Dynamics of Galaxies and Clusters of Galaxies

Dynamics of Galaxies and Clusters of Galaxies
L. A. Aguilar
Instituto de Astronoma, UNAM, Baja California Mexico
Aguilar@astrosen.unam.mx
1 Introduction
The purpose of these lectures is to describe some of the dynamical phenomena
that are important in the evolution of galaxies and clusters of galaxies.
Galactic Dynamics is usually perceived as an arid discipline, whose mathe-
matical formalism make it hard to apply to real astronomical problems. This is
very unfortunate, since gravity is behind all astronomical phenomena and its
resulting dynamics has a role to play. This misconception of Galactic Dynam-
ics stems from the fact that often students are introduced to the subject using
a formal approach, full of mathematical rigor, that leads through a lengthy
path, before reaching astronomical applications. Although mathematical cor-
rectness is gained, a physical grasp of the dynamics behind is sometimes lost.
It is our intention to bypass the usual formal introduction and to use
an eclectic list of topics related to the dynamics of galaxies and clusters of
galaxies, to illustrate the use of Galactic Dynamics. Our main goal is under-
standing, rather than rigor. We will attempt to show the student how, using
some basic equations, we can extract useful dynamical information that can
help in our understanding of the realm of galaxies.
In Sect. 2, Poissons and Boltzmanns equations are presented as the basic
equations of Galactic Dynamics. After briey reviewing what makes a dynam-
ical system collisionless, we introduce in Sect. 3 one of the Jeans equations in
spherical coordinates as a quick tool to derive the dynamical properties of a
model built from a density prole, without having to build a self-consistent
dynamical model. The Navarro, Frenk and White prole is used as an ex-
ample. A Mathematica Notebook version of this section is provided as well.
Students with access to this program can interact with this version and change
parameters in the examples provided. The full Notebook is available from the
Guillermo Haro and the authors web pages.
In Sect. 4 we mention two additional density proles that can be used as
exercises for the reader. We also note that simulations of interacting galaxies
L. A. Aguilar: Dynamics of Galaxies and Clusters of Galaxies, Lect. Notes Phys. 740, 71118
(2008)
DOI 10.1007/978-1-4020-6941-3 3 c Springer Science+Business Media B.V. 2008
72 L. A. Aguilar
tend to give density proles that go as r
4
at large radii, when simu-
lated in isolation, as opposed to those of systems immersed in an expanding
background, which tend to give the gentler r
3
of the Navaro, Frenk &
White (NFW) prole. We show why the former dependence is expected for
an isolated nite mass system that has been perturbed.
In Sect. 5 we talk about the orbital structure of spherical potentials. Al-
though idealized, spherical potentials allow us to introduce basic concepts
about orbits. We describe the use of the Lindblad diagram as a tool to clas-
sify orbits in spherical potentials.
In Sect. 6 we tackle a sticky issue: Dynamical friction, one of the main
culprits responsible for robbing center of mass motion to satellite systems,
causing them to spiral toward the center of the host system. As we will see,
this force is a case of bipolar personality, depending on the speed of the object
being reduced, its behavior can change quite a bit.
In Sect. 7 we examine the tidal force, responsible for truncating small
systems when they move within the gravitational inuence of a larger one.
We will see that, contrary to common opinion, tides do not necessarily stretch
along the radial direction. It all depends on the mass distribution of the tide-
producing object. We then discuss the concept of tidal radius at some length,
going from a very simplied static model of two point particles, to the inclusion
of eects due to the motion of the satellite and host systems, non-circular
orbits and extended mass distributions.
In Sect. 8 we talk about what happens when two galaxies collide. The rapid
variation of the perturbing force pumps orbital energy into the internal degrees
of motion of the interacting galaxies and heats them up, promoting mass loss
above the rate due to tidal radius truncation. We discuss the role of two very
important timescales: the interaction time and the internal dynamical time.
As we will see, the amount of damage the interacting systems suer, depends
to a great extent on the ratio of these two numbers.
In Sect. 9 we present a simple exercise, where the combined action of
dynamical friction and tidal truncation, acting on a Plummer sphere that
spirals, following a path close to circular within a at rotation curve halo, is
modeled. This wraps up much of what has been discussed in the three previous
sections.
Finally, in Sect. 10 we suggest some references to those interested student
that which to pursue, in more detail, the topics covered in these lectures.
Finally, Sect. 11 provides some concluding remarks.
2 Basic Galactic Dynamics
The two most important equations of Stellar and Galactic Dynamics are the
Poissons equation, which relates the gravitational potential with its source
(the mass density function ),
Dynamics of Galaxies and Clusters 73
2
= 4G , (1)
and the Boltzmanns equation, which is a transport equation that describes
the evolution of a dynamical system in phase-space,
f
t
+v
f
x

f
v
=
_
f
t
_
coll
(2)
The phase-space distribution function, f(x, v, t), contains all the dynam-
ical information of the system. Its projection in conguration space gives the
spatial density of the system,
(x, t) =
_
f(x, v, t)d
3
v .
The right hand side of Boltzmanns equation contains the so-called collisional
term. Its eect diers from the term in the left hand side, mainly by the
timescale on which the collisions, it describes, operate. The left hand side
describes the ow of particles in a given parcel of the system as it moves in
phase-space, on a timescale determined essentially by its potential, whereas
the right hand side describes the ow of particles in and out of this parcel
due to 2-body collisions, which usually are so fast, with respect to the for-
mer timescale, that the collisional time derivative appears, to the rest of the
equation, as an instantaneous source and sink term.
The natural timescale of the left hand side of Boltzmanns equation is the
so-called dynamical timescale, which is essentially the orbital time for particles
within the system. As a very rough approximation to a system-wide average
of this timescale, we can use the ratio of the system size and the rms velocity,
t
dyn
R/v
rms
.
The eect of the right hand side of Boltzmanns equation operates on the
so-called collisional timescale. It can be shown that this time is of the form
(e.g. see Sect. 3.2 of [46]),
t
coll
(R/v
rms
)N/ log(N) ,
where N is the number of particles within the system.
The ratio of t
coll
to t
dyn
is a measure of the degree of collisionality of a
dynamical system:
t
coll
/t
dyn
N/ log(N) . (3)
Notice that the parameters that characterize the properties of the system
in physical units cancel out, and the only dependence left is on the number of
particles within the system: the more particles, the larger this ratio is, and so
the less important collisions are. This is a bit counter-intuitive and must be
explained further. One would have thought that the more particles there are
74 L. A. Aguilar
in a system, the more collisions will be, and indeed this is the case, but what
matters at the end is not the number of collisions but rather the eect they
have in the system.
Let us take a system with N particles. If we split each particle in two,
keeping everything else constant in the system, like the total mass, size and
the rms velocity, we double the number of particles and thus the number of
collisions. But since the strength of the gravitational interaction scales as the
product of the masses, and each mass is now one half of what it was before, the
result is a reduction in the eect of collisions. In reality, doubling the number
of particles also reduces the mean inter-particle distance (l N
1/3
), which
has the outcome of increasing the collisional eect, but in 3D space this is
dominated by the mass eect, even when taking into account the increased
number of collisions per unit time.
So, the more particles the more collisions, but the less eect they will
have. When the eect of collisions is negligible over the period of time we are
interested in, we are lead to the collisionless Boltzmann equation:
f
t
+v
f
x

f
v
= 0 . (4)
This is a partial dierential equation that, together with Poissons equa-
tion and proper boundary conditions, presumably set by observations, can in
principle be solved. In practice, there is not enough observational information
and furthermore, the mathematical complexity of the task of solving it, makes
the direct solution an impossibility for realistic cases.
Other alternative approaches have been used. In particular, a fruitful ap-
proach is obtained rewriting the previous equation as a total time derivative,
using the fact that minus the gradient of the potential is just the acceleration,
f
t
+ x
f
x
+ v
f
v
=
Df
Dt
= 0 . (5)
This means that, as we move along dynamical trajectories, the value of the
distribution function does not change. This is a very strong restriction that
can be exploited. If furthermore, the system is in steady state, the trajectories
become invariant orbits and f must be a function of those quantities that are
conserved along them:
f(x, v) = f(I
i
), where I
i
(x, v) is such that DI
i
/Dt = 0 , (6)
with I
i
s the so called integrals of motion. This results in the Jeans theorem
that can be used to nd solutions [12]:
Theorem 1 (Jeans Theorem). Any steady-state solution of the collision-
less Boltzmann equation depends on the phase-space coordinates only through
integrals of motion in the galactic potential, and any function of the integrals
yields a steady-state solution of the collisionless Boltzmann equation.
This lead to a whole industry of model building based on classical integrals
of motion: f = f(E), f = f(E, L), f = f(E, L
2
), f = f(E
J
), etc., where E,
L and E
J
are the energy, angular momentum and the Jacobi energy.
Before leaving this approach, we must note that Jeans theorem is strictly
valid only when the system is integrable (i.e., there is a canonical transfor-
mation to a coordinate system where the motion can be separated in each
of its degrees of freedom). Such systems have orbits completely determined
by isolating integrals of motion and are called regular. The appearance of ir-
regular orbits invalidates Jeans theorem [11]. However, steady-state spherical
systems, like those we examine in these notes, do not have irregular orbits.
We will come back to this subject in Sect. 5.
Another approach has been to extract useful dynamical information with-
out necessarily building a full dynamical model. This method is based in the
Jeans equations, which are just the result of taking velocity moments of Boltz-
manns equation. In spherical coordinates, a very useful equation that we will
be using is the one obtained from the 2nd velocity moment:
d
dr
(
2
r
) +

r
[2
2
r
(
2
+
2
)] =
d
dr
, (7)
where
r
,
and
are the three spherical coordinate components of the

velocity dispersion. Using this equation, we can extract information about
these components for a system for which we know the density prole and the
resulting potential, without building a phase-space distribution function for
it. This is a very useful tool. But we must pay a price for this expedience,
as nding a solution to Jeans equations does not guarantee that a physical
model may exist (i.e., f is positive everywhere).
In the next section we will learn to use this equation to explore some
dynamical properties of the Navarro, Frenk and White density prole, without
building a full self-consistent dynamical model for it.
3 A Case Study: The Navarro, Frenk and White prole
The Navarro, Frenk & White (NFW) prole was proposed as a universal
density prole, produced by hierarchical clustering in cosmological simulations
[40]. Here it is used as an example of how to extract information, using just a
few basic equations.
3.1 Density Prole
The density prole is given by:
(r) =
o
(r/r
o
)
1
(1 +r/r
o
)
2
. (8)
76 L. A. Aguilar
It is convenient to cast it in a dimensionless form, by dening a dimensionless
length:
r/r
o
,
and a dimensionless density:
() (r)/
o
=
1
(1 +)
2
. (9)
Its limits at the center and innity are,
lim
0
()
1
, lim
()
3
0 .
We note that at r = r
o
, the density has a value equal to (r
o
) =
o
/4, and
(r) = 1 = (1 +)
2
= 1 = = 0.465571...
The NFW prole has a varying slope in the loglog plane, it diverges as
1
at the center and goes as
3
for . The midpoint for this slope
variation is at 1 (Fig. 1).
3.2 Cumulative Mass
The mass enclosed within a sphere of radius r, is:
M
r

