A REVIEW OF

GALACTIC DYNAMICS

P.C. van der Kruit

Groningen, 1990-1991
Contents
1 Introduction. 1

2 The fundamentals. 2
2.1 The continuity equation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2.2 Poisson’s equation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.3 Hydrodynamical equations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.4 Virial theorem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.5 Virialization and two-body relaxation time. . . . . . . . . . . . . . . . . . . . . . 6

3 Stellar orbits. 7
3.1 Spherical potentials. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3.2 Axisymmetric potentials. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.3 Third integral and surface of section. . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.4 Differential rotation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.5 Epicycle orbits. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.6 Vertical motion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

4 Ellipsoidal velocity distribution. 12

4.1 The Schwarzschild distribution. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
4.2 Properties of the velocity ellipsoid. . . . . . . . . . . . . . . . . . . . . . . . . . . 13
4.3 Higher order moment equations. . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

5 Isothermal solutions and related results. 17

5.1 Isothermal sphere. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
5.2 Isothermal sheet. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
5.3 Exponential and sech-distributions. . . . . . . . . . . . . . . . . . . . . . . . . . . 18

6 Potential theory. 20
6.1 General axisymmetric theory. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
6.2 Exponential disks. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
6.3 Rotation curves from a general surface density distribution. . . . . . . . . . . . . 22

7 Other potentials. 25
7.1 Plummer model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
7.2 Kuzmin model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
7.3 Toomre models. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
7.4 Logarithmic potentials. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
7.5 Oblate spheroids. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
7.6 Infinitesimally thin disks. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
7.7 Mestel disk. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
7.8 Exponential disk. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
7.9 Disks with finite thickness. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

8 Stäckel potentials. 29
8.1 Coordinate system. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
8.2 The potential and the density distribution. . . . . . . . . . . . . . . . . . . . . . 30
8.3 Velocities and angular momentum. . . . . . . . . . . . . . . . . . . . . . . . . . . 30

i
 8.4 Integrals of motion and the orbits. . . . . . . . . . . . . . . . . . . . . . . . . . . 31
8.5 Stäckel models for the Galaxy. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

9 Instabilities and related topics. 35

9.1 Resonances. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
9.2 Jeans instability. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
9.3 Toomre criterion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
9.4 Goldreich–Lynden-Bell criterion. . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
9.5 Global stability. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
9.6 Tidal radii. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
9.7 Dynamical friction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

ii
1 Introduction.
The following as a description of fundamentals, methods and results in galactic dynamics. It
was originally written (around 1990) as a summary of the principles and fundamental equations
for students following my course on galaxies to give them the necessary background in case they
had not yet had a course in galactic dynamics. Also it serves a reference to have important
equations and results at hand.
I use the notation of the Saas-Fee Course “The Milky Way as a Galaxy” by Gerry Gilmore,
Ivan King and myself. The text is not always original and has been taken or adapted from
various sources. In addition to “The Milky Way as a Galaxy”, I have used “Galactic Dynamics”
by James Binney and Scott Tremaine, Oort’s chapter on “Stellar Dynamics” in Stars and Stellar
Systems, V, and various other publications and friends’ lecture notes.
As a result of the origin of this text, the treatment is restricted to dynamics of flattened
systems with axial symmetry and therefore generally aimed at disk galaxies. A general treatment
of tri-axial systems is beyond the current scope of this overview, but may be attempted some
time in the future.

1
2 The fundamentals.
2.1 The continuity equation.
Studies of galactic dynamics start with two fundamental equations. The first is the continuity
equation, also called the Liouville or collisionless Boltzman equation. It states that obviously in
any element of phase space the time derivative of the distribution function equals the number
of stars entering it minus that leaving it, if no stars are created or destroyed. If the distribution
function is f (x, y, z, u, v, w, t) and Φ the potential then

∂f ∂f ∂f ∂f ∂Φ ∂f ∂Φ ∂f ∂Φ ∂f
+u +v +w − − − = 0. (2.1)
∂t ∂x ∂y ∂z ∂x ∂u ∂y ∂v ∂z ∂w
There is another way of stating this, which does give some fundamental insight. Consider
the equations of motion of an individual star:
dx dy dz du ∂Φ dv ∂Φ dw ∂Φ
= u, = v, = w, =− , =− , =− . (2.2)
dt dt dt dt ∂x dt ∂y dt ∂z
We see by comparing these two equations that along the path of any star in phase space the
total derivative of f is zero, or in other words the density in phase space is constant along any
path a star can follow. So the flow of stars in 6-dimensional phase space is incompressible.
In most applications the system is assumed to be in equilibrium, so that f is not a function
of time and the first term equals zero. The usual method of solving an equation like the Liouville
equation is to form “subsidiary” equations, which in this case are simply the equations of motion
of an individual star, that were already given above. These can be rearranged as follows
dx dy dz du dv dw
(dt =) = = = = = . (2.3)
u v w −∂Φ/∂x −∂Φ/∂y −∂Φ/∂z

The solution of these 5 independent ordinary differential equations is in the form of 5 independent
integrals and then the general solution of the Liouville equation can be written as

f (x, y, z, u, v, w) = F (I1 , I2 , ..., I5 ). (2.4)

The I’s are called the integrals of motion. So we see that the distribution function depends only
on the integrals of motion. One can be identified as the energy, which is always conserved along
an orbit:
I1 = E = 1/2 (u2 + v 2 + w2 ) + Φ(x, y, z) = constant. (2.5)
This is called an isolating integral of motion, because for particular values it isolates hyper-
surfaces in phase space. The others in general are non-isolating and are only implicit in the
numerical integration of an orbit. The important fact then is that the distribution function can
be written as a function only of the isolating integrals. This is so, because in a steady state at
any point on this hyper-surface, f must assume the same value, if a star with that energy is able
to come in its orbit arbitrarily close to that point. This important statement is a fundamental
one in galactic dynamics and is called Jeans’ theorem.
In cylindrical coordinates the distribution function is f (R, θ, z, U, V, W, t) and the Liouville
equation becomes
!
∂f ∂f V ∂f ∂f V2 ∂Φ ∂f UV 1 ∂Φ ∂f ∂Φ ∂f
 
+U + +W + − − + − = 0. (2.6)
∂t ∂R R ∂θ ∂z R ∂R ∂U R R ∂θ ∂V ∂z ∂W

2
In the time-independent case, which I will consider exclusively from here on, and for axial
symmetry this becomes
!
∂f ∂f ∂Φ V 2 ∂f U V ∂f ∂Φ ∂f
U +W − − − − = 0. (2.7)
∂R ∂z ∂R R ∂U R ∂V ∂z ∂W

In the literature one also often finds the velocities (U, V, W ) notated as (VR , Vθ , Vz ), while an old
notation for R is $, pronounced as “pomega” or “curled pi”.
In this case there is a second isolating integral, because the angular momentum in the
direction of the symmetry axis z is also conserved along an orbit. This integral is

I2 = J = RV. (2.8)

Then Jeans’ theorem can be written as

f (R, z, U, V, W ) = F (E, J). (2.9)

In the case of the Galaxy near the plane the R- and z-motions are likely to be decoupled. This
means that the potential can be written as

Φ(R, z) = Φ1 (R) + Φ2 (z) (2.10)

and then there is a third integral for the z-motions:

I3 = 1/2 W 2 + Φ2 (z), (2.11)

which is a decoupled z-energy. There has been a long-standing problem with the question,
whether in the Galaxy there is a third integral for the stellar motions. At small z it is approxi-
mately true, since the potential is separable in r and z, but for higher velocities it has only been
possible to find a better approximate analytical description with the use of Stäckel potentials
(see below).

2.2 Poisson’s equation.

The second fundamental equation is Poisson’s equation, which says that the gravitational po-
tential derives from the combined gravitational forces of all the matter. It can be written as

∂2Φ ∂2Φ ∂2Φ

+ + ≡ ∇2 Φ = 4πGρ(x, y, z), (2.12)
∂x2 ∂y 2 ∂z 2
or in cylindrical coordinates

∂2Φ 1 ∂Φ 1 ∂2Φ ∂2Φ

+ + + = 4πGρ(R, θ, z). (2.13)
∂R2 R ∂R R2 ∂θ2 ∂z 2
For an axisymmetric case this reduces to (in cylindrical coordinates)

∂KR KR ∂Kz
+ + = −4πGρ(R, z). (2.14)
∂R R ∂z

3
 For completeness I also give these basic equations for the case of spherical symmetry, where
we now have velocities VR , Vθ and Vφ :

∂ ν
(νhVR2 i) + {2hVR2 i − Vt2 − h(Vθ − Vt )2 i − hVφ2 i} = νKR (2.15)
∂R R
and
1 ∂ ∂Φ
 
R2 = 4πGρ(R). (2.16)
R2 ∂R ∂R
For plane-parallel layers the basic equations reduce to
d n o
νhW 2 i = νKz (2.17)
dz
and
dKz
= −4πGρ(z). (2.18)
dz
In all of the above ν is the space density of a particular component. In case that we have to
do with a self-gravitating system, ν can be equated to the total density ρ, that gives rise to the
gravitational force through Poisson’s equation. An important problem is to develop techniques to
invert Poisson’s equation, so that the potential can be calculated for a given density distribution
(see below).
The two fundamental equations together completely describe the dynamics of a system.
Especially when it is self-gravitating in principle a distribution function can be found that
satisfies both equations. In practice this is seldom possible. Given a density distribution it is
also often the case that no unique solution exists to the full distribution function. For special
cases (such as isothermal models) solutions to the set of the two equations can be derived, but
this is at the expense of making simplifying assumptions and loss of generality.

