I ARTICLES

m..o./4.

Dynamics

of

Elliptical Galaxies

Elliptical galaxies were once thought to be similar in their structure and dynamics to
rotationally flattened bodies like stars. The discovery that elliptical galaxies rotate much
more slowly than a fluid body with the same shape has led to a qualitative change in our
understanding of the dynamics of these systems. It is now believed that elliptical galaxies
are fully triaxial in shape. Self-consistent triaxial equilibria have been constructed and
appear to be long-lived; they are made possible by the existence of conserved quantities,
or integrals of motion, for galactic potentials without rotational symmetry. Many selfconsistent equilibria are unstable; the nonexistence of elliptical galaxies with axis ratios
more extreme than 3:1 is probably the result of such an instability. There is evidence for
strong central mass concentrations, perhaps massive black holes, at the centers of some
nearby galaxies. Recent observations suggest that many elliptical galaxies formed through
the merger of two or more spiral galaxies.

About one-third of the bright galaxies in

the universe fall into the class of elliptical
galaxies. In contrast to spiral galaxies, like
our own Milky Way, which contain a
prominent disk composed of young stars,
gas, and dust, elliptical galaxies have a
smooth, almost featureless appearance (Fig.
1). The only obvious characteristic that
distinguishes one elliptical galaxy from another-aside from size or total luminosity,
quantities that depend on the assumed distance-is apparent axis ratio. This fact is
reflected in the classification scheme that
was proposed by Hubble in 1936 and is still
widely used (1). Although spiral galaxies
are classified according to the detailed morphology of their spiral arms, elliptical galaxies are distinguished only by their apparent elongation: an elliptical galaxy with
axis ratio b/a, where a is the length of the
semimajor axis and b is the length of the
semiminor axis, has a Hubble type of En,
where n = 10(1 b/a). Thus, an apparently round galaxy has Hubble type EO, and the
most elongated ellipticals are of type E7 (2).
Because we see only the stellar distribution
that is projected along our line of sight, it is
impossible to directly determine the intrinsic
elongation of any single elliptical galaxy or
indeed to decide whether elliptical galaxies
as a class are axisymmetric (oblate or prolate) or fully ellipsoidal ("triaxial" in the
nomenclature of astronomers) (3).
Until about 1975, it was commonly
(though in retrospect, rather loosely) assumed that elliptical galaxies were rotationally flattened and oblate, the stellar-dynamical counterparts to Maclaurin spheroids
(4). Thus, elliptical galaxies were seen as
logical complements to the more rapidly
rotating, and hence more strongly flat-

The author is in the Department of Physics and Astronomy at Rutgers University, Piscataway, NJ 08855.

tened, spirals. However, the first accurate

of the rotation velocities of
elliptical galaxies revealed a significantly
lower rotation rate than expected for an
equivalent fluid body, even one with the
density structure of a real galaxy (5). Elliptical galaxies were revealed to be "hot"
stellar systems, in which most of the support against gravitational collapse comes
from essentially random motions, rather
than "cold" systems, like spiral galaxies, in
which ordered rotation contributes most of
the internal kinetic energy. Two questions
immediately arose from these observations:
What produces the observed flattening;
and, given that rotation plays only a minor
role, are elliptical galaxies axisymmetric or
fully triaxial (6)? Finding the answers to
these questions, and others that have arisen
as the quality of the observational data has
increased, has occupied the majority of
galactic dynamicists for the last decade or
so. This article reviews recent progress in
our understanding of the structure and dynamics of elliptical galaxies.
Stellar systems differ in fundamental
ways from other large-N systems, such as
gases or plasmas. The molecules in a gas
interact by means of short-range forces,
which quickly act to randomize the distribution of molecular velocities at each
point. The stars in a galaxy rarely undergo
physical collisions, except perhaps in regions of atypically high density, such as
galactic nuclei. Furthermore, it is easy to
show that, because of the long-range nature
of the gravitational force, the perturbing
effect of gravitational encounters with nearby stars is typically negligible compared
with the overall gravitational force of the
galaxy. As a result, the gravitational force
on any star will not vary rapidly, and each
star moves smoothly through the force field
produced by the galaxy as a whole. More

measurements

SCIENCE

VOL. 259

26 MARCH 1993

quantitatively, the relaxation time, that is,

the time scale over which gravitational
encounters with individual stars produce
significant changes in a star's orbit, is long
compared to the crossing time, the time
required for a star to move from one side of
the galaxy to the other. For a typical elliptical galaxy, the mean relaxation time is of
order 1015 years, whereas the crossing time
is about 109 years. Astronomers therefore
refer to galaxies as collisionless systems.
The particles in a fully ionized plasma
also interact by means of long-range forces,
and low-density plasmas share with stellar
systems the property that the mean field is
more important than the fields of nearby
particles. However, plasmas contain both
positive and negative charges, so they are
neutral on large scales (7). Because gravity
is always an attractive force, stellar systems
tend to form strongly inhomogeneous equilibria in which the gravitational force is
balanced at every point by the stellar "pressure." This simple fact greatly complicates
the theoretical study of galaxies compared
with that of plasmas because even a zerothorder description of a galaxy typically requires an inhomogeneous model (8). The
problem is particularly severe for hot stellar