_
r
0
4(r
)r
2
dr
= 4
o
_
r
0
_
r
r
o
_
1
_
1 +
r
r
o
_
2
r
2
dr
= 4
o
r
3
o
_

0
(1 +
)
2
= 4
o
r
3
o
[log(1 +) + (1 +)
1
1] .
Again, it is convenient to dene a dimensionless enclosed mass,
() (M
r
/M
) = log(1 +) + (1 +)
1
1 , (10)
0.1 0.2 0.5 1 2 5 10
0.001
0.01
0.1
1
10
()
Fig. 1. NFW density prole
0.1 0.2 0.5 1 2 5 10
0.01
0.02
0.05
0.1
0.2
0.5
1
()
Fig. 2. NFW cumulative mass
where we have also dened a characteristic mass, M
4
o
r
3
o
.
We note the following values and limits:
() = 1 = 5.3054..., ( = 1) = 0.193147...
lim
0
= 0 , lim
= log() ,
thus, the enclosed mass diverges, but only logarithmically (Fig. 2).
3.3 Potential
The potential of a spherical mass distribution can be calculated as,
(r) = 4G
_
1
r
_
r
0
(r
)r
2
dr
+
_

r
(r
)r
dr
_
.
In our case, the rst integral is,
1
(r) =
4G
r
_
r
0
(r
)r
2
dr
=
G
r
_
r
0
(r
) 4r
2
dr
=
GM
r
r
=
GM
r
o
()
= 4G
o
r
2
o
()
,
and the second integral is given by,
2
(r) = 4G
_

r
(r
)r
dr
= 4G
o
r
2
o
_

(1 +)
2
= 4G
o
r
2
o

1
(1 +)
1
.
Putting everything together, we get:
78 L. A. Aguilar
(r) = 4G
o
r
2
o

1
log(1 +) .
It is natural then to dene a dimensionless potential as:
() =
1
log(1 +) , (11)
where (r) =
o
(), and
o
4G
o
r
2
o
.
Since lim
0
() = 1, it is clear that
o
is the depth of the potential
well. We also note that lim
() = 0, so despite the divergent mass, the

potential well has a nite depth. Since
o
< 0, our dimensionless potential is
a positive function (Fig. 3).
3.4 Force
We now compute the magnitude of the force exerted by the NFW prole,
F(r) =
d
dr
=
o
r
o
_
d
d
_
=
o
r
o
T() .
The dimensionless force is given by,
T()
d
d
=
(1 +) log(1 +)
2
(1 +)
. (12)
The limits of the dimensionless force are,
lim
0
T() = 1/2 , lim
T() = 0 .
Notice that the force is discontinuous at the origin. This is due to the
central cusp of the prole (Fig. 4).
0.1 0.2 0.5 1 2 5 10
0.3
0.4
0.5
0.6
0.7
0.8
0.9
()
Fig. 3. NFW potential
0.1 0.2 0.5 1 2 5 10
0.4
0.3
0.2
0.1
0
F ()
Fig. 4. NFW force
3.5 Escape and Circular Velocities
The escape velocity is easily obtained by the condition of null energy:
E = (1/2)v
2
esc
+(r) = 0 = v
2
esc
= 2(r) = 2
o
() .
It is clear that the natural unit of velocity is

o
. We can then dene a
dimensionless escape velocity as:
2
esc
v
2
esc
/
o
= 2 log(1 +)/ (13)
The limits are,
lim
0
2
esc
= 2 , lim
2
esc
= 0 ,
so, despite the innite mass of the model, the escape velocity is nite.
The circular velocity is obtained from the centrifugal equilibrium condi-
tion:
v
2
c
/r = F(r) =
2
c
= T() =
(1 +) log(1 +)
(1 +)
. (14)
The limits in this case are,
lim
0
2
c
/2 0 , lim
2
c
= 0 .
Notice that the rotation curve of the NFW prole rises as

from the
center, reaches a maximum of
max
c
0.465... at 2.16258..., and then goes
down very gently, falling 10% of its peak value for 6.66...
The ratio of escape velocity to circular velocity goes to innity at the
center, while at large radii, it goes to
2, which is the Keplerian value (Fig. 5).

80 L. A. Aguilar
0.1 0.2 0.5 1 2 5 10
0
0.2
0.4
0.6
0.8
1
1.2
1.4
Fig. 5. NFW escape (upper curve) and circular (lower curve) velocities
3.6 Velocity Dispersion
Up to now, all the properties that we have derived from the NFW prole have
not required information about the velocity distribution of the model. Even
the escape and circular velocities that we have derived, are not diagnostics of
the velocity distribution, but rather characterizations of the potential.
The question thus arises as to the range of variations that are possible in
the velocity distribution as a function of position for the NFW prole. This is
an important issue, because at least in the case of luminous elliptical galaxies,
although there is some homogeneity in the surface brightness proles and
isophotal shapes, there is a wider range of variation in the velocity dispersion
proles. This can be interpreted as changes in the velocity distribution, or in
the mass to light ratio. We will explore the rst possibility.
One possible approach is to build appropriate dynamical models by nding,
by whatever means may be available, the range of phase-space distributions
f(r, v), which project onto the same (r) when integrated over velocity space.
Another, more limited but simpler approach which is quite useful, is to
use Jeans equations to impose restrictions, not in the velocity distribution
but in its moments, in particular in the velocity dispersion. We will assume
no net rotation and a velocity distribution that is invariant under rotations.
The two components of the tangential velocity dispersion are then equal:
,
and the velocity ellipsoid everywhere can be characterized by its radial velocity
dispersion,
r
, and an anisotropy parameter:
1
2
/
2
r
. (15)
Notice that is negative when tangential motions dominate, it goes to 0 for
the isotropic case, and can reach up to 1 for the purely radial motion case.
In spherical coordinates, the Jeans equation that corresponds to the 2
nd
moment of Boltzmanns equation is (7) and (15):
1
d
dr
(
2
r
) + 2
2
r
r
=
d
dr
. (16)
Isotropic Case
The rst case we will study is the model whose velocity distribution is isotropic
everywhere. From the Jeans equation in spherical coordinates for an isotropic
(in velocity space) model, we can obtain the 1-dimensional velocity dispersion
as a function of position ((16) with = 0):
1
d
dr
(
2
) =
d
dr
=
2
(r) =
1
(r)
_

r
(r
)
_
d
dr
_
dr
.
In dimensionless form, this equation is,
2
iso
() =
1
()
_

)
_
d
d
_
d
,
where we have dened the dimensionless velocity dispersion as
iso
/
o
.
Using our previously dened dimensionless density and force functions (9)
and (12), we can evaluate this expression:
2
iso
() = (1 +)
2
_

(1 +
)
2
(1 +
) log(1 +
2
(1 +
)
d
= (1 +)
2
_

(1 +
) log(1 +
3
(1 +
)
3
d
. (17)
The integrand is a function, positive everywhere, that diverges as
1
at the origin and approaches zero for large radii. We can not integrate it
analytically but we can do so numerically (Fig. 6).
The velocity dispersion of the isotropic model goes to zero at the origin
and at large radii while it reaches a maximum of
iso
0.30707 at 0.7625.
The shrinking velocity dispersion at the center is a result of the mild
divergence of the density cusp, diverging as
1
. We can see this as
follows: the equation we used to obtain (r) can be written as:
(r)
2
(r) =
_

r
(r
) F(r
) dr
,
with the left hand side being the local amount of kinetic energy per unit
volume, or local pressure. The right hand side is the force per unit volume,
integrated on a radial column from the local position all the way to innity;
this is the force per unit area that the local element has to support. Now, in
82 L. A. Aguilar
0.1 0.2 0.5 1 2 5 10
0.22
0.24
0.26
0.28
0.3
0.32

iso
Fig. 6. NFW isotropic velocity dispersion
a spherical distribution of mass, the force goes as the enclosed mass divided
by radius squared and the enclosed mass goes as the density times the radius
cubed:
F(r)
M
r
r
2

r
3
r
2
r =
_
F dr
2
r
2
,
so the local pressure
2
has to go as
2
r
2
, or
2
r
2
. If we assume that
r
, it is clear that as r 0 we have:

> 2 =
2
0 ,
= 2 =
2
constant ,
< 2 =
2
,
so a cusp steeper than r
2
is required to force a divergent central isotropic
velocity dispersion. Another way of looking at this is that you need to pack a
lot of mass at the center, so that the resulting gravitational force makes the
local velocity dispersion to soar without bound.
Radial Case
As an extreme case, we will now explore the possibility of building an NFW
model in which all orbits are radial. This would maximize the observed central
velocity dispersion. In principle, one can solve the relevant Jeans equation for
the general case ( ,= 0) as follows. We begin multiplying both sides of (16)
by r
2
:
r
2
d
dr
(
2
r
) + 2r (
2
r
) = r
2
d
dr
,
our next step is to realize that,
d
dr
(r
2
2
r
) = r
2
d
dr
(
2
r
) + 2r(
2
r
) .
The rst equation can then be written as,
d
dr
(r
2
2
r
) 2r(
2
r
) (1 ) = r
2
d
dr
=
d
dr
(r
2
2
r
) = r
2
_
2
2
r
r
(1 )
d
dr
_
,
from which we nally obtain ( = 1),
2
r
(r) =
1
r
2
(r)
_

r
r
2
(r
)
d
dr
dr
. (18)
Our next step would be to write the dimensionless form of this equation
and substitute the appropriate functions for the density and potential of the
NFW model. This can be done and, indeed, it gives an answer that diverges
strongly at the center. However, it is important to emphasize that, although
the Jeans equation can be solved, the implied solution may not be physical.
In the particular case of purely radial orbits, we should realize that we are
putting a strong constraint on the central density: since all orbits are radial,
all go through the center, and so the central region must accommodate all
particles, although not at the same time. It can be proved that the solution
obtained from the Jeans equation for the purely radial orbit in the NFW
case, implies a phase-space distribution function that becomes negative at the
center, which something is clearly non-sensical. This is the mathematical way,
of the formal solution, to accommodate all particles on radial orbits within a
central cusp that does not diverge quickly enough.
We will examine in more detail this question and derive a general lower
limit to the rate of divergence that a central cusp must have to accommodate a
population of particles on radial orbits. The relation between the phase-space
distribution function and the spatial density is:
(r) =
_
f(r, v) d
3
v .
In a system that is integrable, the distribution function f(r, v) should be
expressible as a function of the integrals of motion (Jeans Theorem). Now,
if the system is invariant with respect to spatial rotations, then we can use
f = f(E, L
2
), since the energy E and the magnitude (squared) of the angular
momentum L are invariant with respect to rotations. We can then write the
density as,
(r) =
_
f(E, L
2
) d
3
v = 2
_

dv
r
_

0
v
t
dv
t
f(v
2
r
/2 +v
2
t
/2 +, r
2
v
2
t
) ,
where we have separated the integration over velocity space in two parts, one
over the radial direction (v
r
) and the other over the tangential plane (v
t
).
Now, because we are building a model with radial orbits only, the phase-
space distribution function can be written as:
84 L. A. Aguilar
f(E, L
2
) = g(E) (L
2
) = g(v
2
r
/2 +) (r
2
v
2
t
) ,
where the radial velocity dependence is in the g function and that of the
tangential velocity is in the Dirac function:
(x) = 0 x ,= 0,
_
(x) dx = 1 .
The density integral can then be split in two factors:
(r) = 2
_

g(v
2
r
/2 +) dv
r
_

0
(r
2
v
2
t
)v
t
dv
t
.
To integrate on the tangential velocity we change the integration variable:
x = r
2
v
2
t
dx = 2r
2
v
t
dv
t
, to obtain:
_