2.3 Hydrodynamical equations.

From the collisionless Boltzman equation follow the moment or hydrodynamical equations, which
are obtained by multiplying the Liouville equation by a velocity-component (e.g. U ) and then
integrating over all velocities. For the radial direction we then find:
∂ ν ∂
(νhU 2 i) + {hU 2 i − Vt2 − h(V − Vt )2 i} + (νhU W i) = νKR . (2.19)
∂R R ∂z
By assumption we have taken here Vt = hV i and hU i = hW i = 0. This can be rewritten as:
" ( )#
V2 ∂ 1 h(V − Vt )2 i ∂
−KR = t − hU 2 i (ln νhU 2 i) + 1− + hU W i (ln νhU W i). (2.20)
R ∂R R hU 2 i ∂z

The last term reduces in the symmetry plane to

∂ ∂
hU W i (ln νhU W i) = hU W i (2.21)
∂z ∂z
and may then be assumed zero. Note that in the literature we often find velocity dispersions
notated as σRR for hU 2 i1/2 , etc. Since σ is the often used notation for the dispersion of a Gaussian
and since the actual velocity distributions are not necessarily Gaussian, I prefer the notation as
used above. Also the moment equations are independent of the higher order moments of the
velocity distribution.

4
 For the azimuthal direction the moment equation is seldom used, because it only contains
cross-terms of the velocity tensor. It reads
2ν ∂ ∂
hU V i + (νhU V i) + (νhV W i) = 0. (2.22)
R ∂R ∂z
In the vertical direction the moment equation becomes
∂ νhU W i ∂
(νhW 2 i) + + (νhU W i) = νKz . (2.23)
∂z R ∂R
Equations (2.20), (2.23) and (2.14) together are the basic equations to describe the dynamics of
axisymmetric systems and are therefore the starting point of any discussion of disk dynamics.

2.4 Virial theorem.

Alternatively, we may also multiply the collisionless Boltzman equation by a spatial coordinate
and integrate over space. This second procedure also yields some fundamental insight in the
global stability of self-gravitating systems. Write the coordinates as xi with i = 1, 2, 3 and the
position vector as x. Then we get the tensor equation
3
∂(ρVj ) 3 ∂(ρVj Vi ) 3 ∂Φ 3
Z X Z Z
xk d x=− xk d x− ρxk d x. (2.24)
∂t i=1
∂xi ∂xj
The first term on the right can be written as the kinetic energy tensor
3 Z 3
∂(ρVi Vj ) 3
X X Z
− xk d x= δki ρVi Vj d3 x = 2Tkj = 2Tjk , (2.25)
i=1
∂xi i=1

which has a systematic and a random part

Z Z
Tjk = Sjk + 21 Rjk = 1
2 ρhVj ihVk id3 x + ρhVj Vk id3 x. (2.26)

The second term on the right is the potential energy tensor

∂Φ 3
Z
Ωjk = Ωkj = − ρxj d x. (2.27)
∂xk
Average over the kj and jk components, then
d
Z
1
2 ρ(xk hVj i + xj hVk i)d3 x = 2Sjk + Rjk + Ωjk . (2.28)
dt
Define the inertia tensor as Z
Ijk = ρxj xk d3 x. (2.29)
Using the continuity equation if can be shown that
dIjk
Z
= ρ(xk hVj i + xj hVk i)d3 x, (2.30)
dt
and the result is the tensor virial theorem
d2 Ijk
1
2 = 2Sjk + Rjk + Ωjk . (2.31)
dt2
For a system in equilibrium Ijk is time-independent. Taking the trace of the tensor equation
yields the scalar virial theorem, which for a stable system is
2T + Ω = 0. (2.32)

5
2.5 Virialization and two-body relaxation time.
In all of the above we have taken a system in equilibrium. There are various processes for a
galaxy to come into equilibrium and “virialize”, which means that the stellar velocity distribu-
tion randomizes. We can estimate the relaxation time due to two-body encounters as follows.
Suppose that we have a cluster of radius R and mass M , made up of N stars with mass m,
moving with a mean velocity V . If two stars pass at a distance r then the acceleration is about
Gm/r2 . It lasts for about the period when the two stars are a distance r from the closest
approach and therefore for a time 2r/V . The total change in V 2 is then
2
2Gm

2
∆V ∼ . (2.33)
rV

The largest value of r is obviously R. For the smallest, one usually takes that where ∆V 2 is
equal to V 2 itself, since then the approximation breaks down. It is not critical, since we will
need the logarithm of the R divided by this rmin . So we will take
2Gm
rmin = . (2.34)
V2
The density of stars is 3N/4πR3 and the surface density 3N/4πR2 ∼ N/πR2 . The number
of stars with impact parameter r is then the surface density times 2πrdr. After crossing the
cluster once the star has encountered all others and we can calculate the total change in V 2 by
integrating over all r to find
R 2 2
2Gm 2N r 2Gm
Z  
(∆V 2 )tot = dr = 2N ln Λ, (2.35)
rmin rV R2 RV

where Λ = R/rmin . The relaxation time is equal to the number of crossing times it takes for
(∆V 2 )tot to be equal to V 2 . Since a crossing time is of order R/V and since the virial theorem
tells us that V 2 ∼ GN m/R, we find
!1/2
RN R3 N 1
trelax ∼ ∼ . (2.36)
8V ln Λ Gm 8 ln Λ

With the expression above for rmin we find that Λ ∼ N/2 ∼ N and
!1/2
R3 N
trelax ∼ . (2.37)
GM 8 ln N

This ranges from about 109 years for globular clusters to 1012 years for clusters of galaxies. For
galaxies we will need other mechanisms to stabilize them such as violent relaxation, where the
stars virialize due to the rapidly changing gravitational potential field of the collapsing galaxy.
The timescale of this is obiously the collapse time of the galaxy. Clusters of galaxies are probably
not virialized and only approximately in equilibrium.

6
3 Stellar orbits.
3.1 Spherical potentials.
The equation of motion is in vector notation
dΦ
R̈ = − êR . (3.1)
dR
The angular momentum
R × Ṙ = L (3.2)
is constant and the orbit is in a plane. In polar coordinates in this plane we have
dΦ
R̈ − RΦ̇2 = − (3.3a)
dR

R2 Φ̇ = L. (3.3b)
Integrating this we get
1 L2 1 2
2 R2 2 Ṙ +
= Φ(R) = E, (3.4)

where the energy E is constant. If E < 0 then the star is bound between radii Rmax and Rmin ,
which are the roots of
1 L2
+ Φ(R) = E. (3.5)
2 R2
The radial period is the time from Rmin to Rmax and back and follows from
Z Rmax Z Rmax dR
Z Rmax dR
TR = 2 dt = 2 =2 . (3.6)
Rmin Rmin Ṙ Rmin {2[E − Φ(R)] − L2 /R2 }1/2

In the azimuthal direction the angle θ changes in the time TR by

Z TR dθ
Z TR 
L

dR
∆θ = dR = 2 , (3.7)
0 dR 0 R2 Ṙ
which can be evaluated further for the expression for TR for a particular potential. The orbit is
closed if
m
∆θ = 2π , (3.8
n
where m and n are integers. This is not generally true and the orbits then has the form of a
rosette, which visits every point in (Rmin , Rmax ). Even in the simple case of a spherical potential,
the equation of motion of the orbit must be integrated numerically.
There are two special cases:

• The harmonic oscillator; this is the potential of a uniform sphere.

Φ = 12 Ω2 R2 . (3.9)

The orbits are closed ellipses centered on the origin and ∆θ is equal to π in TR .

7
 • The Keplerian potential
GM
Φ=− . (3.10)
R
The orbits are closed ellipses with one focus at the origin:

a(1 − e2 )
R= , (3.11)
{1 + cos(θ − θ◦ )}

where a and e are related to E and L by

L2
a= , (3.12a)
GM (1 − e2 )

GM
E=− . (3.12b)
2a
Further
Rmax , Rmin = a(1 ± e) (3.13)
and s
a3
TR = Tθ = 2π = TR (E). (3.14)
GM
Now ∆θ = 2π in TR .

Galaxies have mass distributions somewhere between these two extremes, so we may expect
that ∆θ is in the range π to 2π in TR .
The Rosette orbit can be closed by observing it from a rotating frame (see below under
resonances), when it is rotating at an angular velocity of

(∆θ − 2π)
Ωp = . (3.15)
TR

3.2 Axisymmetric potentials.

We now have a potential Φ = Φ(R, z), that may be applicable to disk galaxies. The equations
of motion now are
∂Φ
R̈ − RΦ̇2 = − , (3.16a)
∂R
d 2
(R Φ̇) = 0, (3.16b)
dt
∂Φ
z̈ = − . (3.16c)
∂z
Integration of (3.16b) gives
Lz = R2 Φ̇. (3.17)
The motion in the meridional plane then can be described by an effective potential
∂Φeff
R̈ = − , (3.18a)
∂R
∂Φeff
z̈ = − , (3.18b)
∂z

8
where
L2z
Φeff = Φ(R, z) + . (3.19)
2R2
The energy of the orbit is
E = 12 Ṙ2 + 21 ż 2 + Φeff (R, z). (3.20)
The orbit is trapped inside the appropriate contour E = Φeff , which is called the zero-velocity
curve. Only orbits with low Lz can approach the z-axis.
The minimum in Φeff occurs for 5Φeff = 0, or at z = 0 and where
∂Φ L2
= z3 ,
∂R R
which corresponds to the circular orbit with L = Lz . It si the highest angular momentum orbit
that is possible for a given E, or in other words, it has all its kinetic energy in θ-motion.

3.3 Third integral and surface of section.

If E and Lz are the only two isolating integrals, the orbit would visit all points within the zero-
velocity curves. In early numerical integrations it was found that there are limiting surfaces that
seem to forbid the orbit to fill the whole volume within the zero-velocity curves. This behaviour
is very common for orbits in axisymmetric potentials, when the combination (E, Lz ) is not too
far form that of a circular orbit. A third integral is present, although in general its form cannot
be explicitly written down.
For each orbit the energy E(R, z, Ṙ, ż) is an integral, so only three of the four coordinates
can be independent, say R, z and Ṙ. The orbit can visit every point in (R, z, Ṙ)-space as far as
allowed by E.
Now take a slice through (R, z, Ṙ)-space, say at z = 0. This is called a surface of section. The
orbits’ successive crossings of z = 0 generate a set of points inside the region E = 12 Ṙ2 +Φeff (R, 0).
If there is no other integral then these points fill the whole region. If there is another integral,
then its surface IR (R, z, Ṙ) cuts the plane in a curve IR (R, 0, Ṙ) = constant. A periodic orbit
is just a point (or a set of points) on the (R, Ṙ) surface of section. Such curves and points are
called invariant, because they are invariant under the mapping of the surface of section onto
itself generated by the orbit.
Invariant points often have closed invariant curves around them on the surface of section.
These represent stable periodic orbits. Ones where invariant curves cross are unstable periodic
orbits.

