Mon. Not. R. Astron. Soc.

390, 7186 (2008)

Printed 6 December 2008

(MN LATEX style file v2.2)

Measuring the inclination and mass-to-light ratio of axisymmetric

galaxies via anisotropic Jeans models of stellar kinematics
Michele Cappellari
Sub-Department of Astrophysics, University of Oxford, Denys Wilkinson Building, Keble Road, Oxford, OX1 3RH

arXiv:0806.0042v2 [astro-ph] 6 Dec 2008

Accepted 2008 July 23. Received 2008 July 17; in original form 2008 May 30

ABSTRACT

We present a simple and efficient anisotropic generalization of the semi-isotropic (twointegral) axisymmetric Jeans formalism which is used to model the stellar kinematics of
galaxies. The following is assumed: (i) a constant mass-to-light ratio M/L and (ii) a velocity
ellipsoid that is aligned with cylindrical coordinates (R, z) and characterized by the classic
2 . Our simple models are fit to SAURON integral-field
anisotropy parameter z = 1 vz2 /vR
observations of the stellar kinematics for a set of fast-rotator early-type galaxies. With only
two free parameters (z and the inclination) the models generally
provide remarkably good
descriptions of the shape of the first (V ) and second (Vrms V 2 + 2 ) velocity moments,
once a detailed description of the surface brightness is given. This is consistent with previous findings on the dynamical structure of these objects. With the observationally-motivated
assumption that z >
0, the method is able to recover the inclination. The technique can be
used to determine the dynamical mass-to-light ratios and angular momenta of early-type fastrotators and spiral galaxies, especially when the quality of the data does not justify more
sophisticated modeling approaches. This formalism allows for the inclusion of dark matter,
supermassive black holes, spatially varying anisotropy, and multiple kinematic components.
Key words: galaxies: elliptical and lenticular, cD galaxies: evolution galaxies: formation
galaxies: kinematics and dynamics galaxies: structure

1 INTRODUCTION
According to the theory that best reproduces the observations,
galaxy formation proceeds in a hierarchical fashion, driven by gravity, in a Universe dominated by dark matter of unknown nature
(e.g. Springel et al. 2005). The hierarchy of merging is accompanied by changes in galaxy structure and morphology. In particular
early-type galaxies (Es and S0s) are thought to form by the gas-rich
merging of spiral galaxies or by gas starvation of spirals, followed
by subsequent collisionless mergers (e.g. Faber et al. 2007).
Three key global parameters can be used to characterize galaxies structure while studying this sequence of merging of galaxies
and dark matter: (i) The angular momentum, which varies during mergers and increases with the amount of gas dissipation, (ii)
the stellar population, which records the history of star formation
events during the gas-rich mergers, and (iii) the mass-to-light ratio, which is affected by both the population and by the dark-matter
fraction.
The large majority of the galaxies in the Universe are to
first order axisymmetric (except for bars) and posses disks.
This includes fast-rotator early-type galaxies (Emsellem et al.
2007; Cappellari et al. 2007) and spiral galaxies. Both the fast-

E-mail: [email protected]

c 2008 RAS

rotator early-types (Gerhard et al. 2001; Rusin & Kochanek 2005;

Cappellari et al. 2006; Koopmans et al. 2006; Thomas et al. 2007)
and the spiral galaxies (Persic et al. 1996; Palunas & Williams
2000; Bell & de Jong 2001; Kassin et al. 2006) appear dominated by the stellar matter inside one half-light radius.
Observations suggest that they have a dynamical structure
characterized by a flattening of the velocity ellipsoid in the z
direction parallel to the galaxy symmetry axis (Gerssen et al.
1997, 2000; Shapiro et al. 2003; Cappellari et al. 2007;
Noordermeer, Merrifield, & Aragon-Salamanca 2008).
The goal of this paper is to include the knowledge of the structure of the fast-rotator and spiral galaxies, into a realistic but simple dynamical modeling method, which can be applied to the measurement of both the mass-to-light ratio and the amount of rotation
(for which the inclination is needed) in the central regions of these
galaxies, while also allowing for the inclusion of dark matter and
the study of multiple kinematical components. The success of the
adopted models assumptions in describing the kinematics of real
galaxies is verified against integral-field observations of the stellar
kinematics obtained with SAURON (Bacon et al. 2001).
In Section 2 we briefly review the theory and past applications
of the Jeans equations and of the shape of the velocity ellipsoid in
galaxies. In Section 3 we describe our new anisotropic Jeans for-

M. Cappellari

malism, which we apply and test in Section 4. Finally our results

are summarized in Section 5.

2 SOLVING THE AXISYMMETRIC JEANS EQUATIONS

2.1 The collisionless Boltzmann equation
The positions x and velocities v of a large system of stars can
be described by the distribution function (DF) f (x, v). When the
system is in a steady state under the gravitational influence of a
smooth potential , the DF must satisfy the fundamental equation
of stellar dynamics, the steady-state collisionless Boltzmann equation (Binney & Tremaine 1987, hereafter BT; equation [4-13b])

3
X
f
f
= 0.
(1)

vi
xi
xi vi
i=1
Given that f is a function of six variables, an infinite family of solutions satisfies equation (1). Additional assumptions and simplifications are required for a practical application of the equation. One
classic way of constraining the problem consists of drastically reducing it, from that of recovering the DF, to that of studying only
the velocity moments of the DF. This approach leads to the Jeans
equations, which are discussed in the next section.

2.2 The Jeans equations

2.2.1 Summary of derivation
If we rewrite equation (1) in standard cylindrical coordinates
(R, z, ) and we make the important assumption of axial symmetry
(/ = f / = 0) we obtain (cf. BT equation [4-17])
!
v2
f
f f vR v f

f
vR

+vz
+

= 0.(2)
R
z
R
R vR z vz
R v
Multiplying equation (2) respectively by vR and by vz , and integrating over all velocities, we obtain the two1 Jeans equations (Jeans
1922; BT equation [4-29a,c])
2
2
vR
v2
(vR
)
(vR vz )
+
+
R
R
z
2
vR vz
(vz )
(vR vz )
+
+
R
z
R

where we use the notation

Z
vk vj vk vj f d3 v.

=
=

.
z

2.2.2 Closing the axisymmetric Jeans equations

Given that the axisymmetric Jeans equations relate four functions
of the meridional plane (R, z) coordinates, one needs to specify at
least two functions of (R, z) for a unique solution. A natural way to
provide a unique solution for the Jeans equations consists of specifying the shape and orientation of the intersection of the velocity
ellipsoid everywhere in the meridional plane. In fact the three com2
ponents vR
, vz2 and vR vz , of the velocity dispersion tensor, uniquely
describe the equation of the velocity dispersion ellipse, whose shape
can be derived by diagonalizing the tensor. The velocity dispersion
ellipse will have the major axis at an angle with respect to the
equatorial plane
tan 2 =

2 vR vz

(6)

2
vR
vz2

and an axial ratio 0 q 1 given by

q
2
2
vR
+ vz2 (vR
vz2 )2 + 4 vR vz 2
2
q
q =
.
2
2
vR
+ vz2 + (vR
vz2 )2 + 4 vR vz 2

(7)

The specification of the ellipse geometry ( and q) is then sufficient

to write two of the three variables as a function of the remaining
one.
The most common choice which is made consists of assuming2 a circular velocity-ellipsoid in the meridional plane (sometimes called semi-isotropy condition). This assumption implies
2
vR
= vz2 and vR vz = 0, and is sufficient to close the set of
equations to provide a unique solution for the remaining variables
vz2 and v2 . Other common options are discussed in Section 2.4.
A well studied special case of semi-isotropic system is one in
which the 3-integral DF which generally characterizes a stationary
system, is assumed to depend only on the two classical integrals of
motion f = f (E, Lz ), where E is the potential energy, and Lz
is the angular momentum with respect to the symmetry z-axis. For
this reason the semi-isotropic Jeans models are often called twointegral Jeans models.
In this paper we are investigating whether it is possible to make
an alternative assumption on the shape of the velocity ellipsoid,
which retains the simplicity of the widely used semi-isotropic assumption, while also providing a better description of real galaxies
and still leading to an efficient solution of the Jeans equations.

(3)
(4)

(5)

These equations are still quite general, as they derive from the
steady-state Boltzmann equation (1) with the only assumption of
axisymmetry. They do not require self consistency (a potential
generated by the luminosity density ) and they make no assumption on the DF. However, even if one assumes to be known (it
may be derived from the observed via the Poisson equation), the
two equations (3) and (4) are still a function of the four unknown
2
vR
, vz2 , v2 and vR vz and do not uniquely specify a solution.

2.2.3 Past applications of the semi-isotropic Jeans equations

Due to its simplicity, the semi-isotropy assumption has proven
remarkably useful in a large variety of applications of the
Jeans equations. It was used to quantify the amount of rotation
in galaxies (Nagai & Miyamoto 1976; Satoh 1980; Binney et al.
1990; van der Marel et al. 1990), to search for hidden disks in
elliptical galaxies (Rix & White 1992; Cinzano & van der Marel
1994; Rix et al. 1999), to measure their mass-to-light ratio (van der Marel 1991; Statler et al. 1999; Cappellari et al.
2006; Cortes, Kenney, & Hardy 2008) and dark matter profiles
(Carollo et al. 1995), to study the connection between line-strength
in galaxies and the local escape velocities (Davies et al. 1993;
Carollo & Danziger 1994), to study the scatter in the Fundamental Plane (van Albada et al. 1995; Lanzoni & Ciotti 2003;
2

All terms in the third Jeans equation, involving a multiplication by v ,

are zero in axisymmetry.

In the original paper by Jeans (1922) this was not an assumption. Stellar
orbits were thought to conserve only two isolating integrals of motion, in
which case the DF naturally possess that special semi-isotropic form.
c 2008 RAS, MNRAS 390, 7186

2.3 Shape and orientation of the velocity ellipsoid in galaxies

To explore the possibility of making a simple but sufficiently realistic assumption on the shape and orientation of the velocity ellipsoid
in the meridional plane of axisymmetric galaxies, we need to understand what that shape is expected to be.