0
(r
2
v
2
t
)v
t
dv
t
=
1
2r
2
_

0
(x) dx =
1
2r
2
.
Putting this result back in the density integral, we have:
(r) =

r
2
_

g(v
2
r
/2 +) dv
r
= r
2
(r) =
_

g(v
2
r
/2 +) dv
r
.
Lets assume that at the center, the density prole behaves as r
,
then it is clear that r
2
(r) 0 for > 2, forcing the left hand side of our
result to go to zero at the origin. However, the right hand side is an integral
over g(E), which is a positive function, and so the only way that this integral
can be zero is if g(E) = 0, which gives no model! So, we conclude that the
only way to build a dynamical model having only radial orbits, is to have a
central density cusp that diverges at least as fast as 1/r
2
at the center. The
NFW prole does not satisfy this condition and so no radial orbit model is
possible.
Tangential Case
The opposite extreme to a radial orbit model is one with tangential motion
only. In a spherical model, this means that all orbits are circular and so no
radial mixing exists. Such a model is always possible, since we are free to put
as many stars as required by the density prole at each radius. It is easy to
see that in this case, the phase-space distribution function is:
f(r, v
r
, v
t
) =
1
(r) (v
r
) (v
2
t
v
2
c
) , (19)
where v
c
is the local circular velocity (14).
In this case, (7) reduces to
2
t
= r(d/dr) and so, the tangential velocity
dispersion is simply the circular velocity,
2
tan
() =
2
c
(20)
A word of caution is appropriate here: just because we can nd a solution,
does not imply that it is stable. This is particularly critical for models built
with circular orbits only.
Line of sight
z
2
=

r
2

R
2
r
z
R
Fig. 7. Projection on the plane of the sky, with r the 3D radius, R the projected
component on the plane of the sky radius and z the corresponding along the line of
sight
3.7 Projected Properties
So far we have obtained 3D information about the NFW prole. What is
observed, however, is projections on the plane of the sky. So we now proceed
to obtain the projected versions of the density and velocity dispersion proles.
Projected Density
Having computed the velocity dispersion for the isotropic and tangential ver-
sions of the NFW prole, we can now compute the line of sight velocity dis-
persion. Our rst step is to obtain the projected density as an integral over
the line of sight:
(R) =
_

(r) dz = 2
_

R
(r)
r dr
r
2
R
2
,
where R is the projected distance on the plane of the sky and z is along the
line of sight (see Fig. 7).
Dening a dimensionless projected radius as R/r
o
, and using our
dimensionless functions, we cast the previous equation in dimensionless form:
(r) = 2
_

o
() r
o
r
o
_
2
r
o
d = 2
o
r
o
_

()
_
2
d .
We can now dene a dimensionless projected density as,
() (R)/(2
o
r
o
) .
In the case of the NFW prole, we get (Fig. 8):
() =
_

d
(1 +)
2
_
2
=
1
2
1

arcsec()
(
2
1)
3/2
. (21)
Caution should be used for < 1, where numerator and denominator in
the second term are imaginary, but the result is real and nite. In this range,
it is preferable to use an equivalent form with only real values:
1
1
To get the second form we use arcsec(z) = i ln[(1 +
1 z
2
)/z].
86 L. A. Aguilar
0.1 0.2 0.5 1 2 5 10
0.01
0.02
0.05
0.1
0.2
0.5
1
2
()
Fig. 8. NFW projected density
() =
1
2
1
+
1
(1
2
)
3/2
ln
_
1 +
_
1
2
_
, 0 1 . (22)
We also note that as 1, 1/3. The projected density has the following
limits:
lim
0
() = , lim
() = 0 .
Projected Velocity Dispersion
We can now compute the projected velocity dispersion for the isotropic and the
purely tangential orbits cases. This can be done using the following expression:
2
p
(R) =
2
(R)
_

R
(r)
2
los
(r, R) r
r
2
R
2
dr ,
where
los
(r, R) is the line of sight velocity dispersion, on a volume element
at distance r from the center along the line of sight, at projected distance R.
Elementary geometry (Fig. 9) shows that it can be written as,
2
los
(r, R) =
_
1
R
2
r
2
_

2
r
(r) +
_
R
2
2r
2
_

2
t
(r) ,
Line of sight

r
los
r
z
R
Fig. 9. Projection along the line of sight
los
of the local radial
r
and tangential
t
velocity dispersions
where
r
(r) and
t
(r) are the radial and tangential velocity dispersions at r.
We see that in the isotropic case,
2
t
= 2
2
r
=
2
los
(r, R) =
2
r
.
The projected squared velocity dispersion in the isotropic case is then,
2
piso
(R) =
2
(R)
_

R
(r)
2
r
(r) r
r
2
R
2
dr ,
or in dimensionless form,
2
piso
() =
1
()
_

()
2
iso
()
_
2
d , (23)
where
2
iso
is given by (17).
In the tangential case we have,
2
ptan
(R) =
2
(R)
_

R
(r) (R
2
/2r
2
)
2
t
(r) r
r
2
R
2
dr ,
and in dimensionless form,
2
ptan
() =

2
2()
_

()
2
tan
()
2
d , (24)
where
2
tan
is given by (20).
In both cases the projected squared velocity dispersion rises from the cen-
ter to a maximum at 0.6 (isotropic case) or 2.5 (tangential case) and
then decreases steadily for larger distances (Fig. 10).
0.1 0.2 0.5 1 2 5
0.02
0.04
0.06
0.08
0.1

2
p
Fig. 10. NFW projected squared velocity dispersion for the isotropic (solid) and
tangential (dashed) models
88 L. A. Aguilar
4 Other Interesting Proles
There are two very useful, but still simple proles, that have been used ex-
tensively. These are the Jae ( [33, 37]) and the Hernquist ( [6, 28]) proles:
J
(r) = (M/4r
3
o
) (r/r
o
)
2
(1 +r/r
o
)
2
, (25)
H
(r) = (M/2r
3
o
) (r/r
o
)
1
(1 +r/r
o
)
3
. (26)
Here M is the total mass and r
o
is a scale-length whose physical meaning
is dierent in each case. Notice that these two proles behave as r
4
at large radii, as opposed to the shallower NFW prole. This behaviour has
been found in N-body simulations of perturbed galaxies when they are not
part of a cosmological expansion [4], and has also been observed in some real
elliptical galaxies [35].
What is the reason for a r
4
prole at large radii? It is clear that any
nite mass, power-law prole, should be of the form 1/r
3+
, with > 0
at large radius, however, why the particular 4 exponent?
The distribution function of an isolated, steady-state, galaxy is such that
the number of stars N(E) with energy between E and E+dE goes to zero at
E = 0, where the energy boundary of the system lies. However, when a galaxy
suers an external perturbation, like the tidal force of a passing galaxy, the
external layers of the perturbed galaxy heat up and a continuous non-zero
distribution of stars at the zero energy boundary develops. We will show that
this ensures a r
4
tail at large galacto-centric distances (Fig. 11).
Proposition 1. If a spherical galaxy with nite mass, no rotation and
isotropic velocity distribution, develops a nite, non-zero population of par-
ticles at E = 0, then the tail of the density prole at large radii will exhibit a
r
4
behaviour.
Proof. The energy distribution is given by,
2
N(E=0) 0
N(E)
E
0
Fig. 11. Initial (lower) and perturbed (upper) energy distribution
2
It is important to distinguish between f(E) and N(E). The former is the fraction
of stars per unit phase-space volume, d
3
rd
3
v, while the latter is the fraction of
stars per unit interval in E. f may be very large at some energy, but if the
corresponding volume between E and E + dE is tiny, N will be small.
N(E)dE =
_
E
f(r, v) d
3
r d
3
v = 16
2
_
E
f(E) r
2
dr v
2
dv , (27)
where
E
is the volume in phase-space with energy between E and E + dE
(Fig. 12), and we have used the isotropy of f in r and v.
As f is a function of E only, it is convenient to change the integral to
N(E)dE = 16
2
_
E
f(E) r
2
v
2
(r, v)
(r, E)
drdE .
Since the Jacobian is equal to 1/v, we get
N(E)dE = 16
2
f(E)dE
_
r
E
0
r
2
v dr = f(E) A(E) dE ,
where we have taken f(E) out of the integral, since it is done at xed energy,
r
E
is the largest radius that a particle of energy E can reach, and
A(E) 16
2
_
r
E
0
r
2
_
2[E (r)] dr ,
where we have used v =
_
2[E (r)]. A(E) is the area of the constant
energy surface in phase-space [10].
Now, at suciently large radii, any spherical nite mass distribution has a
potential that approaches the Keplerian limit 1/r. Using this asymptotic
dependence we get,
A(E) r
2
r
1/2
r = r
5/2
E
5/2
.
Then, if N(E 0)dE = f(E 0)A(E 0)dE, is nonzero and nite, we
should have f(E 0) E
5/2
. The density can be written as,
Fig. 12. Two constant energy surfaces in phase-space for the Kepler potential. The
thick axis is the radial distance and the orthogonal plane is the velocity space. The
volume in between the surfaces is
E
. They have been cut open for clarity
90 L. A. Aguilar
(r) = 4
_
0
(r)
f(E)
_
2[E (r)] dE
= E
5/2
E
1/2
E = E
4
r
4
.
This result does not apply to the NFW prole, whose mass diverges.
5 The Orbital Structure of Spherical Potentials
We now study the orbital structure of spherical mass models. Why should
we be interested in the orbital structure of a model? There are interesting
problems where it is important to nd out, for instance, the radial region
spanned by individual orbits: we may be interested in the fraction of stars in
a galactic bulge that plunge within the radius of inuence of a central massive
black hole, or the eect of radial mixing in galactic metallicity gradients.
There is another, more fundamental, reason for studying orbits. At the
beginning of the XX century, the mathematician Emmy Noether proved a
result that implies that, when a potential presents a symmetry (i.e., its func-
tional form is unchanged by a spatial and/or temporal transformation), there
is a corresponding physical quantity that is conserved, when moving along the
orbits supported by the potential. For instance, energy is conserved for orbits
in potentials that are time-invariant, linear momentum is conserved when the
potential is invariant under a spatial translation and angular momentum is
conserved when we have rotational invariance.
Now, in Sect. 2 we introduced the concept of integrals of motion and
saw that the distribution function is a function of them. Clearly, the con-
served physical quantities in Noethers result are the integrals of motion of
the collisionless Boltzmann equation (in fact, energy, and linear and angular
momentum are the so-called classical integrals of motion). Since orbits are
the set of points in phase-space where the integrals keep a constant value, the
distribution function must depend on the orbital structure of the potential.
Indeed, for steady-state, collisionless systems, orbits are the basic bricks used
to assemble them in phase-space.
5.1 A Phase-Space Portrait of Orbits
Each integral of motion denes a hyper-surface in phase-space, and orbits
move along the intersection of all of them. A system with N degrees of freedom
has a 2N-dimensional phase-space. Each integral of motion lowers by one the
dimensions of the allowed region, and thus a system with M integrals has
orbits restricted to a region of dimension 2NM. Figure 13 shows this for the
Kepler potential. Since the direction of the angular momentum is xed, the
motion is restricted to a plane in conguration space and we only need 2 spatial
coordinates to describe the motion (say polar coordinates r, ). Since we can
Fig. 13. Two integral of motion surfaces in phase-space. The view is the same as
in Fig. 12, with the tangential velocity axis running vertically. The funnel is the
constant energy surface (drawn open for clarity). The folded plane is the constant
angular momentum surface. The orbit lies at the intersection (thick line)
not draw the 4-dimensional phase-space, one coordinate must be sacriced.
Since the potential is symmetric in , we drop it from the gure, knowing that
whatever we get, must be wrapped around in to obtain the full picture.
The intersection of the funnel-like energy surface and the bent plane angu-
lar momentum surface denes a loop in Fig. 13. It is clear that at xed energy,
if we increase angular momentum, the bent plane moves upward and the loop
shrinks to a point that corresponds to the maximum angular momentum orbit:
the circular orbit.
Figure 13 describes orbits in any spherical potential, since E and L are
always conserved in them (using a dierent potential only changes details).
To get the orbit in conguration space, we must introduce the missing
direction sacricing one of the velocity axes. This is shown in Fig. 14, where
the base plane is conguration space and the vertical axis is the radial velocity.
Wrapping the loop in results in a torus in phase-space: the invariant orbital
torus. Its projection in conguration space gives a rosette limited by two
circles whose radii are the periapsis and apoapsis.
3
It can be shown that all
regular orbits move on orbital torii in phase-space, although their form, and
corresponding projection, can be very complicated.
5.2 The Lindblad Diagram
Since energy and angular momentum dene the shape of orbits, in spherical
potentials, we can use them as labels to catalogue them. This is precisely the
idea behind a diagram rst used by B. Lindblad in 1933.
3
In Celestial Mechanics, the apsis of an orbit is the point of maximum or minimum
distance from the center of attraction. Periapsis is the minimum distance point
while apoapsis is the maximum distance point. In this and the next section, we
will also use these terms as proxies for the actual distance at these points, as no
generally accepted term exists for them.
92 L. A. Aguilar
Vr
Fig. 14. Invariant orbital torus. The base plane is the conguration space while
the vertical axis is the radial velocity. The orbit wraps around the torus while the
projection in ordinary space traces a rosette
A Lindblad diagram is like a chart that allows us to see in one glance,
the orbital make up of any spherical model. In this section we will illustrate
its use by building the Lindblad diagram of a simple spherical model with a
at rotation curve: the singular, truncated logarithmic potential. Its density
prole, potential function and circular velocity curve are given by,
(r) =
_
v
2
o
/4Gr
2
, r < r
t
0, r r
t
(28)
(r) =
_
v
2
o
[1 log(r/r
t
)], r < r
t
v
2
o
(r
t
/r), r r
t
(29)
v
2
c
(r) =
_
v
2
o
, r < r
t
v
2
o
(r
t
/r), r r
t
(30)
where v
o
is the constant circular velocity within the truncation radius r
t
.
Since at a xed energy, orbits can range in angular momentum from the
radial to the circular orbit (see 5.1) our rst job is to nd the locus of circular
orbits in the Lindblad Diagram, where E will be plotted on the horizontal
axis and L on the vertical one. All possible orbits will lie beneath this curve.
The energy and angular momentum of a circular orbit of radius r
c
is given by,
E
c
= (r
c
) +v
2
c
/2, L
c
= r
c
v
c
,
where v
c
is the local circular velocity. To nd E
c
vs. L
c
, we eliminate r
c
from
these equations and substitute the potential function. This gives,
E
c
=
_
v
2
o
[(1/2) log(L
c
/r
t
v
o
)], r < r
t
(1/2) (r
t
v
2
o
/L
c
)
2
, r r
t
It is convenient to have the inverse relation as well,
L
c
=
_
r
t
v
o
exp[(E
c
/v
2
o
) + (1/2)], r < r
t
(r
t
v
2
o
/2
_
[E
c
[), r r
t
.
We can now plot the locus of circular orbits in the Lindblad diagram
(Fig. 15). The axes are shown in dimensionless units. The lower horizontal
axis shows the radial position at which the potential function is equal to the
energy in the upper axis. The envelope of circular orbits goes from L
c
0 at
the innitely deep center of the potential well, to L
c
at innite distance
(E 0), denoted by the vertical dotted line.
We now introduce the concept of the characteristic parabola. Let us take
a spherical shell of radius r
o
and consider all orbits that touch it, but do not
cross it (Fig. 16). The condition for this to happen is (v
r
= 0 at r = r
o
),
E
= (r
o
) +v
2
t
/2 = (r
o
) + (L
2
/2r
2
o
) , (31)
where E
and L
are the energy and angular momentum of these orbits. This