3.4 Differential rotation.

From simple kinematics it can be derived that, seen from the sun, stars at small distances r
have the following components of velocity as a result of the differential rotation of the Galaxy:
Vrad = Ar sin 2l cos2 b, (3.21a)
Vtan
= 4.74 µ = {A cos 2l + B}cos2 b, (3.21b)
r
where l is galactic longitude. A and B are the Oort constants
1 Vrot dVrot
 
A= − , (3.22a)
2 R dR
1 Vrot dVrot
 
B=− + . (3.22b)
2 R dR

9
3.5 Epicycle orbits.
For small deviation from the rotation, stars move in epicyclic orbits. If R0 is a fudicial distance
from the centre and if the deviation R − R0 is small compared to R0 , then we have in the radial
direction
d2 R d2 V 2 Vrot
2
= (R − R 0 ) = − = 4B(A − B)(R − R0 ), (3.23)
dt2 dt2 R R
where the last approximation results from a Taylor expansion of Vrot at R0 and ignoring higher
order terms. Similarly we get for the tangential direction
dθ V Vrot,0 A−B
= − = −2 (R − R0 ), (3.24)
dt R R0 R0
where θ is the angular tangential deviation seen from the galactic centre. These equations are
easily integrated and it is then found that the orbit is described by
U0
R − R0 = sin κt, (3.25a)
κ
U0
θR0 = − cos κt (3.25b)
2B
and the orbital velocities by
U = U0 cos κt, (3.26a)
U0 κ
V − Vrot,0 = sin κt. (3.26b)
−2B
The period in the epicycle equals 2π/κ and the epicyclic frequency is

κ = 2{−B(A − B)}1/2 . (3.27)

For a flat rotation curve we have

√ Vrot
κ= 2 . (3.28)
R
Through the Oort constants and the epicyclic frequency, the parameters of the epicycle
depend on the local forcefield, because these are all derived from the rotation velocity and its
radial derivative. The direction of motion in the epicycle is opposite to that of galactic rotation.
The ratio of the velocity dispersions in the plane for the stars can also be calculated as

hV 2 i −B
2
= . (3.29)
hU i A−B

With this result equation (2.20) can then be reduced to the so-called asymmetric drift equa-
tion
∂ ∂ B
  
2 2 2 2
Vrot − Vt = −hU i R ln ν + R lnhU i + 1 − . (3.30)
∂R ∂R B−A
If the asymmetric drift (Vrot − Vt ) is small, the left-hand term can be approximated by
2
Vrot − Vt2 ≈ 2Vrot (Vrot − Vt ). (3.31)

10
3.6 Vertical motion.
For the vertical motion the equivalent approximation is also that of a harmonic oscillator. For
a constant density with z we have

d2 z
Kz = = −4πGρ0 z. (3.32)
dt2
Integration gives
W0
z= sin λt (3.33)
λ
and
W = W0 cos λt. (3.34)
The period equals 2π/λ and the vertical frequency is

λ = (4πGρ0 )1/2 . (3.35)

For the solar neighbourhood we have Vrot ≈ 220 km/sec, R ≈ 8.5 kpc and ρ0 ≈ 0.1 M pc−3 .
Then the epicyclic period is about 1.7 ×108 years and the vertical period about 8.1 ×107 years.
It would be interesting to see where in galaxies these two might be equal.

11
4 Ellipsoidal velocity distribution.
4.1 The Schwarzschild distribution.
The distribution of space velocities of the local stars can be described with the so-called el-
lipsoidal distribution. This was first introduced by Karl Schwarzschild and is therefore also
sometimes called the Schwarzschild distribution. The distribution is Gaussian along the princi-
pal axes, but has different dispersions. This anisotropy was Schwarzschild’s explanation of the
“star-streams” that were discovered by Kapteyn.
The general equation for the Schwarzschild distribution is

8hU 2 ihV 2 ihW 2 i

f (R, z, U, V, W ) = ν
π 3/2
" #
U2 (V − Vt )2 W2 UV UW (V − Vt )W
exp − − − − − − . (4.1)
2hU 2 i 2hV 2 i 2hW 2 i 2hU V i 2hU W i 2hV W i
There is an interesting deduction that can be made from this ellipsoidal velocity distribution, as
has been shown by Oort in 1928. If one puts the distribution in the asymmetric drift equation,
adds the condition that z = 0 is a plane of symmetry, one gets an equation in terms of velocities
and multiplications thereof that has to be identical. So after sorting the terms all of these need
to be zero. This then gives
2hU 2 i = C1 + 12 C5 z 2 , (4.2)
2hV 2 i = C1 + C2 R2 + 21 C5 z 2 , (4.3)
2hW 2 i = C4 + 21 C5 z 2 , (4.4)
2hU W i = −C5 Rz, (4.5)
hU V i = hV W i = 0, (4.6)
C3 R
Vt = . (4.7)
C1 + C2 R2 + 21 C5 z 2
The constants C1 to C5 are positive constants. The density distribution follows from

∂ln ν C 2 R2 + (2C1 C32 − C1 C2 )R C1 R

= 2C1 KR + 2 2 2
− , (4.8)
∂R (C2 R + C1 ) C5 R2 + 2C4

at z = 0, and
" #
∂ln ν 2(C2 + 2C32 )R2 + C5 z 2 + 2C1 1
= (C5 R2 + 2C4 )Kz − C5 z RKR + 2 2 2
+ 2
. (4.9)
∂z (2C2 R + C5 z + 2C1 ) C5 z + 2C1

As it turns out, this does at best describe the solar neighbourhood approximately. The Schwarz-
schild distribution does e.g. not allow for high-velocity stars. Oort’s derivation only holds if the
stellar velocity distribution is exactly Gaussian. So, these equations are of historical interest
only. However, it is interesting to see that Oort assumed that C5 = 0. This uncoupled the radial
and vertical motion (as for a third integral).

12
4.2 Properties of the velocity ellipsoid.
For the solar neighbourhood, but probably anywhere in galactic disks, the velocity distribution
of the stars is very anisotropic.
• The ratio of the radial versus tangential velocity dispersions is governed by the local dif-
ferential rotation and can be described by the epicycle approximation. For that we had
hV 2 i −B
2
= . (4.10)
hU i (A − B)

• The ratio of the vertical to radial velocity dispersion is unconstrained, as a result of the
third integral. However, the existence of a third integral does not necessarily imply that
the velocity distribution has to be anisotropic. However, if no third integral would exist,
the velocity distribution would have to be isotropic, according to Jeans.
• The long axis of the velocity ellipsoid in the plane should point to the center. However, it
does not in practice. This is called the “deviation of the vertex” and presumably is due to
local irregularities in the Galactic gravitational field.
• The long axis of the velocity ellipsoid perpendicular to the plane has an unknown orienta-
tion. This has been a longstanding problem, also sometimes referred to as the “tilt” of the
velocity ellipsoid. Oort assumed it to be pointing parallel to the Galactic plane (C5 = 0),
but later also assumed it to be pointing always towards the Galactic center. In the Poisson
equation
∂KR KR ∂Kz
+ + = −4πGρ(R, z) (4.11)
∂R R ∂z
for a flattened disk, the first two terms are near the plane z = 0
∂KR KR
+ ≈ 2(A − B)(A + B). (4.12)
∂R R
For a flat rotation curve we have A = −B, so this is zero and the equation reduces to that
for a plane-parallel case. On this basis one may expect the long axis to be parallel to the
plane.

4.3 Higher order moment equations.

The hydrodynamical equations were obtained by multiplication of the Liouville equation with
velocities and then integrating over all velocity space. These are useful equations, but the
system is not complete (there is a “closure problem”): there are only three equations for eight
unknowns. These unknows are as a function of position the density, the rotation velocity,
the three velocity dispersions and three “cross-dispersions”. In principle, taking higher order
moments (by multiplying the Jeans equations with velocities and again integrating over all
velocities) should make matters worse. However, with reasonable assumptions Vandervoort and
in particular Cuddeford and Amendt have recently been able to make progress. In analogy to
the second moment
1
Z
σab (R, z) = hVa Vb i = (Va − hVa i)(Vb − hVb i)f d3 V, (4.13)
ν
define the third and fourth moments as
1
Z
Sabc (R, z) = hVa Vb Vc i = (Va − hVa i)(Vb − hVb i)(Vc − hVc i)f d3 V, (4.14)
ν

13
 1
Z
Tabcd (R, z) = hVa Vb Vc Vd i = (Va − hVa i)(Vb − hVb i)(Vc − hVc i)(Vd − hVd i)f d3 V. (4.15)
ν
For a Gaussian the “skewness” (e.g. SRRR /(σRR )3/2 ) is zero, since it is completely symmetric.
The fourth moment is related to the “kurtosis” (e.g. TRRRR /(σRR )2 ), which decribes how peaked
the distribution is and a Gaussian has a kurtosis of 3. The assumptions of Cuddeford and Amendt
were:
– All parameters can be expanded in terms of a small parameter , which is the ratio of the
radial velocity dispersion to the rotation velocity.
– The ordering scheme of these remains such that only terms in the leading order have to be
taken.
– The velocity distributions are Gaussian (Schwarzschild) up to one more order than required
by the equations.
This means that the assumption is that we have a cool, highly flattened and quasi-isothermal
system. Then the system can be closed and five more equations result after a lot of algebra:

hV W i = 0, (4.16)
" #
∂ln ν ∂hW 2 i ∂hU W i ∂hW 2 i ∂hU W i
hW 2 i + 2hU W i + hU W i + 2hW 2 i
∂z ∂R ∂R ∂z ∂z
hU W i ∂(R2 Vt2 ) 1 ∂(Vt2 hW 2 i)
=− 1 − , (4.17)
R3 2R
R ∂z
∂hW 2 i ∂hW 2 i
hW 2 i + hU W i = 0, (4.18)
∂z ∂R
" #
∂ln ν ∂hU 2 i ∂hU W i ∂hU 2 i ∂hU W i
hU W i + 2hU 2 i + hW 2 i + 2hU W i
∂z ∂R ∂R ∂z ∂z
4Vt2  2 2
 2 ∂(Vt2 hU 2 i) 2 ∂(Vt2 hU W i)
=− hU i − hV i − − , (4.19)
R2 R ∂R R ∂z
"
1 ∂(R2 Vt2 ) ∂hU 2 i ∂hU 2 i   ∂hU W i
hW 2 ihU W i + hW 2 i2 + 2 hU 2 ihW 2 i − 2hU W i2
4R ∂R ∂R ∂z ∂R
#
2 ∂hU W i ∂hW 2 i ∂hW 2 i
hW ihU W i − 2hU W ihU 2 i − 2hU W i2
∂z ∂R ∂z
" #
2 ∂hV 2 i hU W i ∂(R2 hV 2 i)
2
= −hW i hW i + Vt2 . (4.20)
∂z R2 ∂R
These equations can be used to derive further information on the velocity ellipsoid in cool,
flattened galaxies (i.e. in disks). There are a few applications.
The first is the tilt of the velocity ellipsoid. From the equations it can be found that the tilt
can be described as follows by the derivative of the cross-dispersion term near the plane
!
∂hU W i hU 2 i − hW 2 i
(R, 0) = λ(R) (R, 0), (4.21)
∂z R

with  !−1 
∂3Φ ∂Φ ∂2Φ ∂2Φ
λ(R) = R2 3 + R 2 − 4R 2  (R, 0). (4.22)
∂R∂z 2 ∂R ∂R ∂z