0.8

0.6
z

Riciputi et al. 2005), to measure the deviations of the gas kinematics from circular velocities (Bertola et al. 1995; Corsini et al.
1999; Cinzano et al. 1999; Young et al. 2008), and to estimate the masses of supermassive black holes (Magorrian et al.
1998; van der Marel et al. 1998; Cretton & van den Bosch 1999;
Joseph et al. 2001). Although these models have been largely superseded by Schwarzschild (1979) orbit-superposition method, when
good kinematic data are available and the maximum generality is required, they are becoming useful to study the mass-tolight ratios and rotation of galaxies at high-redshift, where the
data quality still does not justify more sophisticated approaches
(van der Marel & van Dokkum 2007a,b; ?).

Anisotropic axisymmetric Jeans models

0.4

0.2

0
0.2

0.4

0.6

0.8

R
2.3.1 Theory
A qualitative insight into the orientation of the velocity ellipse in
real galaxies can start from the analysis of the special case of separable potentials (de Zeeuw 1985). In an axisymmetric oblate separable potential the equations of motion for the stellar obits can be
separated in a prolate spheroidal coordinates system (, , ) which
also defines the boundaries of the orbits. In other words the orbital
motions can be written as the linear combination of two independent oscillations in and , plus a non-uniform rotation around the
symmetry axis. For this reason the velocity ellipse is aligned with
the coordinate system and the cross term v v vanishes (Eddington
1915; de Zeeuw & Hunter 1990). The prolate spheroidal coordinates are characterized by the fact that they tend to be aligned with
the cylindrical coordinates (R, z, ) at small radii and with the polar ones (r, , ) at large radii. Separable potentials are characterized by a central constant-density region. The cylindrical alignment
happens in that region, as the stars there tend to move like an harmonic oscillator. For these reasons one expects the velocity dispersion ellipse to be cylindrically aligned at small radii and radially
aligned at large radii.
The gravitational potential of real galaxies is not of the separable form. Separable potentials are in fact necessarily characterized
by smooth analytic centers, while real galaxies possess central singularities due to cusped density profiles (e.g. Ferrarese et al. 1994;
Lauer et al. 1995) and supermassive black holes (Magorrian et al.
1998; Gebhardt et al. 2000; Ferrarese & Merritt 2000). Still, numerical integrations of orbits in non-separable axisymmetric potentials show that most of them are still bounded by curves, which
qualitatively resemble the spheroidal coordinates (e.g. Richstone
1982; Dehnen & Gerhard 1993; Cretton et al. 1999). For this reason one can expect the velocity dispersion ellipse to behave in real
galaxies in a way that is qualitatively similar to the separable case.
Numerical calculations confirm this fact in non-separable triaxial and axisymmetric potentials (Merritt 1980; Dehnen & Gerhard
1993). In the limit of a point mass, the velocity ellipsoid has to be
spherically aligned for symmetry. This suggests that at small radii,
where the supermassive black hole or the stellar cusp dominates,
the velocity ellipsoid in real galaxies should be more spherically
oriented than in the separable case.
c 2008 RAS, MNRAS 390, 7186

Figure 1. Qualitative description of the shape and orientation of the velocity

ellipsoid in the meridional plane (R, z) of the fast-rotator early-type galaxies (derived from Cappellari et al. 2007). The ellipses show the intersection
of the velocity ellipsoid with the (vR , vz ) plane. The solid lines show a
representative prolate spheroidal system of coordinates. The dashed line is
a representative iso-density contour. The ellipsoids are aligned in spheroidal
coordinates, but the axial ratio of the ellipses varies with the polar angle in
such a way that the shape and orientation of the ellipsoids are similar on
both the equatorial plane and the symmetry axis.

2.3.2 Observations
The DF of a stationary system is a function of the three separable
integrals of motion (Jeans 1915). In general it cannot be recovered
from real galaxies without at least another three-dimensional observable quantity. This quantity is now being provided by integralfield observations of the stellar kinematics (e.g. Emsellem et al.
2004), which allow the stellar line-of-sight velocity-distribution to
be measured at every position of the galaxy image on the sky.
We used these observations, in combination with orbit-based threeintegral axisymmetric models, to measure the shape and orientation of the velocity ellipsoid at every position within the meridional
plane (within one half-light radius Re ), for a sample of 25 earlytype galaxies (Cappellari et al. 2007). Fig. 1 qualitatively summarizes the main findings of that paper regarding the shape of the velocity dispersion ellipsoid for the fast-rotator galaxies: (i) The ellipsoids qualitatively behave like in separable potentials and already
near 1Re they are essentially spherically oriented; (ii) the axial ratio of the ellipsoids varies gradually as a function of the polar angle,
in such a way that the ellipsoid has nearly the same shape on both
the equatorial plane (where the density is the highest) and the symmetry axis. The net effect is that to first order the global anisotropy
of the galaxies can be described as a simple flattening of the veloc2
ity ellipsoid in the z-direction (vz2 < vR
).

2.4 Choice of the coordinate system for the Jeans solution

One can think of three physically motivated choices of the coordinates system in which to align the velocity ellipsoid, for a unique

M. Cappellari

solution of the axisymmetric Jeans equations. In this section we examine advantages and problems of each one in turn.
2.4.1 Prolate spheroidal coordinates
The observed behavior of the ellipsoid could be approximated
by solving the Jeans equations in prolate spheroidal coordinates
in a generic axisymmetric potential (Dejonghe & de Zeeuw 1988;
Evans & Lynden-Bell 1991; Arnold 1995; van de Ven et al. 2003).
To qualitatively reproduce the behavior of Fig. 1, the simplest realistic model would need at least four free parameters. Two parameters are required to define the coordinate system. A third parameter could define the shape of the velocity ellipse on the equatorial plane and another parameter would describe the angular variation. A problem of this approach is that the general solution, to derive the observables for realistic galaxy potentials, requires at least
a computationally-expensive triple3 numerical quadrature. The derived equations are also rather cumbersome and not justified by the
fact that these models can at best provide a qualitative description of
real galaxies. For these reasons those solutions have currently only
been applied to a handful of analytic potentials, and no application
to real galaxies exist.
2.4.2 Spherical coordinates
Given that the velocity ellipsoids in Fig. 1 are nearly spherically aligned, a much simpler choice would consist of solving the
anisotropic Jeans equation in spherical coordinates. A solution for
this case was presented by Bacon et al. (1983) and applied to real
galaxies by Bacon (1985) and Fillmore (1986). However even under this spherical assumption the solution still has the same form as
in the spheroidal coordinates. The computation still involves a triple
numerical quadrature. This is in fact a generic feature of the axisymmetric Jeans solution: even in the simple case in which the density is
assumed to be stratified on similar oblate spheroids, one quadrature
is needed to obtain the potential, a second quadrature provides the
intrinsic velocitys second moments and a third quadrature finally
gives the projected observables. In the spherically-oriented case the
coordinates system has no free parameters and the simplest realistic model to reproduce Fig. 1 could have two parameters (anisotropy
and its angular variation).
2.4.3 Cylindrical coordinates
A more radical alternative is to assume the velocity ellipse is
oriented with the cylindrical coordinate system. This option was
tested by e.g. Fillmore (1986), however until now, for the reasons described in Section 2.4.1, it was not considered a sensible choice to describe the shape of the velocity ellipsoid in real
galaxies. Our new results, based on the SAURON integral-field observations (Cappellari et al. 2007), make this choice worth exploring again. This option cannot provide a formally accurate representation of Fig. 1, however it is accurate near the equatorial
plane, where the density is at its maximum, and near the minor
axis. Models with cylindrically-oriented velocity ellipsoid provide
a good qualitative description of the empirical observation that
the global anisotropy in fast-rotator galaxies is best characterized
as a flattening of the velocity ellipsoid in the vertical z-direction

(Cappellari et al. 2007, their Fig. 2). The near-cylindrical orientation of the velocity ellipsoid may be due to the presence of disks,
where this orientation appears natural. For this reason the cylindrical orientation is certainly appropriate to describe the dynamics
of spiral galaxies (Gerssen et al. 1997, 2000; Shapiro et al. 2003;
Noordermeer, Merrifield, & Aragon-Salamanca 2008).
In the next section we show that the assumption of
a cylindrically-aligned velocity-dispersion ellipsoid, combined
with the powerful Multi-Gaussian Expansion (MGE) method of
Emsellem et al. (1994), can generate simple solutions that well reproduce the integral-field kinematics of real galaxies, and also allow
for variable mass-to-light ratios (e.g. dark matter), spatially varying
anisotropy and multiple kinematical components. All this while still
requiring a single numerical quadrature to predict the observables
on the sky plane (or any other projection).

3 ANISOTROPIC JEANS SOLUTIONS

3.1 Axisymmetric case
3.1.1 General solution
We start from the general axisymmetric Jeans equations (3) and (4)
and we make the following two assumptions: (i) the velocity ellipsoid is aligned with the cylindrical coordinate system (R, z, ) and
2
(ii) the anisotropy is constant and quantified by vR
= b vz2 . In this
case the Jeans equations reduce to
b vz2 v2
(b vz2 )

+
=
(8)
R
R
R
(vz2 )

=
,
(9)
z
z
which corresponds to the semi-isotropic case (two-integral) when
b = 1. With the boundary condition vz2 = 0 as z the
solution reads
Z

dz
(10)
vz2 (R, z) =

z
z

(vz2 )
v2 (R, z) = b R
+ vz2 + R
.
(11)
R
R
A general caveat regarding the Jeans equations is that the existence
of a solution does not guarantee the existence of a corresponding
physical positive DF. This can only be verified using different techniques.
3.1.2 Summary of MGE formalism
To derive solutions for the Jeans equations we make an explicit
choice for the parametrization of the stellar density and the total
density (which can include dark matter and a central black hole).
We adopt for both the MGE parametrization of Emsellem et al.
(1994) due to its flexibility in accurately reproducing the surfacebrightness of real galaxies and the availability of robust routines4 to
fit the galaxy photometry (Cappellari 2002). If the x-axis is aligned
with the photometric major axis, the surface brightness at the location (x , y ) on the plane of the sky can be written as

N
X
Lk
y 2
1
2
(x , y ) =
x
+
,
(12)
exp

2k2 qk
2k2
qk2
k=1

For typical accuracies, every additional numerical quadrature generally

increases the computation time by about two orders of magnitude.