is the equation of a parabola in the Lindblad diagram that opens up to the
right and crosses the E axis at (r
o
).
Figure 15 shows one such characteristic parabola. Point A is the radial
orbit that just reaches r = r
o
, before plunging back to the center. Point B is
the circular orbit at this radius and point C is the parabolic orbit that comes
3.0 2.0 1.0
0
0.5
1.5
1.0
(L/r
T
v
o
)
(E/v
2
)
(r/r
T
)
III
II
A
I
B
C
0
0.1 0.2 0.3 0.4 0.6 0.8
10
10
2 1
o
Fig. 15. Lindblad diagram for the singular, truncated, logarithmic potential. The
thick upper envelope is the locus of circular orbits. The thin lower curve that goes
through points A, B and C is a characteristic parabola
94 L. A. Aguilar
Fig. 16. Orbits in Kepler potential that lie along the characteristic parabola that
corresponds to the thick red circle
from innity and reaches the r
o
radius before going back to innity. Points
along the characteristic parabola between A and B are orbits that share r
o
as their apoapsis (green orbits in Fig. 16), while points between B and C are
orbits that share r
o
as their periapsis (blue orbits in same gure).
3.0 2.0
R
Apo
R
Peri
0.2 0.3 0.4 0.50.6 0.8 1.0 1.5 2 4 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1.25
1.5
2
1.0
0
0.5
1.5
1.0
(L/r
T
v
o
)
(r/r
T
)
0
0.1 0.2 0.3 0.4 0.6 0.8
10
10
2
1
(E/v
2

)
o
Fig. 17. Lindblad diagram with grid of iso-periapsis (blue) and iso-apoapsis (green)
extrema
A characteristic parabola splits the allowed region of the Lindblad diagram
in three regions (Fig. 15). Region I contains orbits enclosed completely by the
spherical shell of radius r
o
, while region III contains those that are always
outside this shell. Region II then is where the orbits that cross this shell lie.
If we draw characteristic parabola for several radial distances in the Lind-
blad diagram (Fig. 17), we produce a grid of iso-apoapsis with the segments of
the parabola between the E axis and the point where they touch the circular
orbit envelope, and iso-periapsis for the segments beyond to the E = 0 bound-
ary. Each point in the Lindblad diagram, below the circular orbit envelope
and to the left of the zero energy boundary, represents a unique bound orbit.
4
The two unique characteristic parabola that go through it, dene its radial
extrema. The Lindblad diagram thus provides us with a unique and complete
catalogue for all orbits in a spherical potential, arranged according to orbital
characteristics. From it, we can easily gure out the radial range of individual
orbits. Furthermore, if we actually have the form of the distribution function
as a function of E and L, we can compute from this diagram the fraction of
the model that shares some particular orbital characteristics (for this we need
to get from f(E, L) to N(E, L) in a way analogous to (27).
6 A Sticky Story: Dynamical Friction
When a massive object moves in a sea of background particles, the gravita-
tional force of the former stirs the latter; since the energy invested in producing
the stirring must come from somewhere, the impinging object losses kinetic
energy. This is a classic tale of pumping ordered motion energy into a thermal
bath.
5
A classic cartoon model depicts an overdense wake in the background,
trailing the moving object and decelerating it with its own gravity.
6.1 Chandrasekhars Formula
S. Chandrasekhar [19] thought about this in 1942 and came with a very famous
formula that describes the situation when a massive point of mass M, moves
along a straight path with velocity v, within a uniform background of non
self-interacting and uncorrelated point particles, of individual mass m and
density
f
, with zero mean motion and velocity distribution given by f(v
f
),
(dv/dt) = 16
2
G
2
ln()
f
(m +M)
_
v
0
f(v
f
)v
2
f
dv
f
(v/v
3
) . (32)
The is the so-called Coulomb term
6
and it is given by,
4
Except for the spatial orientation of the orbital plane, line of apses and sense of
rotation.
5
If Maxwells demon worked here, we would talk about dynamical acceleration!
6
The name comes from Plasma Physics.
96 L. A. Aguilar
= p
max
v
2
/G(m+M) , (33)
where p
max
is formally the maximum impact parameter.
It is important to take a second look at the paragraph that precedes (32).
There are a lot of caveats attached to Chandrasekhars result. We emphasize
this, because this equation has been used, and abused, under many dier-
ent physical situations that clearly invalidate those caveats, and yet, it has
been found to provide an adequate description on many dierent situations,
provided some details are taken into account.
There is a rich literature in this subject, we will just mention a few ex-
amples. Tremaine and Weinberg [51], have reviewed the derivation of the dy-
namical friction formula in the context of a satellite in a bound orbit within
a spherical system. The key question here is that, in contrast to the innite
background medium used by Chandrasekhar, background particles in reso-
nant orbits with the satellite play a very prominent role in slowing down the
satellite and can lead to unusual eects. If the satellites orbit decays rapidly
enough, however, Chandrasekhars formula remains approximately valid. Bon-
tekoe and Van Albada [14], as well as Zaristky and White [55], debated about
the global response that a satellite in a bound orbit produces in the halo of
the host galaxy and its consequences upon dynamical friction. They concluded
that the purely local description of the eect, as given by (32), is adequate.
Cora et al. [17] worried about the eect of chaotic orbits for the background
particles, and came to a similar conclusion.
Since dynamical friction is so important in the accretion process of a small
system into a larger one, we will examine it in some detail. There are important
issues here: How fast is the orbital decay? Does the orbit becomes circularized?
Can the accreting system make it all the way to the center of the host?
We begin with the Coulomb term. What exactly is p
max
? In deriving (32),
an integration over impact parameters must be performed. Unfortunately,
however, such an integration diverges at the upper end. Chandrasekhar [18]
argued that an upper cut-o should be used, since distant collisions are not
isolated binary encounters, as modeled in his derivation. He used the mean
particle separation, while other authors have advocated the distance where
the average density signicantly drops, some others, the radial distance of
the spiraling system, and yet others have used the size of the host system.
What should one do? Fortunately enters into the dynamical friction for-
mula through its logarithm and this waters down our ignorance about it. In
fact, a fractional error in , translates into a fractional error / ln() in the
computed deceleration. Let us compute a specic example: (33) in astronom-
ical units can be written as,
= 2.32 (p
max
/kpc) (v/10
2
km/s)
2
/ (M/10
9
M
) .
For a 10
9
M
satellite moving at 300 km/s, the logarithm of the Coulomb

