k
~ s
k
(r) exp i
_
r
k
r
(r
/
) dr
/
imj io
k
t
_ _
c:c:;
(1)
where ~ s
k
(r) is the real amplitude, which is the slowly
varying in space, k
r
(r) is the real radial wavenumber, m is
the nonnegative azimuthal mode number, o
k
= Ro
k
iIo
k
is the complex frequency of excited waves, sufxes k
denote the kth Fourier component, and c:c: means the
complex conjugate. The rapidly varying part of s
1
is
absorbed in its phase. The radial wavenumber is presumed
to be of the form
k
r
(r) = AC(r),
where Ais a large parameter and C(r) is a smooth, slowly
and monotonically varying function of r, i.e.,
dlnk
r
=dlnr = O(1), and [
_
k
r
dr[b1. The WKB method
is a powerful approach that can be used in a large class of
wave problems (Mikhailovskii, 1974; Alexandrov et al.,
1984; Krall and Trivelpiece, 1986; Swanson, 1989). The
WKB treatment leads to a relatively simple phaseintegral
ARTICLE IN PRESS
E. Griv / Planetary and Space Science 55 (2007) 547568 550
type relation as a dispersion relation from which in many
cases stability boundaries can be obtained analytically.
Generally, the WKB approach allows the determination of
the necessary and sufcient condition for instability.
Evidently, in Eq. (1) s
1
is a periodic function of j, and
hence m must be an integer. The criteria for stability differ
for each m, and must be determined by a detailed analysis.
The assumption that ~ s
k
has a weak spatial dependence
corresponds to the quasiclassical approximation in quan
tum mechanics and to the approximation of geometrical
optics in the propagation of light in an inhomogeneous
medium (Alexandrov et al., 1984, p. 245). In the linear
theory, one can select one of the Fourier harmonics:
s
1
= ~ s(r) exp i
_
r
k
r
(r
/
) dr
/
imj iot
_ _
c:c:: (2)
A disk is considered to be a superposition of different
oscillation modes. A disturbance in the disk will grow until
it is limited by some nonlinear effect. It is remarkable that
the results of WKB theory are in good agreement with the
computer simulations (Mayer et al., 2002).
The analysis is restricted to a treatment of perturbations
which are symmetric with respect to the z = 0 equatorial
plane of the disk (which do not cause it to bend) (Fig. 1b).
This type of vertical motions does not deform the
horizontal disk plane z = 0, because the vertical velocity
v
z
in a density wave is odd in z: v
z
(z) = v
z
, so that in the
plane z = 0 it is equal to zero. The perturbed pressure,
density, gravitational potential, and horizontal velocity
components are even functions of z, while the perpendi
cular velocity is odd in z. Such even sausagelike
perturbations can release gravitational energy and are
subject to ordinary Jeans instability. These perturbations
are associated with such phenomena as, for example, the
appearance of the spiral structure of galaxies and the ne
scale of the order of 100 m or even less irregular wave
structure of Saturns rings (Griv et al., 1999, 2002; Griv
and Gedalin, 2003, 2005, 2006). See Kulsrud et al. (1971)
and Bertin and Casertano (1982) for a discussion of the
problem.
The second type of the vertical motions makes the disk
bend in the same way as the plane of an oscillating
membrane does. The vertical velocity of such motions is an
even function of z: v
z
(z) = v
z
, and in the plane z = 0 it is
not equal to zero (Fig. 1c). Contrary to the case of sausage
like even perturbations, the perturbed pressure, density,
gravitational potential, and horizontal velocity compo
nents are odd functions of z. These perturbations of the
bending type do not release gravitational energy and,
consequently, they are expected to be Jeansstable (Bertin
and Casertano, 1982). The bending type of motions can be
either caused by tidal inuence of a satellite, or excited by
the socalled rehose instability. Apparently, the rehose
type bending instability of a sufciently thin stellar disk has
been predicted by Toomre (1966) by using the simplied
theory based on moment equations. Toomre considered
the collisionless analog of the KelvinHelmholtz instability
in an innite, twodimensional, nonrotating sheet of
stars. (See also Kulsrud et al., 1971; Toomre, 1983 for a
discussion.) Note that this is the usual way to discuss
the conditions of the rehose instability in plasma physics
(e.g., Krall and Trivelpiece, 1986). It has been demon
strated by Toomre that the bending instability is driven by
the stellar pressure anisotropy: the source of free energy
in the instability is the intrinsic anisotropy of a velocity
dispersion (temperature). Raha et al. (1991) and Griv
and Chiueh (1998) have used particle simulations to study
the rehosetype bending instabilities in galaxies.