14
For a flat rotation curve this gives
!
2πGR3 ∂ρ
λ(R, 0) = 2 2
(R, 0). (4.23)
Vt − 8πGR ρ ∂R

The second application has to do with the radial dependence of velocity dispersions. A
solution of the equations then has the following form
!
∂hU 2 i
f1 (R) (R, 0) + f2 (R)hU 2 i(R, 0) = f3 (R). (4.24)
∂R

The complication is in the functions f :

∂hW 2 i
f1 = 4(α + β)(2α + β)(4α + β) , (4.25)
∂R
∂ 2 hW 2 i 3 ∂hW 2 i
f2 = (4α + β)(4α2 + 3αβ + 2β 2 ) + αβ(4α + β) −
∂R2 R ∂R
2 2 ∂hW 2 i
2 3 2 2 3 ∂ hW i 2
(12α + 15α β + 14αβ + 2β ) + (3 − 4γ)(α + β)(2α + β)(4α + β) +
hW 2 i ∂R2 R ∂R
1 dγ 4 ∂hW 2 i
 
αβ(4α + 7β) + (γ − 1) , (4.26)
γ dR R ∂R
!2
∂ 2 hW 2 i ∂hW 2 i 3 ∂hW 2 i
f3 = −α(4α+β)(4α+7β)hW 2 i +2α 2
(4α+13β) + αβ(4α+β)hW 2 i −
∂R2 ∂R R ∂R
4 ∂hW 2 i 1 dγ 4 ∂hW 2 i
 
(3 − 4γ)(4α + β)α(α + β)hW 2 i + αβ(4α + 7β) + (γ − 1) hW 2 i . (4.27)
R ∂R γ dR R ∂R
The parameters α, β and γ are related to the potential and therefore to the local kinematics:
!
∂2Φ
α=− (R, 0) = −λ2 , (4.28)
∂z 2

where λ is the vertical frequency,

! !
∂2Φ 3 ∂Φ 1 ∂(R2 Vt2 )
 
β= (R, 0) + (R, 0) = 3 (R, 0) = −κ2 , (4.29)
∂R2 R ∂R R ∂R

with κ the epicyclic frequency,

( ! −1 ) !
1 ∂2Φ ∂Φ hV 2 i
γ= R + 3 (R, 0) = (R, 0), (4.30
4 ∂R2 ∂z hW 2 i

which is the anisotropy in the velocity distribution.

This can be solved for a given potential; the most realistic solution is with a logarithmic-
exponential potential
R
 
Φ(R, z) = Aln R − BR − C 2 z exp − , (4.31)
h
which has !
∂2Φ R
 
(R, 0) = 2C exp − (4.32)
∂z 2 h

15
and thus an exponential density profile.
The resulting distributions show:
– the radial velocity dispersion hU 2 i to decrease more or less exponentially with radius,
– the velocity anisotropy hU 2 i/hW 2 i to be roughly constant in the inner regions at least, and
– Toomre Q to be constant with radius, except near the center.
A further application is the following equation, which can be derived from the new set of
hydrodynamic equations
! " ! #
∂ 2 hW 2 i hU 2 i − hW 2 i ∂ln hW 2 i
(R, 0) = −λ(R) (R, 0). (4.33)
∂z 2 R ∂R

Since λ(R) > 0; hU 2 i > hW 2 i and hW 2 i decreasing with R, the righthand side of the equation
has to be positive. That means that hW 2 i has a minimum in the plane. So disks are not
strictly isothermal in z and numerical values suggest less peaked in density than the exponential
function.
The final application gives a more accurate estimate of the velocity anisotropy in the plane
through (
hV 2 i 1 ∂ln Vt Sθθθ 1 ∂R2 νSRRθ
= 1 + − + +
hU 2 i 2 ∂ln R Vt hU 2 i νRVt hU 2 i ∂R
)
R ∂SRθz Vt2 − Vrot
2 TRRθθ
2
+ 2 2
SRRθ + . (4.34)
Vt hU i ∂z Vt hU i hU 2 i2
In practice this can be approximated as

hU 2 i 1 ∂ln Vt TRRθθ
 
2
= 1+ + . (4.35)
hV i 2 ∂ln R hU 2 i2

This constitutes a small correction to the classical result

hU 2 i 1 ∂ln Vt −B
 
2
= 1+ = . (4.36)
hV i 2 ∂ln R A−B

16
5 Isothermal solutions and related results.
For simple geometries, such as spherical density distributions or density distributions on strat-
ified layers and isothermal velocity distributions (that is equal velocity dispersions at all posi-
tions), full solutions for the distribution function to the set of two fundamental equations can
be obtained.

5.1 Isothermal sphere.

The equations for a (non-rotating) isothermal sphere are
1 ∂
(R2 KR ) = −4πGρ(R) (5.1)
R2 ∂R
and
∂ν
hV 2 i = νKR . (5.2)
∂R
The solution is given by
ρ(R) = ρ0 exp(−Φ), (5.3)
where Φ follows from a numerical integration of
1 d dΦ
 
exp(−Φ) = χ2 (5.4a)
χ2 dχ dχ
!1/2
hV 2 i
χ= R. (5.4b)
4πGρ0
For large R this becomes
hV 2 i −2
ρ(R) = R . (5.5)
2πG
The “core radius” is
−1/2
4πGρ0
 
r0 = B . (5.6)
9hV 2 i
King-models are adapted isothermal spheres with a tidal radius Rt and a corresponding upper
boundary in the velocity distribution. The total mass is
2 2 Rt
 
M (Rt ) = hV ir0 f (5.7)
G r0
and the central surface density
Rt
 
σ 0 = ρ0 r 0 g . (5.8)
r0
The functions f and g can only be calculated numerically and are given in the literature. From
the above it follows that the velocity dispersion is
ρ M (R )
hV 2 i1/2 ∝  0  t . (5.9)
Rt Rt
f r0 g r0

Since for elliptical galaxies it is observed that log(Rt /r0 ) is about constant at 2.2 and the central
surface brightness also (“Fish’s law”), we find for a constant M/L the Faber-Jackson relation,
which is the equivalent of the Tully-Fisher relation for spirals:
L1/4 ∝ hV 2 i1/2 . (5.10)

17
5.2 Isothermal sheet.
For an isothermal sheet the basic equations become
∂Kz
= −4πGρ(z) (5.11)
∂z
and
∂ν
hW 2 i
= νKz . (5.12)
∂z
In the self-gravitating case we may write ν = ρ. The two equations can be combined into

d2 ρ(z)
 
2
−4πGρ(z) = hW i 2 ln . (5.13)
dz ρ(0)

The solution is
hW 2 i z
 
ρ(z) = sech2 . (5.14)
2πGz02 z0
The corresponding surface density is
σ = 2z0 ρ0 (5.15)
and the relation to the velocity dispersion is

hW 2 i = πGσz0 . (5.16)

The vertical force results from integration of Poisson’s equation as

hW 2 i z
 
Kz = −2 tanh . (5.17)
z0 z0
Usefull approximations are
!
z z2
 
2
sech = exp for z  z0 , (5.18a)
z0 z02

z 2z
   
2
sech = 4 exp for z  z0 . (5.18b)
z0 z0
For a second isothermal component of negligible mass and different velocity dispersion in this
force-field we find
z
 
ρII (z) = ρII (0)sech2p , (5.19)
z0
where
hW 2 i
p= . (5.20)
hW 2 iII

5.3 Exponential and sech-distributions.

The isothermal z-distribution is only an approximate description of the vertical mass distribution
in disks of galaxies. There is a range of generations of stars, each with their own velocity
dispersion.

18
 Often used is the exponential distribution, since it is a convenient fitting function. Since
the velocity dispersion now varies with z we have to write the equation in terms of the velocity
1/2
dispersion in he plane hW 2 i0 . The equations corresponding to this case are:

hW 2 i0 z
 
ρ(z) = 2
exp − , (5.21)
2πGZe ze

σ = 2ze ρ0 , (5.22)
hW 2 i0 = πGσze , (5.23)
z
  
Kz = −2πGσ 1 − exp − . (5.24)
ze
If an isothermal component of negligible mass moves in this force field, then
2pz z
   
ρII (z) = ρII (0)exp − + 2p 1 − exp − , (5.25)
ze ze
where now
hW 2 i0
p= . (5.26)
hW 2 iII

As an intermediate case between the isothermal solution and the exponetial it is also possible
to use the sech-distribution. This corresponds probably closest to reality. The equations then
are:
2hW 2 iII z
 
ρ(z) = 3 2 sech , (5.27)
π Gze ze
σ = πρ0 ze , (5.28)
π2
hW 2 i0 = Gσze , (5.29)
2
z
  
Kz = −4Gσarctan sinh . (5.30)
ze
For the second isothermal component we now get
8 z
  
ρII (z) = ρII (0)exp − pI , (5.31)
π2 ze
where Z y
I(y) = arctan(sinhx)dx. (5.32)
0
This integral can be evaluated easily by numerical methods or through a series expansion.