factor varies only from 5.3 to 7.6, if we take 10 or 100 kpc for p
max
, respectively.
So you can plug your favorite value for p
max
and compute the dynamical
friction deceleration without worrying much for (some authors have turned
the problem around and measured from N-body simulations [14]).
The next thing to notice in (32) is the
f
(m+M) dependence. Although
both, the mass of the object being slowed down and that of the individual
background particles appear, it is usually the case the the former is much
bigger than the latter. In this limit, we can forget about m and just say that
the deceleration is directly proportional to the mass of the massive object and
the background density. If a globular cluster decays in the galactic halo, it
really does not matter whether the halo is made up of subatomic particles,
brown dwarfs, or stellar remnants.
The nal actor, in the dynamical friction play, is the integral over the
velocity distribution of the background particles. There are a couple of things
that must be noted from the outset: Firstly, dynamical friction tries to bring
massive objects to a stand-still with respect to the velocity centroid of the
background particles. This can have important consequences. Secondly, the
integral limit only goes up to the massive object speed: It is only the slower
background particles that contribute to dynamical friction (faster particles can
overtake the massive object and do not necessarily contribute to the overdense
wake).
The shape of the velocity distribution function depends on the details of
the background model, however, the overall shape is pretty much like that of
a Gaussian distribution: it peaks at the origin and then drops smoothly for
increasing velocity. Assuming a Gaussian of dispersion , we get Fig. 18 for
the velocity part of (32). And here is where the split personality of dynamical
friction arises: in the low velocity regime (v ), f is approximately constant
and the integral goes as v
3
; plugging this into Chandrasekhars formula we
get a deceleration that goes as v, just like in Stokes formula for the drag on a
solid object moving inside a viscous uid. However, in the high velocity regime
(v ), f drops precipitously and the integral is approximately constant.
This leads to a deceleration that goes as 1/v
2
, opposite to your run-of-the-mill
0 1 2 3 4 5
v/
0
0.05
0.1
0.15
0.2
(
1
/
v
2
)
v
2
f
(
v
)
d
v
Fig. 18. Velocity dependence of dynamical friction assuming a Gaussian distribution
of velocity dispersion for the background particles
98 L. A. Aguilar
frictional force. So dynamical friction has bipolar disorder, which makes it
work at low speeds and it becomes more ecient the faster you try to go; but
if you are going too fast, the faster you go, the less ecient it becomes.
7
This
is a characteristic that is important to consider, as we examine now.
6.2 Dynamical Friction in an NFW Prole
An elongated orbit goes fastest near its periapsis and slowest near its apoapsis.
Indeed, it needs to go faster that the local circular velocity at periapsis to
climb up from that point. Similarly, it goes slow with respect to the local
circular velocity at apoapsis, to fall down from this point. If the whole span
in velocity of an elongated orbit is such that it remains to the left of the
peak in Fig. 18, dynamical friction will be more ecient at periapsis and the
orbital eccentricity will increase. If on the contrary, it remains to the right,
the eccentricity will decrease. Orbits that straddle the peak can go either way.
Let us examine the situation for the NFW prole that we explored in
Sect. 3. Figure 19 shows the escape and circular velocities, normalized to the
velocity dispersion for the isotropic model. If a satellite spirals, following an
orbit close to circular, it will move from right to left along the circular velocity
curve. We can see that as its orbit shrinks from far away, dynamical friction
behaves inversely with velocity and its eciency increases as v
c
/ steadily
approaches peak eciency at 0.771...; past this point, dynamical friction
enters the Stokes regime, but its eciency shrinks to zero as v
c
/ goes to
zero at the center.
If we now have a satellite that plunges to its periapsis along a parabolic
orbit, it will follow the escape velocity line in Fig. 19. In this case, dynamical
friction will remain in the regime that scales inversely with velocity. Since v/
0.1 0.2 0.5 1 2 5 10
1
2
3
4
5
6
V
esc
/, V
c
/
Fig. 19. Escape (solid line) and circular (long dashed line) velocities, normalized
to the local isotropic velocity dispersion, for a spherical NFW model. The constant
short dashed line at v/ = 1.3688, is where the peak of Fig. 18 occurs
7
If you are going very fast, the overdense wake trails far behind. The faster you
go, the more distant it becomes and thus its retarding eect diminishes.
is very high at all times, dynamical friction is very inecient. Furthermore, the
deeper the satellite penetrates on its rst pass, the less drag it will experience.
A satellite on a bound, but elongated orbit, will move between the escape
and the circular velocity curves in Fig. 19. For > 0.771, dynamical friction
will work in the inverse velocity regime, where orbits are circularized. Past
this point, the situation gets murky.
There is, however, a big caveat we must make before leaving this subject. In
this section we have only considered the eect of velocity in Chandrasekhars
dynamical friction formula. The variation of the other parameters that enter
must be accounted for, before any conclusion is reached. The variation in
background density, in particular, is important. We also have that as the
satellite moves in, it gets peeled layer by layer by the tidal force and so its
mass also changes. We will say more about this latter eect in the next section.
Before leaving the subject of dynamical friction, we want to stress the very
important role that (32) has had, and continues to have, in our understanding
of stellar and galactic dynamics. The original Chandrasekhar article [19] was
chosen to be part of a collection of fundamental papers published during the
XX century, and reprinted as the centennial volume of ApJ [1]. We encour-
age the reader to take a look at F. Shus commentary on Chandrasekhars
contribution [45].
7 The Eect of Tides
We will now take a look at a subject very familiar to sailors and careless
galaxies that venture too close to others: the tidal force. The idea behind it,
is very simple: an extended object moving within an external gravitational
eld, will experience a changing force across its body. If we sit at its center of
mass, we will see a dierential force acting on the rest of the body, and it is
up to the self-gravity of the object to keep itself together. From this cartoon
model it is apparent that the tidal force depends on two factors: the rate of
change of the underlying force and the size of the object:
8
F
tid
(dF/dr)r.
In a Kepler potential the force decays with distance, this produces tidal
stretching along the radial direction. It is usually assumed that tides always
produce radial stretching, however, it all depends on the sign of the force
gradient. Let us assume a spherical mass model with a power-law density
prole: r
. In this case, the enclosed mass within radius r goes as M

r

r
3+
. The gravitational force then goes as F M
r
/r
2
r
1+
. It is clear that
for < 1 we have the usual radial stretching. However, > 1 produces
radial squeezing and = 1 produces no radial deformation!
8
The gravitational force is the vector eld that results from applying the /x
i
operator to the potential scalar eld. The tidal force is described by the 2nd rank
tensor that results from applying the
2
/x
i
x
j
operator to the potential. This
results in a force that may produce distortions in all directions. For simplicity,
we focus here on the radial direction only.
100 L. A. Aguilar
Can this happen in astronomical objects? Well, a constant density core, like
that of a King model, produces radial squeezing (can this help star formation
at the center of at-core galaxies?). The NFW spherical model has a cusp
( = 1) that produces no radial deformation, while its outer part ( = 3)
produces the usual radial stretching.
7.1 The Tidal Radius
An important concept is the so called tidal radius, usually dened as the radius
beyond which, a test particle in orbit around a satellite becomes unbound to it
and ies apart within the gravitational force eld of the larger system. Direct
evidence for the existence of such radius is provided by the sharp cut-o in
the density proles of globular clusters in our galaxy and of satellite galaxies,
like M32, trapped deep within the potential well of a larger system.
How do we compute the tidal radius? One possibility is to compute the
point at which the tidal force and the self-gravity of the object are equal. If
we assume point masses for the tide-producing object (M) and tide-distorted
object (m), we get (Fig. 20):
F
g
=
Gm
r
2
F
t
=
d
dR
_
GM
R
2
_
r =
_
GM
R
3
_
r
_
F
t
= F
g
= 2 (r
R
/R)
3
= m/M . (34)
r
R
is called the Roche limit
9
and is the largest radius that a self-gravitating
object can have when immersed within an external gravitational force.
Now, is this the tidal radius? We said that it denes the region beyond
which it is not possible to have bound orbits around m; isnt this r
R
?, well
no. The Roche limit is a stationary concept, no information whatsoever about
the motion of M and m is used in its derivation.
If we consider m and M to be in circular orbit around their center of
mass, besides their combined forces, we must include a centrifugal term that
arises in the co-rotating frame. The energy dened in the this frame, using
the combined potential functions of both objects, together with that of the
centrifugal term, is conserved. This is the Jacobi energy. The combined po-
tential produces a series of spherical equipotentials that surround each object
M
m R r
Fig. 20. Geometry for computation of tidal radius
9
First derived by