Fig. 1. A large elliptical galaxy, M8(, in the core

of the Virgo galaxy cluster. The center of the
image is burned out through overexposure,
whereas the envelope is underexposed and
almost invisible. The bright knots on one side
are produced by a jet of hot gas that is being
ejected from the nucleus, perhaps by a massive central black hole. This is one of the few
elliptical galaxies with obvious substructure.
[Photo by A. R. Sandage with the 200-inch Hale
telescope; courtesy of J. Bedke at the Space
Telescope Science Institute]
1867

Downloaded from www.sciencemag.org on March 25, 2013

David Merritt

-I. . . .

closed form, or can even be shown mathematically to exist, is very small. The mathematician Stickel first classified, in 1890,
the full set of potentials for which the
Hamilton-Jacobi equation becomes separable in curvilinear coordinates to yield three
exact integrals of motion (11). Remarkably, it required a further 80 years for the
realization that certain ellipsoidally stratified mass models have potentials of the
Stickel form (12). The most general such
mass distribution is

The dynamics of stellar systems are closely

tied to the forms of the trajectories, or
orbits, followed by individual stars. The
classical theory of orbits, dating back to
Kepler and Newton, is based on the inverse-square force law that describes two
interacting point masses. In a galaxy, the
force field derives from all the stars (as well
as nonluminous components, such as dust
and dark matter), and the dependence of
the gravitational acceleration on distance
from the galactic center is rarely close to
inverse-square. The orbits are therefore
very different from the familiar closed ellipses of the two-body problem.
In an arbitrary spherical potential, the
orbital motion can be well approximated as
a precessing ellipse that fills a plane, annular region. In axisymmetric potentials,
these annular orbits precess around the
symmetry axis, producing time-averaged orbital densities that are roughly doughnutshaped. These shapes can be discovered by
direct integration of trajectories on a computer, but they can also be predicted
through study of the mathematical form of
the integrals of motion associated with realistic galactic potentials.
An integral of motion is any function of
the phase-space coordinates (x, v) that is
constant along orbits. Examples are the
so-called "classical" integrals: the total energy, which is always an integral of motion
if the gravitational field is not time-varying,
and the angular momentum, which is conserved about any axis of rotational symmetry. Because integrals of motion are relations between the phase-space variables,
they restrict the volume in which a star can
move; for instance, an orbit respecting only
the energy integral would fill a volume defined by an equipotential surface. Numerical
integration reveals that the equations of
motion in many three-dimensional potentials seem to respect integrals beyond the
classical ones: the orbits confine themselves
to regions much smaller than the regions
defined by the classical integrals (10).
Unfortunately, the class of three-dimensional mass models for which the nonclassical integrals of motion can be expressed in
1868

p(m2) =2Pa 2)2

P

Orbits and Shapes

p=
m 2=

x2

y2

-2

Z2
+

C-2

(1)

The mass density, p, is constant on surfaces

of constant m, which are ellipsoids. These
ellipsoids can have arbitrary axis ratios.
The orbits in the force field generated by
the mass model of Eq. 1 fall into four
families (Fig. 2). Three of these families are
"tubes," orbits that respect an integral similar to the angular momentum around one
of the three axes, which causes them to
avoid the center. Orbits of the fourth family
are called "boxes," which are unique to
triaxial potentials. At low energies, and
thus small amplitudes, the box orbits are
rectangular parallelepipeds confined to the
core, where the equations of motion are
those of a three-dimensional, incommensurate harmonic oscillator. At large energies,

Fig. 2. The four families of orbits in an integrable, nonrotating, triaxial potential. The pictures
are perspective views of solid bodies that envelope the volumes filled by the four types of
orbits. The symmetry axes of the solids coincide with the symmetry axes of the galaxy.
[Reprinted from (14) with permission American Astronomical Society]
SCIENCE