In weakly inhomogeneous medium in which [k
r
L[b1,
where [L[ is the scale of inhomogeneity, in Eq. (2) the
amplitude and the radial wavenumber may be expanded in
a Taylor series around the point r
~ s(r
/
) = ~ s(r) (r
/
r) d~ s=dr , (3)
k
r
(r
/
) = k
r
(r) (r
/
r) dk
r
=dr , (4)
where [(r
/
r)=r[51 and r is the direction of the disk
inhomogeneity. After integration
_
k
r
(r
/
) dr
/
, Eq. (4) leads
to
(r
/
r)k
r
(
1
2
)(r
/
r)
2
dk
r
=dr . (5)
For the longwavelength perturbations with [(r
/
r)
dln~ s=dr[51 and [(r
/
r)
2
dk
r
=dr[51, in the sums (3) and
(5) only the rst terms may be retained. This is the lowest,
or the socalled local approximation of the WKB method.
Thus, in the lowest WKB approximation in the radial
direction, the expression for the perturbation of the surface
density takes the simple form
s
1
(r; j; t) = ~ s exp(ik
r
r imj io
+
t) c:c:; (6)
where we now call ~ s a constant amplitude, k
r
a constant
radial wavenumber in a weakly inhomogeneous disk, and
l
r
= 2p=k
r
is a constant radial wavelength. The perturbed
ARTICLE IN PRESS
z
(a)
(b)
(c)
+h
Fig. 1. Sketch of perturbations of a threedimensional protostellar disk. In
(a) a section of the disk is shown edgeon. In (b) an even sausagelike
(Jeanstype) perturbation is shown (the dashed line). In (c) an odd
(bending rehosetype; Kulsrud et al., 1971; Bertin and Casertano, 1982;
Raha et al., 1991; Griv and Chiueh, 1998) perturbation is illustrated (the
dashed line).
E. Griv / Planetary and Space Science 55 (2007) 547568 551
potential of the disk F
1
(r; t) and the perturbed pressure
P
1
(r; t) are also of this form. The meaning of localized
solution has been discussed in plasma physics (Mikhai
lovskii, 1974; Alexandrov et al., 1984; Krall and Trivel
piece, 1986; Swanson, 1989). In the local approximation
the wave is considered to be plane, i.e., by investigating the
particular case of a medium which is only weakly
inhomogeneous on the scale of the oscillation wavelength,
all derivatives of ~ s(r) and k
r
(r) or all terms of the order
1=k
r
L and of higher orders are neglected. When calculating
the terms of higher ordero qk
r
=qr, q~ s=qr, etc.one can
simultaneously solve the eld equations with any desired
degree of accuracy. It has been demonstrated that localized
solutions seem to describe the physical situation in a
natural way and give results consistent with the more exact
results of the global modes theory (Alexandrov et al.,
1984). Solution (6) represents a spiral wave with m arms or
a ring (m = 0). The imaginary part of o corresponds to a
growth (Io40) or decay (Ioo0) of the components in
time, s
1
o exp(Iot), and the real part to a rotation with
constant angular velocity O
p
= Ro=m. When Io40, the
medium transfers its energy to the growing wave and
oscillation buildup occurs.
In the lowest WKB approximation it is assumed that the
wave vector and the wavefrequency vary continuously. By
utilizing the more accurate nonlocal WKB approximation,
it may be shown that in fact the characteristic oscillation
frequencies of an inhomogeneous disk must be quantized,
i.e., must pass through a discrete series of values
(Alexandrov et al., 1984, p. 249). Ru diger and Kitchatinov
(2000) have studied the stability of a selfgravitating
innitesimally thin gaseous disk rotating around a central
mass by using a nonlocal formulation. Both axisymmetric
and lowm nonaxisymmetric excitation has been analyzed.