Eduard Roche in 1848.
Fig. 21. Roche lobes for two point masses in circular orbit
close to them. As we move away, the equipotentials become elongated until we
nally arrive at the particular equipotential that goes through a saddle point,
where both lobes touch each other
10
(Fig. 21). Any higher energy equipo-
tential surrounds both masses. Since this particular equipotential is the least
bound that encloses separately either object, no particle with Jacobi energy
less than it can escape.
The distance from either point mass to the saddle point is its so-called
Roche lobe radius (it is also called the Hill radius, since George Hill discovered
it independently of Roche). Its magnitude is given by,
3 (r
H
/R)
3
= m/M . (35)
Comparing with (34), we see that the Roche limit and Hill radius have exactly
the same functional form and dier only by a factor of 2/3. The important
thing to consider is that both can be cast as an argument that relates densities:
the average density of M, within the orbit of m, and the average density of
m within its limiting radius:
11
M
(R))
m
(r
R,H
)).
Which one should we use? Well, the Hill radius has additional dynamical
information and we may be tempted to use it as a bona de tidal radius.
Unfortunately, there are additional complications to consider. First of all, the
criteria that dene the Roche limit and the Hill radius lead to non-spherical
regions, so strictly speaking, these radii are direction dependent additionally,
the dynamics in the co-rotating frame is complicated by the appearance of
the Coriolis term, which can not be expressed as the gradient of a potential
function. As a result of this, an orbit whose Jacobi energy exceeds that of the
Roche lobe does not necessarily escapes. In fact, the region just beyond the
Roche Lobe becomes so complex, that some authors have characterized it as
fractal [39] (see Fig. 2 in this reference for an example of a hovering orbit
that ends up escaping). In fact, it was a study of the three-body problem,
that lead H. Poincare to discover what we now call chaos.
10
This is the L
1
Lagrange point. In the combined potential of a two body system
there are ve points where the forces cancel. These are the Lagrangian points
rst determined by Lagrange in 1772.
11
Although we are considering point masses, we dene an average density within r
simply as m/r
3
.
102 L. A. Aguilar
7.2 The Coriolis Eect
The eect of the Coriolis term gives rise to a distinction between prograde
and retrograde orbits for particles orbiting the satellite. The distinction is
dened by the relative orientation of the orbital angular momentum of the
satellite and the particle. If they are parallel, it is a prograde orbit, if they are
anti-parallel, it is a retrograde orbit (see Fig. 22).
Now, the Coriolis acceleration is of the form a
c
(v), where is the
satellites angular velocity and v is the instantaneous velocity of the particle,
in the co-rotating frame. It is easy to see, using your right hand, that this
acceleration opposes the self-gravity of the satellite for prograde orbits, while
it reinforces it for retrograde orbits.
This eect makes prograde orbits more fragile and leads to net retrograde
rotation in the outer parts of a tidally pruned satellite [34], from which it
becomes necessary to dene separate tidal radii for prograde and retrograde
orbits [44].
7.3 Non-Circular Orbits and Extended Mass Distributions
Even if we take into account all that we said before, our job is not nished. As
soon as the satellite moves in a non-circular orbit, the problem becomes time-
dependent and away goes conservation of energy for the orbiting particle. The
so-called elliptical restricted three-body problem, where a test particle moves
in the same plane of two point masses in an elliptical orbit around each other,
is very rich in history and has been investigated very thoroughly [49]. Although
the situation may seem hopeless, in reality it is not. It turns out that, except
at the very edge of the system, lingering orbits that end up escaping, occupy a
vanishingly small volume in phase-space and, although a bit fuzzy, a boundary
between bound and unbound orbits can be dened. A working denition of
the tidal radius can be the instantaneous Hill radius, or its value at peripasis,
where the tidal eect is largest and mass loss peaks [38]. A tidal radius valid
for point masses in an elliptical orbit around each other is [32],
r
t
R
Peri
=
2
3
_
m
(3 +e)M
_
1/3
, (36)
Fig. 22. Prograde and retrograde orbits around a satellite
where R
Peri
is the distance between M and m at periapsis and e is the orbital
eccentricity. Comparing with (35), we see that in the limit of circular orbits,
r
t
is a factor 2/3 smaller that r
H
. This is because our calculation of r
H
was
done along the radial direction, whereas r
t
represents an average value.
Another complication is that in our discussion of the tidal radius we have
assumed point masses, and this is not the case when a globular cluster, or satel-
lite galaxy, move inside the halo of a large galaxy. Fortunately, this problem is
easier to remedy. The potential produced by the extended mass distributions
should be used when calculating the tidal radius. An expression valid for a
point mass satellite, moving within a at rotation curve galaxy is [32],
r
t
R
Peri
=
2
3
_
1 ln
_
2R
Peri
R
Peri
+R
Apo
__
1/3
_
m
2M
P
_
1/3
. (37)
R
Apo
is the separation between the centers of the galaxy and the satellite at
apoapsis and M
P
is the galaxys mass enclosed within R
Peri
. An even more
general formula, valid for galaxies with power-law density proles, can be
derived [44].
Before leaving this section, we mention that for very elongated orbits, the
tidal force varies abruptly near periapsis. Besides the mass shed due to the
much diminished tidal radius, a new eect appears: tidal shocking. We will
examine this phenomenon in the next section.
8 Tidal Encounters
Anyone who has been unfortunate enough to have suered a collision in the
highway will tell you that it is a shocking experience. It is the same for galaxies.
However, a key dierence between collisions in the highway and collisions in
your galactic backyard, is that in the former, we have physical contact and the
larger the relative velocity, the bigger the damage. In the astronomical case, we
have long range interactions, where the larger the relative velocity, the shorter
the interaction time and less damage will occur. This is an important lesson:
if you are a galaxy, zip by very fast when traversing crowded environments,
like rich clusters of galaxies, if you want to survive the experience.
Both, dynamical friction and the tidal force dissipate encounter orbital
energy and dump it within the thermal reservoir of internal motions of the
hapless galaxies. The action of the tidal force, in this context, is a bit dierent
from what we saw in the last section. In that case, we examined the static
eect of the tidal force, which basically leads to the imposition of a boundary.
When we have an encounter, the time-varying part of the tidal force heats up
the interacting systems (this is often called tidal shocking [41]). In any case,
at this point the distinction between these forces becomes a semantic issue
and we must look for a more holistic description.
A gravitational encounter is a very complicated phenomenon [8]. It is il-
luminating just listing the reasons for this. Imagine a perturber passing close
104 L. A. Aguilar
to a target system (it doesnt matter which one is which). As the perturber
passes nearby, its time-varying gravity perturbs the motion of the stars in the
target, which in turn perturb the perturber in a manner dierent to what
you would have guessed, if you had not taken into account this back reaction.
Since the situation is symmetric, the same goes when you place yourself in the
other galaxy. It is this coupling of perturbations that makes any analytical
treatment hopeless (except for very specic settings) and we must turn to
N-body simulations.
With simulations the problem is availability of computer time (as always!).
Even with the impressive advances in computational technology, predicated
by Moores law, we lack computer cycles to tackle some basic issues in a
comprehensive manner that may give us a proper statistical description. You
may have seen a lot of computer simulations of galaxy collisions out there, but
do you know what is the range of parameters that lead to mergers when two
galaxies interact (the capture cross section)? If you search in the literature
you will nd that a lot of parameter space remains to be explored, particularly
in the intermediate mass ratio regime [15]. The basic problem is the vastness
of the parameter space that uniquely denes a galactic encounter [7], and so
it is dicult to undertake a proper exploration that could give us a good
statistical description of the problem.
12
So, fortunately for us, a lot of work
remains to be done.
8.1 A Simplied Picture: The Impulse Approximation
Even if a full N-body experiment may be needed to study galactic encounters,
some basic physical insight can be gained by using an approximation, with
a rich history in stellar and galactic dynamics since it was introduced: the
impulse approximation [47].
In this approximation, we assume that a perturber of mass M
p
moves at
constant speed v
col
along a straight path, so fast, that we can neglect the
motion of stars in the target system. Figure 23 gives the basic geometry.
The impulse in the velocity of the target star is obtained by integrating
the component of the gravitational force of the perturber that is orthogonal
to the path, along the entire path:
v
=
_

dt = 2
_

o
GM
P
r
2
p
r
dt .
Simple geometry allows us to compute this integral,
v
=
_
2GM
P
pv
col
_
. (38)
12
The dynamics of clusters of galaxies may benet from the approach used to
understand the dynamics of globular clusters, where all the complicated dynamics
of binary scattering was synthesized in a few interaction cross sections [30].
M
p
p
r
v
col
Fig. 23. Encounter geometry for impulse approximation
An easy way to remember this result, and give physical meaning to it, is
splitting it as the product of two factors, whose physical units are acceleration
and time, respectively:
v
=
_
GM
P
p
2
_ _
2p
v
col
_
. (39)
The rst factor is just the acceleration at the point of closest distance,
the second is the time it takes the perturber to travel a distance of twice the
impact parameter (Fig. 24), we will call it, the interaction time. So, here you
have it: the longer the interaction time, the larger the impulse.
There is, however, an undesired quirk in (39): what happens in the case
of a head-on collision (p = 0)? The undesired behavior arises because we
assumed a point mass for the perturber. This is easy to solve, and doing a
proper calculation for an extended mass, spherical perturber, gives the same
result, but multiplied by a correction factor given by [3],
f(p) =
_