VOL. 259

26 MARCH 1993

the box orbits become elongated in the

direction of the long axis of the ellipsoid.
Box orbits are distinguished from tube orbits in that they pass arbitrarily close to the
center and have an alternating sense of
rotation. Numerical integrations show
that-with some important exceptions discussed below-these families of orbits are
the main ones in a fairly wide class of
triaxial potentials (13).
A first step in determining whether real
elliptical galaxies are triaxial (as opposed to
axisymmetric) is to show that self-consistent
triaxial equilibria exist. One simple, although approximate, way to accomplish this
is by numerically combining a discrete set of
orbits like those in Fig. 2 in such a way as to
reproduce the mass distribution of Eq. 1.
Such experiments suggest that triaxial equilibria do indeed exist for essentially any
choice of axis ratios (13, 14). These models
typically contain sizable contributions from
each of the four basic families; the box
orbits, which are elongated in the same
direction as the galaxy and are concentrated
near the center, are often the dominant
component. An alternative way to verify the
existence of triaxial equilibria is to use an
N-body program to simulate the relaxation
of a set of stars from nonequilibrium initial
conditions, such as an extended protogalactic cloud or a pair of subsystems in the process
of merging. One finds that the final state is
often substantially triaxial and that many of
the stellar orbits have the same basic properties as the orbits in integrable potentials (15).
Taken together, these two sorts of experiments suggest that triaxial ellipsoids are natural, and perhaps long-lived, shapes for slowly
rotating galaxies (16).
Real elliptical galaxies have density profiles that are much more centrally condensed than that of the integrable model
described by Eq. 1. High-resolution observations of nearby ellipticals and the central
bulges of spiral galaxies, including the
Milky Way, often reveal core luminosity
profiles that rise very steeply into the center
with no discernible limiting density (17).
The bulge of the nearby spiral galaxy M31
contains a dense nucleus of almost stellar
appearance (18), and many elliptical galaxies may contain massive black holes at their
centers, as discussed below. The gravitational potentials of triaxial models with density
profiles similar to these are far from Stickellike, and, although many of the orbits in
such model potentials appear to respect
three integrals of motion, the orbital shapes
are often quite different from those in the
analytically separable potentials.
Most strongly affected are the box orbits,
which pass arbitrarily close to the center
after sufficient time. Numerical orbit integrations show that most box orbits are
destroyed when the stellar density increases

Downloaded from www.sciencemag.org on March 25, 2013

systems like elliptical galaxies, in which the

stars pass through regions of very different
gravitational force in one crossing time. The
collective gravitational effects that determine
the structure and evolution of a hot stellar
system are therefore usually too complex to be
described by simple mathematical models,
and most studies of these systems must be
based on computer simulations. Recent advances in the art of studying the gravitational
N-body problem have contributed substantially to progress in this field (9).

,I
p..'It
. co..

more rapidly than r-1 near the center, as

appears to be the case for many resolved
galaxies. Some of these box orbits become
"stochastic," or chaotic, filling volumes
that are roughly bounded by equipotential
surfaces. Other box orbits are replaced by
so-called "boxlets," orbits similar to boxes
but which curve in such a way as to avoid
the center (19). It is unknown at present
whether self-consistent triaxial models with
central cusps can be constructed from these
orbits. If the answer is no, this would imply
that many elliptical galaxies are axisymmetric, or perhaps that their triaxial figures are
slowly evolving toward axisymmetry.
An interesting problem that has not been
resolved by orbital studies is why elliptical
galaxies are so nearly ellipsoidal in shape.
The class of integrable potentials includes
mass models with isodensity contours that
are strongly nonellipsoidal, both "disky" and
"boxy" (20). Triaxial models formed during
the N-body evolution of nonequilibrium or
unstable initial conditions are often found to
be extremely boxy, even peanut-shaped (21,
22). The ubiquity of boxy models is perhaps
not surprising in consideration of the fact
that most orbits in triaxial potentials have
strongly flaring shapes, as shown in Fig. 2
(23). In contrast, however, real elliptical
galaxies have isophotal shapes that are almost precisely elliptical: the deviations are
typically less than 1% and are as often in the
direction of diskiness as of boxiness (24). It
is unclear what, if anything, this fact might
be telling us about the formation history of
elliptical galaxies (25).

(26). A spherical model in which the orbits

are strongly eccentric, or radial, evolves
quickly into a triaxial bar. The instability
arises because highly eccentric orbits are
nearly closed, and nonradial forces can
realign them around a barlike perturbation.
The radial-orbit instability implies an upper
limit to the degree of velocity anisotropy in
spherical or spheroidal models and can
sometimes be used to constrain the dynamical state of observed galaxies (27). This
instability may also play a role in galaxy
formation because collapse tends to result
in eccentric orbits. The N-body simulations
of such systems verify that collapse from a
wide range of initial conditions leads to