In galaxies, discrete spiral modes have already found in
stellar population by Rix and Zaritsky (1995), Rudnick
and Rix (1998), and Block and Puerari (1999). In the near
infrared, the morphology of older stardominated disk
indicates a simple classication scheme: the dominant
Fourier mmode. A ubiquity of lowm modes has been
conrmed.
3. Perturbed motion
The disk is subject to the equations of motion along the
radial and azimuthal directions and to the continuity
equation. The equations of twodimensional motion of the
gas element in the frame of reference rotating with angular
velocity O at the reference position r
0
can be written in
Hills approximation as (Spitzer and Schwarzschild, 1953;
Julian and Toomre, 1966):
dv
r
dt
2Ov
j
2r
0
r
1
O
dO
dr
=
qF
1
qr
c
2
s
s
0
qs
1
qr
, (7)
dv
j
dt
2Ov
r
=
1
r
0
qF
1
qj
c
2
s
r
0
s
0
qs
1
qj
. (8)
Motions in the z direction do not affect motions in the
(r; j)plane, in the rst approximation, and are ignored
(Spitzer and Schwarzschild, 1953). In Eqs. (7) and (8), v
r
and v
j
are the radial and azimuthal velocities, respectively,
r
0
is the radius of the chosen circular orbit in the (r; j)
plane, O = O(r
0
), c
s
= (qP=qs)
1=2
0
is the local sound speed
(in the numerical experiments using the particle code, c
s
is
represented by the rootmeansquare random velocity
dispersion), r
1
and j
1
are small perturbations of the
coordinates, r = r
0
r
1
, and [r
1
=r
0
[51. The inuence of
central star enters through O(r). In Eqs. (7) and (8), the
following expansion is used:
P = P
0
s
1
qP
qs
_ _
0
and
s
1
P
0
qP
qs
51.
Eqs. (7) and (8) must be solved simultaneously with the
linearized continuity equation, which gives the (Eulerian)
density perturbation s
1
(Lin and Lau, 1979, p. 159)
5
:
s
1
s
0
=
qr
1
qr
0
imj
1
r
1
d
dr
0
ln(r
0
s
0
), (9)
where v
r
= dr
1
=dt and v
j
= r
0
dj
1
=dt are now the
perturbed (forced) velocities, and relatively small term
s
0
v
r
=r
0
is omitted, i.e., the curvature effect is neglected.
This is a valid approximation if [k
r
[r is large (Lau and
Bertin, 1978; Lin and Lau, 1979; Sellwood and Kahn,
1991).
For such a form of F
1
(Eq. (6)) the particular solution of
the system of Eqs. (7)(8) is (Lin and Lau, 1979; Griv et al.,
1999)
v
r
=
o
2
+
k
2
o
+
k
r
i2O
m
r
_ _
, (10)
v
j
=
o
2
+
k
2
4O
2
k
2
o
2
+
o
+
m
r
i2Ok
r
_ _
, (11)
where the subscript on r
0
is dropped as we are considering
linearized quantities, o
+
= o mO is the Dopplershifted
(in a circular rotating frame) wavefrequency, k
j
m=r is
the azimuthal wavenumber, O(r) is the angular velocity of
differential rotation at the distance r from the center,
k = 2O 1
r
2O
dO
dr
_ _
1=2
\O
is the epicyclic frequency, o
+
a0, o
2
+
k
2
a0, and
= F
1
c
2
s
s
1
=s
0
. Solutions (10) and (11) describe the
forced velocities of the gas element in the radial and
azimuthal directions under the action of the small gravity
perturbation, [v
r
[ and [v
j
[5rO. If both the expressions are
inserted into Eq. (9), a relation between the uctuations in
density and potential is obtained. A second relation of this
kind is supplied by the Poisson equation.
Thus, the present theory suggests some systematic radial
and azimuthal motions of the gas element distributed in the
ARTICLE IN PRESS
5
Lin and Lau (1979) as well as Lau and Bertin (1978) represented a
perturbation in a slightly different form o exp(io
+
t ik
r
r imj).