19
6 Potential theory.
6.1 General axisymmetric theory.
Much attention has been paid to inverting Poisson’s equation (for the axisymmetric case)

∂2Φ 1 ∂Φ ∂ 2 Φ
+ + = 4πGρ(R, z), (6.1)
∂R2 R ∂R ∂z 2
so that the potential (and the forces) can be calculated when the density distribution is given.
This is a limited problem in that it does not involve the continuity equation and the distribution
function and therefore is not a general solution for a dynamical system, such as the isothermal
solutions above.
At the basis lies the Hankel transform, which in the radial direction for the density is
Z ∞
ρ̃(k, z) = uJ0 (ku)ρ(u, z)du, (6.2)
0

where J0 is the Bessel function of the first kind. The important property, why this is useful, is
that the transform can be inverted to
Z ∞
ρ(u, z) = kJ0 (kR)ρ̃(k, z)dk.
0

If we take this transform in the radial direction for both sides of the Poisson equation we get

∂2
−k 2 Φ̃(k, z) + Φ̃(k, z) = 4πGρ̃(k, z). (6.3)
∂z 2
This linear non-homogeneous ordinary differential equation can be solved to give
2πG
Z ∞
Φ̃(k, z) = − exp (−k|z − v|)ρ̃(k, v)dv. (6.4)
k −∞

Using this, Poisson’s equation can then be inverted to

Z ∞Z ∞
Φ(R, z) = −2πG J0 (kR)ρ̃(k, v)e−k|z−v| dv dk. (6.5)
0 −∞

Then Z ∞Z ∞Z ∞
Φ(R, z) = −2πG uJ0 (kR)J0 (ku)ρ(u, v)e−k|z−v| dv du dk. (6.6)
0 0 −∞
The forces follow by taking the negative derivatives of the potential in the radial and vertical
directions.
∂Φ(R, z)
Z ∞Z ∞Z ∞
KR (R, z) = − = −2πG ukJ1 (kR)J0 (ku)ρ(u, v)e−k|z−v| dv du dk, (6.7)
∂R 0 0 −∞

and
∂Φ(R, z)
Z ∞Z ∞Z ∞
Kz (R, z) = − = −2πG uJ0 (kR)J0 (ku)ρ(u, v)sign(z − v)e−k|z−v| dv du dk.
∂z 0 0 −∞
(6.8)
The integrations are somewhat simpler, if the density is separable as

ρ(R, z) = ρR (R)ρz (z). (6.9)

20
6.2 Exponential disks.
There are various ways of proceeding from here. The first is by taking an analytical form for
the density distribution. Kuijken and Gilmore have done this for exponential disks. If the radial
density distribution is exponential

ρR (R) = ρ0 exp (−R/h), (6.10)

then the Hankel transform becomes

∞ ρ0 h 2
Z
ρ0 J0 (ku)ue−u/h du = (6.11)
0 (k 2 h2 + 1)3/2

and the potential

Z ∞Z ∞ J0 (kR)
Φ(R, z) = −2πGh2 ρz (v)e−k|z−v| dv dk. (6.12)
0 −∞ (k 2 h2 + 1)3/2

Kuijken and Gilmore first solve for an exponential z-distribution:

ρz = exp (−|z|/ze ). (6.13)

First note that if ρz (z) is symmetric around z = 0, then

Z ∞
Iz (k, z) = ρz (v)e−k|z−v| dv
−∞
Z |z| Z ∞
= 2ek|z| ρz (v) cosh(kv)dv + 2 cosh(kz) ρz (v)e−kv dv
0 |z|
Z |z| Z ∞ Z ∞
−k|z| −kv −k|z|
=e kv
ρz (v)e dv + e k|z|
ρz (v)e dv + e ρz (v)e−kv dv. (6.14)
0 |z| 0

Solving this for the exponential z-distribution gives

∞ J0 (kR) e−k|z| − ze ke−|z|/ze
Z
Φ(R, z) = −4πGρ0 h2 ze dk. (6.15)
0 (k 2 h2 + 1)3/2 1 − k 2 ze2

The possible term for which the denominator is zero (kze = 1) is still finite; the last quotient is
then
1
(1 + k|z|)e−k|z| . (6.16)
2ze k
The forces are
∞ J1 (kR) e−k|z| − ze ke−|z|/ze
Z
KR (R, z) = −4πGρ0 h2 ze k dk, (6.17)
0 (k 2 h2 + 1)3/2 1 − k 2 ze2

and
∞ J0 (kR) e−k|z| − e−|z|/ze
Z
Kz (R, z) = −4πGρ0 h2 ze k sign(z) dk. (6.18)
0 (k 2 h2 + 1)3/2 1 − k 2 ze2
Next they go further and assume that the density distribution is given by

ρ(R, z) = ρ0 exp (−R/h) sechn (z/nze ). (6.19)

21
For n = 0 we have again the exponential z-distribution with vertical, exponential scaleheight
ze . For n = 2 we have the locally isothermal disk of van der Kruit and Searle and for n = 1 the
“sech-disk” proposed by van der Kruit. Kuijken and Gilmore then show that the potential can
be written as Z ∞
Φ(R, z) = −4πGρ0 h2 ze 2n J0 (kR)(k 2 h2 + 1)−3/2 ×
0
∞
−n (1 + 2m/n) exp (−k|z|) − ze k exp [−(1 + 2m/n)|z|/ze ]
X  
dk. (6.20)
m=0
m (1 + 2m/n)2 − k 2 ze2
The possible term, for which m = n(kze − 1)/2, has a zero denominator and must be written as
1 −n
 
(1 + k|z|)e−k|z| . (6.21)
2ze k m
The binomial with the upper coefficient negative can be written as follows
−n (−n)(−n − 1)........(−n − m + 1)
 
=
m m!
m+n−1 (m + n − 1)!
 
= (−1)m = (−1)m . (6.22)
n−1 (n − 1)!m!
So the potential is in this case expressed as a sum of those for exponential z-distributions. This
is essentially related to the fact that the sech is written as a sum of exponentials:
∞
(−1)j e−(2j+1)|x| .
X
sech x = 2 (6.23)
j=0

This well-known expansion suffers from the fact that it does not work for x = 0, because then
the terms are alternatingly +1 and –1. This does not necessarily make it unsuitable, because
after integration each term gets divided by −(2j + 1) and the series will converge even for x = 0.
However, it may remain slow for small x. For example the sum for x = 0
∞
X (−1)j π
2 = (6.24)
j=0
2j + 1 2

takes 32 steps to reach an accuracy of 1%.

Similar expressions as above can be found for the forces, but this will not be fully written
out here.

6.3 Rotation curves from a general surface density distribution.

Casertano has derived an expression for the potential in the plane in order to find the rotation
curve of a disk with a general density distribution. He uses the radial force in the plane and
performs the integration over k first (rather than over u). The equation for the radial force in
the plane for a symmetrical z-distribution is
Z ∞Z ∞Z ∞
KR (R, 0) = −4πG ukJ1 (kR)J0 (ku)ρ(u, v)e−kv dv du dk. (6.25)
0 0 0

It helps to have the same order Bessel functions and get rid of the linear factor k by integrating
by parts
∞ ∞ ∞

u 1 ∂ρ(u, v)
Z Z
uJ0 (ku)ρ(u, v)du = J1 (uk)ρ(u, v) − uJ1 (uk) du. (6.26)
0 k 0 k 0 ∂u

22
Then Z ∞Z ∞Z ∞ ∂ρ(u, v) −kv
KR (R, 0) = −4πG uJ1 (kR)J1 (uk) e dv dk du, (6.27)
0 0 0 ∂u
and this can be solved to give
Z ∞Z ∞r u ∂ρ(u, v)
KR (R, 0) = 8G [K(p) − E(p)]du dv, (6.28)
0 0 Rp ∂u
where
R 2 + u2 + v 2
p
p=x− x2 − 1,
. x= (6.29)
2Ru
K and E are the complete elliptic integrals of the second and first kind respectively for which
good approximations are known. For the z-dependence of the density one can take an exponential
or the isothermal distribution.
Casertano’s work can be extended to the potential, vertical force and the radial force out of
the plane. First start with KR at arbitrary z. At a general position we had
Z ∞Z ∞Z ∞
KR (R, z) = −2πG ukJ1 (kR)J0 (ku)ρ(u, v)e−k|z−v| dv du dk. (6.30)
0 0 −∞

As Casertano we can do the integration over k (after integration by parts) and obtain
∞ (2 − p2 )K(p) − 2E(p)
Z
J1 (kR)J1 (uk)e−k|z−v| dk = √ , (6.31)
0 πp Ru
where √
Ru
p = 2p . (6.32)
(z − v)2 + (R + u)2
This is the same as Casertano found (except that he had z = 0), but he chose to rework it
further to the form above. The formula for p has a singularity at R = u = z = 0. Note however
that for R = u = 0 we already have p = 0 for all z, so that we should take p = 0 also for z = 0.
Of course this only occurs when evaluating the force in the center.
The radial force now becomes
∞Z ∞ (2 − p2 )K(p) − 2E(p) ∂ρ(u, v)
Z
KR (R, z) = 2G √ du dv. (6.33)
0 −∞ p Ru ∂u
For the vertical force and the potential itself we have a product of Bessel functions of equal
order before the integration by parts, but this of different order after that. When then the
integration over k is done, we get expressions which contain the Heuman Lambda function.
This can be rewritten only in forms that involve incomplete elliptic integrals of the first and
second kind or the elliptic integral of the third kind, but these are much more difficult to evaluate
numerically. Also the integrals over u must then be written as the sum of two different integrals,
one from 0 to R and one from R to ∞. So it is better to start with the forms before the
integration by parts.
For the vertical force we start with
Z ∞Z ∞Z ∞
Kz (R, z) = −2πG uJ0 (kR)J0 (ku)ρ(u, v)sign(z − v)e−k|z−v| dv du dk. (6.34)
0 0 −∞

