-
rotating and/or stratified, we in this work idealize the lations for the orbital decay of SMBHs take quite large
gaseous medium as being initially static and uniform.
 Nonlinear Dynamical Friction 3

values (up to a few pc) for rs (e.g., Escala et al. 2004; circular orbit, which is absent for perturbers on straight-
Mayer et al. 2007), although the realistic values are prob- line trajectories. For future purposes, we define a dimen-
ably of order of the Schwartzschild radius (∼ 1 AU sionless parameter
for Mp = 2 × 107 M⊙ ). Grid-based simulations of the B
common-envelope phase of binaries also represent a com- ηB ≡ , (6)
panion using a point mass with size set by the grid spac- M2 − 1
ing, which is larger than the actual companion size by similarly to ηA in equation (2). To explore the paramet-
one or two orders of magnitude (e.g., Sandquist et al. ric dependence of the drag force on R, M, and B (or
1998; Ricker & Taam 2008). If the drag force on circular- equivalently A), we run a total of 65 three-dimensional
orbit perturbers depends on rs similarly to the linear- simulations with R varying in the range of 0.025–0.4, M
trajectory cases, simulations with such large rs would in 0.5–4, and B from 5 × 10−4 to 4 (A from 0.01 to 80).
significantly underestimate the real decay time since the Our fiducial model has R = 0.05, M = 2, and B = 0.5.
induced density wake would erroneously be in the linear We use the orbital time torb = 2πRp /Vp = 2πtcross /M
regime. If the drag force instead turns out insensitive to as the time unit of our presentation.
rs in circular-orbit cases, a large value of rs would be We integrate the basic equations of ideal hydrody-
reasonable as long as it does not change the drag force namics using a modified version of the ZEUS code
much. (Stone & Norman 1992), parallelized on a distributed-
The paper is organized as follows: In §2, we describe memory platform. Our simulations domain is a rectangu-
numerical methods and models we adopt. In §3, we run lar box which spans the region −10Rp ≤ x, y, z ≤ 10Rp .
models for low-mass perturbers and compare the result- We construct a logarithmically spaced Cartesian grid
ing drag forces with the analytical predictions to find with 5123 zones, with spacings of 0.0075Rp, 0.018Rp , and
a proper relationship between rs and rmin . In §4, we 0.11Rp at |x|/Rp = 0, 1, and 10, respectively. For some
present evolution and quasi-steady distributions of fully selected parameters, we have also run low-resolution sim-
nonlinear density wakes and the associated drag forces ulations with 2563 zones and checked that the resulting
on massive, circular-orbit perturbers. Finally in §5, we drag forces agree, within a few percents, with those from
summarize our findings and discuss their implications on the high-resolution models. We adopt the outflow bound-
the proper choice of the perturber size. ary conditions at all the boundaries and run the simula-
tions typically up to 15tcross . The simulation outcomes
2. NUMERICAL METHOD are insensitive to the box size and the boundary condi-
We consider a massive perturber moving on a circular tions we adopt since it takes about ∼ 20tcross for sound
orbit through a gaseous medium, and study the gravita- waves to travel across the simulation box, while the drag
tional response of the gas to the perturber and the result- forces saturate typically within ∼ 0.5torb ∼ (2 − 3)tcross .
ing drag force. Similar work for low-mass perturbers on
3. LINEAR CASES
circular orbits and massive perturbers on straight-line
trajectories was presented in KK07 and KK09, respec- The time-dependent linear-perturbation analyses for
tively. Since the DF timescale is usually much longer the wakes of low-mass perturbers usually consider a
than the orbital time, the idealized circular orbit is point-mass object with vanishing rs in equation (4),
a reasonable approximation to real curvilinear orbits. which in turn requires to introduce the cut-off radius
We assume that the background medium is unmagne- rmin in the linear DF force formula (e.g, Just & Kegel
tized, non-self-gravitating, and initially static and ho- 1990; Ostriker 1999; KK07). On the other hand, one
mogeneous with density ρ0 and adiabatic sound speed has to adopt a non-zero value for the softening radius
a0 . We adopt an adiabatic equation of state with index in numerical simulations, making the DF force depen-
γ = 5/3. The perturber with mass Mp moves with a dent on rs in the linear regime. To explore how the per-
constant angular velocity Ωp along a circle with radius turber mass affects the drag force in comparison with the
Rp on the z = 0 plane; the corresponding linear veloc- linear theory, therefore, it is necessary to find a proper
ity is Vp = Rp Ωp . We represent the perturber using a relationship between rs and rmin that makes the numeri-
Plummer potential cal results consistent with the analytic predictions when
B ≪ 1. For this purpose, we present in this section the
GMp results of numerical simulations for low-mass perturbers
ΦP (x, t) = − , for t ≥ 0, (4)
(|x − xp (t)|2 + rs2 )1/2 with B = 5 × 10−4 (A = 0.01) and R = 0.05 as functions
of M.
where rs is the softening radius and xp (t) =
Figure 1 shows the snapshots of density wakes on the
Rp (cos Ωp t, sin Ωp t, 0) is the perturber location at time
x–y plane at t/torb = 1 for M = 0.8, 1.5, 2, and 3 cases.
t in the Cartesian coordinates (x, y, z).
The perturber in each frame is moving along the white
We take Rp , a0 , and tcross = Rp /a0 as the units
circle in the counterclockwise direction. For subsonic
of length, velocity, and time, respectively, in our sim-
perturbers, the circular orbit makes the density wakes
ulations. Then, our models are completely character-
curved along their orbits that would otherwise remain
ized by three dimensionless parameters: R ≡ rs /Rp ,
symmetric with respect to the line of motion. For super-
M ≡ Vp /a0 , and
sonic models, the perturber is able to overtake its own
GMp wake, creating a strong trailing tail bounded by Mach
B≡ = RA, (5) waves (KK07). As the Mach number increases from
a20 Rp
unity, the opening angle of the head of the curved Mach
which is the ratio of the Bondi radius to the orbital ra- cone deceases, while the tail thickens. The dotted lines
dius. Note that R is a new parameter introduced by the for the supersonic models plot equation (B2) of KK07
4 W.-T. Kim