1
P
(p)
2
_
2
1
d , (40)
where
P
(r) is the mass fraction of the perturber within a radius r. It is
easy to see that f has the right asymptotic behavior (approaches zero for an
innitely spread perturber, and goes to unity for a point mass).
Now, to estimate the tidal heating we must obtain the change in energy
suered by the star within the target. Since the motion of the star is neglected
in the impulse approximation, the only change is due to v
,
E
= (m
v
2
/2) = m
(v
) +m
v
2
/2
1 1
t/t
col
a a
max
Fig. 24. Impulse acceleration as a function of time (thick line). The equivalent step
function (thin line), has the same area under it
106 L. A. Aguilar
The rst term averages to zero when integrating over all stars in the tar-
get.
13
so we are left with the quadratic term. Integrating it, we get,
E =
1
3
_
2GM
P
f(p)
p
2
v
col
_
2
r
2
) , (41)
where r
2
) is the mean of the squared radial distance of all stars in the target.
8.2 A Tale of Two Timescales: Adiabatic and Impulsive Regimes
Spitzer did not introduce (41), he went a bit beyond. He knew that the motion
of stars in the target system should be considered, even if very crudely. He
modeled these stars as harmonic oscillators and casted the encounter as a
standard problem in Classical Mechanics: the forced harmonic oscillator [48].
He found that the eciency for energy transfer to the oscillator varies
dramatically as a function of an adiabatic parameter given by,
=
2p
v
col
_
GM
s
r
3
_
1/2
, (42)
where M
s
is the target mass. This parameter is proportional to the ratio of
the interaction time to the oscillator period (local orbital time). And here
is where our tale of two timescales begins: When < 1, the encounter is
fast with respect to the natural period of the oscillator, while for > 1, the
encounter is slow. Spitzer found that the energy transfer eciency could be
described by a multiplicative factor that decays exponentially with [48].
Weinberg has reviewed this problem, removing the 1-dimensional treatment
used by Spitzer and taking into account the eect of resonances. He found
a function that decays as a power-law [52]. Both corrections approach unity
for 0, this is the impulsive regime that we saw in Sect. 8.1. For >
1, both correction factors shrink rather quickly, this is the adiabatic regime
(Fig. 25).
Equation (41) should then be multiplied by a net shock eciency, obtained
as a mass-weighted average of over the whole target [5],
=
4
r
2
)M
s
_
r
t
0
s
(r)()r
4
dr (43)
Why is that the amount of energy transfer diminishes so rapidly as we
enter the adiabatic regime? This has to do with adiabatic invariants, a subject
discussed by Einstein at the rst Solvay Conference in 1911. The basic idea is
that when a dynamical system undergoes a change that is slow with respect
to its internal timescale, there are properties of the motion that will remain
invariant, these are the actions (see [25], Sect. 12.5). Here is a cartoon answer:
13
For this argument to work, we must include escaping stars.
0.1 0.5 1 5 10 50 100
0.0001
0.001
0.01
0.1
1
Fig. 25. Adiabatic correction of Spitzer (solid line) and Weinberg (dashed line) [24]
when a perturber encounters a slow star, the star sweeps only a fraction of
its orbit while the encounter takes place, which can then be distorted. On the
other hand, a fast star covers many periods and so it appears to the perturber,
not as a point mass, but as a mass spread over the whole orbit. The perturber
then can only push the orbit as a whole (Fig. 26).
This eect can clearly be seen in Fig. 27, which shows an old N-body sim-
ulation of the eect of a hyperbolic encounter on a de Vaucouleurs galaxy. The
perturber was launched directly across the frame but is deected downward
during the encounter. The halo of the target reacts impulsively and absorbs
energy that leads to its heating up and expanding without any appreciably
overall displacement. The inner core of the target, on the contrary, hardly
expands but is displaced as a whole from its original position, leaving behind
the halo that becomes unbound. This is called tidal stripping.
At the end, we must get a copy of an N-body code and run our own
simulations. However, a close examination of the impulse approximation and
Fig. 26. The same encounter can be impulsive for a star in the outskirts of a target
galaxy and adiabatic for one in its central region. In the rst case, the orbit of the
star gets distorted, in the second, it gets pushed as a whole
108 L. A. Aguilar
a b c
d e f
Fig. 27. N-body simulation of the hyperbolic encounter of two de Vaucouleur galax-
ies. The perturber is represented by the lled circle of radius equal to its eective
radius. The frames are set 2 crossing times (at the eective radius) apart [3]
knowledge about what to expect in the impulsive and adiabatic regimes, can
help us understand the results we obtain.
14
9 Putting Things Together: The Orbital Decay
of a Satellite within an Extended Halo
In this section we put together a bit of what we have learned about dynamical
friction and tidal truncation, to follow a satellite whose orbit decays as it
moves within an extended halo. This problem has been treated by Binney and
Tremaine [12] (see their Sect. 7.1.1.a) for the case of a point mass satellite of
xed mass, that spirals down a singular isothermal halo. A similar calculation
has also been made elsewhere for a constant density halo [27]. The treatment
we present here is inspired by that of Binney and Tremaine, but with the
distinction that we allow for an extended satellite and describe the shrinking
tidal radius, as it plunges into deeper layers of the host halo. Ours is a simple
model that wraps up what we have learned in Sects. 6 and 7.
9.1 The Cast of Characters
We need a model for the satellite and the host halo. Due to its simplicity, we
choose a Plummer [43] model for the satellite:
s
(x) =
_
3M
s
4r
3
o
_
(1 +x
2
)
5/2
, (44)
14
Computers are useless, they can only give you answers. Pablo Picasso.
s
(x) =
GM
s
r
o
(1 +x
2
)
1/2
, (45)
/
s
(x) = M
s
x
3
(1 +x
2
)
3/2
, (46)
Here
s
,
s
and /
s
are the density prole, potential function and cumulative
mass. M
s
is the total initial mass of the satellite and x r/r
o
, where r
o
is
the nuclear radius (it contains 35% of the mass). This is a model with an
approximately at density core within r
o
. As we will soon see, this is a very
important feature.
For the host halo, we use a mass model that results in a at rotation curve
(see (28), (29) and (30)). This is the same halo used by Binney and Tremaine.
Its cumulative mass, within a galacto-centric distance R, is given by:
M
R
= (v
2
o
/G) R , (47)
with v
o
being the constant circular velocity of the halo. Notice that both of
our models are spherical. We will use r to denote radial distances within the
satellite and R for the radial position of the satellite within the halo.
9.2 Tidal Truncation
We can now use the tidal truncation condition to nd the tidal radius of
the satellite r
t
as a function of position within the halo. We will make the
simplifying assumption that, as the satellite travels inward within the halo
and its tidal radius shrinks, the satellite inside r
t
remains unaected and it is
just the layers outside it, that are lost.
We saw in Sect. 7 that the tidal radius can be written as a condition that
relates the average density of the satellite inside its tidal radius, with the
average density of the underlying halo within the satellite orbit:
r
t
/R = [/
s
(r
t
)/M
R
]
1/3
, (48)
where = 2
2/3
/3 0.529, for the tidal radius imposed by a singular isother-
mal halo (37).
15
We will assume that the satellite decays following a spiraling
orbit very close to circular, and so we only consider the case (e = 0).
Substituting the mass prole of the satellite and halo (46) and (47) in the
tidal radius condition, we get:
r
t
R
=
_
GM
s
Rv
2
o
_
1/3
x
t
_
1 +x
2
t
. (49)
The left hand side term can be written as,
r
t
R
=
r
t
/r
o
R/r
o
=
x
t
R/r
o
,
15
This equation assumes a point mass satellite. We will have to live with this.
110 L. A. Aguilar
where x
t
r
t
/r
o
is the dimensionless tidal radius.
Substituting this back in (49), we get,
x
t
=
R
r
o
_
GM
s
Rv
2
o
_
1/3
x
t
_
1 +x
2
t
.
This can be easily manipulated to yield,
x
2
t
=
2
_
GM
s
R
2
r
3
o
v
2
o
_
2/3
1 . (50)
A bit more algebra leads to r
t
,
r
t
=
2
_
GM
2
s
v
2
o
_
2/3
r
2
o
. (51)
We have thus obtained the satellite size as a function of its position within
the halo. Notice that the size shrinks to zero at a particular position R
t
,
which occurs when the radical vanishes. This condition can be cast as a tidal
condition:
r
o
/R
t
= [M
s
/M
R
t
]
1/3
, (52)
comparing with (48), we see that a Plummer model, dropped within a at
rotation curve halo, will survive until its Roche lobe size (computed using
its original mass) is equal to its core radius. What is behind the existence
of a terminal galacto-centric position? As we mentioned at the beginning of
this section, the tidal radius criterion can be interpreted as a condition that
relates densities: If the satellite does not have a central density higher than
the corresponding value for the halo, it will be destroyed before reaching the
center of the halo. The density of the Plummer model does not vary much
within its core radius. Once the Roche lobe reaches it, the satellite will be
gone very soon.
The terminal galacto-centric position can be obtained from the previous
equation. It is equal to,
R
t
= v
o
r
3
o
3
GM
s
. (53)
Finally, the remaining satellite mass can be computed using (50) and (46) to
evaluate /
s
(x
t
).
In astronomical units, appropriate for our problem, (51) and (53) are,
(r
r
/pc) =
1, 595
_
(M
s
/10
6
M
)(R/kpc)
2
(v
o
/10
2
kms
1
)
2
_
2/3
_
r
0
pc
_
2
, (54)
(R
t
/kpc) = 5.868 10
4
(v
o
/10
2
kms
1
)
(r
o
/pc)
3
(M
s
/10
6
M
)
. (55)
100
10
1
10
4
10
5
Pal 4
10
6
10
8
10
7
10
9
1

k
p
c
1
0
0

p
c
1
0

p
c
r
o

(
p
c
)
0.1
M
s
(M )
Fig. 28. Terminal galactocentric distance for Plummer models as a function of core
radius r
o
and satellite mass. Three lines of constant terminal distance are shown.
The dots correspond to globular clusters in our Galaxy. Clusters that lie on the
lower horizontal axis are core-collapsed clusters
We present in Fig. 28 the terminal galacto-centric distance as a function
of satellite mass and core radius. The halo has a characteristic velocity of
v
o
= 220 km s
1
. Data for globular clusters in our galaxy are shown too [23].
It is clear from this gure that the vast majority of clusters are dense enough
at their centers, to survive all the way to the central region of our Galaxy, the
exception being a group of three, or maybe ve clusters, of which the most
extreme example is Palomar 4.
16
In Fig. 29 we show the cluster tidal radius as a function of galactocentric
position, for satellites of mass 10
6
and 10
9
M
, and in each case, for core

radii of 1, 10 and 100 pc. It is clear that all clusters of a given mass shrink
steadily in size as R
2/3
until they reach close to their terminal galactocen-
tric distance, at which point, they are very quickly destroyed, this terminal
distance depending on the core radius: the smaller it is, the deeper the satellite
survives.
Finally, Fig. 30 shows the satellite mass as a function of galactocentric
position. The same three core radii of Fig. 29 are used for a satellite mass of
10
6
M
. We note that the r

o
= 1 and 10 pc curves coincide with those of a
10
9
M
satellite with r
o
= 10 and 100 pc, respectively.
Given the form of the cumulative mass distribution of the Plummer model,
it is clear that our satellites loose little mass until they approach their terminal
position, at which point they are quickly destroyed. It is then a reasonable
16
We must remember that we have modeled the dark halo only. The bulge and
galactic disk have not been taken into account.
112 L. A. Aguilar
1000
100
10
0.01 0.1 1 10 100
R (kpc)
1 pc
100 pc
10 pc
r
c
1
0
9
M
1
0
6
M
r
t

(
p
c
)
1
Fig. 29. Satellite tidal radius as a function of galactocentric position. Two sets of
curves are shown for satellite masses of 10
6
and 10
9
M
. In each set we show the

results for three core radii (1, 10 and 100 pc). Notice the abrupt destruction when
the satellite reaches its terminal galactocentric distance
approximation to assume that the satellites evolve at xed mass. We will use
this approximation in the next section.
9.3 Orbital Decay
We now compute the rate of orbital decay for the satellites. This decay is
driven by dynamical friction, which produces a force that opposes the motion
of the satellite (32). To apply this formula we need the halo density, which is
1
0.1
0.01 0.1 1 10 100
R (kpc)
1 pc
100 pc
10 pc
r
c
M
s
(
t
)

/

M
s

(
t
i
)
0.01
Fig. 30. Fraction of satellite mass as a function of galactocentric position. The same
core radii as in Fig. 29 are shown, for a 10
6
M
satellite
given by (28), and the velocity distribution at each radial distance, here using
a Maxwellian (see Fig. 18). These assumptions are identical to the model of
Binney and Tremaine, so we refer to their (7.23) for the retarding force:
F
DF
0.428 ln()
GM
s
2
R
2
,
where is the Coulomb term (33) and R is the galactocentric position of the
satellite.
This force produces a torque = F
DF
R, for our case of a circular orbit.
The change in orbital angular momentum is then,
d
dt
(RM
s
v
o
) = 0.428 ln()
GM
2
s
R
.
Strictly speaking, we should now take into account the variation in M
s
due to tidal truncation that we saw in Sect. 9.2. This, however, would greatly
complicate our model, which we want to keep simple. Furthermore, as we saw
in Fig. 30, assuming a constant mass for the satellite is not too bad, so we
will do so. Solving the resulting equation we get the galactocentric position
of the satellite, as a function of time,
R(t) =
_
R
2
i
0.856 ln()
GM
s
v
o
t . (56)
In astronomical units, this can be written as:
R(t)
kpc
=
_
R
i
kpc
_
2
0.0171 ln()
_
M
s
10
6
M
__
v
o
220 kms
1
_
1
_
t
Gyrs
_
.
(57)
0 2 4 6 8 10
t
Gyrs
2
4
6
8
10
12
14
16
R
kpc
Fig. 31. Satellite galactocentric position as a function of time. The satellite masses
are 10
7
(solid lines) and 10
8
M
(dashed lines). Initial positions of 2, 4, 8 and 16

kpc, have been considered
114 L. A. Aguilar
Fig. 31 shows the evolution of 10
7
and 10
8
M
satellites launched from several