Available from http://www-astro.physics.ox.ac.uk/mxc/idl/

c 2008 RAS, MNRAS 390, 7186

Anisotropic axisymmetric Jeans models

where N is the number of the adopted Gaussian components, having total luminosity Lk , observed axial ratio 0 qk 1 and dispersion k along the major axis.
The deprojection of the surface brightness to obtain the intrinsic luminosity density is not unique unless the axisymmetric galaxy
is seen edge-on (i = 90 ) (Rybicki 1987; Kochanek & Rybicki
1996), and the degeneracy becomes especially severe when the
galaxy is seen at low inclinations (Gerhard & Binney 1996;
Romanowsky & Kochanek 1997; van den Bosch 1997; Magorrian
1999). The MGE method provides a simple possible choice for the
deprojection (Monnet et al. 1992). One of the advantages of the
MGE method is that one can easily enforce the roundness of the
model (Cappellari 2002), thus producing realistic densities, which
look like real galaxies when projected at any angle. The method
cannot eliminate the intrinsic degeneracy of the deprojection and
this has to be considered when interpreting results of galaxies that
are close to face-on. The deprojected MGE oblate axisymmetric luminous density can be written as

N
X
Lk
z2
1
2

(R, z) =
,
(13)
exp 2 R + 2
2k
qk
( 2 k )3 qk
k=1
where the individual components have the same luminosity Lk and
dispersion k as in the projected case (12), and the intrinsic axial
ratio of each Gaussian becomes
p 2
qk cos2 i
qk =
,
(14)
sin i
where i is the galaxy inclination (i = 90 being edge-on).
The total density can be generally described by a different
set of M Gaussian components

M
X
Mj
1
z2

(R, z) =
exp 2 R2 + 2
.
(15)
2j
qj
( 2 j )3 qj
j=1
In the self-consistent case the Gaussians are the same as in equation (13) and one has M = N , j = k , qj = qk and Mj = k Lk ,
where k is the mass-to-light ratio, which can be different for different components. In the non-self-consistent case the density can
be described with the sum of two sets of Gaussians: the first derived by deprojecting the surface brightness with equation (13), and
the second e.g. obtained by fitting a (one-dimensional) MGE model
to some adopted analytic parametrization for the dark matter (e.g.
Navarro et al. 1996).
The gravitational potential generated by the density of equation (15) is given by (Emsellem et al. 1994)
Z 1X
M
p
Mj Hj (u)
(R, z) = 2/ G
du,
(16)
j
0 j=1
where G is the gravitational constant and with

2
u2
z2
exp 2 2 R + 1(1q2 )u2
j
j
q
Hj (u) =
.
2
2
1 (1 qj )u

(17)

A supermassive black hole can be modeled by adding a Keplerian

potential
(R, z) =

GM
R2 + z 2

(18)

to equation (16). However, as pointed out by Emsellem et al.

(1994), an even simpler approach consists of modeling the black
hole as a small Gaussian in equation (16), having mass Mj = M ,
c 2008 RAS, MNRAS 390, 7186

qj = 1 and 3j . rmin , where rmin is the smallest distance from

the black hole that one needs to accurately model (e.g. one could
choose rmin psf ).
3.1.3 MGE Jeans solution
Now we apply the MGE formalism to the solution of the axisymmetric anisotropic Jeans equations of Section 3.1.1. Our derivation is an extension of what was done in the semi-isotropic selfconsistent case (bk = 1 and Mj = Lk ) in section 3.4 of
Emsellem et al. (1994). As already pointed out by Jeans (1922), his
equations can be used to model the kinematics of different dynamical tracers, as long as they all move in the same potential (e.g.
Rix & White 1992; Cinzano & van der Marel 1994). To maintain
generality we will then write the solution for the individual N luminous Gaussian components, which can then be assumed to have
different anisotropy. This fact can be used e.g. to model anisotropy
gradients, or to study the anisotropy of kinematical subcomponents
in galaxies (e.g. bulge and disk). Substituting equations (13) and
(16) into equations (10) and (11), the integral in z can be performed
analytically and we obtain
2
[vR
]k = bk [vz2 ]k
Z 1X
M
k2 qk2 k qj 0j Hj (u) u2
[vz2 ]k = 4G
du
1 Cu2
0 j=1

[v2 ]k = bk [vz2 ]k
Z 1X
M
k qj 0j Hj (u) u2
D R2 du
+ 4G
1 Cu2
0 j=1
= 4G

M
1X
j=1

(19)
(20)

(21)

k qj 0j Hj (u) u2 `
D R2 + bk k2 qk2 du,
1 Cu2

where we defined k = k (R, z), 0j = j (0, 0) and

k2 qk2
j2

D = 1 bk qk2 (1 bk ) C + (1 qj2 ) bk u2 .
C = 1 qj2

(22)
(23)

In all the equations of this paper the index k refers to the parameters,
or the anisotropy, of the Gaussians describing the galaxys luminosity density (equation [13]), while the index j refers to the parameters of the Gaussians describing the total mass (equation [15]), from
which the potential is obtained.
When bk is not the same for the individual luminous Gaussians, the total luminosity-weighted anisotropy at a certain spatial
location (R, z) of an MGE model is given by the standard definition
(Binney & Mamon 1982), combined with equation (19):
P
P
[vz2 ]k
k
v2
1 P k
z (R, z) 1 z = 1 P k
. (24)
2
2
vR
k bk k
k bk [vz ]k

The last approximation comes from the fact that [vz2 ]k , being mostly
a function of the total MGE potential, varies relatively little for the
different Gaussians, while k can be completely different and varies
by many orders of magnitude for the various luminous MGE components. This allows the global anisotropy of an MGE model, at a
certain spatial location in the meridional plane, to be approximately
estimated from a simple luminosity-weighted sum of bk . A similar
reasoning applies to the estimation of the total parameter in in
Section 3.1.5 and the total parameter in Section 3.2.2.

M. Cappellari

3.1.4 Line-of-sight integration of the velocity second moment

The intrinsic quantities have to be integrated along the line-of-sight
(LOS) to generate the observables which can be compared with the
galaxy kinematics. For this we define a system of sky coordinates
with the z axis along the LOS and the x axis aligned with the
galaxy projected major axis (see equation [12]). The galaxy coordinates (x, y, z) are related to the ones in the sky plane by
10 1
1 0
0
x
1
0
0
x
@ y A = @ 0 cos i sin i A @ y A ,
(25)
0
sin i
cos i
z
z

where the z-axis coincides with the galaxy symmetry axis and the
cylindrical radius is defied by R2 = x2 + y 2 . The projected sec2
ond velocity moment along the line-of-sight vlos
vz2 , for one
luminous Gaussian component, is then given by5
Z n
2
[ vlos
]k =
[vz2 ]k cos2 i

2
]k sin2 + [v2 ]k cos2 sin2 i dz ,
(26)
+ [vR
where cos = x/R, while the total observed second moment, for
the whole MGE model, will be
2
vlos
=

N
X

2
[ vlos
]k .

(27)

k=1

After substitution of equations (19)(21), the z integral can be written explicitly. Summing over all the N luminous Gaussian components we obtain the final expression
Z 1X
N X
M
2
vlos
(x , y ) = 4 3/2 G
0k qj 0j u2
0

k=1 j=1

k2 qk2 cos2 i + bk sin2 i + D x2 sin2 i

q

(1 Cu2 ) (A + B cos2 i) 1 (1 qj2 )u2

(A + B)y 2
exp A x2 +
du,
A + B cos2 i

where we defined 0k = k (0, 0) and

1
1 u2
+
A=
2 j2
k2
(
)
(1 qj2 )u4
1 1 qk2

B=
+ 2
2
k2 qk2
j 1 (1 qj2 )u2

holes (representing the point mass with a small gaussian as described in the last paragraph of Section 3.1.2).
Simple expressions, involving a single quadrature, can be derived for the second velocity moments also for the case of the Keplerian potential (18) of a black hole, as done in the semi-isotropic
case in appendix A of Emsellem et al. (1994). The black hole moments can then be quadratically co-added to the ones in equation (28) to obtain the observed velocity moments for the galaxy.
We found no advantage in speed or accuracy when performing a
separate calculation for the black hole and the galaxy potential. For
this reason we will then not give separate expressions for the Keplerian case. Using equation (28) with a black hole it is however
important to use a quadrature routine which samples the integrand
function at a sufficiently high number of initial points, to properly
recognize the sharp peak in the integrand near u = 0. Here we used
the vectorized adaptive quadrature algorithm of Shampine (2008).
2
The second moments vlos
provided by equation (28) are a good
2
approximation for the observed quantity Vrms
= V 2 + 2 , where
V is the stellar mean velocity and is the velocity dispersion. In
Cappellari et al. (2006) we used realistic semianalytic dynamical
2
models of galaxies and found that, to extract vlos
from the simulated data, one should use a single Gaussian LOSVD and adopt as
V and the mean velocity and dispersion of that Gaussian. Due to
the sensitivity of the second moments to the uncertain wings of the
LOSVD, this approach is preferable than trying to extract a more
complex LOSVD, e.g. by fitting the Gauss-Hermite parametrization (van der Marel & Franx 1993; Gerhard 1993) or a fully non2
parametric LOSVD, and numerically integrate vlos
from that.

3.1.5 Line-of-sight integration of the velocity first moment

The projected first velocity moments vlos vz are given by
Z
vlos =
(31)
v cos sin i dz .

(28)

(29)

(30)

As expected equation (28) reduces to equation (61) of

Emsellem et al. (1994) when bk = 1 and Mj = Lk . As in
the semi-isotropic case this formula is very quick to evaluate as it
still requires a single numerical quadrature and involves no special functions. All this starting directly from a fit to the galaxy
surface-brightness, without the need for numerical deprojection or
PSF deconvolution, as required in other approaches. The formula
has various possible applications, as it can be used to model realistic galaxies with variable anisotropy or multiple kinematic components, dark-matter, variable stellar M/L and supermassive black

In this case, the two assumptions we made in Section 3.1.1 are not
sufficient any more to provide a unique prediction and therefore
additional assumptions are needed. The Jeans equations (8) and (9)
in fact only give a prediction for v2 and one has to decide how
the second moments separate into the contribution of ordered and
random motion, as defined by
v2 = v 2 + 2 .