Fig. 1.— Steady-state distributions of the perturbed density in logarithmic color scale on the x–y plane at t/torb = 1 for low-mass
perturbers with R = 0.05 and B = 5 × 10−4 (A = 0.01). Colorbar labels log(ρ/ρ0 − 1). The white circle in each panel denotes the orbit,
while the small black circle near (x, y) = (Rp , 0) marks the softening radius of the perturber. The spiral dotted lines for the supersonic
cases draw the outer edges of the tails from the linear theory, eq. (B2) of KK07, which are in good agreement with the numerical results.

for the shape of the tail1 , which are in excellent agree- temporal evolution of F (t) for the cases shown in Fig-
ment with the outer boundaries of the tails produced in ure 1. The friction force saturates to a constant value
the simulations. Nevertheless, the finite perturber size in in less than ∼ 2tcross ; the amplitudes of fluctuations in
the latter makes the tail boundaries broader compared F are less than 1% of the mean values. This is unlike
to the point-mass counterparts with rs = 0. in the linear-trajectory models where the DF force on a
For a given density wake ρ(x, t), it is straightforward supersonic perturber increases logarithmically with time,
to calculate the gravitational drag force on the perturber since the whole density wake growing in size with time is
by evaluating the integral located behind the perturber. The region of influence
in the circular-orbit models expands at a sonic speed
GMP (|x − xp |)ρ(x, t)(x − xp ) · ϕ̂ 3
Z
F (t) = d x, (7) from the orbit center. Consequently, the far-field wake in
|x − xp |3 these models is more or less spherically symmetric and
thus does not contribute to the DF force much. Fig-
where MP (r) = Mp r3 /(r2 + rs2 )3/2 is the mass distribu- ure 2b plots as solid circles the steady-state drag forces
tion of the Plummer sphere and ϕ̂ = ẑ × x̂p is the unit against M for low-mass perturbers with B = 5 × 10−4
vector in the azimuthal direction. Figure 2a shows the (A = 0.01) and R = 0.05. The solid line represents the
1 Equation (B2) in KK07 reduces to R/R = M−1 s for R/R ≫ fit using equation (1) with
p p
1, with s being the azimuthal angle, suggesting that the wake tail
at large radii resembles an Archimedean spiral. rmin /rs = 1.5 + 0.7(M − 1.8)2 , (8)
 Nonlinear Dynamical Friction 5

Fig. 2.— (a) Temporal evolution of the dimensionless DF force for R = 0.02, B = 5 × 10−4 (A = 0.01), and M = 0.8, 1.5, 2.0, and 3.0.
The drag force reaches a steady-state value within ∼ 2tcross . (b) The mean drag force (solid circles), averaged over t/tcross = 2–5, against
the Mach number. The standard deviation is smaller than the size of the circles. The solid line plots the linear force formula (eq. [1]) with
Vp t = 2Rp and rmin /rs = 1.5 + 0.7(M − 1.8)2 in good agreement with the numerical results.

which is in good agreement with the numerical results. For the linear-trajectory cases, KK09 has shown that
In what follows, we will use equation (8) as the cut-off the counterstream slows down due in part to the strong
radius when we compare the numerical DF forces with pressure gradient established near the perturber and in
the analytic estimates. part to a buoyant expansion of the vortex ring in the
lateral direction. The interaction of the counterstream
4. NONLINEAR CASES with the shocked gas causes the bow shock to move back
and forth around an equilibrium position. During this
4.1. Wake Evolution
time, the vortex ring also undergoes an overstable oscil-
We now focus on the cases with very massive per- lation whose amplitude grows secularly with time. When
turbers. Figure 3 plots the snapshots of logarithmic the vortex ring moves out beyond a half of the Hoyle-
density on the x–y plane for our standard model with Lyttleton radius rHL = 2GMp /Vp2 , it becomes less gravi-
R = 0.05, M = 2, and B = 0.5 (A = 10). In each tationally bound and soon swept downstream by the ram
panel, the white line draws the perturber orbit, while the pressure of the incident gas, leaving a near-hydrostatic
small circle in black marks the perturber location with envelope surrounding the perturber. This is how a den-
its size corresponding to the softening radius rs . The sity wake enters into a quasi-steady state in the linear-
perturber is initially located at (x, y) = (Rp , 0) and or- trajectory cases. For the circular-orbit cases, however,
bits in the counterclockwise direction. Note that the size the line of perturber motion keeps changing with time.
of the region displayed in Figure 3 differs from panel to With the perturber displaced from the direction of the
panel for clear presentation. The evolution of the den- counterstream, the counterstream is virtually unhindered
sity wake at early phase is qualitatively similar to the and travels almost straight in the direction tangent to
linear-trajectory cases described in KK09. An introduc- the orbit, producing the protruding region in the wake
tion of the perturber at t = 0 provides the background at x ≈ y ≈ 0.85Rp shown in Figure 3b . Consequently,
gas with strong perturbations that readily develop into the vortex ring manifesting its presence by low density
a bow shock. The incident gas along the line of motion at x/Rp = 0.6–0.9 and y/Rp = 1.1–1.2 in Figure 3c that
is shocked and gathered near the perturber. This sets was carried by the counterstream decouples from the per-
up a strong pressure gradient in the region between the turber, promptly leaving a quasi-static density distribu-
shock and the perturber, which in turn causes the shock tion around the perturber. Note that the time to form a
to slowly move away from the perturber in the upstream near-hydrostatic envelope is about ∼ 0.5torb ≈ 30rs /a0 ,
direction until it reaches an equilibrium position. Some which is shorter than in the linear-trajectory cases by
gas flowing with a non-zero impact parameter is grad- about an order of magnitude.
ually pulled by the perturber, adding to the density at While the density distribution near the perturber is
the rear side (Fig. 3a). Due to the strong gravity, the nearly spherical, a part of the wake outside the orbit is
material piled up at the backside is pulled back toward pushed radially outward by the counterstream, develop-
the perturber, creating a strong counterstream along the ing a trailing tail on the x–y plane. As the counterstream
line of motion as well as a surrounding vortex ring (see weakens, the low-density regions associated with the vor-
KK09).
6 W.-T. Kim