The integration over k yields

∞ |z − v|p3
Z
kJ0 (kR)J0 (ku)e−k|z−v| dk = p E(p), (6.35)
0 4π(1 − p2 ) (uR)3

23
and we get

G ∞Z ∞ u|z − v|p3 E(p)

Z
Kz (R, z) = − sign(z − v) p ρ(u, v)dv du. (6.36)
2 0 −∞ (1 − p2 ) (uR)3

For the potential we start with

Z ∞Z ∞Z ∞
Φ(R, z) = −2πG uJ0 (kR)J0 (ku)ρ(u, v)e−k|z−v| dv du dk. (6.37)
0 0 −∞

The integration over k now yields

Z ∞ p
J0 (kR)J0 (ku)e−k|z−v| dk = √ K(p). (6.38)
0 π uR
The potential then is given by
Z ∞Z ∞ upK(p)
Φ(R, z) = −2G √ ρ(u, v)dv du. (6.39)
0 −∞ uR
Kent has used yet another method by starting from a straightforward integration of all
contributions to the forces over the whole volume of the system. This goes as follows

∂Φ
Z ∞ Z 2π Z ∞ u(R − u cos θ)
=G ρ(u, v) dv dθ du, (6.40)
∂R 0 0 −∞ [R2 + u2 + (z − v)2 − 2Ru cos θ]3/2

∂Φ
Z ∞ Z 2π Z ∞ u(z − v)
=G ρ(u, v) dv dθ du. (6.41)
∂z 0 0 −∞ [R2 + u2 + (z − v)2 − 2Ru cos θ]3/2
This formulation suffers from the fact that the denominators have singularities for R = u. Kent
avoids this by changing to another independent variable. As expected he also ends up with
elliptic integrals. The expression for the radial force is similar to the one found by Casertano.
The one for the vertical force contains two terms, each with an elliptic integral of the third kind.
The expressions will not be repeated here, but it is clear that the ones derived above are better
usable in practice.

24
7 Other potentials.
There are in the literature many particular potentials that can be used to describe galaxies, but
are not isothermal. The most important ones will be summarized here. These are not solutions
of the Liouville and Poisson equation. Rather they are convenient expressions for the potential
or density distribution that can be inserted analytically in Poisson’s equation. Equilibrium
solutions exist for rotating, incompressible fluids, such as the Maclaurin spheroids for the
axisymmetric case. Tri-axial solutions also exist as Jacobi ellipsoids and Riemann ellipsoids.
These will not be given here.

7.1 Plummer model.

This was originally used to describe globular clusters. The potential has the simple spherical
form
GM
Φ(R) = − √ . (7.1)
R2 + a2
The corresponding density distribution is
!−5/2
3M R2
 
ρ(R) = 1+ 2 . (7.2)
4πa3 a

7.2 Kuzmin model.

This derives from the potential
GM
Φ(R, z) = − p 2 . (7.3)
R + (a + |z|)2
This is an axisymmetric potential that can be used to describe very flat disks. The corresponding
surface density is
aM
σ(R) = . (7.4
2π(R + a2 )3/2
2

7.3 Toomre models.

These are models that derive from the Kuzmin model by differentiating with respect to a2 . The
n-th model follows after (n − 1) differentiations:
!
R2
σn (R) = σ(0) 1 + 2 2 . (7.5)
4n a

The corresponding potential can be derived by differentiating the potential an equal number of
times. It can be seen that Toomre’s model 1 (which has n = 1) is Kuzmin’s model. The limiting
case of n → ∞ becomes a Gaussian surface density model.

7.4 Logarithmic potentials.

These are made to provide rotation curves that are not Keplerian for large R. Since these can
be made flattened they provide an alternative to the simple isothermal sphere. The potential is
!
V2 z2
Φ(R, z) = 0 ln r02 + R2 + 2 . (7.6)
2 c

25
V0 is the rotation velocity for large radii and c controls the flattening of the isopotential surfaces
(c ≤ 1). The density distribution is

V02 (2c2 + 1)r02 + R2 + 2z 2 [1 − 1/(2c2 )]

ρ(R, z) = . (7.7)
4πGc2 (r02 + R2 + z 2 /c2 )2

At large radii R  r0 the isodensity surfaces have a flattening

 2
b
= c4 (2 − c−2 ). (7.8a)
a
In the inner regions R  r0 it is
 2
b 1 = 4c2
= . (7.8b)
a 2 + 3c−2
The rotation curve is
V0 R
Vrot = q . (7.9)
r02 + R2

7.5 Oblate spheroids.

Assume that all iso-density surfaces are confocal ellipsoids with axis ratio c/a and therefore
excentricity s
c2
e = 1 − 2. (7.10)
a
Let the density along the major axis be ρ(R). Define

z2
α2 (R, z) = R2 +
1 − e2
Then inside the spheroid the forces and potential are
√ Z sin−1 e
4πG 1 − e2
KR = − R ρ(α) sin2 βdβ. (7.11a)
e3 0
√ Z sin−1 e
4πG 1 − e2
Kz = − z ρ(α) tan2 βdβ. (7.11b)
e3 0
√ "Z #
4πG 1 − e 2 δ Z a
Φ(R, z) = ρ(α)αβdα + sin−1 e ρ(α)αdα . (7.11c)
e 0 δ

Here
z2
δ 2 = R2 + , (7.12)
1 − e2
and
R2 sin2 β + z 2 tan2 β
α2 = . (7.13)
e2
Outside the spheroid we have
√ Z γ
4πG 1 − e2
KR = − R ρ(α) sin2 βdβ, (7.14a)
e3 0

26
 √ Z γ
4πG 1 − e2
Kz = − z ρ(α) tan2 βdβ, (7.14b)
e3 0
√
4πG 1 − e2 a
Z
Φ(R, z) = ρ(α)αβdα. (7.14c)
e 0
Here γ follows from
R2 sin2 γ + z 2 tan2 γ = a2 e2 . (7.15)

7.6 Infinitesimally thin disks.

This is analogous to the treatment of general disk potentials above. The potential can be written
as Z ∞ Z ∞
Φ(R, z) = −2πG exp (−k|z|)J0 (kR) σrJ0 (kr)r dr dk (7.16)
0 0
and the rotation velocity Z ∞
Vc2 (R) = −R S(k)J1 (kR) k dk, (7.17)
0
where Z ∞
S(k) = −2πG J0 (kR)σ(R)dR. (7.18)
0
Also it may be useful to calculate the surface density corresponding to a known rotation curve
Vc (R). Using the inversion of (7.16) it can be shown that
" #
1 1 R dVc2 r ∞ 1 dVc2 R
Z   Z  
σ(R) = 2 K dr + K dr , (7.19)
π G R 0 dr R R r dr r

where K is the complete elliptic integral. Note that there is a contribution from the part of the
disk beyond R. This also holds for disks with finite thickness as long as the density distribution
is not described by spheroids. Only in the case that isodensity surfaces are spheroids do the
forces at radii larger than R cancel. In the general case the rotation curve of a disk depends on
the surface density at all radii.

7.7 Mestel disk.

This has the surface density distribution
R0
σ(R) = σ0 . (7.20)
R
The corresponding rotation curve is flat and has

GM (R)
Vc2 (R) = 2πGσ0 R0 = , (7.21)
R
where M (R) is the mass interior to R.

27
7.8 Exponential disk.
The surface density is
R
 
σ(R) = σ0 exp . (7.22)
h
The corresponding potential is from (7.16)

R R R R
        
Φ(R, 0) = −πGσ0 R I0 K1 − I1 K0 . (7.23)
2h 2h 2h 2h

Here I and K are the modified Bessel functions. The rotation curve is using (7.17)
2 
R R R R R
        
Vc2 (R) = 4πGσ0 h I0 K0 − I1 K1 . (7.24)
2h 2h 2h 2h 2h
The total potential energy of the disk is

Ω ≈ −11.6Gσ02 h3 . (7.25)

7.9 Disks with finite thickness.

This has been treated in section 6. Here I will only give the result for the exponential sech-disk
with density distribution
R z
   
ρ(R, z) = ρ0 exp − sech . (7.26)
h ze
Then
∞ ∞ −k|z| − kz e−(1+2m)|z|/ze
J0 (kR) m (1 + 2m)e
Z
2
X e
Φ(R, z) = −8πGρ0 h ze (−1) 2
dk.
0
2 2 3/2
(k h + 1) m=0 (1 + 2m) − k 2 ze2
(7.27)

28
8 Stäckel potentials.
Stäckel potentials are potentials that can be written as separable functions in ellipsoidal coor-
dinate systems. I will here only treat the axisymmetric case with oblate density distributions
(which means a prolate potential distribution), which applies to disk galaxies. In that case the
coordinate system is spheroidal and it can be seen as a further generalisation of the axisymmetric,
plane-parallel case, where the potential is separable in R and z.

8.1 Coordinate system.

The new coordinate system is (λ, φ, ν). The relation with the axisymmetric system (r, φ, z) is,
that λ and ν are the two roots for τ of
r2 z2
+ = 1, (8.1)
τ +α τ +γ
with
0 ≤ ν ≤ λ. (8.2)
The constants α and γ are sometimes also given in the form

α = −a2 , γ = −c2 . (8.3)

These correspond to a focal distance

∆ = (|γ − α|)1/2 = (|a2 − c2 |)1/2 . (8.4)

Note that λ and ν have a dimension of length2 . The coordinate surfaces are spheroids for
constant λ and hyperboloids for constant ν with the z-axis as rotation axis. The case for
flattened disks obtains, when −α > − γ, so that −γ = c2 ≤ ν ≤ −α = a2 ≤ λ. Spheroids
of constant λ then are prolate, while the hyperboloids of constant ν have two sheets. On each
meridional plane of constant φ we then have elliptical coordinates (λ, ν) with foci on the z-axis
at z = ±∆. Note that the mass distribution is oblate, although the coordinate system is prolate.
Other relations between the two coordinate systems are

(λ + α)(ν + α)
r2 = , (8.5a)
α−γ

(λ + γ)(ν + γ)
z2 = , (8.5b)
γ−α
and q
1 2 2 1
λ, ν = 2 (r + z − γ − α) ± 2 (r2 − z 2 + γ − α)2 + 4r2 z 2 . (8.6)
Also
λ + ν = r2 + z 2 − α − γ, (8.7a)
λν = αγ − γr2 − αz 2 . (8.7b)
Note that ν and λ occupy different, but contiguous parts of the positive real line. In the
plane we have ν = −γ, λ = r2 − α and on the z-axis ν = z 2 − γ, λ = −α for 0 ≤ |z| ≤ ∆ and
ν = −α, λ = z 2 − γ for |z| ≥ ∆.