E. Griv / Planetary and Space Science 55 (2007) 547568 552
form of a spirallike ow eld which is a small correction to
the basic circular, equilibrium motion (cf. Yuan, 1969;
Rohlfs, 1977). Equilibrium is described by the following
equation:
rO
2
=
qF
0
qr
c
2
s
s
0
qs
0
qr
,
where the term o c
2
s
is a small correction. Equilibrium is
established in a simple manner in such a disk, i.e., it is
governed mainly by the balance between the centrifugal
and gravitational forces. It is this equilibrium model of a
disk that is to be examined for stability in the present
investigation.
4. Asymptotic surfacedensitypotential relation
For such a form of perturbation (Eq. (6)), the Poisson
equation
q
2
F
qr
2
1
r
qF
qr
1
r
2
q
2
F
qj
2
q
2
F
qz
2
= 4pGsd(z), (12)
where d(z) is the Dirac deltafunction with respect to the
spatial coordinate z, becomes
d
2
dz
2
k
2
_ _
F
1
= 4pGs
1
d(z), (13)
k =
k
2
r
k
2
j
_
, k
j
= m=r is the total wavenumber and
k
2
r
bk
2
j
. In the vacuum (z40 and zo0), Eq. (12) is reduced
to the Laplace equation
DF
2
= 0,
and, therefore, in these regions the solutions are
F
F
1;z40
= C
1
e
[k[z
; F
F
1;zo0
= C
2
e
[k[z
, (14)
where C
1
and C
2
are constants. By integrating Eq. (13)
over z, the boundary conditions relating F
1
to F
2
on the
surface of the diskvacuum partition (z = 0) are found:
F
= F
[
z=0
;
qF
qz
qF
qz
_ _
z=0
= 4pGs
1
. (15)
(The potential F must be a continuous function for all z
and, in particular, when z = 0, as otherwise the z
component of the force would be innite.) On substituting
the solutions (14) into the boundary conditions (15), one
obtains the required connection between the perturbed
potential F
1
(r; t) and the perturbed surface density s
1
(r; t)
of the innitesimally thin disk
F
1
=
2pGs
1
[k[
e
[k[[z[
c:c:; (16)
which reduces to F
1
= 2pGs
1
=[k[ c:c: on z = 0 where
the disk lies (see also Lau and Bertin, 1978; Lin and Lau,
1979), showing that density maxima correspond to
potential minima. We have just obtained that Eq. (12),
an integral relation for arbitrary values of [k
r
[r, becomes a
local relation (16) in the shortwave, or WKB limit
[k
r
[rb1.
5. Dispersion relation
Upon substituting Eqs. (10)(11) into Eq. (9) and
paralleling the analysis leading to Eq. (D12) in Lin and
Lau (1979), it is straightforward to show that
s
1
s
0

o
2
+
k
2
k
2
r
4O
2
k
2
m
2
r
2
2O
o
+
m
rL
_ _
c:c:; (17)
where s
1
(t o) = 0, so by considering only growing
perturbations we neglected the effects of the initial
conditions, and
[L[ =
d
dr
ln(s
0
Ok
2
)
1
(18)
is the radial scale of spatial inhomogeneity. In Eq. (17), the
expansion 2O=k  1 (r=4O)(dO=dr) is used and only the
most important lowfrequency nonresonant ([o
2
+
[tk
2
and o
+
a0, respectively) perturbations developing between
the inner and outer Lindblad resonances are considered
(Griv et al., 1999, 2002; Griv and Gedalin, 2003, 2006).
Generally, in Eq. (17) the term o L
1
is a small
correction. As is known, this term corresponds to a
waveuid resonance o
+
0, and sufciently far from
the resonance this term may be omitted (Lau and Bertin,
1978; Lin and Lau, 1979, p. 159). However, as I will show
in Section 7 below the existence of spatial inhomogeneity is
critically important for the exchange of angular momen
tum in the wavegas system, and, therefore, the term o L
1
must be retained in Eq. (17). Notice that the waveuid
resonance has been studied by Lovelace and Hohlfeld
(1978) and Morozov (1980).