Fig. 3.— Density shapshots on the x–y plane for a supersonic model with M = 2, R = 0.05, and B = 0.5 (A = 10). The size of the
region displayed differs from panel to panel for clarity. The white circle or part of it in each panel denotes the orbit, while the small black
circle marks the softening radius of the perturber. In (f ), the black dotted line plots equation (B2) of KK07 for the linear tail, while the
white dashed line that matches the outer edge of the nonlinear tail is drawn by rotating the former by 75◦ in the counterclockwise direction.
Colorbar labels log(ρ/ρ0 − 1). See text for detail.
 Nonlinear Dynamical Friction 7

Fig. 4.— Quasi-steady distributions of the density wake and velocity field, seen in the inertial frame, on the x/Rp = 1 (bottom right),
y = 0 (top left), and z = 0 (bottom left) planes for a supersonic model with M = 2, R = 0.05, and B = 0.5 (A = 10) at t/torb = 2. The
small black circle marks the softening radius of the perturber, while the white curve in the bottom left panel draws the perturber orbit.
The length of the line segment shown in the upper left corner of each panel, corresponding to twice the sound speed, measures the size of
the velocity vectors. Colorbar labels log(ρ/ρ0 − 1).
8 W.-T. Kim

Fig. 5.— Distributions of quasi-steady density wakes with differing M on the x–y plane, with the perturber located at (x, y, z) = (Rp , 0, 0).
All the models have R = 0.05 and B = 1. The white circle in each panel shows the orbit. Colorbar labels log(ρ/ρ0 − 1).

tex ring first merge together and are then stretched along of each panel, corresponding to twice the sound speed,
the orbit as the perturber continues the orbital motion measures the amplitudes of the velocity vectors. The
(Fig. 3d, e). In the low-mass perturber cases with A ≪ 1, density wake consists of a spherical envelope surrounding
the wake tail is created by the overlapping of the Mach the perturber, a detached bow shock, and an extended
cone and the sonic sphere (KK07). In the nonlinear cases, low-density region at the rear side. The envelope is al-
however, it is the low-density regions produced by the most hydrostatic, as evidenced by low-amplitude velocity
counterstream that separate the tail from the rest of the vectors in the y = 0 plane to which the perturber mo-
wake. The black dotted curve in Figure 3f plots the tion is perpendicular instantaneously. The bow shock
shape of the linear tail, as in Figure 1, which does not standing outside the orbit, in the z = 0 plane, is just a
match the outer edge of the nonlinear tail in our fiducial curved version of that in the linear trajectory counter-
model. Instead, the latter is well matched by rotating the part and extends to large radii R = (x2 + y 2 )1/2 in the
former by 75◦ in the counterclockwise direction, shown orbital plane. On the other hand, the (almost planner)
as the dashed curve. shock located inside the orbit gradually weakens at small
4.2. Quasi-steady Density Wakes
R and terminates at R/Rp ∼ 0.5–0.6 since the speed of
the density wake corotating with the perturber becomes
Figure 4 shows the quasi-steady distributions of the subsonic there. The low-density regions delineating the
density as well as the velocity field, seen in the inertial tail seen at t/torb = 1 (Fig. 3f ) are less apparent in Fig-
frame, on the x/Rp = 1, y = 0, and z = 0 planes for ure 4 as they diffuse out over time by pressure gradients.
our fiducial model at t/torb = 2. The colorbar labels While the velocity just ahead of the perturber is parallel
the perturbed density in logarithmic scale. The size of to the direction of the instantaneous perturber motion,
the short line segment shown in the upper left corner
 Nonlinear Dynamical Friction 9

Fig. 6.— Density profiles of steady-state wakes as functions of the distance l from the perturber center along the azimuthal direction
for (a) models with fixed R = 0.05 and varying B and (b) models with fixed B = 1 and varying R. All the models have M = 2. In (a),
the dotted lines give the respective density distributions under the assumption of hydrostatic equilibrium, while the numerical results are
drawn as the solid lines. Note that varying R < ∼ 0.1 changes the density only in the very central parts of the hydrostatic envelopes.