29
8.2 The potential and the density distribution.
Now suppose that the potential Φ, which is minus the usual potential Φ and therefore always
positive, can be separated as follows

(λ + γ)G(λ) − (ν + γ)G(ν)
Φ(λ, ν) = . (8.8)
λ−ν
Such potentials are called (axi-symmetric) Stäckel potentials. For models with a finite mass M
the potential should tend to zero for large radii, which means that for λ → ∞ we get
GM
G(λ) ∼ . (8.9)
λ1/2
The density ρ, which is defined such that ρ dx dy dz is the mass in the volume element dx dy dz,
can be calculated from Poisson’s equation, which has the complicated form

∂2Φ  3 1
 ∂Φ
πGρ(λ, ν)(ν − λ) = (λ + α)(λ + γ) + 2 λ + 2 α + γ −
∂λ2 ∂λ
∂2Φ  3 1
 ∂Φ
(ν + α)(ν + γ) − 2 ν + 2 α + γ . (8.10)
∂ν 2 ∂ν
The Kuzmin equations give the properties, when the density on the z-axis are given. Assume
that this density is ϕ(τ ), where τ = λ, ν and note from above that on the z-axis we always have
τ = z 2 − γ for all z. Then the density is

ρ(z) = ϕ(z 2 − γ) = ϕ(τ ). (8.11)

Define the primitive function of ϕ(τ ) as

Z τ
ψ(τ ) = ϕ(σ) dσ. (8.12)
−γ

Then
2 2
λ+α (λ + α)(ν + α) ψ(λ) − ψ(ν) ν+α
 
ρ(λ, ν) = ϕ(λ) − 2 + ϕ(ν). (8.13)
λ−ν (λ − ν)2 λ−ν λ−ν

The total mass is

Z ∞ σ + 2γ − α
Z ∞
M = 2π √ ϕ(σ) dσ = 4π (z 2 + ∆2 )ϕ(z) dz. (8.14)
−γ σ+γ 0

The potential follows from

2πG
Z τ σ+α
G(τ ) = 2πGψ(∞) − √ ψ(σ) dσ. (8.15)
τ +γ −γ 2(σ + γ)3/2

8.3 Velocities and angular momentum.

In order to convert velocities we write
1/2
(ν + α)(λ + γ)

cos Θ = , (8.16a)
(α − γ)(λ − ν)

30
 1/2
(λ + α)(ν + γ)

sin Θ = . (8.16b)
(γ − α)(λ − ν)
Velocities are related for the oblate mass models (γ − α > 0) as

Vr = Vλ cos Θ − Vν sin Θ, (8.17a)

sign (z) Vz = Vλ sin Θ + Vν cos Θ, (8.17b)

and
Vλ = Vr cos Θ + sign (z) Vz sin Θ, (8.18a)
Vν = −Vr sin Θ + sign (z) Vz cos Θ. (8.18b)
Note that Vλ and Vν are the velocities in the local Cartesian system and do not describe the
changes in the coordinates λ and ν.
For the momenta we need the coefficients of the coordinate system
λ−ν
P2 = , (8.19a)
4(λ + α)(λ + γ)

ν−λ
R2 = . (8.19b)
4(ν + α)(ν + γ)
The momenta then are
pλ = P V λ , pφ = rVφ , pν = RVν . (8.20)
The angular momenta are

Lx = y ż − z ẏ = rVz sin φ − z(Vr sin φ + Vφ cos φ), (8.21a)

Ly = z ẋ − xż = −rVz cos φ + z(Vr cos φ − Vφ sin φ), (8.21b)

Lz = xẏ − y ẋ = rVφ . (8.21c)
The total angular momentum L is

L2 = (r2 + z 2 )Vφ2 + (rVz − zVr )2 . (8.22)

8.4 Integrals of motion and the orbits.

The Hamiltonian H is
p2λ p2φ p2ν
H= + + − Φ(λ, ν). (8.23)
2P 2 2r2 2R2
It can then be shown that there are three integrals of motion, namely

E = −H, (8.24a)

I2 = 12 L2z , (8.24b)
G(λ) − G(ν)
 
I3 = 12 (L2x + L2y ) + (γ − α) 1 2
2 Vz − z2 . (8.24c)
λ−ν
The equations of motion then are
1 I2 I3
 
p2λ = G(λ) − − −E , (8.25a)
2(λ + α) λ+α λ+γ

31
 p2φ = 2I2 , (8.25b)
1 I2 I3
 
p2ν = G(ν) − − −E . (8.25c)
2(ν + α) ν+α ν+γ
In the meridional plane the orbits are restricted to the area defined by

−γ ≤ ν ≤ ν0 , λ1 ≤ λ ≤ λ 2 , (8.26)

where the turning points ν0 , λ1 and λ2 are the values for ν and λ for which respectively Vν and
Vλ are zero. The case ν = −γ corresponds to z = 0. The turning points are the three solutions
τ1 ≤ τ2 ≤ τ3 of
I2 I3
G(τ ) − − − E = 0, (8.27)
τ +α τ +γ
where in general there should be one solution τ1 ≤ −α, which is ν0 , and two −α ≤ τ2 ≤ τ3 ,
which are λ1 and λ2 . In the case of an oblate mass distribution (prolate coordinate system)
all orbits are “short axis tubes”, bounded by two prolate spheroids and one hyperboloid of one
sheet.
Stäckel potentials were used for galactic dynamics by Eddington in 1915, who showed that
the velocity dispersion tensor is diagonal in the coordinate system in which the potential is
separable. Hence, the principal axes of the velocity ellipsoid everywhere line up with the local
coordinate system. It is this property, plus the fact that three isolating integrals of motion can
be written explicitly, that makes Stäckel potentials most useful. In the tri-axial case studied by
de Zeeuw, it can be shown that stellar orbits can be classified in four families, namely box, inner
long axis, outer long axis and short axis tube orbits. In axisymmetric potentials only the last
family exists as a set of stable orbits.

8.5 Stäckel models for the Galaxy.

A few Stäckel models have appeared in the literature, but these have a number of shortcomings.
They will be reviewed here briefly.
• Statler (1989) used a model, which is useful in local applications. He starts from a flat
rotation curve, so that with rotation velocity Vc

dΨ(r, 0) dλ dΨ(λ, −γ)

−r = −r = Vc2 . (8.28)
dr dr dλ
Note that in the plane Ψ(r, 0) = G(λ). Also r2 = λ + α and then

Vc2
G(λ) = − ln(λ + α) + constant. (8.29)
2
2
Now let the flat rotation curve extend out to λ = Λ = rhalo − α and let the mass within this
radius be M . Then
GM = Vc2 rhalo = Vc2 (Λ + α)1/2 . (8.30)
Then for λ ≥ Λ we need
GM
G(λ) = , (8.31)
(λ + α)1/2
so that
G(Λ) = Vc2 . (8.32)

32
Then we get for λ ≤ Λ
1 λ+α
  
G(λ) = Vc2 2 − ln . (8.33)
2 Λ+α
In order to describe the z-distribution Statler proposed
 !1/2 
∆2 S2 ν+γ S
G(ν) = αVc2  +S − . (8.34)
(ν + γ)(ν + α) C2 ∆2 C

This obviously will not work on that part of the z-axis, where ν = −α or |z| ≥ ∆. In the plane,
where ν = −γ, we have
(ν + γ)G(ν) = 0. (8.35)
As Statler states in his paper, the constant C relates to the density in the plane and the constant
S to the integrated surface density. Both monotonically, but not proportionally. In a sense then
S/C is something like the vertical scaleheight, which can be taken independent of galactocentric
radius. However, S is only locally constant and really G(ν) has an R-dependence implicit in the
constant S. So Statler’s formula is only a local approximation and it ignores the local radial
density gradient. This model can only be used locally, because G(τ ) is not continuous at τ = −α.
In Statler’s model ∆ = 0.1 kpc.

• Sommer-Larsen and Zhen (1990) build the Galaxy from two components. The disk (and
the central bulge) are the oblate perfect ellipsoid. This has
!−2
r2 z2
ρ(r, z) = ρ0 1+ 2 + 2 . (8.36)
a c

Then s
2GM τ +γ
G(τ ) = (τ + γ)−1/2 arctan , (8.37)
π −γ
where the total mass M is
M = π 2 a2 cρ0 . (8.38)
Again this is only useful locally, because the radial profile is not exponential and the thickness
not constant. The surface density is
!−3/2
πρ0 c r2
σ(r) = 1+ 2 , (8.39)
2 a

and the equivalent thickness

!1/2
πc r2
z0 = 1+ 2 (8.40)
2 a
The dark halo is the s = 2 model of de Zeeuw, Peletier & Franx, which has
1
ρ(0, z) ∝ , (8.41)
z 2 + c2
for which "
2 ∆2 + c2 τ + γ + ∆2 τ + c2 + γ
G(τ ) = 4πGρ0 c ln − ln +
c2 2(τ + γ) c2

33
 !#
∆2 + c2
r
1 τ +γ 1 ∆
√ arctan 2
− arctan . (8.42)
c τ +γ c ∆ c
For large r this gives a flat rotation curve with an amplitude

Vc2 = 4πGρ0 c2 . (8.43)

Sommer-Larsen and Zhen use ∆ = 4.0 kpc. The difficulty is, that the disk component is flaring
to increasing thickness with galactocentric radius, which is in contradiction with observation
and is not exponential.
The use of the “flattened isochrone” (Evans et al., 1990) instead of the perfect ellipsoid does
not improve things. The general equation
GM (τ + γ)
G(τ ) = p √ (8.44)
(−α) + τ
is used in the extreme case γ = 0. Then the density distribution has a central axis ratio zero
and this ratio becomes (3/8)1/4 ≈ 0.78 at large radii. Then we have
M a2 X + 2aY + a2
ρ(λ, ν) = (8.45)
2π Y 3 (X + aY + a2 )2
with
X = a|z| and Y 2 = r2 + (a + |z|)2 . (8.46)
The surface density is
Ma 1
σ(r) = , (8.47)
π Y (Y + a)2
which is ∝ r−3 for large r. The equivalent thickness is
2Y 2
z0 = , (8.48)
2Y + a
which again is ∝ r for large r. So again the surface density is not at all exponential and the
disk flares at larger galactocentric radii.