The time behavior of the quantity s
0
Ok
2
in self
gravitating disks has been considered by Berman and
Mark (1977). They have proved that in the axisymmetric
case
d
dt
s
0
O
k
2
_ _
= 0, (19)
that is, vorticity is carried along with the uid. Eq. (19) is a
consequence of the detailed conservation of mass and
angular momentum (in the axisymmetric case). Lovelace
and Hohlfeld (1978) have pointed out that in Eq. (18)
f (r) = s
0
Ok
2
has the role of the distribution function for
angular momentum. In selfgravitating disks in equili
brium, f (r) is a decreasing function over a range of r; thus,
in main parts of protostellar disks Lo0. Values of O(r),
k(r), and s
0
(r) derived from the oftenly used equilibrium
models of disks, namely Hunters (1965) n = 1 model,
exponential disk model (Binney and Tremaine, 1987,
p. 78), and Hunters n o model are used to calculate
f (r) for the three cases. The f (r) curves are shown in Fig. 2.
As is seen, in all cases f (r) is a decreasing function of r in
the main parts of the systems.
Equating the density s
1
(Eq. (17)) to the perturbed
density s
1
= [k[F
1
=2pG c:c: given by the asymptotic
(k
2
r
bm
2
=r
2
) solution of the Poisson equation (Eq. (16)),
ARTICLE IN PRESS
E. Griv / Planetary and Space Science 55 (2007) 547568 553
one obtains the generalized LinShu dispersion relation
o
3
+
o
+
o
2
J
4pOGs
0
(m=r[k[L) = 0, (20)
where [o
+
[k, [kL[b1, k
2
c
2
s
52pGs
0
[k[,
o
2
J
= k
2
2pGs
0
(k
2
+
=[k[) k
2
+
c
2
s
(21)
is the squared Jeans frequency,
k
2
+
= k
2
{1 [(2O=k)
2
1] sin
2
c]
is the squared effective wavenumber, and c =
arctan(m=rk
r
) is the perturbation pitch angle. Eq. (20)
differs from the standard LinShu dispersion relation
o
2
+
= k
2
2pGs
0
[k
r
[ k
2
r
c
2
s
(22)
(Lin et al., 1969; Binney and Tremaine, 1987, p. 359) by the
appearance of the total k and effective k
+
wavenumbers,
which originate from the consideration of the nonaxisym
metrical modes o c, and by the factor o L
1
, which
originate from the consideration of the effects of inhomo
geneity (see Morozov, 1980, 1985; Griv et al., 1999, 2002
for a discussion).
From Eq. (20) in the most important frequency range
[o
+
[
3
[o
J
[
3
b4pOGs
0
(m=r[k[L), (23)
we determine the dispersion law for the Jeans branch of
oscillations:
o
+1;2
 p[o
J
[ 2pGs
0
O
o
2
J
m
r[k[L
, (24)
where p = 1 for gravitystable perturbations with
o
2
+
 o
2
J
40, p = i for gravityunstable perturbations with
o
2
+
 o
2
J
o0, and the term involving L
1
is the small
correction. Eq. (24) determines the spectrum of oscilla
tions. Accordingly, an inhomogeneity will not inuence the
stability condition of Jeans modes of oscillations.
From Eq. (24), in the gravityunstable case (o
2
+

o
2
J
o0), the local equilibrium parameters of the disk
s
0
(r
0
), O(r
0
), c
s
(r
0
), and L(r
0
) determine the pattern speed
of unstable nonaxisymmetric (ma0) perturbations (in a
local rotating reference frame):
O
p
Ro
+
m
 2pGs
0
O
[o
2
J
[
1
r[k[L
, (25)
where 2pGs
0
[k[O
2
, [o
J
[
2
O
2
, rk
2
[L[b1, and, therefore,
O
p
O=rk
2
L5O. As is seen, the typical pattern speeds of
spiral structures in unstable disks are only a small fraction
of some average angular velocity O
av
. Because O
p
does not
depend on m, each Fourier component of a perturbation in
an inhomogeneous system will rotate with the same
constant angular velocity. The theory states that in
homogeneous ([L[ o) disks O
p
= 0.