the direction of gas motions at the perturber location solid lines in Figure 6a give the simulation results for
and the immediate behind it are inclined about 30◦ –50◦ perturbers with fixed R = 0.05 and differing B. The
from the tangential direction. This implies that the gas sharp discontinuity in each curve is due to the presence
in the downstream side retains the information on the of the bow shock. The dotted lines draw the density
direction of perturber motion in the past. distributions of the hydrostatic spheres
Figure 5 displays changes in the quasi-steady density
wakes with varying M on the orbital plane for models 
(γ − 1)a20 B

rs
1/(γ−1)
with R = 0.02 and B = 1 (A = 20). In each panel, ρ(l) = ρc 1 + −1 ,
the perturber located at x/Rp = 1 and y = 0 is mov- a2c R (l2 + rs2 )1/2
ing along the white circle in the counterclockwise direc- (9)
tion. When M = 0.5, the density distribution is circular where ρc and ac denote the density and adiabatic sound
near the perturber and becomes eccentric away from it, speed at l = 0 (i.e., the perturber center), respectively
with the minor axis parallel to the line of motion. Still, (KK09). These are in good agreement with the den-
there exists a slight excess of the perturbed density in sity profiles of the steady-state wakes in our simulations,
the rear side, giving rise to a non-vanishing (but small) suggesting that the envelopes are indeed almost hydro-
DF force. Compared to the M = 0.5 model, the model static. Figure 6b shows how the density profile varies
with M = 0.8 has a larger density excess at the back- with R. When B = 1, the nonlinear effect is not signif-
side. While the perturber is subsonic in this model, a icant in models with R > 0.33 since they have ηA < 1;
strong gravitational pull is able to accelerate the back- the density jump occurring at l ∼ 0 in the R = 0.4
ground gas to a near-transonic speed, creating a weak model is more like a Mach wave rather than a shock. As
shock outside the orbit. The shape of the density wake R decreases (or ηA increases), the density wake becomes
in the M = 1.2 model is similar to that for the M = 2.0 increasingly nonlinear, producing a hydrostatic envelope
model, although the former has a larger peak density and and a detached bow shock. Note that the shock loca-
a larger distance of the detached bow shock. As M in- tion is unchanged as long as R < >
∼ 0.2 (or ηA ∼ 2). When
creases further, both the detached shock distance and the the density is highly nonlinear, the strong gravity as-
size of the hydrostatic envelope decrease. The ring-like sociated with small rs increases the density only in the
low-density gap at R/Rp ∼ 2 in Figure 5d is a remnant small regions within < ∼ (1 − 2)rs . Since this core region
of the counterstream that was stretched along the orbit. is spherically symmetric, the resultant DF force becomes
With large values of M and B, the counterstream in this independent of R provided ηA > ∼ 2. We will show this
model was so strong that the deep gap has not been filled directly in §4.4 below.
in yet. Unlike in the linear-trajectory cases where a perturber
In order to check whether the density wake near the travels through an undisturbed medium, it is inevitable
perturber is in hydrostatic equilibrium, we plot in Figure that a circular-orbit perturber enters the backside of its
6 the density profiles of steady-state wakes as functions own wake before completing one orbit. In the circular-
of distance l measured from the perturber location along orbit cases, therefore, the density ahead of the bow
the azimuthal direction in models with M = 2. The shock differs from the undisturbed density ρ0 and de-
pends considerably on B, as Figure 6a illustrates. The
10 W.-T. Kim

some fluctuations. The quasi-steady value of δ becomes

larger as the perturber mass increases.
For massive perturbers moving on a straight-line tra-
jectory, KK09 found that the detached shock distance is
given empirically by
GMp
δ∞ = rs ηA = . (11)
a20 (M2
− 1)
This can be interpreted by the balance between the post-
shock thermal energy (∝ a20 M2 ) and the gravitational
potential energy (∝ −GMp /δ) at the shock location for
strong shocks (M ≫ 1). To study how the circular obit
affects the standoff distance of the shock, we take a time
average of δ(t) over t/torb ∼ 1–3 and plot δ/δ∞ in Fig-
ure 9 as various symbols. Again, errorbars indicate the
standard deviations of δ(t) during this time interval. The
numerical results show that the detached shock distances
in the circular orbits are in general smaller than those in
the linear-trajectory cases. This is likely because the con-
Fig. 7.— Modified background density ρ1 ahead of the bow tinuous change in the direction of perturber motion ef-
shock in models with R = 0.05. Various symbols and errorbars fectively reduces the forward momentum of the incident
give the means and standard deviations of the preshock density in gas near the perturber, as illustrated by the spatially-
the regions with ∆l/Rp = 0.5–1.5 from the detached shock. The
solid line plots the best fit given by equation (10). varying velocity field in the wake shown in Figure 4. In
this case, the ram pressure of the incoming flow is able
enhanced preshock density can alternatively be inter- to push the shock front toward the perturber, making
preted as arising from the response of the background δ smaller. The modified background density and sound
gas to the orbit-averaged
Rt gravitational potential of the speed discussed in §4.2 might affect δ as well. We fit our
perturber hΦP i ≡ 0 orb ΦP (t)dt/torb , which is centered results using
at the orbit center. In this sense, the enhanced preshock δ