Dark Matter and Central

Black Holes

Stability
Any proposed stellar-dynamical configuration should be tested for stability before
being accepted as a model of a real galaxy.
The importance of instabilities in rapidly
rotating stellar systems, such as spiral galaxies, has been recognized for at least three
decades; this work has antecedents in the
classic papers of Jacobi, Dirichlet, and others
on the equilibrium and stability of rotating
fluid bodies. In contrast, the importance of
instabilities in galaxy models with little or
no rotation was appreciated only quite recently. In part, the reason was that the
stellar velocity vectors in a hot stellar system
are pointed in all directions, so a perturbation in the stellar density might be expected
to rapidly attenuate as the stars move along
their respective orbits. However, it turns out
that the stellar motions in a variety of
interesting hot models are sufficiently correlated to induce instability, and these instabilities often grow on a time scale that is
short compared to the age of the universe.
The first such instability to be discovered, and probably still the most important,
is the so-called "radial-orbit instability"

final states whose axis ratios are determined

by a process similar to the radial-orbit instability (28).
Another important instability is illustrated in Fig. 3. Models that are sufficiently
elongated are unstable to global bending
modes, which arise when the centrifugal
force experienced by a star that moves
along the bend overwhelms the gravitational restoring force (29). This instability is
similar to the "firehose" instability of plasma physics (30) and is usually called by the
same name. The N-body integrations of a
family of prolate models suggest that galaxies with axis ratios more extreme than
about 3:1 are unstable to these bending
modes, which quickly act to "puff up" the
short axis and reduce the flattening (21).
This instability is a plausible explanation
for why elliptical galaxies never have Hubble types more extreme than E7 (31). The
firehose instability may also play a crucial
role in the formation of the central bulges
of spiral galaxies: N-body integrations of
rapidly rotating disks show that the central
regions, after forming a rotating bar
through an instability in the disk plane,
immediately bend out of the plane, producing a hot, spheroidal system with properties
similar to those of galactic bulges (32).
Little is known yet about the stability of
triaxial models.

Fig. 3. The firehose instability in a nonrotating,

prolate galaxy model (21). Frames are separated by about two crossing times. The final model
is substantially more "boxy" than real elliptical
galaxies; the simulation was not carried far
enough to determine whether this boxiness
would persist for a time corresponding to the
age of the universe. This N-body simulation
contained 104 particles and was executed on
the Cray Y-MP at the University of Pittsburgh.
SCIENCE

VOL. 259

26 MARCH 1993

The model building described above is

based on the premise that the gravitational
potential in which the stars move is generated by the stars themselves. There is good
reason to suspect, however, that much of
the matter in the universe is dark (33). The
most secure evidence for dark matter comes
from studies of the rotation curves of spiral
galaxies. The stars in these galaxies lie
mainly in a thin disk and travel on nearly
circular orbits around the galactic center.
By comparing the observed centripetal acceleration, v2/r, with the calculated gravitational acceleration attributable to the luminous mass, one infers that stars make up
less than 10% of the total mass in spiral
galaxies and that the dark matter is more
extended spatially than the stars, in what is
commonly called a dark halo. The shapes of
these halos are not well constrained observationally but are thought to be spherical or
ellipsoidal (34). The composition of the
dark matter is a completely open question.
It is reasonable to expect that elliptical
galaxies are also surrounded by extended
dark halos. Their presence is more difficult
to infer than in the case of spiral galaxies,
however, because elliptical galaxies are not
spherical and the orbits in elliptical galaxies
are more complex than those in spiral
1869

Downloaded from www.sciencemag.org on March 25, 2013

--

d(v&)
dr

= -

d42-(2)
dr

Here v (r) is the number density of stars at

radius r, &2(r) is their mean square velocity
in one dimension, and (r) is the gravitational potential, which may result from
both stars and dark matter. Observations
yield the projected number density and
velocity-dispersion profiles from which v(r)
and a(r) can be inferred. Equation 2 then
gives the potential (r) and, hence, the
mass M (r) = G `r2d4'/dr, where G is the
gravitational constant. This sort of analysis,
applied to stars and other kinematical samples, suggests that dark matter may be
present around at least some elliptical galaxies (35). But even in a spherical galaxy,
the distribution of stellar velocities need
not be isotropic, which leads to large formal
uncertainties in the estimated F(r) (36,
37). If we permit the postulated dark-matter halos to be nonspherical, the uncertainties increase even more.
Recent work suggests that it is possible
to map dark matter in hot stellar systems
with the use of the full distribution of
line-of-sight velocities of some test sample
at a number of different positions, rather
than just the velocity dispersion that appears in Eq. 2 (38). One way to obtain this
additional information is to measure the
Doppler-shifted velocities of a very large
sample of discrete objects (stars, planetary
nebulae, globular clusters, and so on) that
surround the galaxy and construct a velocity histogram at a variety of radii. Such
observations are greatly facilitated by a new
generation of detectors that use optical
fibers to observe hundreds of objects at once
(39). Another route is to measure the
integrated spectrum of the stars at a given
point in a galaxy and ask what velocity
distribution function would yield the observed line shapes after convolution with
the spectrum of a single star (40). Such
studies are only beginning but have the
potential to place much stronger constraints on the dynamical state of elliptical
galaxies than was possible in the past.
Another piece of evidence for dark matter in elliptical galaxies comes from kinematical studies of galactic nuclei. In 1978, a
group demonstrated the possible existence
of a black hole or other compact object
with a mass of about 5 x 109 solar masses at
1870