initial galactocentric distances.
Our model is very simple but fullls our pedagogical goal of introducing
the main actors that play a role in bringing down and trimming the unfor-
tunate accreting satellites. Given the observational evidence and prevailing
cosmological scenario, this is an important problem to study. In particular,
a proper coupling of mass loss due to tidal truncation and dynamical fric-
tion, must be modeled carefully. We refer the interested reader to other, more
sophisticated models, that have been introduced recently [50, 56].
10 Further Reading
For those interested in digging up some more, here is a list of recommended
readings. The literature is very rich and we have chosen just a few references
to get somebody interested started.
For Sects. 3 and 4 we recommend the following:
Fully self-consistent anisotropic models in phase-space have been obtained
by [6] for the Hernquist prole and by [37] for the Jae prole.
A very general class of models that includes the Jae and Hernquist mod-
els, as special cases, is presented by [20]. Most of these models can be
expressed as combinations of elementary functions.
Many properties of spherical models with NFW proles, including phase-
space distribution functions with various kinematics is given by [36].
A class of self-consistent models for disk galaxies that consist of an NFW
dark halo, an exponential disk, a Hernquist bulge and even a massive
central black hole, has been introduced by [54].
For Sect. 5, we suggest the following references,
An approximation to orbits in the Hernquist model based in epicycloids
has been obtained by [2]
Sections 1 and 2 of [16] contain a good introduction to orbits, integrals of
motion and orbital torii.
A good discussion of the Lindblad diagram and its application to generate
dynamical models of various degrees of eccentricities for their orbits can
be found in [37].
For Sect. 6 there is a very extensive literature. Our list of suggested refer-
ences is as follows,
For an illuminating discussion about the subtleties and perils involved in
simulating dynamical friction with N-body codes, and whether the global
response of the background invalidates the classical Chandrasekhar for-
mula, take a look at [14, 53] and [55].
An examination of the eect on dynamical friction of velocity anisotropy
is found in [9] and [42].
The eect of chaotic orbits on dynamical friction is studied by [17].
An investigation of whether dynamical friction produces orbit circulariza-
tion can be found in [26].
For Sect. 7 there is also a very extensive literature. There is not, un-
fortunately, a good reference for the calculation of tidal forces produced by
extended objects.
Descriptions of the tidal force and computation of the tidal radius can be
found in Chap. 5 of [48] and Sect. 7.3 of [12].
A basic introduction to the three-body problem can be found in almost
any book on Celestial Mechanics [13]. For a lighter description check [25].
To gain the historical perspective in the three-body problem look at [22].
This reference presents a very nice account of Poincares musings about
the three-body problem and the discovery of chaos.
A gallery of orbits that illustrates the intricacies of the three-body problem
can be found in Fig. 20.7 of [31]
A description of the three-body problem and the tidal force, although in
the context of planetary dynamics, can be found in Sects. 2.2 and 2.6,
respectively, of [21]
An illustrative reference that plots the magnitude of the Coriolis, centrifu-
gal and gravitational terms in the co-rotating frame of reference for the
restricted three-body problem is [34]
For an up-to-date re-examination of the tidal radius take a look at [44]
For Sect. 8, we suggest the following,
The rst N-body simulation of an encounter between galaxies was done
by E. Holmberg in 1941. It is interesting to see how this author was able
to do the simulation at a time when no computers were available to him
[29].
A review of the eld of N-body simulations of galactic encounters can be
found in [8]
A detailed description of the impulse approximation together with Spitzer
correction for adiabaticity can be found in Sect. 5.2 of [48]
An example of the use of the impulse approximation, with Weinberg cor-
rection for adiabaticity for the problem of globular cluster survival in our
Galaxy, can be found in [24]
For Sect. 9, we recommend these references,
Our model for orbital decay has followed [12]. A similar model, but for a
at density halo motivated by dwarf galaxies, is given by [27]
More sophisticated, but still analytical models of the orbital decay of a
satellite, have been presented by [50] and [56]
116 L. A. Aguilar
Finally, [12] is the standard graduate level reference for Galactic dynamics.
It is very thorough and it may contain far more than what you may need for
just a quick computation, but if you really want to dig up what is behind the
topics we have been discussing, you will nd it here.
11 Some Final Words
It has been my job to give you a theoretical view of clusters of galaxies, or
rather a glimpse. I guess that if we stretch the electromagnetic spectrum a
bit, my talk ts within the title of this school: A Panchromatic View . . . .
The organizers certainly think so, and asked me to do this job. We live in
the golden era of Astronomy. I am sure that you have heard this so many
times that by now it may sound like a cliche, however, it is true. Those of
us old enough to remember the time when an IR image meant laboriously
taking readings with your bolometer on a 10 10 sampling array on the sky,
will tell you so. However, with the current deluge of observational information
comes the task of making sense of all of it, and for this we need theoretical
understanding. It is appropriate here to quote a Philosopher of old times,
O , oo o,
o o Aoo
o o o

,
o o . . .
, To
Innocent light minded men,
who think that Astronomy can be learnt
by looking at the stars without knowledge
of Mathematics will, in the next life, be birds . . .
Plato (Timaeus)
Dynamics is one of the pillars of Classical Physics. Given the spread of
Gravity in the Universe, its application in Astronomy is only natural. In fact, a
good dynamical intuition ought to be second nature to Astronomers. I hope
that these lectures help students a bit to move in this direction, motivating
them to explore dynamical aspects of Galaxies, even if, and specially if, they
are not theoreticians.
I want to thank Omar L opez, Manolis Plionis and David Hughes for invit-
ing me and organizing a wonderful Gullermo Haro school. INAOE for pro-
viding a beautiful environment to ponder these matters about far away galax-
ies, within a setting rich in history and cultural heritage. I also want to thank
the students who had to suer my lectures and then wait for me to produce
these notes.
References
1. Abt, H., ed. 1999, The Astrophysical Journal American Astronomical Society
Centennial Issue, Univ. of Chicago Press
2. Adams, F., Bloch, A.: ApJ 629, 204 (2005)
3. Aguilar, L.A., White, S.D.M.: ApJ 295, 374 (1985)
4. Aguilar, L.A., White, S.D.M.: ApJ 307, 97 (1986)
5. Aguilar, L.A., Hut, P., Ostriker, J.P.: ApJ 335, 720 (1988)
6. Baes, M., Dejonghe, H.: AA 393, 485 (2002)
7. Barnes, J.E.: ApJ 393, 484 (1992)
8. Barnes, J.E.: Galaxies: Interactions and Induced Star Formation, Saas-Fee Ad-
vanced Course 26. Swiss Society for Astrophysics and Astronomy XIV, In: Ken-
nicutt, R.C., Schweizer, F., Barnes, J., Friedli, D., Martinet, L., Pfenniger, D.
(eds.) p. 275. Springer-Verlag, New York (1998)
9. Binney, J.: MNRAS 181, 735 (1977)
10. Binney, J.: MNRAS 200, 951 (1982)
11. Binney, J.: MNRAS 201, 15 (1982)
12. Binney, J., Tremaine, S.: Galactic Dynamics. Princeton University Press,
Princeton (1987)
13. Boccaletti, D., Pucacco, G.: Theory of Orbits, Vol. 1, Ch. 4, Springer-Verlag,
New York, Astronomy & Astrophysics Library (1996)
14. Bontekoe, T., van Albada, T.S.: MNRAS 224, 349 (1987)
15. Bournaud, F., Jog, C.J., Combes, F.: AA 437, 69 (2005)
16. Carpintero, D.D., Aguilar, L.A.: MNRAS 298, 1 (1998)
17. Cora, S., Vergne, M., Muzzio, J.C.: ApJ 546, 165 (2001)
18. Chandrasekhar, S.: Principles of Stellar Dynamics, p. 56. University of Chicago
Press, Chicago (1942)
19. Chandrasekhar, S.: ApJ 97, 255 (1943)
20. Dehnen, W.: MNRAS 255, 250 (1993)
21. de Pater, I., Lissauer, J.: Planetary Sciences. Cambridge University Press,
Cambridge (2001)
22. Diacau, F., Holmes, P.: Celestial Encounters, Ch. 1, Princeton University Press,
Princeton (1996)
23. Djorgovski, S., Meylan, G. (ed.): Structure and Dynamics of Globular Clusters,
ASPCS, Vol. 50 (Astron. Soc. Pacic, Provo) (1993)
24. Gnedin, O.Y., Ostriker, J.P.: ApJ 474, 223 (1997)
25. Goldstein, H., Poole, C., Safko, J.: Classical Mechanics, 3rd ed., Sec. 3.12,
Addison-Wesley (2002)
26. Hashimoto, Y., Funato, Y., Makino, J.: ApJ 582, 196 (2003)
27. Hern andez, X., Gilmore, G.: MNRAS 297, 517 (1998)
28. Hernquist, L.: ApJ 356, 359 (1990)
29. Holmberg, E.: ApJ 94, 385 (1941)
30. Hut, P., Inagaki, S.: ApJ 298, 502 (1985)
31. Hut, P., Heggie, D.: The Gravitational Million-Body Problem. Cambridge Uni-
versity Press, Cambridge (2003)
32. Innanen, K., Harris, W., Webbink, R.F.: AJ 88, 338 (1983)
33. Jae, W.: MNRAS 202, 995 (1983)
34. Keenan, D.W., Innanen, K.A.: AJ 80, 290 (1975)
35. Kormendy, J.: ApJ 218, 333 (1977)
118 L. A. Aguilar
36. Lokas, E., Mamon, G.: MNRAS 321, 155 (2001)
37. Merritt, D.: MNRAS 214, 25 (1985)
38. Meza, A., Navarro, J.F., Abadi, M., Steinmetz, M.: MNRAS 359, 93 (2005)
39. Murison, M.: AJ 98, 2346 (1989)
40. Navarro, J., Frenk, C.S., White, S.D.M.: ApJ 490, 493 (1997)
41. Ostriker, J.P., Spitzer, L., Chevalier, R.: ApJ 176, L51 (1972)
42. Pe narrubia, J., Andrew, J., Kroupa, P.: MNRAS 349, 747 (2004)
43. Plummer, H.: MNRAS 76, 107 (1905)
44. Read, J.I., Wilkinson, M.I., Evans, N.W., Gilmore, G., Kleyna, J.T.: MNRAS
366, 429 (2006)
45. Shu, F.: ApJ 525, 347 (2000)
46. Sparke, L.S., Gallagher, J.S.: Galaxies in the Universe. Cambridge University
Press, Cambridge (2000)
47. Spitzer, L.: ApJ 127, 17 (1958)
48. Spitzer, L.: Dynamical Evolution of Globular Clusters, Princeton University
Press, Princeton (1987)
49. Szebehely, V., Giacaglia, G.: AJ 69, 230 (1964)
50. Taylor, J.E., Babul, A.: ApJ 559, 716 (2001)
51. Tremaine, S., Weinberg, M.: MNRAS 209, 729 (1984)
52. Weinberg, M.D.: AJ 108, 1403 (1994)
53. White, S.D.M.: ApJ 274, 53 (1983)
54. Widrow, L.M., Dubinski, J.: ApJ 631, 838 (2005)
55. Zaritsky, D., White, S.D.M.: MNRAS 235, 289 (1988)
56. Zhao, H.: MNRAS 351, 891 (2004)

Dynamics of Galaxies and Clusters of Galaxies

Uploaded by

Copyright:

Available Formats

Dynamics of Galaxies and Clusters of Galaxies

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Dynamics of Galaxies and Clusters of Galaxies

Uploaded by

Copyright:

Available Formats

Dynamics of Galaxies and Clusters of Galaxies

are the three spherical coordinate components of the

() = 0, so despite the divergent mass, the

2, which is the Keplerian value (Fig. 5).

, it is clear that as r 0 we have:

are the energy and angular momentum of these orbits. This

satellite moving at 300 km/s, the logarithm of the Coulomb

. In this case, the enclosed mass within radius r goes as M

, and in each case, for core

. We note that the r

. In each set we show the

(dashed lines). Initial positions of 2, 4, 8 and 16

satellites launched from several

You might also like