1 − 0.8Λ, Λ < 0.1,
density ρ1 acts as a modified background density to a = (12)
δ∞ 0.14 + 0.2(1 − log Λ)2 , 0.1 < Λ < 10,
circular-orbit perturber with the gravitational potential
ΦP − hΦP i. To quantify ρ1 , we take a spatial average of with Λ ≡ ηB /(M2 − 1)0.3 , which is plotted as dotted
the preshock density in quasi-steady state over the re- lines. As expected, δ/δ∞ → 1 as Λ → 0 corresponding
gion with (l − δ)/Rp = 0.5–1.5 along the orbit, where to Rp → ∞, although models with Λ < ∼ 0.1 are weakly
δ denotes the location of the detached shock. Figure 7 nonlinear and have δ comparable to or less than rs . Since
plots the resulting ρ1 for models with R = 0.05, with er- δ∞ /Rp = ηB , equation (12) indicates that δ increases
rorbars representing the standard deviations. The solid with Λ, although its increasing rate slows down as Λ →
line is the best fit 10 where δ becomes comparable to Rp .
ρ1 0.46B 1.1
=1+ , (10) 4.4. Drag Force
ρ0 (M2 − 1)0.11
For our nonlinear models, we calculate the DF force on
showing that the change in the background density to a perturber due to its own induced wake. Figure 8b il-
circular-orbit perturbers is almost linearly proportional lustrates the temporal changes of the dimensionless drag
to B. Since ρ1 is quite insensitive to M, the M- force I ≡ F/[4πρ0 (GMp /a0 )2 ] in models with M = 2
dependence of ρ1 in equation (10) can be ignored. and R = 0.02 and varying B. The drag force reaches
a quasi-steady value in less than one orbit, although it
4.3. Detached Shock Distance exhibits some fluctuations for large B. This is markedly
In laboratory experiments and hydrodynamic theo- different from the cases of straight-line trajectories where
ries, supersonic flows over a blunt-body generate a de- the drag force increases logarithmically with time. The
tached bow shock when the maximum angle allowed in quasi-steady value of I is almost independent of B when
the postshock flows is smaller than the nose angle (e.g., B< < <
∼ 0.05, decreases with B for 0.05 ∼ B ∼ 2, and becomes
Liepmann & Roshko 1957; Shu 1992). Although a per- again insensitive to B when B > ∼ 2. The behavior of I
turber in our models does not hold any defined surface with B is a consequence of the circular orbit combined
and merely provides gravitational perturbations to the with the nonlinear effect. The size of a hydrostatic en-
background gas, a spherical envelope formed around it velope in a nonlinear density wake, as measured by the
nevertheless sends off pressure waves into the upstream detached shock distance, becomes larger with increas-
direction, acting as an obstacle to the incoming gas. ing B, which makes the drag force smaller owing to the
Since the nose-angle of a spherical body is 90◦ , the shock front-back symmetry in the wake. At the same time, the
should be detached in our nonlinear supersonic models. circular orbit draws the background material toward the
Figure 8a plots the time evolution of the detached shock orbit center and increases the density in the preshock
distance δ for some selected models with R = 0.02 and region, which in turn tends to increase the drag force.
M = 2. Note that δ initially increases rapidly and then For models with B < <
∼ 2 (so that ρ1 /ρ0 ∼ 2), the nonlinear
saturates at t/torb ∼ 0.5 to an equilibrium value with effect dominates and I decreases with B. When B > ∼ 2,
 Nonlinear Dynamical Friction 11

Fig. 8.— Temporal evolution of (a) the detached shock distances δ and (b) the DF forces F for models R = 0.05 and M = 2. Both δ and
F increase rapidly with time initially and then saturate at t/torb ∼ 0.5 to equilibrium values with some fluctuations. The dimensionless
drag force decreases with B for 0.05 < B < 2 due to the nonlinear effect.

instead of ρ0 . The dotted line marked with ηA = 2 de-

marcates the parameter space into two parts such that
the models shown in the lower-left region are fully non-
linear with ηA > 2, while those in the upper-right region
are in the linear or weakly nonlinear regime. The nor-
malized drag force on weakly-nonlinear perturbers with
0.7 < ηA < 2 is usually larger than that on low-mass per-
turbers with ηA < 0.7 because the density wakes in the
former are distributed closer to the perturber (KK09).
In highly-nonlinear models with ηA > 2 (or ηB > 2R),
on the other hand, the DF force becomes essentially in-
dependent of R. This is of course because the change
in rs affects the wake only within a small region close to
the perturber where the density is spherically symmet-
ric, making a negligible contribution to the drag force
(see §4.2).
Figure 10b plots F/Flin versus ηA for the same models
shown in Figure 10a. Note again that ρ0 in equation (1)
for Flin is replaced by ρ1 . For given B (so that increas-
ing ηA corresponds to decreasing R), F/Flin decreases
Fig. 9.— Quasi-steady values of the detached shock distances δ of
circular-orbit perturbers relative to those, δ∞ , of linear-trajectory logarithmically with increasing ηA > 2. This is because
perturbers. Various symbols and errorbars give the means and ηA ∝ R−1 for fixed B and M, leading to Flin ∝ ln ηA ,
standard deviations of δ/δ∞ over t/torb = 1–3. The dotted curves, while F is constant for ηA > 2. For fixed R (so that in-
expressed as equation (12), are the best fits to the numerical results. creasing ηA corresponds to increasing B), F/Flin ∝ ηA −0.5
,
as illustrated by two dotted lines connecting the numer-
on the other hand, the enhancement of the background ical results for R = 0.05 and 0.025, respectively. Figure
density becomes significant enough to compensate for the 11a plots F/Flin for all of our supersonic models with
nonlinear effect, making I unchanged with B. As we will R = 0.05. Obviously, F/Flin ≈ 1 for low-mass perturbers
show below, the DF force scales well with B when ρ1 /ρ0 with ηA < 0.7. The dotted line is the best fit
is taken into account. F
Compared to the linear-trajectory cases, the circular −0.5
= 1.1ηA , for ηA > 2, (13)
orbit introduces an additional length scale, Rp , so that Flin
it is interesting to explore the dependence of the drag to the numerical results for nonlinear, circular-orbit per-
force upon R. Figure 10a plots as various symbols and turbers, which is very similar to equation (3) for massive,
errorbars the mean values and standard deviations of the linear-trajectory perturbers.
drag forces averaged over t/torb = 2–4 for models with In order to apply equation (13) to estimate the de-
M = 2. Note that the drag force is normalized by ρ1 cay timescale, one needs to specify the softening radius
12 W.-T. Kim