the center of the giant elliptical galaxy M87

(41). The evidence was a moderately steep
rise in the stellar velocity dispersion in the
inner kiloparsec of the galaxy, which implied, with the use of Eq. 2, a large central
mass. Very soon after, a number of workers
stressed the nonuniqueness of this model
because of uncertainties in the orbital distribution, and the most recent work suggests that no strong case can be made for a
significant nonluminous component in the
nucleus of this galaxy (42). However, it has
long been clear that dead quasars should be
roughly as common as giant galaxies, and it is
natural to expect that the massive black holes
that are thought to power the quasars should
reside in galactic nuclei. This argument has
prompted a search for dynamical signatures of
massive black holes in other galaxies. To
date, reasonably strong kinematic evidence
for central mass concentrations has been obtained for a handful of nearby galaxies (43).
The most convincing cases are the central
bulges of spiral galaxies, which often exhibit a
rapid rise in the velocity of stellar streaming
very near the center. These nuclei are probably highly flattened and may even consist of
disks of stars that formed from gas that fell in
from much larger radii. However, the resolution limits imposed by the Earth's atmosphere
have so far precluded the unambiguous determination of the shapes of these nuclei, and
the resulting uncertainties in the inferred
black hole masses are generally comparable to
the masses themselves (44). It is not clear
whether improvement of this situation is possible with ground-based instruments (45).

Formation of Elliptical Galaxies

One remarkable regularity in the morphology of elliptical galaxies is the form of their
surface brightness profiles, that is, the dependence of the stellar number density on
radius. Aside from scaling and shape factors,
this dependence is very nearly the same for
all elliptical galaxies. One commonly used
parametrization is the so-called "r1/4 law"
(3)
I(R)X exp [-7.67 (R/Re)0.251
first proposed by de Vaucouleurs in 1948
(46); here l(R) is the luminosity density on
the plane of the sky at apparent radius R,
and Re is the projected radius containing
one-half of the total light. Equation 3 is
accurate over at least three decades in radius
for many bright elliptical galaxies (47). Although the detailed form of Eq. 3 is almost
certainly without physical signfficance, the
existence of a characteristic density law cries
out for an explanation. Attempts to derive
the rl/4 law from a maximum-entropy principle have been largely unsuccessful (48).
Nevertheless, computer simulations of galaxy formation demonstrate that a density
profile similar to Eq. 3 is a very common
SCIENCE * VOL. 259

26 MARCH 1993

outcome, as long as the evolution satisfies

two general conditions: (i) there must be
strong exchange of energy between the stars
and the overall gravitational field, so that
the initial conditions are forgotten, and (ii)
the initial state must contain an appreciable
fraction of matter at high phase-space densities, to allow the formation of a highdensity core (49). One simple formation
scenario that has received much attention is
collapse from a cold, that is, low velocity
dispersion, initial state. Computer simulations (28), as well as simple analytical arguments (50), confirm that such collapses produce galaxies with density profiles very similar to the r1/4 law. However, it is clear from
the numerical simulations that many other
models of galaxy formation can produce
good r0/4 laws (51), and it does not seem
especially likely that real galaxies formed
through cold collapse.
One model for the formation of elliptical
galaxies that has gained many adherents in
recent years is the merger hypothesis proposed by Toomre (52). According to this
hypothesis, most or all elliptical galaxies
formed from the coalescence of spiral galaxies, either in the distant past or more
recently. The initial motivation for the
merger hypothesis was the observation that
the rate of apparent merger events in the
current universe is large enough to have
formed all the observed elliptical galaxies
over the course of 1010 years (52). This
argument was strengthened by the discovery of dark-matter halos around spiral galaxies, which imply a much higher probability for a merger than if galaxies were no
bigger than their luminous parts (53). More
recent evidence in favor of the merger
hypothesis includes the discovery of a number of fine-structure features, including ripples, twists, imbedded disks, and tidal tails,
that may be indicative of past mergers (54).
A large fraction of the elliptical galaxies
that are examined closely seem to have
such fine structure (55). Although the interpretation of this fine structure is still a
matter of debate, it seems clear at least that
elliptical galaxies are far less uniform in
their structure and formation histories than
was previously thought.