(32)

This need for extra assumptions on the tangential anisotropy is a

fundamental limitation of the first-moments equations, and it is the
reason why one should fit the more general equation (28) to Vrms ,
and only subsequently fit any extra parameter of the first moment
solution to V , instead of simultaneously fitting the Jeans solutions
to both V and .
The first moment equations however are very useful to quantify the amount of rotation in galaxies and for this reason have been
used in the past (Section 2.2.3). One can think of two natural options to parameterize the separation of random and ordered streaming rotation around the symmetry axis. The first option consists of
assuming a constant anisotropy, for each Gaussian component, in
the (vR , v ) coordinates, analogously to equation (19)

Completely analogue expressions can be found for the proper motions,

using e.g. the formulas in appendix A of Evans & de Zeeuw (1994). We
found that all three components of the projected proper motion dispersion
tensor can be written via single quadratures and using no special functions.
6 As in Cappellari et al. (2006) we corrected the typo in the expression for
B in equation (63) of Emsellem et al. (1994).

2
[2 ]k = ck [vR
]k ,

which would imply

1/2
2
[v ]k = [v2 ]k ck [vR
]k
.

(33)

(34)
c 2008 RAS, MNRAS 390, 7186

Anisotropic axisymmetric Jeans models

A practical limitation of this approach is that the mean velocity
would depend in a non-linear way on the ck parameter. For this reason nearly all previous authors adopted, in the semi-isotropic case,
a second approach, introduced by Satoh (1980). This consists of
defining a constant k which quantifies how much the model velocity
field is scaled with respect to that of the isotropic rotator (|k| = 1).
In this way the first moment can be computed only once, for the
isotropic rotator, and is scaled linearly with k.
We adopt here the analogue, in our anisotropic case, of Satoh
(1980) approach. It has the added advantage of providing a direct
measure of the amount of rotation, which is a more meaningful
quantity than the tangential anisotropy. We define for each Gaussian component

1/2
2
[v ]k = k [v2 ]k [vR
]k
.
(35)

Here k = 0 when the k-th Gaussian component is not rotating

and |k | = 1 when its velocity ellipsoid is a circle in the (vR , v )
plane. If bk = 1 then k reduces to Satoh (1980) parameter and
in this case |k | = 1 implies isotropy (the velocity ellipsoid is a
sphere everywhere). An upper limit to |k | is set by the physical
requirement that 2 > 0 (equation [32]). When summed over all
luminous Gaussian components of the MGE model7
v 2 =

N
X

[v 2 ]k ,

(36)
8

so equation (35) implies

v =

N
X

k=1

2k

[v2 ]k

2
[vR
]k

#1/2

(37)

Substituting equation (37) into equation (31), and using equations (19)(21), we obtain the projected first velocity moment of the
whole MGE model

vlos (x , y ) = 2 G x sin i
(38)
#1/2
Z " Z 1X
N X
M
2k k qj 0j Hj (u) u2 D

du dz .

2
1

Cu

0
j=1
k=1

When both bk = |k | = 1 and Mj = Lk , this equation reduces to a self-consistent isotropic rotator as in equation (59) of
Emsellem et al. (1994). A double quadrature seems unavoidable
here, but when is assumed to be constant for the whole MGE,
this integral has to be evaluated only once with k = 1, at the best
fitting (i, z , ) parameters previously determined from a fit to the
more general second moment equation (28), and then vlos can be
linearly scaled by to fit the data.9

As v 2 is defined as a quadratic difference of second moments it is not

P
correct to write v = N
k=1 [v ]k , as one would normally expect for a
first moment.
8 To allow for counter-rotating Gaussian components, namely to keep track
of the sign of k in equation (37) and (38), one should multiply each
term of the k-summation by sgn(k ) k /|k | and then compute
sgn(w) |w|1/2 , where w = [. . .] is the expression inside the big square
brackets.
9 When the alternative definition for the tangential anisotropy of equation (33) is used instead of equation (35), the term D in equation (38) becomes [D + (1 ck ) bk qk2 k2 /R2 ].
c 2008 RAS, MNRAS 390, 7186

3.2 Spherical case

The cylindrically oriented assumption for the shape of the velocity ellipsoid discussed in Section 3.1 is likely to be unrealistic for
slow-rotator elliptical galaxies. As a class these objects are weakly
triaxial, and indeed some of them show clear twists in their kinematical axes (Kormendy & Bender 1996; Emsellem et al. 2007;
Cappellari et al. 2007; Krajnovic et al. 2008). The most general and
realistic models for these objects are triaxial and orbit- or particlebased (de Lorenzi et al. 2007; van den Bosch, & van de Ven 2008).
However also in the triaxial approximation a degeneracy in the
recovery of the galaxy shape is still generally present so that no
unique solution can be obtained. When one is only interested in
global galaxy quantities, or to test the results of more general models, it is still useful to construct simpler and approximate models.
The isophotes of the slow rotators are in projection close to circular,
especially in their central parts where the kinematics is generally
obtained, and this implies they must be intrinsically not far from
spherical. This suggests that one can use simple spherical models
as a first order approximation to the dynamics of at least some of
these objects. The spherical solution of the anisotropic Jeans equations, in the MGE formalism, will be discussed in the following
sections.
3.2.1 General solution
When the Boltzmann equation (1) is written in spherical coordinates (r, , ), by analogy with our derivation of the axisymmetric Jeans equations one can obtain the Jeans equations in spherical
symmetry (Binney & Mamon 1982; equation [4-54] of BT)

k=1

"

d(vr2 )
2 vr2
d
+
=
,
dr
r
dr

(39)

where v2 = v2 for symmetry and we defined = 1 v2 /vr2 . The

solution of this linear first-order differential equation with constant
anisotropy and the boundary condition vr2 = 0 as r is
given by (e.g. van der Marel 1994)
Z
d(u) 2
vr2 (r) = r 2
(u)
u du
du
r
Z
(u)M (u)
= G r 2
du,
(40)
u22
r
considering that d/dr = GM/r 2 . After projection along the lineof-sight z , we obtain the observed second velocity moment (see
equation [4-60] of BT)

Z
v 2 (r)r
R2
2
vlos
(R) = 2
1 2 r
dr
(41)
r
r 2 R2
R
#
"
Z
Z 12 ` 2
r
r R2
(u)M (u)

du dr,
= 2G
u22
r 2 R2
r
R

where R is the projected radius, measured from the galaxy center.

Integrating by parts the double integral can be reduced to a single
quadrature, involving special functions
Z
F(r)(r)M (r)
2
vlos
(R) = 2G
dr,
(42)
r 22
R
where

1 1
1 1
R12
Bw + ,
Bw ,
2
2 2
2 2
`
#
3
1
( 2 ) 2
,
+
()

F(r) =

(43)

M. Cappellari

with w = (R/r)2 , is the Gamma function and Bw is the incomplete Beta function (equation [6.6.1] of Abramowitz & Stegun
1965). In the isotropic limit
p
lim F(r) = r 2 R2 ,
(44)

(k ) for each MGE Gaussian with dispersion k . This simple

method works because in a spherical MGE each k-th Gaussian produces a significant contribution to the total luminosity density only
near its k (Cappellari 2002; see also equation [24]).

and equation (42) reduces to the spherical isotropic form of equation (29) of Tremaine et al. (1994). Some numerical implementations provide the incomplete Beta function only for positive arguments (e.g. Press et al. 1992), for negative values ( < 1/2) one
can use its analytic continuation via Gausss Hypergeometric function 2 F1 (equation [6.6.8] of Abramowitz & Stegun 1965)
Bw (a, b) =

wa
2 F1 (a, 1 b; a + 1; w),
a

3.3 Availability
The Jeans Anisotropic MGE (JAM) package of IDL 11 procedures,
providing a reference implementation for the equations described
in this section, together with other routines to evaluate auxiliary
quantities, like the circular velocity, from MGE models, is available
online from http://www-astro.physics.ox.ac.uk/mxc/idl/.

(45)

for which efficient routines10 exist (Shanjie & Jianming 1996). For
= 1/2, the Bw function is divergent, but for all practical purposes we found it is sufficient to perturb the value by a negligible
amount to avoid the singularity.

3.2.2 MGE spherical Jeans solution

We proceed as in Section 3.1.3 to derive an explicit solution for the
spherical Jeans equation using the MGE parametrization for both
the luminosity density and the total density. The expressions for the
surface brightness, luminosity density and total density are as in
equation (12), (13) and (15), with qj = qk = 1 (Bendinelli 1991).
For a single Gaussian component they read

Lk
R2
k (R) =
exp

,
(46)
2k2
2k2

r2
Lk
(47)
exp 2
k (r) =
2k
( 2 k )3

r2
Mj
exp 2 .
(48)
j (r) =
2j
( 2 j )3
The mass of a Gaussian contained within the spherical radius r is
q
2
3

2
r

r
r
Mj (r) = Mj 4erf

exp 2 5 ,
(49)
j
2j
2 j

with erf(x) the error function (equation [7.1.1] of

Abramowitz & Stegun 1965). Using equation (42) and summing over all Gaussian components with equation (27), we obtain
the projected second velocity moment for the whole MGE model
#
"
Z X
M
N
X
Fk (r) k (r)
2
vlos (R) = 2G
Mj (r) dr, (50)
M +
r 22k
R k=1
j=1

where k (r) and Mj (r) are given by equation (47) and (49) respectively, and Fk (r) is obtained by replacing the parameter in
equation (43) with the anisotropy k of each luminous Gaussian
component. We explicitly included the mass M of a central supermassive black hole. As in the axisymmetric case, this formula involves a single numerical quadrature. It can be used to model nearly
spherical objects with a variable anisotropy profile, variable stellar
M/L, a supermassive black hole and dark matter.
An approximate way to construct a model with a certain
smooth radial profile of anisotropy (r), consists of defining k =