Fig. 10.— Normalized DF forces averaged over t/torb = 2–4 for Fig. 11.— (a) DF forces on supersonic perturbers with R = 0.05
models with M = 2 and various B and R as functions of (a) R and normalized by the linear drag force as a function of ηA . Various
(b) ηA . The dotted line corresponding to ηA = 2 in (a) separates symbols and their errorbars give the means and standard deviations
the domain into two parts such that the models at the lower-left of F/Flin for t/torb = 2–4. The dotted line indicates a slope of
part are fully nonlinear. Dotted lines in (b) have a slope of −0.5 −0.5. (b) The same data as in (a) but with different normalization
and connect the numerical results for R = 0.05 and 0.025 models and as a function of ηB . The dotted line shows the best fit to the
quite well. numerical results for ηB > 0.1 and is given by equation (14).
which is not directly observable in most astronomical sys- to be uniform with density ρ0 and the sound speed a0 ,
tems. In addition, equation (13) is applicable only when adiabatic with index γ = 5/3, non-self-gravitating, non-
R = 0.05 since F on highly massive perturbers is inde- rotating, and unmagnetized. The perturber is repre-
pendent of rs , while Flin diverges as rs → 0. For prac- sented by a Plummer sphere with softening radius rs .
tical use in various situations, therefore, it is desirable Our models are characterized completely by three di-
to have an expression of F that does not rely on rs . For mensionless parameters: R = rs /Rp , M = Vp /a0 , and
this purpose, we plot in Figure 11b the dimensionless DF
forces as a function of ηB for all supersonic models with B = GMp /(a20 Rp ). We run a total of 65 models that
R = 0.05. The dotted line is our fit to the nonlinear part differ in R, M, and B, and obtain the dependence of the
using drag force on these parameters.
For our supersonic models, the density wake of a mas-
4πρ1 (GMp )2 0.7
 
sive perturber reaches a quasi-steady state typically in ∼
F = , for ηB > 0.1, (14) 0.5torb , about 10 times faster than the linear-trajectory
Vp2 ηB0.5
cases. A quasi-steady density wake consists of a spheri-
with ρ1 given by equation (10). For the models with cal envelope surrounding the perturber, a detached bow
ηB < 0.1, equations (1) and (8) with Vp t = 2Rp provide shock in the upstream direction, and an extended low-
a good estimate of F that depends on rs . density region in the rear side. The envelope is almost
hydrostatic, providing a negligible contribution to the
5. SUMMARY AND DISCUSSION DF force. Compared to the linear-trajectory cases, an
DF due a gaseous medium may play a central role in orbit-averaged gravitational potential of a circular-orbit
removing angular momentum from astronomical objects perturber is able to gather the background gas toward
in orbital motions, causing them to spiral in toward the the orbit center, effectively increasing the background
orbit center. In this paper, we use three-dimensional density ahead of the bow shock. Equation (10) describes
hydrodynamic simulations to explore the gravitational the modified preshock density ρ1 , showing that density
wake and the associated drag force on a very massive enhancement in the preshock region is almost linearly
perturber with mass Mp moving at speed Vp on a circu- proportional to B and depends weakly on M.
lar orbit with radius Rp . This work extends our pre- Strong perturbations sent off from a massive perturber
vious studies that considered low-mass, circular-orbit develop into a bow shock through which the upstream su-
perturbers (KK07) and high-mass, linear-trajectory per- personic flow becomes subsonic. Since the density wake
turbers (KK09). The background medium is assumed in a quasi-steady state forms a pattern that is corotating
 Nonlinear Dynamical Friction 13