At the limit of gravitational stability, the two conditions
qo
2
J
=qk = 0 and o
2
J
X0 are fullled. The rst condition
determines the most unstable wavelength (the modied
JeansToomre wavelength)
l
crit

2c
2
s
Gs
0
, (26)
corresponding to the minimum on the dispersion curve
(21). Use of the second condition determines the critical
sound speed for the stability of arbitrary but not only
axisymmetric perturbations (Lau and Bertin, 1978; Lin and
Lau, 1979; Morozov, 1985; Griv et al., 1999, 2002).
From Eq. (21), the disk is Jeansunstable to both
axisymmetric (radial) and nonaxisymmetric (spiral) pertur
bations if c
s
oc
T
, where
c
T
=
pGs
0
k
(27)
is the ordinary SafronovToomre (Safronov, 1960, 1980;
Toomre, 1964, 1977) critical sound speed to suppress the
instability of axisymmetric m = 0 perturbations only. On
the other hand, writing the sound speed as c
s
=
RT
_
, the
SafronovToomre instability condition reads
ToT
crit
p
2
G
2
s
2
0
Rk
2
, (28)
where T is the local gas temperature. Both radial and spiral
instabilities can occur in any region of a disk that becomes
ARTICLE IN PRESS
0 0.5 1
0
0.2
0.4
0.6
0.8
1
r/R
f
(
r
)
,
0
,
V
r
o
t
(
a
r
b
i
t
r
a
r
y
u
n
i
t
s
)
f
(
r
)
,
0
,
V
r
o
t
(
a
r
b
i
t
r
a
r
y
u
n
i
t
s
)
f
(
r
)
,
0
,
V
r
o
t
(
a
r
b
i
t
r
a
r
y
u
n
i
t
s
)
f (r)
V
rot
0
(a)
0 2 4 6
0
0.2
0.4
0.6
0.8
1
r
0
V
rot
f (r)
(b)
0 0.5 1
0
0.2
0.4
0.6
0.8
1
r/R
f (r)
V
rot
0
(c)
Fig. 2. The distribution function for angular momentum f (r) = s
0
Ok
2
(Lovelace and Hohlfeld, 1978, p. 55) vs. a normalized radius r=R (or ar) for the
equilibrium disk models calculated from the surface density s
0
(r), the angular rotation velocity O(r) = V
rot
=r, and the epicyclic frequency k(r): (a) Hunters
n = 1 model; (b) the exponential disk model; and (c) Hunters n o model.
E. Griv / Planetary and Space Science 55 (2007) 547568 554
sufciently cool, ToT
crit
. Thus, disks with Toomres Q
c
s
=c
T
parameter o1 (or ToT
crit
, respectively) break up
into fragments of preferred mass
M
frag
pl
2
crit
s
0
= 4pc
4
s
=G
2
s
0
. (29)
Similar results hold for a selfgravitating Safronov
Toomre unstable disk of nite but small thickness in
hydrostatic equilibrium with the critical wavenumber
k
crit
= 2p=l
crit
reduced by a factor of 2 (Morozov, 1981).
The m40, i.e., nonaxisymmetric instabilities in a
differentially rotating (dO=dra0, that is, 2O=k41) disk
is more difcult to stabilize; stability is achieved only for
sufciently large sound speed (although still of the order
of c
T
)
c
s
\
2O
k
c
T
 2c
T
, (30)
or for sufciently large temperature T\(2O=k)
T
crit
 2T
crit
, respectively (Lin and Lau, 1979; Morozov,
1980, 1985; Griv et al., 1999, 2002; Griv and Gedalin,
2003). Thus, nonaxisymmetric disturbances in a nonuni
formly rotating system are more difcult to suppress
than the axisymmetric ones, in general agreement with
the work by Goldreich and LyndenBell (1965) and
Julian and Toomre (1966). The free kinetic energy
associated with the differential rotation of the system
under study is one possible source for the growth of the
energy of these spiral Jeanstype perturbations, and
appears to be released when angular momentum is
transferred outward. See Lau and Bertin (1978), Lin and
Lau (1979), Morozov (1985), and Montenegro et al. (1999)
for a discussion of the problem.