4 APPLICATIONS AND TESTS

The formalism derived in Sections 3.1.4 and 3.1.5 generalizes to
anisotropy the widely used semi-isotropic formalism for the solution of the Jeans equations while maintaining its simplicity. It
was motivated by our modeling results on the orbital distribution
of a sample of real galaxies (Cappellari et al. 2007). However the
method would be of limited usefulness if it did not describe the
dynamics of fast-rotator early-type galaxies well. In this section
we show that the simple assumptions we made provide a remarkable good description of the main features of the kinematics12
(Emsellem et al. 2004) of the fast rotator early-type galaxies, as observed with the SAURON integral-field spectrograph (Bacon et al.
2001) as part of the SAURON survey (de Zeeuw et al. 2002).
4.1 Comparison with more general models
4.1.1 Anisotropy comparison
In this section we compare the anisotropy derived from the Jeans
models, with the results for the global anisotropy obtained with the
more general Schwarzschild (1979) method on the same galaxies
and from the same data in Cappellari et al. (2007). We use equation (28), with constant anisotropy bk = b and constant mass-tolight ratio for the whole MGE model, to predict the velocity
second moments. Following standard practice we parameterize the
2
anisotropy with the variable z 1 vz2 /vR
= 1 1/b. The
problem becomes a function of two nonlinear variables (i, z ) and
of the scaling factor . To further limit the number of free variables
in the model, and to eliminate the degeneracy in the deprojection of
the surface brightness, we select for the comparison the galaxies for
which the photometry already constrains the inclination to be close
to edge-on (i = 90 ). For this we select from the MGE models
in Table B1-B2 of Cappellari et al. (2006), the fast-rotator galaxies
for which the flattest Gaussian has q < 0.35, which forces a strict
limit on the inclination i > 70 . We further exclude the galaxy
NGC 4550, due to the presence of two counter-rotating stellar disks,
which complicate the interpretation of the models, and NGC 4526,
which has a strong dust disk affecting the observed stellar kinematics. This selection leads to the four galaxies NGC 821, NGC 3377,
NGC 4621 and NGC 5845.
We evaluated the MGE model predictions given by equation (28), adopting i = 90 as in Cappellari et al. (2007), for a
direct comparison with the anisotropy determinations. The model
predictions were convolved with the PSF and integrated over the
11

10

Available from http://jin.ece.uiuc.edu/routines/routines.html

12

http://www.ittvis.com/idl/
Available from http://www.strw.leidenuniv.nl/sauron/.
c 2008 RAS, MNRAS 390, 7186

Anisotropic axisymmetric Jeans models

Figure 2. Data-model comparison for the second velocity moments of edge-on galaxies. From left to right, thecolumns show the galaxies NGC 821, NGC 3377,
NGC 4621 and NGC 5845. The top row visualizes the bi-symmetrized SAURON observations for Vrms = V 2 + 2 . The subsequent rows show the model
2 )1/2 as given by equation (28) for = 0 (semi-isotropic model; second row) and anisotropic models with = 0.1, 0.2 and 0.3
predictions for (vlos
z
z
respectively. The color levels are the same for the five panels in each column. NGC 4621 appears well reproduced by a weakly anisotropic model z 0.1.
NGC 812 and NGC 5845 are well described by a model with z 0.2, while only the strongest anisotropy z 0.3 can qualitatively describe the observed
shallow gradient of Vrms along the projected minor axis of NGC 3377. These estimates are more accurately quantified in Fig. 3.

pixels, before comparison with the observables, as described in Appendix A. For each galaxy we computed the model predictions for
2 1/2
(vlos
)
inside each Voronoi bin on the sky, at different anisotropy
values z . At every anisotropy the best fitting mass-to-light ratio
2
as a linear
is obtained from the simple scaling relation vlos
least-squares fit. The best-fitting scaling factor for the model velocities is s = cos |d|/|m|, where is the cosine of the angle
between the data d and model m vectors, which implies:13

2
dm
=
,
(51)
mm
where the vectors d and m have elements dn = [Vrms /Vrms ]n
2
2 1/2
and mn = [(vlos
) /Vrms ]n , with [Vrms ]n
Vn + n2
the measured values for the P Voronoi bins from Emsellem et al.
2 1/2
(2004) and [Vrms ]n the corresponding errors, while [(vlos
) ]n
are the model predictions for = 1. A qualitative comparison between the observed velocity second moments and the model ones,
for different values of the anisotropy parameter z is shown in
Fig. 2.
To ease the visual comparison, given that the models are
bi-symmetric by construction, the observations have been bisymmetrized with respect to the kinematical major axis PAkin given
in Cappellari et al. (2007). The actual fits and 2 calculations are
always performed using the original data and errors. It is clear that
the isotropic models (z = 0) do not provide a good description

13

We corrected a typo in the corresponding equation (2) of Cappellari et al.

(2006).
c 2008 RAS, MNRAS 390, 7186

Figure 3. Best-fitting Jeans anisotropy. 2 = 2 2min , describing the agreement between the data and the models, is plotted against the
anisotropy parameter z . The four panels show the galaxies NGC 821,
NGC 3377, NGC 4621 and NGC 5845. The solid line with open circles represent the 2 of the Jeans models. The dashed horizontal line indicates
the level 2 = 9, which corresponds to the 3 confidence level for one
degree-of-freedom. The filled square with error bars shows the value of the
global anisotropy measured using Schwarzschild models in Cappellari et al.
(2007).

10

M. Cappellari

for the observed second velocity moments of any of the four galaxies14 , the disagreement being strongest for NGC 3377 and weakest for NGC 4621. Using the models presented in this paper we
can now move away from the isotropic assumption and quantify
that the model of NGC 4621 requires a small amount of anisotropy
to appear qualitatively like the data: any z >
0.1 produces in fact
two vertical lobes along the minor axis, which are not observed.
Both NGC 821 and NGC 5845 require a more significant anisotropy
2 1/2
z 0.2 to reduce the amount of rotation (high [vlos
] ) along
the major axis to the level of the observations. A larger z 0.3 is
strongly excluded by the data, as the model immediately produces a
significant elongation along the minor axis, which is not observed.
Finally, in the case of NGC 3377, only the most extreme anisotropy
2 1/2
shown (z 0.3) is able to reduce the gradient of (vlos
)
along
the minor axis to the shallow level of the observations.
To quantify more precisely the findings of the previous paragraph, in Fig. 3 we show the plots of 2 = 2 2min , describing
the agreement between the data and the models, as a function of z .
We also show, as a filled square with error bars, the value of the
global anisotropy we measured using more general axisymmetric
Schwarzschilds models in Cappellari et al. (2007). The 2 plots
confirm the visual impression from the maps of Fig. 2 and the fact
that the Jeans models give a tight constrain to the anisotropy. The
small size of the Jeans formal errors is of course illusory, as it is
due to the restrictive model assumptions. However the comparison
with the more general results show that the simple anisotropic Jeans
models are able to recover the anisotropy within the errors. There
seems to be a systematic difference of z 0.05 between the
anisotropic Jeans and the Schwarzschild determinations, with the
Jeans results being lower. This difference is at the level of the errors
in the Schwarzschild models (see section 4.3 in Cappellari et al.
2007), and not particularly surprising given the radically different
modeling approaches adopted. If real, it may be due to the fact
that the Jeans models force a constant anisotropy everywhere in
the galaxy model, while the Schwarzschild ones allow for a more
realistic and more spherical velocity ellipsoid at intermediate latitudes ( 45; see Fig. 1). The Jeans models may need a lower
anisotropy to compensate the mismatch with the data at these intermediate latitudes.

Figure 4. M/L versus anisotropy. The different symbols, connected by

lines, show the variation with anisotropy z in the dynamical M/L (Iband) for each of the Jeans models shown in Fig. 2. Also shown for
reference, with horizontal lines using the same style and color, is the
(M/L)Schw , determined from the same data in Cappellari et al. (2006), using three-integral Schwarzschild models. The symbols represent NGC 821
(green circles), NGC 3377 (blue upward triangles), NGC 4621 (black downward triangles) and NGC 5845 (red squares). The filled symbols correspond
to the best-fitting z from Fig. 3. The large cross shows the characteristic
error in z and (M/L)Schw .

own: it has an small size with Re 4. 6 (Cappellari et al. 2006) and

constitute a unique example of an elliptical galaxy with integralfield stellar kinematics out to 3Re . Our standard self-consistent
model still well reproduces the SAURON kinematics without the
need to invoke a dark matter halo.
The small dependence of M/L on the anisotropy explains the
fact that, in a comparison between the semi-isotropic (M/L)Jeans
and the three-integral (M/L)Schw for a sample of 25 early-type
galaxies (including the ones in Fig. 4), we found an excellent agreement between the two determinations. We did not detect any significant bias in the Jeans values and the differences could be explained
by random errors at the 6 per cent rms level (Cappellari et al. 2006,
their fig. 7).

4.1.2 M/L comparison

When studying the stellar population or the dark-matter content of
galaxies, one is interested more in the M/L determination than
in the anisotropy. It is important to test whether an error in the
anisotropy can lead to a significant error in M/L. For this in
Fig. 4 we show the variation in the dynamical M/L (I-band) for
each of the models shown in Fig. 2. Also shown for reference
is the (M/L)Schw determined from the same SAURON data in
Cappellari et al. (2006), using three integral Schwarzschild models. The comparison shows that the M/L varies very little for the
ranges of anisotropy observed in fast rotators early-type galaxies.
In all cases the Jeans M/L at the best-fitting z represents an improvement over the M/L determined from semi-isotropic models
(already given in Table 1 of Cappellari et al. 2006). With the exception of NGC 5845, the improvement is generally at the level of the
measurement errors. The latter galaxy is an interesting case on its

14

The difference between the isotropic models of NGC 821 and

NGC 3377, and the observations, has already been shown in figure A1 of
Cappellari et al. (2007).