with the perturber at a constant angular speed, the in- these authors showed that SMBHs inspiral rapidly in
cident flow outside the orbit is supersonic to a perturber ∼ 106 − 107 yrs due to the gaseous DF force, they
with M > 1. This naturally produces a bow shock out- also found that the DF force in supersonic models is
side the orbit that is simply a curved version of that in ∼ 1.5 times smaller, and depends on the black hole
the linear-trajectory counterpart. Formation of a weak mass less sensitively, than the analytical predictions of
bow shock outside the orbit is possible even for a sub- Ostriker (1999) formula (e.g., Escala et al. 2004, 2005).
sonic perturber with M > ∼ 0.8 for which the incident flow The discrepancies between the numerical and analytical
is accelerated to near-transonic speeds. On the other results are most likely due to the nonlinear effect. For
hand, the motion of the steady-state wake is subsonic to instance, one of the models considered by Escala et al.
the gas near the orbit center, even when M > 1. This (2005) has a black hole with mass Mp = 5 × 107 M⊙ and
makes the bow shock gradually weaken at small radii, softening radius rs = 4 pc moving initially with speed
eventually terminating somewhere inside the orbit. The Vp ∼ 100 km s−1 on a circular orbit with Rp = 200 pc
detached shock distance measured along the tangential through a medium with sound speed a0 ∼ 70 km s−1 ,
direction is generally smaller than in the liner-trajectory corresponding to B = 0.2, M = 1.4, and R = 0.02.
counterpart since the continuous change in the direction For this model, equation (14) compared with equation
of the perturber motion effectively reduces the forward (1) would give Flin /F ∼ 2, consistent with the results of
momentum of the wake; equation (12) gives the alge- Escala et al. (2005). Since the orbital decay timescale
braic fits to the numerical results for the detached shock is given by τdecay = Mp Vp /F , equation (14) predicts
distance. τdecay ∝ Mp−0.5 for sufficiently massive perturber, which
Unlike in the linear-trajectory cases where the drag is more consistent with τdecay ∝ Mp−0.3 inferred from
force increases logarithmically with time, the DF force in
the circular orbits approaches a quasi-steady value in less the results of Escala et al. (2004, 2005) than the linear
than one orbit. The quasi-steady drag force is essentially prediction τdecay ∝ Mp−1 . The difference between the re-
independent of R as long as ηA > 2 (or ηB > 2R) since sults of the current paper and Escala et al. (2004, 2005)
the change in rs modifies the density wake only within is presumably due to the effects of density stratification,
∼ (1 − 2)rs from the perturber center, where the wake is rotation, self-gravity in the background medium, which
almost spherically symmetric and thus has a negligible are not considered in the present work.
contribution to the drag force. For sufficiently massive Another important issue regards the proper choice of
perturbers, the presence of a near-hydrostatic envelope the softening length of a point-mass perturber employed
in the density wake also makes the nonlinear drag force in numerical simulations for the gaseous DF. As men-
smaller than the linear estimate. Since the choice of rs tioned in §1, the limited numerical resolution commonly
is uncertain in many practical applications, we fit the requires to take rs a few orders of magnitude larger than
nonlinear drag force on a circular-orbit perturber using the realistic size. Are such large values of rs acceptable
equation (14) that does not involve rs . in calculating the orbital decay time accurately? Based
That the nonlinear drag force in circular orbits is in- on our numerical results, the answer is yes, provided
sensitive to the softening radius seems inconsistent with ηA > 2 (or ηB > 2R). Otherwise, a large value of rs
the results of KK09 for linear-trajectory perturbers. But, would cause the gravity near the black hole to be re-
it does not. For the latter case, rs is in fact a lone length duced significantly. The resulting density wake would
scale relative to which other length scales such as the then remain in the linear regime, making the drag force
Bondi radius and the distance traveled by the perturber overestimated considerably. For the decay of SMBHs,
are measured. It is then difficult to separate the effect of the models considered in Escala et al. (2004, 2005) have
rs on the drag force from that of Mp because the models ηA ∼ 10, well above the required lower limit. Some mod-
depend on them only through the dimensionless param- els with Mp ∼ 106 − 107 M⊙ in Dotti et al. (2007) and
eter A. Had we run simulations by using dimensional Mayer et al. (2007) have ηA ∼ 1, so that their choices of
quantities in KK09, we would probably have gotten the rs ∼ 0.1 − 0.2 pc are marginally acceptable. For the evo-
results that F ∝ Mp1.5 independent of rs for models with lution of a main-sequence companion with M = 0.4 M⊙
ηA ≫ 1, similarly to the results of the present paper. in a common-envelope binary, Sandquist et al. (1998)
Given that Flin ∝ Mp2 ln(Vp t/rs ), the time-averaged val- took rs ∼ 2.3 × 1011 cm, about 10 times larger than
the real size. Since this corresponds to ηA ∼ 20, our re-
ues of F/Flin , when expressed in terms of ηA in KK09,
sults suggest that the softening radius did not affect the
somehow picked up the power-law dependence on Mp
decay time in their numerical simulations.
more clearly than the weak logarithmic dependence on
Finally, we remark on the several assumptions made
rs . The circular orbit breaks the degeneracy between rs
in the current work. First, we consider a single per-
and Mp by introducing another length scale, Rp , allow-
turber moving on a circular orbit. If the resulting drag
ing to separate the effect on the DF force of rs from that
force (eq. [14]) is to be applied to the decay of double
of Mp . The similarity between equation (13) obtained
black holes, one has to take into account the drag force
for fixed R in the current work and the result of KK09
from the wake of its companion located at the other side
suggests that the latter should be interpreted as the de-
of the orbit, as well. Kim et al. (2008) found that for
pendence of F on Mp for fixed rs .
equal-mass perturbers, the ratio of the DF force from
Now we compare the results of our idealized mod-
the companion wake to that from its own induced wake is
els with those from realistic simulations for the orbital
about ∼ 10 − 50% for supersonic cases, depending on the
decay of SMBHs at galaxy centers (e.g., Escala et al.
Mach number, when the perturbers have so low masses
2004, 2005; Dotti et al. 2006, 2007; Mayer et al. 2007; see
that the wakes are in the linear regime. The nonlinear
also Colpi & Dotti 2009 and references therein). While
effect on massive perturbers has yet to be explored. Sec-
14 W.-T. Kim

ond, the current study assumes an adiabatic equation of We also have ignored the rotation, self-gravity, buoyancy,
state with index γ = 5/3. It is not well known what and turbulent motions of the background medium in the
the most appropriate value of γ is in the background present paper. It is interesting to see what effect each
medium, but it would be γ = 5/3 if the gas is fully of these physical ingredients makes, which will direct our
ionized or γ = 7/5 for preferentially molecular gas, un- future research.
der the assumption that radiative heating and cooling
are unimportant. The numerical results of Mayer et al.
(2007) indicate that the orbital decay is more effective for The author is grateful to E. C. Ostriker and H. Kim for
a softer equation of state, suggesting that the DF force helpful discussions and the anonymous referees for stim-
may depend sensitively on γ. Third, while we consider ulating comments. This work was supported by the Na-
a static background medium with uniform density, it is tional Research Foundation of Korea (NRF) grant funded
more likely to be stratified in real situations. For a col- by the Korean government (MEST), No. 2009-0063616.
lisionless background, Just & Peñarrubia (2005) found Simulations were performed by using the supercomput-
that a density gradient induces an additional drag force ing resource of the Korea Institute of Science and Tech-
in the lateral direction of the perturber motion, amount- nology Information through the grant KSC-2009-S02-
ing to about 10% of the drag in the backward direction. 0008.