Downloaded from www.sciencemag.org on March 25, 2013

galaxies. The basic principle behind the

estimation of the gravitational potential in
an elliptical galaxy can be illustrated by the
case of a spherical and nonrotating galaxy
in which the stellar velocities are everywhere isotropic. The first velocity moment
of the collisionless Boltzmann equation
yields an equation similar to the equation of
hydrostatic equilibrium

REFERENCES AND NOTES

1. E. P. Hubble, The Realm of the Nebulae (Yale
Univ. Press, New Haven, CT, 1936).
2. Many galaxies classified as E6 or E7 are probably
disk galaxies.
3. With the assumption of random orientations, the
distribution of intrinsic elongations can be estimated from a sufficiently large sample of galaxies [E. Hubble, Astrophys. J. 64, 321 (1926)] and
is only weakly dependent on the assumed geometry. One finds that most elliptical galaxies
have axis ratios of about 3:5 and that spherical
galaxies are rare [for example, D. Merritt, in
Morphological and Physical Classification of Galaxies, G. Longo, M. Capaccioli, G. Busarello, Eds.
(Kluwer Academic, Norwell, MA, 1992), p. 309].

4. Such a view of elliptical galaxies is implicit in

some texts, for instance, K. F. Ogorodnikov, Dynamics of Stellar Systems (Pergamon, Oxford,
1965), among others.
5. F. Bertola and M. Capaccioli, Astrophys. J. 200,
439 (1975); J. J. Binney, Mon. Not. R. Astron. Soc.
183, 501 (1978).
6. It is difficult to understand in retrospect why the
observation of low rotational velocities was very
surprising because it had been known for many
years that nonrotating oblate models could be
constructed with any degree of flattening with the
use of the two classical integrals in axisymmetric
potentials [see, for example, G. G. Kuzmin and S.
A. Kutuzov, Bull. AbastumaniAstrophys. Obs. 27,
82 (1962); D. Lynden-Bell, Mon. Not. R. Astron.
Soc. 123, 447 (1962)].
7. The equations that describe the evolution of a
nonneutral plasma free of magnetic fields are
more similar to those of self-gravitating systems;
see R. C. Davidson, Physics of Nonneutral Plasmas (Addison-Wesley, Reading, MA, 1990).
8. Most treatments of plasma phenomena are only
valid on length scales that are long compared to
the Debye length, scales at which it is common to
assume that the plasma is homogeneous. The
gravitational analog to the plasma Debye length is
the Jeans length, which is roughly equal to the
size of a galaxy. This is one reason why the results
of plasma physics cannot usually be carried over
into galactic dynamics.
9. For a recent review, see J. Sellwood, Annu. Rev.
Astron. Astrophys. 25, 151 (1987).
10. A. J. Lichtenberg and M. A. Lieberman, Regular
and Stochastic Motion (Springer-Verlag, New

York, 1989).
11. P. Stackel, Math. Ann. 35, 91 (1890).
12. G. G. Kuzmin, in Dynamics of Galaxies and Clusters, T. B. Omarov, Ed. (Akademie Nauk Kazakh
SSR, Alma Ata, 1973), p. 71. Kuzmin's discovery
appeared before the realization that elliptical galaxies were slowly rotating. His work was not
translated into English until 1987. In the meantime, P. T. de Zeeuw and D. Lynden-Bell [Mon.
Not. R. Astron. Soc. 215, 713 (1985)] independently rediscovered the Stackel nature of Eq. 1,
which they christened "The Perfect Ellipsoid."
13. M. Schwarzschild, Astrophys. J. 232, 236 (1979);
ibid. 263, 599 (1982).
14. T. S. Statler, ibid. 321, 113 (1987).
15. S. Aarseth and J. J. Binney, Mon. Not. R. Astron.
Soc. 185, 227 (1978); R. H. Miller, Astrophys. J.
223, 122 (1978); A. Wilkinson and R. A. James,
Mon. Not. R. Astron. Soc. 199, 171 (1982); J.
Barnes, Astrophys. J. 393, 484 (1992).
16. Triaxiality has certain observable consequences,
such as an apparent twisting of the isophotes in
galaxies for which the axis ratios vary with radius.
The observational evidence for triaxiality is only
moderately convincing; see the reviews by F.
Bertola [in Morphological and Physical Classification of Galaxies, G. Longo, M. Capaccioli, G.
Busarello, Eds. (Kluwer Academic, Norwell, MA,
1992), pp. 115-125], M. Franx [in ibid., pp. 2338], and D. Merritt [in ibid., pp. 309-320].
17. D. A. Allen, A. R. Hyland, T. J. Jones, Mon. Not. R.
Astron. Soc. 204, 1145 (1983); J. L. Tonry, Astrophys. J. 322, 632 (1987).
18. E. S. Light, R. E. Danielson, M. Schwarzschild,
Astrophys. J. 194, 257 (1974).
19. 0. E. Gerhard and J. J. Binney, Mon. Not. R.
Astron. Soc. 216, 467 (1985); J. Miralda-Escud6
and M. Schwarzschild, Astrophys. J. 339, 752

(1989).