4.2 First and second velocity moments model examples

In the previous section we showed that the anisotropic Jeans formalism introduced in this paper provides a good qualitative description of the integral-field second velocity moments observed
with SAURON. We also showed that for edge-on galaxies the recovered anisotropy agrees well with the one derived from more
general models. Here we present additional examples of model fits
to the second moments and we show that, using the simple Satoh
(1980) assumption for the splitting of the random and ordered rotation (Section 3.1.5), a remarkably good prediction of the galaxies
mean velocity can be obtained as well, in many cases.
In Fig. 5 the data-model comparison is presented for both the
first and second velocity moments. The model fits were determined
with the following procedure:
(i) For each pair of the nonlinear model parameters (i, z ) we
2 1/2
evaluated the predicted second moments (vlos
)
using equation (28), with bk = 1/(1 z ) and Mj = Lk .;
(ii) The model predictions were scaled by the best-fitting massc 2008 RAS, MNRAS 390, 7186

Anisotropic axisymmetric Jeans models

11

Figure 5. Data-model comparison for three E and three S0 fast-rotator galaxies. From left to right, the columns show the fast-rotators NGC 821 (E; i =
82 , z = 0.20, = 0.75), NGC 2549 (S0; i = 90 , z = 0.17, = 0.99), NGC 4546 (S0; i = 71 , z = 0.05, = 1.00), NGC 4660 (E;
i = 67 , z = 0.23, = 0.93), NGC 5308 (S0; i = 87 , z = 0.28, = 1.02) and NGC 5845 (E; i = 75 , z = 0.18, = 0.95). The top row

2 )1/2 as given
visualizes the bi-symmetrized SAURON observations for Vrms = V 2 + 2 . The second row show the best-fitting model predictions for (vlos
by equation (28). The third row presents the observed SAURON mean velocity V . The bottom row shows the best fitting model first velocity moment vlos as
given by equation (38). The color levels are the same for data and model. These galaxies are all constrained by the photometry to be quite close to edge-on,
so the models can vary essentially only the single parameter z to fit the shape of the observed first and second moments. The kinematics varies widely for
different galaxies, yet this single parameter is sufficient to correctly predict the main features of a pair of two-dimensional functions (Vrms and V ), once the
observed surface brightness is given.

to-light ratio via equation (51) and the 2 , describing he agreement of data and models was determined;
(iii) Step (i)(ii) were repeated for a grid of (i, z ) parameters
and the best fitting value was found. In the grid, the inclination was
sampled at equally spaced intervals in the intrinsic axial ratio of the
flattest MGE component, while z was sampled linearly;
(iv) The Gaussians describing the mass of the MGE (equation [15]) were scaled by the best fitting and a prediction for the
first velocity moment vlos was computed with equation (38), with
k = 1;
(v) The model velocity field vlos was scaled by the factor
PP
Fn |xn Vn |
,
(52)
= PP n=1

n=1 Fn |xn [vlos ]n |

where P is the number of Voronoi bins in the observed SAURON
data and Fn is the corresponding flux. Vn and [vlos ]n are the observed and model velocities respectively, and xn are the bin coordinates (the x axis being along the projected major axis).

The last step ensures that the model has the same projected angular momentum as the observed galaxy, within the observed region. Doing a normal least-squares fit would heavily underestimate the amount of rotation of the model with respect to the data,
in cases where the galaxy contains counter-rotating stellar components, which are clearly excluded by the simple assumption of
a constant factor for the whole MGE model. The angular momentum of the model could be computed by numerically integrating Lz = Rv over the galaxy volume using equation (21).
More useful may be to use the fitted galaxy inclination to deprojected the observed galaxy stellar angular momentum per unit
mass (Emsellem et al. 2007). To model kinematically distinct stellar components one could allow for different k components in
c 2008 RAS, MNRAS 390, 7186

equation (38), and perform a standard least-squares fit as in equation (51).

For the examples of Fig. 5 we selected three ellipticals and
three lenticulars fast-rotator showing a range of kinematical properties. For all the galaxies the MGE models15 were fitted to the observed photometry using the public software4 of Cappellari (2002).
Although some differences between the data and models are visible, the 2 per degree-of-freedom in the fit presented, computed
from original non-symmetrized data, is close to one for all the fits.
This indicates that the models are generally consistent with the data,
within the measurements errors, which is remarkable for what is essentially a one parameter model! A more general statistical investigation of the kinematic parameters for a larger sample of galaxies
goes beyond the scope of the present paper and will be presented
elsewhere. Here we note that for the galaxies presented in Fig. 5
and 6 the parameter 1 within a few percent accuracy (except for NGC 821). This appears a general characteristics also of
2
the other fast-rotators we modeled. It implies 2 vR
and confirms the results of more general models. It indicates that the velocity ellipsoid of fast-rotator early-type galaxies tends to be oblate
(Cappellari et al. 2007, their fig. 2).
Interestingly Fig. 5 shows that even the complex double-hump
structure observed in the velocity fields of NGC 2549, NGC 4660 or
NGC 5845, which seems to imply a complex dynamical structure in
the galaxy, can be well reproduced by these one-parameter models,
once an accurate MGE parametrization for the surface brightness is
given. The structures are explained by the presence of thin stellar
disks, already visible in the photometry as strongly disky isophotes.

15

The MGE parameters for NGC 2549, NGC 4546 and NGC 5308 are
taken from Scott et al. (in preparation), while the MGE models for the other
galaxies are given in Cappellari et al. (2006).

12

M. Cappellari

There is no evidence for a significantly different anisotropy between

the bulge and the disk of these galaxies, as a constant anisotropy
well reproduces the observations (within 1Re ). These examples
show that a great deal of information on the kinematical structure
of the fast-rotators is contained in the photometry alone!

4.3 Recovery of the galaxy inclination

We have shown in the previous sections that the anisotropic Jeans
axisymmetric models presented in this paper, varying essentially
only one free parameter z , provide a good qualitative description of the observed shape of the first and second velocity moments of fast-rotator early-type galaxies, when they are already constrained by the photometry to be close to edge-on. We also confirmed that the fast-rotators are characterized by a velocity ellipsoid
which is flattened in the z direction so that the anisotropy z >
0
(Cappellari et al. 2007). In this section we study the variations of
the models when the inclination is allowed to vary.

4.3.1 The inclination degeneracy

In Krajnovic et al. (2005) we showed, using real data and analytic
tests, that there appears to be a degeneracy in the recovery of the
galaxy inclination (or its corresponding shape) using general threeintegral axisymmetric models. This is true even in the case of stateof-the-art integral-field data, and assuming the galaxy potential is
accurately known at every inclination and can be uniquely recovered from the surface brightness. This result was confirmed for
a larger sample of galaxies in Cappellari et al. (2006), and in the
oblate limit, but using a triaxial modeling code, by ?. In a realistic
situation both the galaxy potential and the stellar luminosity density
are only approximately known, due to the important intrinsic degeneracy in the deprojection of the surface brightness, due to the possible presence of bars, and due to the likely contribution of a (small)
fraction of dark-matter. Moreover the kinematical data are always
affected by low-level systematics, which are difficult to control and
quantify. The dynamical models are also affected by numerical approximations and discretisation effects. All this makes it unlikely
that any inclination derived from dynamical models of the stellar
kinematics can be trusted, unless further assumptions are made. In
this section we explore whether the anisotropic Jeans models, combined with realistic assumptions on the galaxy anisotropy, can be
used to recover the galaxy inclination.

4.3.2 Inclination recovery with the Jeans models

For our experiments on the recovery of the inclination we selected
the same four galaxies with SAURON integral-field kinematics that
we presented for the same reason in appendix A of Cappellari et al.
(2006). These galaxies have a photometry which, under the axisymmetric assumption, can be deprojected for a wide range of inclinations. Most importantly they possess disks of gas or dust, which
allow the inclination to be determined independently from the stellar dynamics. In that paper we found that the inclination derived
using semi-isotropic models appeared in agreement with the one
from the dust disks. Here we try to understand the reason for the
good agreement.
Our prototypical test case is the fast-rotator E4 galaxy
NGC 2974. The inclination of this galaxy was studied extensively
by a number of authors using the kinematics of the gas (e.g.
Weijmans et al. 2008). From accurate models of the SAURON gas

kinematics in the regions where we have our stellar kinematics

Krajnovic et al. (2005) find a best fitting inclination i = 60 3.
A sequence of model predictions for the velocity second moments
2 1/2
(vlos
)
as a function of anisotropy z and inclination i is shown
in Fig. 6. A strong similarity is evident between the effect on the
models of the anisotropy and of the inclination. Another general
feature one can see from the figure is that the second moments are
more weakly sensitive to an anisotropy variation when the galaxy
is intrinsically flat (i = 55 , q 0.32) than when it is rounder
(i = 90 , q 0.63). A thin disk can be more strongly anisotropic
without producing the vertical elongation of the second moments
(see i = 90 , z = 0.3), which is never observed in real galaxies.
The observed behavior can be qualitatively understood as follows: A more face-on view makes the model intrinsically flatter and
disk-like, causing the second moments to approach the | cos | behavior along the galaxy isophotes as one would expect for a thin
rotating disk, with a peak on the projected major axis and a strong
minimum on the minor axis. A similar increase of the second moments on the projected major axis, with respect to the minor axis,
can be produced by reducing the fraction of stars on radial orbits
(lowering z ), thus correspondingly increasing the stars on tangential orbits. Due to projection these in fact produce zero motion along
the minor axis and show a peak on the major axis. For example in
Fig. 6 the model predictions with (i, z ) = (63.8 , 0.3) are very
similar to the ones for (i, z ) = (57.5 , 0.0). The contours of the
2 , describing the agreement between the Jeans models and the observed SAURON Vrms are shown in Fig. 7. Within a 3 confidence
level any inclination between i = 56 and 61 , and correspondingly any anisotropy between z = 0.2 and 0.1 is equally consistent with the data.
In Cappellari et al. (2007, their fig. 2) we found that all
the fast-rotators have anisotropy z >
0.05. For the nearly edgeon galaxies we confirmed this determination using Jeans models in Section 4.1.1 and Section 4.2. Forcing this observationallymotivated constraint, the inclination of NGC 2974 becomes tightly
constrained to the range i = 56 to 58 . The qualitative behavior
seen in NGC 2974 is representative of what we observed for the 48
galaxies we modeled using this method from the SAURON sample
(Scott et al. in preparation). In Fig. 7 we show the 2 contours for
the three additional galaxies. In the case of NGC 4150 the inclination is well constrained by the data independently of anisotropy,
while in the case of NGC 4459 and NGC 524 the best fitting location has z < 0 and the inclination is heavily dependent on the
assumed anisotropy. For these two galaxies the un-physical location
of the 2 minimum is likely due to the degeneracy in the deprojection of the surface brightness (Section 3.1.2). These two galaxies
have a quite low inclination and the deprojected mass distribution is
probably significantly in error even at the true inclination, which is
then formally excluded by the data. Like all the fast-rotators, these
galaxies likely possess stellar disks, which cannot be detected at
these low inclinations (Rix & White 1990; Magorrian 1999). This
would explain the general fact that we tend to better reproduce the
kinematics of the fast-rotators which are close to edge-on than the
kinematics of the more face-on ones. Nonetheless, in all the four
cases, if one enforces the constrain z >
0.05, the recovered inclination lies within the error bars of the independent gas/dust determinations. This indicates that one can use these anisotropic Jeans
models to recover the inclinations of the fast-rotators by enforcing the requirement z > 0.05, and this explains the good agreement with the determinations from isotropic models we found in
Cappellari et al. (2006).
c 2008 RAS, MNRAS 390, 7186

Anisotropic axisymmetric Jeans models

13

2 )1/2
Figure 6. Inclination and anisotropy variation for NGC 2974. Each of the five columns show the model predictions for the second velocity moment (vlos
computed at different values of z = 0.3, 0.15, . . . , 0.3. Different rows correspond to different inclinations i, equally spaced in the axial ratio of the
flattest Gaussian in the MGE model of NGC 2974. The model mass-to-light ratio was optimized for each fit using equation (51). The values for
the input
parameters (z , i) and for the best fitting are printed on top of each panel. The bottom row shows (i) the bi-symmetrized SAURON data Vrms = V 2 + 2 ;
2 )1/2 ; (iii) the bi-symmetrized SAURON data for V and (iv) the best-fitting model for v
(ii) the best fitting model for (vlos
los . The colormap has the same scaling
for all the panels.