• The model of Dejonghe and de Zeeuw (1988) uses the function

f (τ ) = (τ + γ)G(τ ), (8.49)

so that the potential is

f (λ) − f (ν)
Ψ(λ, ν) = . (8.50)
λ−ν
The function f (τ ), which is monotonically increasing, is then expanded into an interpolating
formula
k
X
ln f (τ ) = Ai t i , (8.51)
i=0
where
2 ln τ − ln λm − ln(−γ)
t= . (8.52)
ln λm − ln(−γ)
They use ∆ = 0.88 kpc and 12 constants Ai . Sommer-Larson and Zhen have shown, that the
model does not adequately represent the effect of the disk, mainly because of the choice of a
small value for ∆.

34
9 Instabilities and related topics.
9.1 Resonances.
First we will look at resonances. The most important ones are between epicyclic frequency and
some other frequency that we will call pattern speed Ωp . The inner Lindblad resonance occurs
for
κ
Ωp = Ωrot (R) − , (9.1)
2
where Ωrot (R) is the angular rotation speed. This resonance occurs at the radius, where –in a
rotating frame with angular velocity Ωp – the particle goes through 2 epicycles in the same time
is it goes once around the centre. The resulting orbit in that frame then is closed and has an oval
shape. It goes back to Lindblad’s discovery that the property Ωrot (R) − κ/2 in the inner Galaxy
is roughly constant with R. The pattern speed may be identified with that of the rotating frame
in which the spiral pattern (not the spiral arms as physical structures themselves) is stationary
or with the body rotation of a bar or oval distortion.
Equivalently we have the outer Lindblad resonance
κ
Ωp = Ωrot (R) + (9.2)
2
and co-rotation
Ωp = Ωrot (R). (9.3)
Higher order Lindblad resonances (involving κ/n) sometimes also play a role.

9.2 Jeans instability.

Now we will look at local stability. We then start with the Jeans instability in a homogeneous
medium. There are various ways of describing it to within an order of magnitude. The first is
to make use of the “virial theorem”
2 Tkin + Ω = 0 (9.4)
for stability against gravitational contraction. In a uniform, isothermal sphere the kinetic energy
is
Tkin = 1/2 M hV 2 i (9.5)
and the potential energy
3 GM 2
Ω=− . (9.6)
5 R
So the sphere is unstable, when its mass M is larger than the Jeans mass MJeans , which then
comes out as !1/2
5 3/2 3 1/2 hV 2 i3
   
MJeans = . (9.7)
3G 4π ρ
A method that gives roughly the same result, but is easier to adapt to the two-dimensional
case is the following, which starts by calculating the free-fall time of a homogeneous sphere.
Anywhere the equation of motion is

d2 r G M (r) 4π
2
=− 2
= − Gρr. (9.8)
dt r 3

35
Solve this and apply for r = 0, then
1/2
3π

tff = . (9.9)
32Gρ
The free-fall time is independent of the initial radius and depends only on the density. Now, if
there were no gravity a star will move out to the radius of the sphere R in a time
R
t= 2 i1/2
. (9.10)
hV
For marginal stability the two have to be equal and it follows that the Jeans length is
!1/2
3π hV 2 i
RJeans = . (9.11)
32 Gρ
This can be done in a similar manner for an infinitely flat disk. The equation of motion now is
(writing immediately R for the radius of the circular area considered)
d2 R
= −πGσ. (9.12)
dt2
The free-fall time then becomes 1/2
2R

tff = (9.13)
πGσ
and the Jeans length
2hV 2 i
RJeans = . (9.14)
πGσ

9.3 Toomre criterion.

Now look at the differentially rotating disk. For each element with radius R0 , the average
angular velocity Ω equals obviously B as a result of differential rotation and the specific angular
momentum R02 B. This gives rise to a centrifugal force Fcf = R0 Ω2 . Now let it contract to a
radius R. The angular momentum is conserved, so the angular velocity becomes Ω = R02 B/R2
and the centrifugal force Fcf = RΩ2 = R04 B 2 /R3 . If the contraction is dR, then we have
dFcf 3R4 B 2
= − 04 . (9.15)
dR R
The gravitational attraction is approximately (within a factor less than two, since we now have
a flat distribution) Fgrav = −GπR02 σ/R2 and
dFgrav 2πGR02 σ
= . (9.16)
dR R3
The critical condition then is that at R = R0 the two must compensate each other and the
critical radius then is
2πGσ
Rcrit = . (9.17)
3B 2
So for lengthscales larger than this critical radius the disk is stabilized by differential rotation.
Toomre realized then that the disk is stable only if this critical radius is smaller than the local
Jeans length and therefore the velocity dispersion has to exceed a critical value, which then is
1/2 π Gσ
hV 2 icrit = √ . (9.18)
3 B

36
In practice −B ≈ A and then the equation can be written as
 1/2
2 1/2 2 Gσ Gσ
hV icrit ≈ 2π = 5.13 . (9.19)
3 κ κ
This is Toomre’s criterion, who used a more precise analysis to arrive at a constant of 3.36.

9.4 Goldreich–Lynden-Bell criterion.

This can also be extended to the criterion, that Goldreich and Lynden-Bell derived for stability
of gaseous disks of finite thickness against sheared instabilities. With the equations of the
isothermal sheet we can express the velocity dispersion as a function of ρ0 and z0 . Then for the
critical case
π σ2
G ≈ ρ0 B 2 . (9.20)
6 z02
Equating σ/z0 and ρ0 to a mean density ρ and using (B − A) ≈ 2B, we get
π ρ
G ∼ 1. (9.21)
3 B(B − A)
From a detailed discussion Goldreich and Lynden-Bell found for stability
πGρ < 1. (9.22)
4B(B − A) ∼
These sheared instabilities were proposed as a possible mechanism for the formation of spiral
structure. More recently, Toomre has studied the process in stellar disks and finds an instability
based on shear due to differential rotation, that he called “swing amplification”. This process is
prevented when
Rκ2 >
X= 3, (9.23)
2πmGσ ∼
where m is the number of arms. For −B ≈ A (a flat rotation curve) this can be written as
QVrot >
∼ 3.97 m. (9.24)
hU 2 i1/2
This is Toomre’s local stability citerion if the velocity dispersion is replaced by 0.22 Vrot /m.

9.5 Global stability.

For global stability there is a global condition due to Efstathiou, Lake and Negroponte from
numerical experiments, which reads
1/2
h

Y = Vrot >
GMdisk ∼ 1.1. (9.25).

For a pure exponential disk without any dark halo Y = 0.59. For a flat rotation curve it is then
easy to show that the condition implies that within the disk radius of 4 to 5 scalelengths the
mass in the halo should exceed that of the disk by a factor of about 3.5.
For a flat rotation curve and an exponential disk Y can be rewritten as
1/2
QRVrot R
  
Y = 0.615 exp (9.26)
hhU 2 i1/2 2h

37
and this gives
QVrot >
∼ 7.91. (9.27)
hU 2 i1/2
Comparing this to (9.24) we see that for spirals that are stable against global modes, swing
amplification is possible for all modes with m ≥ 2, at least at those radii where the rotation
curve is flat.
Ostriker and Peebles have also found from numerical experiments a general condition for
global stability. Stability occurs only when the ratio of kinetic energy in rotation S to the
potential energy Ω is less than a certain value. This is related to the criterion above. They
found that stability requires
S <
t= 0.14. (9.28)
|Ω| ∼
The virial theorem says that 2S + R + Ω = 0 and since R/S > 0, we would have expected t
to have the range 0 to 0.5 available. The criterion translates into R/S ∼ > 5, while for the local

Galactic disk it is about 0.15. So disk galaxies require additional material with high random
motion in order to conform to the criterion, either in the disk itself (e.g. the stars in the central
region) or in the dark halo.

9.6 Tidal radii.

Globular clusters have tidal radii due to the force field of the galaxy. These radii can be estimated
as follows. Assume two point masses M (the Galaxy) and m (the cluster) and a separation R
in a circular orbit (the following can be adapted to elliptical orbits as well with R the smallest
separation). The angular velocity of the globular cluster around the center of gravity is
1/2
G(M + m)

Ω= , (9.29)
R3

while the center of gravity is at a distance M R/(M +m) from the cluster. Take a star at distance
r from the center of the cluster in the direction of M and calculate where the total force on that
star is zero. Thus
M M M +m MR
 
− 2 − − r = 0. (9.30)
(R − r)2 r R3 M +m
Since r is much less than R we may expand the first term
M M r
 
≈ 2 1+2 . (9.31)
(R − r)2 R R

Then the tidal radius is the solution for r of this equation

1/3
m

rtidal ∼R , (9.32)
3M
where m has been taken small compared to M .

9.7 Dynamical friction.

As a star moves through a background of other stars, the small deflections will give a small
overdensity behind the star and consequently induce a drag. Suppose that a body of mass m
moves in a circular orbit with radius R through a background of bodies with mass M at a speed

38
Vc and assume that the background is an isothermal sphere with Vc the circular speed (and Vc /2
the velocity dispersion). Then the loss of angular momentum is about

dJ Gm2
∼ −0.4 ln Λ, (9.33)
dR R
where
Rc Vc2
Λ= . (9.34)
G(m + M )
Rc is the core radius of the isothermal sphere (the typical lengthscale of the background density
distribution). The timescale of dynamical friction for the body to spiral into the center is then

R 2 Vc
tdf ∼ . (9.35)
Gm ln Λ
This timescale is large and only relevant for globular clusters in the inner few kpc of the halo or
for galaxies in the central parts of clusters of galaxies. The effect may contribute to respectively
the formation of galactic nuclei and the central cD or gE galaxies in clusters through cannibalism.
The timescale for the Magellanic Clouds to be drawn in into the Galaxy is about another 1010
years.

39