20. T. de Zeeuw, R. Peletier, M. Franx, Mon. Not. R.

Astron. Soc. 221, 1001 (1986).
21. D. Merritt and L. Hernquist, Astrophys. J. 376, 439

(1991).

22. S. Udry, "Nbody Equilibrium Figures of EarlyType Galaxies," Geneva Observatory preprint, 21
April 1992.
23. In fact, the first self-consistent triaxial model had a
decidedly peanutlike shape; see (13).
24. R. Bender, S. Dobereiner, C. MIllenhoff, Astron.
Astrophys. Suppl. Ser. 74, 385 (1988).
25. Even though there is no dynamical basis for the

--

.1

I.

preference of ellipsoidal shapes, it is commonly

assumed that elliptical galaxies are naturally ellipsoidal and that any deviations of the isophotes
from ellipses can be attributed to some special
mechanism. Boxy elliptical galaxies are generally
interpreted to be systems that have undergone
mergers or other interactions [see J. J. Binney
and M. Petrou, Mon. Not. R. Astron. Soc. 214, 449
(1985)], and disky isophotes are taken as evidence for imbedded disks [H. W. Rix and S.
White, Astrophys. J. 362, 52 (1990)]. Observational support for both interpretations has been adduced [R. Bender, P. Surma, S. Dobereiner, C.
Mollenhoff, R. Madejsky, Astron. Astrophys. 217,
35 (1989)].
26. The radial-orbit instability was first postulated in
1973 by V. A. Antonov [in Dynamics of Galaxies
and Clusters, T. B. Omarov, Ed. (Akademie Nauk
Kazakh SSR, Alma Ata, 1973), p. 139] in the same
volume in which G. G. Kuzmin's discovery of the
integrable triaxial mass models first appeared.
Antonov's proof is probably not valid; it seems to
apply to the stability of a single radial orbit, rather
than to collective instability. Nevertheless, the
instability is real, as shown by the Nbody integrations of J. Barnes [in The Dynamics of Star Clusters, Proceedings of the 11 3th Symposium of the
International Astronomical Union, Princeton, NJ,
29 May-1 June 1984; J. Goodman and P. Hut,
Eds. (Kluwer Academic, Norwell, MA, 1985), p.
297] and D. Merritt and L. A. Aguilar [Mon. Not. R.
Astron. Soc. 217, 787 (1985)] and the numerical
normal-mode calculations of P. Saha [ibid. 248,
494 (1991)] and M. Weinberg [Astrophys. J. 368,
66 (1991)].
27. D. Merritt, Astrophys. J. 319, 55 (1987).
28. L. Aguilar and D. Merritt, ibid. 354, 33 (1990).
29. A. Toomre, in Notes on the 1966 Summer Study
Program in Geophysical Fluid Dynamics at the
Woods Hole Oceanographic Institution (1966),
pp. 111-114.
30. E. N. Parker, Phys. Rev. 109, 1874 (1958). A
better analogy is with beads sliding along a string;
the centrifugal force of the beads as they follow
the curvature of the string tends to increase the
amplitude of the wave. In a sufficiently flattened
stellar system, the gravitational force acts as the
analog of the force that confines the beads to the
string.
31. A. M. Fridman and V. L. Polyachenko, Physics of
Gravitating Systems, A. B. Aries and 1. N. Poliakoff, Transl. (Springer-Verlag, New York, 1984),
vol. 2, p. 159.
32. N. Raha, J. A. Sellwood, R. A. James, F. D. Kahn,
Nature 352, 411 (1991).
33. For a recent review, see S. Tremaine, Phys. Today
45, 28 (February 1992).
34. One way to constrain the shapes of dark halos is
through studies of the so-called polar-ring galaxies, disk galaxies that have a ring of gas orbiting
in a plane perpendicular to that of the stellar disk.
By comparing the orbital velocities of the stars
and gas in the two disks, one can place limits on
the flattening of the gravitational potential that is
produced by the dark halo. The most careful
study to date of a polar-ring galaxy [P. Sackett
and L. Sparke, Astrophys. J. 361, 408 (1990)]
concludes that only weak constraints can be
placed on the elongation of its halo. Simulations of
the formation of structure in the universe, based
on the cold, dark-matter hypothesis, suggest that
most dark halos should be nearly prolate and very
elongated [J. Dubinski and R. Carlberg, Astrophys. J. 378, 496 (1991)].
35. T. de Zeeuw, in Morphological and Physical Classification of Galaxies, G. Longo, M. Capaccioli, G.
Busarello, Eds. (Kluwer Academic, Norwell, MA,
1992), p. 139.
36. H. Dejonghe and D. Merritt, Astrophys. J. 391,
531 (1992).
37. A way to avoid the problem of velocity anisotropy
is for one to use hot gas to gauge the gravitational
potential, by measuring the dependence of gas
pressure on radius and applying the equation of
hydrostatic equilibrium. Instrumental shortcomings have so far strongly limited the efficacy of this

SCIENCE * VOL. 259 * 26 MARCH 1993

91 'I

"'

38.