4.4 Relation with the tensor virial theorem

The Jeans equations are useful when one needs a predictions for the
spatial variation of the kinematics in galaxies. However the stars
in galaxies also satisfy more general global relations between kinematic observables, the most important of which are the virial equations (Binney & Tremaine 1987, sec. 4.3). These are a tensor generalization of the well known virial theorem 2K = W , which
relates the total kinetic energy K and the potential energy W of an
isolated stellar system in equilibrium.
The small dependence of the measured M/L on the anisotropy
(Section 4.1.2) can be qualitatively understood as a consequence of
the tensor virial equations. For an axisymmetric galaxy in equilibrium the total kinetic energy is proportional to the galaxy mass. The
ratio of the kinetic energy Kzz along the direction of the symmetry z-axis and the kinetic energy Kxx along a direction orthogonal to it, is only a function of the galaxy shape (equation [4-91] of
Binney & Tremaine 1987). The effect of increasing the anisotropy
z is simply that of redistributing the same kinetic energy Kxx
from the tangential to the radial direction. This reduces the velocc 2008 RAS, MNRAS 390, 7186

ity second moments along the galaxy projected major axis, while
correspondingly increasing them along the minor axis. The projected luminosity-weighted second moments, from which the M/L
is measured, remain nearly unchanged when averaged over the full
galaxy image.
The virial equations also provide a qualitative understanding of
why the observations can constrain the inclination of an axisymmetric galaxy (Section 4.3.2). In fact they state that the ratio Kxx /Kzz
increases with galaxy flattening. In other words, for a given projected surface brightness, a flat system needs more kinetic energy
in a direction parallel to its the equatorial plane, to be in hydrostatic
equilibrium.
The fast rotator NGC 2974 of Fig. 6 shows high velocity second moments along the projected major axis. Although the galaxy
does not appear very flat in projection, its kinematics is already an
indication that the system is likely to be intrinsically flat. A flat
model has more kinetic energy in the equatorial plane and will be
able to easily reproduce the high Vrms along the major axis, without invoking extreme tangential anisotropy. A quantitative example,
using both the first and second velocity moments, of how the ten-

14

M. Cappellari

Figure 7. The inclination-anisotropy degeneracy. Contours of 2 =

2 2min , describing the agreement between the data and the anisotropic
Jeans models, are plotted as a function of the anisotropy parameter z and
the galaxy inclination i in degrees. The lowest levels correspond to the formal 1, 2 and 3 (thick red line with 2 = 11.8) confidence levels for two
degrees of freedom. Additional contours are characterized by a factor of 2
increase in 2 . The two horizontal blue solid lines enclose the region of
acceptable inclinations as determined from the gas kinematics or dust geometry (appendix A of Cappellari et al. 2006). The vertical solid green line
indicates isotropy (z = 0). The different panels correspond to NGC 4150,
NGC 2974, NGC 4459 and NGC 524 respectively. There is a clear degeneracy between inclination and z anisotropy. However if one enforces the
observationally-motivated constraint z > 0.05, the recovered inclination
always lies within the error bars of the gas/dust determinations.

sor virial equations can be used to constrain the galaxy inclination

is presented in fig. 11 of Cappellari et al. (2007). There we showed
that the (V /, ) diagram of Binney (2005) can be used to provide a
qualitative estimate of the inclination of fast-rotators galaxies, given
some assumptions on the anisotropy.
Although the virial equations can be used to estimate the M/L
and the inclination of galaxies, they have some obvious limitations:
(i) They do not allow one to make full use of the spatial information contained in the integral-field observations. The whole twodimensional kinematics is reduced to one number; (ii) They do not
allow the limited spatial extent of the kinematics to be properly
taken into account; (iii) They do not easily deal with non-similar
isophotes and multiple photometric components in galaxies.

5 SUMMARY
We present a generalization of the widely used semi-isotropic (twointegral) axisymmetric Jeans modeling method to describe the stellar dynamics of galaxies. Our method uses the powerful MultiGaussian Expansion (MGE) technique to accurately parameterize
the galaxies photometry. It represents an anisotropic extension of
what was presented in the semi-isotropic case by Emsellem et al.
(1994), and it maintains its simplicity and computational efficiency.

We assume (i) a constant mass-to-light ratio and (ii) a velocity ellipsoid which is aligned with the cylindrical (R, z) coordinates and
has a flattening quantified by the classical z-anisotropy parameter
2
z = 1 vz2 /vR
, where z is the galaxy symmetry axis.
We test the technique using SAURON integral-field observations of the stellar kinematics (Emsellem et al. 2004) for a small
set of fast-rotator galaxies with a variety of kinematical properties. For galaxies that are constrained by the photometry to be
close to edge-on we find that, although in the semi-isotropic limit
(z = 0) the models do not provide a good description of the data,
the variation of the single global anisotropy z is generally sufficient to accurately
predict the shape of both the first (V ) and sec
ond (Vrms = V 2 + 2 ) velocity moments, once an detailed MGE
parametrization of the photometry is given. An accurate description of the photometry, including ellipticity variations and disky
isophotes, appears crucial to reproduce in detail the features of the
kinematics.
2
2
In all cases we find that z >
0.05, while generally vR
with good accuracy. This confirms previous findings on the dynamical structure of these galaxies, showing that their velocity ellipsoid tends to be oblate (Cappellari et al. 2007). The anisotropy derived with our anisotropic Jeans dynamical modeling method agrees
within the errors with the one previously measured using a more
general Schwarzschild approach.
For fast-rotator galaxies that are not constrained by the photometry to be close to edge-on, we find that in general the inclination i (or the corresponding galaxy shape) and the anisotropy
z are highly correlated and cannot be independently determined.
However, if we introduce the observationally-motivated constraint
z >
0.05, the inclination becomes constrained to a narrow range
of values and it agrees with independent determinations when those
are available.
We are applying this method to determine the inclination, the
mass-to-light ratio and the amount of rotation of a large sample of
galaxies for which integral field kinematics are available. For galaxies close to edge-on, the global anisotropy or the dynamical structure of different galaxy subcomponents (e.g. bulge and disk) can
also be investigated. We are using this method to test independent
determinations of the masses of supermassive black holes in the nuclei of fast-rotator galaxies. This technique is ideal to study the dark
matter content and the anisotropy of disks of spiral galaxies.

ACKNOWLEDGEMENTS
I acknowledge fruitful discussions and feedback from Eric Emsellem, Davor Krajnovic and Glenn van de Ven. I am grateful to
Roger Davies, Tim de Zeeuw and Remco van den Bosch for constructive criticism on the draft. I thank Nicholas Scott for providing
the MGE model of three of the galaxies presented in Fig. 5, before
publication. I am grateful to the anonymous referee for useful comments. I acknowledge support from a STFC Advanced Fellowship
(PP/D005574/1). The SAURON observations were obtained at the
William Herschel Telescope, operated by the Isaac Newton Group
in the Spanish Observatorio del Roque de los Muchachos of the
Instituto de Astrofsica de Canarias.

REFERENCES
Abramowitz M., Stegun I. A., 1965, Handbook of mathematical
functions with formulas, graphs, and mathematical tables. Dover
c 2008 RAS, MNRAS 390, 7186