REFERENCES
Adams, F. C., Ruden, S. P., & Shu, F. H. 1989, ApJ, 347, 959 Liepmann, H., & Roshko, A. 1957, Elements of Gasdynamics,
Barausse, E. 2007, MNRAS, 382, 826 Galcit Aeronautical Series (New York: Wiley, 1957)
Balbus S. A., Soker N., 1990, ApJ, 357, 353 Lubow, S. H., & Ida, S. 2010, to appear in Exoplanets, ed. S.
Begelman, M. C., Blandford, R. D., & Rees, M. J. 1980, Nature, Seager, Univ. Arizona Press; arXiv:1004.4137
287, 307 Lufkin, E. A., Balbus, S. A., & Hawley, J. F. 1995, ApJ, 446, 529
Binney, J., & Tremaine, S. 2008, Galactic Dynamics, 2nd ed. Mastrodemos, N., & Morris, M. 1999, ApJ, 523, 357
(Princeton: Princeton Univ. Press) Maxted, P. F. L., Gänsicke, B. T., Burleigh, M. R., Southworth,
Chambers, J. E. 2009, Annu. Rev. Earth Planet. Sci. 2009, 37, 321 J., Marsh, T. R., Napiwotzki, R., Nelemans, G., & Wood, P. L.
Chandrasekhar, S. 1943, ApJ, 97, 255 2009, MNRAS, 400, 2012
Colpi, M., & Dotti, M. 2009, to appear on Advanced Science Mayer, L., Kazantzidis, S., Madau, P., Colpi, M., Quinn, T., &
Letters; arXiv:0906.4339 Wadsley, J. 2007, Science, 316, 1874
Conroy, C., & Ostriker, J. P. 2008, ApJ, 681, 151 Menou, K., Haiman, Z., & Narayanan, V. K. 2001, ApJ, 558, 535
Cuadra, J., Armitage, P. J., Alexander, R. D., & Begelman, M. C. Milosavljević, M., & Merritt, D. 2003, ApJ, 596, 860
2009, MNRAS, 393, 1423 Narayan, R. 2000, ApJ, 536, 663
Dokuchaev, V. P. 1964, Soviet Astron., 8, 23 Namouni, F. 2010, MNRAS, 401, 319
Dotti, M., Colpi, M., & Haardt, F. 2006, MNRAS, 367, 103 Nordhaus, J., & Blackman, E. G. 2006, MNRAS, 370, 2004
Dotti, M., Colpi, M., Haardt, F., & Mayer, L. 2007, MNRAS, 379, Ostriker, E. C., Shu, F. H., & Adams, F. C. 1992, ApJ, 399, 192
956 Ostriker, E. C. 1999, ApJ, 513, 252
Edgar, R. 2004, New Astro. Rev., 48, 843 Rephaeli, Y., & Salpeter, E. E. 1980, ApJ, 240, 20
El-Zant, A. A., Kim, W.-T., & Kamionkowski, M. 2004, MNRAS, Ricker, P. M., & Taam, R. E. 2008, ApJ, 672, L41
354, 169 Ruderman, M. A., & Spiegel, E. A. 1971, ApJ, 165, 1
Escala, A., Larson, R. B., Coppi, P. S., & Mardones, D. 2004, ApJ, Ruffert, M. 1993, A&A, 280, 141
607, 765 Sandquist, E. L., Taam, R. E., Chen, X., Bodenheimer, P., &
Escala, A., Larson, R. B., Coppi, P. S., & Mardones, D. 2005, ApJ, Burkert, A. 1998, ApJ, 500, 909
630, 152 Sánchez-Salcedo, F. J., & Brandenburg, A. 2001, MNRAS, 322, 67
Ferrarese, L., & Ford, H. C. 2005, Space Sci. Rev., 116, 523 Shu, F. H. 1992, The Physics of Astrophysics. II. Gas Dynamics
Just, A., & Kegel, W. H. 1990, A&A, 232, 447 (Mill Valley: Univ. Science Books)
Just, A., & Peñarrubia, J. 2005, A&A, 431, 861 Stone, J. M., & Norman, M. L. 1992, ApJS, 80, 753
Kim, H., & Kim, W.-T. 2007, ApJ, 665, 432 (KK07) Taam, R. E., Sandquist, E. L. 2000, ARA&A, 38, 113
Kim, H., & Kim, W.-T. 2009, ApJ, 703, 1278 (KK09) Tanaka, H., Takeuchi, T., & Ward, W. R. 2002, ApJ, 565, 1257
Kim, H., Kim, W.-T., & Sánchez-Salcedo, F. J. 2008, ApJ, 679, Villaver, E., & Livio, M. 2009, ApJ, 709, L81
L33 Ward, W. R. 1997, ICARUS, 126, 261
Kim, W.-T. 2007, ApJ, 667, L5
Kim, W.-T., El-Zant, A. A., & Kamionkowski, M. 2005, ApJ, 632,
157