39.
40.
41.
42.

technique [G. Fabbiano, Annu. Rev. Astron. Astrophys. 27, 87 (1989)].

0. Gerhard, "Line-of-Sight Velocity Profiles in
Spherical Galaxies," preprint, Landessternwarte
Heidelberg, 13 June 1992; D. Merritt, "Dynamical
Mapping of Hot Stellar Systems," Rutgers Astrophysics preprint No. 131, 1 July 1992.
S. C. Barden, Ed. Fiber Optics in Astronomy, vol.
3 of Astron. Soc. Pacific Conf. Series (Astronomy
Society of the Pacific, San Francisco, CA, 1988).
R. Bender, Astron. Astrophys. 229, 441 (1990).
W. L. W. Sargent et al., Astrophys. J. 221, 731
(1978); P. Young et al., ibid., p. 721.
A. Dressier and D. 0. Richstone, ibid. 348, 120

(1 990).

,_ ibid. 324, 701 (1988); J. Kormendy, ibid.

335, 40 (1988); __ and D. 0. Richstone, ibid.
393, 559 (1992).
44. 0. E. Gerhard, in Physics of Active Galactic
Nuclei, W. J. Duschl and S. J. Wagner, Eds.
(Springer-Verlag, Berlin, 1992), p. 273.
45. The Hubble Space Telescope was supposed to
have settled this question by providing kinematical information, at very high spatial resolution,
about the nuclei of nearby galaxies. The spherical
aberration in the primary mirror has so far precluded such observations. Aberration-corrected
images from the Hubble Space Telescope do
provide somewhat higher resolution than groundbased telescopes, and, with the combination of
stellar luminosity profiles from the Hubble Space
Telescope and ground-based kinematical data,
the case for massive nuclear black holes in two
galaxies can be strengthened somewhat [T.
Lauer et al., Astron. J. 103, 703 (1992); T. Lauer et
al., ibid. 104, 552 (1992)].
46. G. de Vaucouleurs, Ann. Astrophys. 11, 247
(1948).
47. __ and M. Capaccioli, Astrophys. J. Suppl.
Ser. 40, 699 (1979).
48. The Boltzmann entropy, S = - f fin fd3xd3v, is
precisely conserved in a collisionless stellar system. It is nevertheless common to assume that
something like the entropy is maximized during
the chaotic period of galaxy formation. However,
the maximization of S subject to fixed values of the
total mass and energy leads to a stellar distribution function that is incompatible with finite mass
and energy [(4), p. 302]; thus, stellar systems
admit no configuration of stationary entropy.
49. T. S. van Albada, Mon. Not. R. Astron. Soc. 201,
939 (1982); A. May and T. S. van Albada, ibid.
209,15 (1984). The "charming" oxymoron violent
relaxation was coined by D. Lynden-Bell [ibid.
136, 101 (1967)] to describe the formation of a
galaxy in a rapidly varying gravitational field.
50. D. Merritt, S. D. Tremaine, D. Johnstone, ibid. 236,
829 (1989); D. Spergel and L. Hernquist, Astrophys. J. 397, L75 (1992).
51. Examples are (21); (49); R. B. Larson, Mon. Not.
R. Astron. Soc. 166, 585 (1974); and D. Merritt
and M. Stiavelli, Astrophys. J. 358, 399 (1990).
52. A. Toomre, in The Evolution of Galaxies and
Stellar Populations, B. M. Tinsley and R. B. Larson, Eds. (Yale University Observatory, New Haven, CT, 1977), p. 401. Perhaps the first clear
statement of the merger hypothesis was in A.
Toomre and J. Toomre, Astrophys. J. 178, 623
(1972).
53. S. Tremaine, in The Structure and Evolution of
Normal Galaxies, S. M. Fall and D. Lynden-Bell,
Eds. (Cambridge Univ. Press, Cambridge, 1981),
p. 67.
54. F. Schweizer, Astrophys. J. 252, 455 (1982); D. F.
Malin and D. Carter, ibid. 272, 534 (1983); R.
Bender, Astron. Astrophys. 202, L5 (1988).
55. F. Schweizer, in Structure, Dynamics, and Chemical Evolution of Early-Type Galaxies, J. Danziger,
Ed. (European Southern Observatory, Garching,
Germany, in press).
56. thank the referees, A. Toomre and S. Tremaine,
for their criticisms and suggestions. J. Bronzan, J.
Sellwood, S. Udry, and H. Zapolsky made numerous comments. thank J. Bedke and T. Statler for
providing Figs. 1 and 2. Supported by NSF grant
AST 90-16515.

43.

1871

Downloaded from www.sciencemag.org on March 25, 2013

1-11