Anisotropic axisymmetric Jeans models

Books on Advanced Mathematics, New York: Dover, 1965, Corrected edition, edited by Abramowitz, Milton; Stegun, Irene A.
Arnold R., 1995, MNRAS, 276, 293
Bacon R., 1985, A&A, 143, 84
Bacon R., et al., 2001, MNRAS, 326, 23
Bacon R., Simien F., Monnet G., 1983, A&A, 128, 405
Bell E. F., de Jong R. S., 2001, ApJ, 550, 212
Bendinelli O., 1991, ApJ, 366, 599
Bertola F., Cinzano P., Corsini E. M., Rix H.-W., Zeilinger W. W.,
1995, ApJ, 448, L13
Binney J., 2005, MNRAS, 363, 937
Binney J., Mamon G. A., 1982, MNRAS, 200, 361
Binney J., Tremaine S. D., 1987, Galactic Dynamics. Princeton
Univ. Press, Princeton, NJ
Binney J. J., Davies R. L., Illingworth G. D., 1990, ApJ, 361, 78
Cappellari M., 2002, MNRAS, 333, 400
Cappellari M., et al., 2006, MNRAS, 366, 1126
Cappellari M., et al., 2007, MNRAS, 379, 418
Carollo C. M., Danziger I. J., 1994, MNRAS, 270, 523
Carollo C. M., de Zeeuw P. T., van der Marel R. P., Danziger I. J.,
Qian E. E., 1995, ApJ, 441, L25
Cinzano P., Rix H.-W., Sarzi M., Corsini E. M., Zeilinger W. W.,
Bertola F., 1999, MNRAS, 307, 433
Cinzano P., van der Marel R. P., 1994, MNRAS, 270, 325
Corsini E. M., et al., 1999, A&A, 342, 671
Cortes J. R., Kenney J. D. P., Hardy E., 2008, ApJ, 683, 78
Cretton N., de Zeeuw P. T., van der Marel R. P., Rix H.-W., 1999,
ApJS, 124, 383
Cretton N., van den Bosch F. C., 1999, ApJ, 514, 704
Davies R. L., Sadler E. M., Peletier R. F., 1993, MNRAS, 262,
650
de Lorenzi F., Debattista V. P., Gerhard O., Sambhus N., 2007,
MNRAS, 376, 71
de Zeeuw P. T., et al., 2002, MNRAS, 329, 513
de Zeeuw P. T., Hunter C., 1990, ApJ, 356, 365
de Zeeuw T., 1985, MNRAS, 216, 273
Dehnen W., Gerhard O. E., 1993, MNRAS, 261, 311
Dejonghe H., de Zeeuw T., 1988, ApJ, 333, 90
Eddington A. S., 1915, MNRAS, 76, 37
Emsellem E., et al., 2007, MNRAS, 379, 401
Emsellem E., et al., 2004, MNRAS, 352, 721
Emsellem E., Monnet G., Bacon R., 1994, A&A, 285, 723
Evans N. W., de Zeeuw P. T., 1994, MNRAS, 271, 202
Evans N. W., Lynden-Bell D., 1991, MNRAS, 251, 213
Faber S. M., et al., 2007, ApJ, 665, 265
Ferrarese L., Merritt D., 2000, ApJ, 539, L9
Ferrarese L., van den Bosch F. C., Ford H. C., Jaffe W., OConnell
R. W., 1994, AJ, 108, 1598
Fillmore J. A., 1986, AJ, 91, 1096
Gebhardt K., et al., 2000, ApJ, 539, L13
Gerhard O., Kronawitter A., Saglia R. P., Bender R., 2001, AJ,
121, 1936
Gerhard O. E., 1993, MNRAS, 265, 213
Gerhard O. E., Binney J. J., 1996, MNRAS, 279, 993
Gerssen J., Kuijken K., Merrifield M. R., 1997, MNRAS, 288, 618
Gerssen J., Kuijken K., Merrifield M. R., 2000, MNRAS, 317, 545
Jeans J. H., 1915, MNRAS, 76, 70
Jeans J. H., 1922, MNRAS, 82, 122
Joseph C. L., et al., 2001, ApJ, 550, 668
Kassin S. A., de Jong R. S., Weiner B. J., 2006, ApJ, 643, 804
Kochanek C. S., Rybicki G. B., 1996, MNRAS, 280, 1257
c 2008 RAS, MNRAS 390, 7186

15

Koopmans L. V. E., Treu T., Bolton A. S., Burles S., Moustakas

L. A., 2006, ApJ, 649, 599
Kormendy J., Bender R., 1996, ApJ, 464, L119+
Krajnovic D., Cappellari M., Emsellem E., McDermid R. M., de
Zeeuw P. T., 2005, MNRAS, 357, 1113
Krajnovic D., et al., 2008, MNRAS, 390, 93
Lanzoni B., Ciotti L., 2003, A&A, 404, 819
Lauer T. R., et al., 1995, AJ, 110, 2622
Magorrian J., 1999, MNRAS, 302, 530
Magorrian J., et al., 1998, AJ, 115, 2285
Merritt D., 1980, ApJS, 43, 435
Monnet G., Bacon R., Emsellem E., 1992, A&A, 253, 366
Nagai R., Miyamoto M., 1976, PASJ, 28, 1
Navarro J. F., Frenk C. S., White S. D. M., 1996, ApJ, 462, 563
Noordermeer E., Merrifield M. R., Aragon-Salamanca A., 2008,
MNRAS, 388, 1381
Palunas P., Williams T. B., 2000, AJ, 120, 2884
Persic M., Salucci P., Stel F., 1996, MNRAS, 281, 27
Press W. H., Teukolsky S. A., Vetterling W. T., Flannery B. P.,
1992, Numerical recipes in FORTRAN. The art of scientific computing. Cambridge: University Press, 1992, 2nd ed.
Qian E. E., de Zeeuw P. T., van der Marel R. P., Hunter C., 1995,
MNRAS, 274, 602
Richstone D. O., 1982, ApJ, 252, 496
Riciputi A., Lanzoni B., Bonoli S., Ciotti L., 2005, A&A, 443, 133
Rix H.-W., Carollo C. M., Freeman K., 1999, ApJ, 513, L25
Rix H.-W., White S. D. M., 1990, ApJ, 362, 52
Rix H.-W., White S. D. M., 1992, MNRAS, 254, 389
Romanowsky A. J., Kochanek C. S., 1997, MNRAS, 287, 35
Rusin D., Kochanek C. S., 2005, ApJ, 623, 666
Rybicki G. B., 1987, in de Zeeuw P. T., ed., Structure and Dynamics of Elliptical Galaxies Vol. 127 of IAU Symposium, Deprojection of Galaxies - how much can BE Learned. p. 397
Satoh C., 1980, PASJ, 32, 41
Schwarzschild M., 1979, ApJ, 232, 236
Shampine L. F., 2008, J. Comp. Appl. Math., 211, 131
Shanjie Z., Jianming J., 1996, Computation of Special Functions.
New York, Wiley
Shapiro K. L., Gerssen J., van der Marel R. P., 2003, AJ, 126, 2707
Springel V., et al., 2005, Nature, 435, 629
Statler T. S., Dejonghe H., Smecker-Hane T., 1999, AJ, 117, 126
Thomas J., Saglia R. P., Bender R., Thomas D., Gebhardt K.,
Magorrian J., Corsini E. M., Wegner G., 2007, MNRAS, 382,
657
Tremaine S., Richstone D. O., Byun Y.-I., Dressler A., Faber
S. M., Grillmair C., Kormendy J., Lauer T. R., 1994, AJ, 107,
634
van Albada T. S., Bertin G., Stiavelli M., 1995, MNRAS, 276,
1255
van de Ven G., Hunter C., Verolme E. K., de Zeeuw P. T., 2003,
MNRAS, 342, 1056
van den Bosch F. C., 1997, MNRAS, 287, 543
van den Bosch R. C. E., van de Ven G., Verolme E. K., Cappellari
M., de Zeeuw P. T., 2008, MNRAS, 385, 647
van den Bosch R. C. E., van de Ven G., 2008, MNRAS, submitted
(arXiv:0811.3474)
van der Marel R. P., 1991, MNRAS, 253, 710
van der Marel R. P., 1994, MNRAS, 270, 271
van der Marel R. P., Binney J., Davies R. L., 1990, MNRAS, 245,
582
van der Marel R. P., Cretton N., de Zeeuw P. T., Rix H.-W., 1998,
ApJ, 493, 613

16

M. Cappellari

van der Marel R. P., Franx M., 1993, ApJ, 407, 525
van der Marel R. P., van Dokkum P. G., 2007a, ApJ, 668, 738
van der Marel R. P., van Dokkum P. G., 2007b, ApJ, 668, 756
van der Wel A., van der Marel R. P., 2008, ApJ, 684, 260
Weijmans A.-M., Krajnovic D., van de Ven G., Oosterloo T. A.,
Morganti R., de Zeeuw P. T., 2008, MNRAS, 383, 1343
Young L. M., Bureau M., Cappellari M., 2008, ApJ, 676, 317

(Ly /2) y
(Ly /2) + y

erf
+ erf
.
2i
2i

(A6)

In practice the integral of equation (A5) can be limited to the region

where K(x, y) is significantly nonzero:
<
(Lx /2) 3 max{i } <
x (Lx /2) + 3 max{i }
<
<
(Ly /2) 3 max{i } y (Ly /2) + 3 max{i }.

(A7)
(A8)

APPENDIX A: SEEING AND APERTURE CONVOLUTION

The first and second velocity moments have to be convolved with
the instrumental point-spread-function (PSF) and integrated over
the aperture used in the observations, before making comparisons
with the observed quantities. The PSF effects are particularly important when studying the nuclear regions of galaxies (e.g. to estimate the mass of supermassive black holes). The PSF-convolved
velocity moments are given by the general formulas (e.g. equations [51]-[53] of Emsellem et al. 1994)
obs = PSF

(A1)

[vlos ]obs =

(vlos ) PSF
obs

(A2)

2
[vlos
]obs =

2
) PSF
(vlos
,
obs

(A3)

with representing convolution and where the normalized PSF is

generally described as the sum of Q Gaussians

Q
X
R2
Gi
exp

,
(A4)
PSF(R) =
2i2
2i2
i=1
P
with Q
i=1 Gi = 1. To efficiently evaluate numerically the above
convolutions up to large radii, and even for very flat models, one can
2
and vlos on a grid linear in
evaluate the models predictions vlos
the logarithm of the elliptical radius and in the eccentric anomaly.
This is done by defining a logarithmically-spaced radial grid Rj and
then computing the moments at the coordinate positions (x , y ) =
(Rj cos k , q Rj sin k ), for linearly spaced k values, with q a
characteristic (e.g. the median) observed axial ratio of the MGE
model. The model is then re interpolated onto a fine Cartesian grid
before the convolution with the PSF using fast Fourier methods.
Finally the model is rebinned into the observed apertures.
Alternatively, especially when the apertures sparsely sample
the plane of the sky, instead of performing the convolutions of equations A1A3 on a regular grid, one can evaluate the same convolution integrals only for the observed apertures, while also including
the integration onto the apertures (appendix D of Qian et al. 1995).
Given a rectangular aperture of sides Lx and Ly , aligned with the
coordinate axes at position (x , y ) on the sky, and assuming a PSF
given by equation (A4), the PSF-convolved observable S(x , y )
inside the aperture is
Z Z
Sobs (x , y ) =
S(x , y ) K(x x , y y ) dxdy, (A5)

2
where S has to be replaced by , vlos , or vlos
respectively, and
2
]obs recorrespondingly Sobs becomes obs , [vlos ]obs , or [vlos
spectively. The kernel is given by

Q
X
(Lx /2) + x
(Lx /2) x
Gi

+ erf
K(x, y) =
erf
4
2i
2i
i=1

c 2008 RAS, MNRAS 390, 7186