Thus, if the disk is thin, c
s
5rO, and dynamically cold,
c
s
oc
T
(Toomres stability parameter Qo1 or ToT
crit
),
then such a model will be gravitationally unstable, and it
should almost instantaneously (see below for a time
estimate) take the form of a cartwheel, that is, a structure
of spirals and rings (Fig. 3a). (One understands, however,
that in a nonlinear state of instability there will be some
exchange of energy and angular momentum between the
radial and spiral modes which will give rise to a pattern
more complex than the cartwheel shown in Fig. 3a.)
Clearly, in the case of both axisymmetric and nonaxisym
metric excitation, the distribution of the surface density
along the spiral arms is not uniform, but describes a
sequence of maxima that might be identied with forming
planets (Fig. 4). The number of maxima (hot spots)
N
spots
seen in Fig. 4 is roughly
N
spots
=
R
l
crit
, (31)
where l
crit
is given by Eq. (26), ToT
crit
, and T
crit
is the
averaged critical SafronovToomre temperature (Eq. (39)).
This dynamical instability is driven by a strong nonreso
nant interaction of the gravity uctuations (e.g., those
produced by a spontaneous disturbance or, in rare cases, a
satellite system) with the bulk of the particle population,
and the dynamics of Jeans perturbations can be character
ized as a nonresonant interaction, that is, in Eq. (17),
o
+
lka0. Toomres Qparameter that is o1 suggests
that the disk is likely subject to both radial and spiral
instabilities and might, therefore, be clumpy: if the local
SafronovToomre instability criterion Qo1 is satised, the
disk should break up into discrete blobs of matter
distributed in spirals around the spin axis (Fig. 4).
Interestingly, Greaves (2005) and Greaves et al. (2005)
have detected arcandlump debris disks orbiting nearby
solartype stars. Both optical and nearinfrared observa
tions of premainsequence stars of intermediate mass have
also revealed the structure of spirals and rings, and thus
presumably the Jeans instability of ring and spiral
perturbations, in the circumstellar disk with structure more
than 100 AU from the parent star (Grady et al., 2001;
Clampin et al., 2003; Fukagawa et al., 2004, 2006).
Observations at submillimeter wavelengths have been
conrmed that the dust around Vega is distributed in a
clumpy ring (Koerner et al., 2001). Both the James Clerk
Maxwell Telescope and Keck images of the dust have
revealed the inner rings with azimuthal nonaxisymmetric of
Eridani, Fomalhaut, and b Pictoris (Greaves et al., 1998;
Holland et al., 2003; Macintosh et al., 2003; Wahhaj et al.,
2003). One can suggests that all these disks are the Jeans
unstable to both axisymmetric and nonaxisymmetric
perturbations. Observations may provide an indication
on whether the Qvalue for those disks is indeed compar
able to or less than unity.
6
Note that the present model is
based on the selfexcited (i.e., intrinsic), nonresonant,
almost aperiodic gravitational instability in the circum
stellar disks. To explain warps, radial and spiral structures,
and other azimuthal and radial asymmetries of the resolved
disks around mainsequence stars, alternative models have
been also suggested by Takeuchi and Artymowicz (2001),
Augereau and Papaloizou (2004), Kenyon and Bromley
(2004), Ardila et al. (2005), and Quillen et al. (2005) by
exploring the roles of radiation, gas drag forces from the
gas disk, and tidal forces from nearby stars, brown dwarfs,
or planets in creating and maintaining the spiral structures
in resolved disks. The observed properties of circumstellar
disks around young stars have been summarized by Dutrey
et al. (2004).
Contrary, if the disk is thin and warm, QX1 but
Qt2O=k  2 (or T
crit
pTt2T
crit
), then such an uncooled
model will be unstable only with respect to spiral
perturbations (Fig. 3b) and cannot be therefore fragment.
An uncooled hot model with Q\2O=k  2 (or T42T
crit
) is
Jeansstable. In the latter cases the formation of planets
can only be imagined via gradual accumulation of solid
ARTICLE IN PRESS
6
The direct estimation of Q value, which in turn the measurements of
surface density, is still very difcult, partly because we cannot well resolve
the disk around premainsequence stars in millimeter wavelengths. Also,
the dust opacity can be a source of large uncertainty, as particles grow.
Anyway, the future interferometers will greatly improve the situation with
much higher resolution (Fukagawa, 2005).
E. Griv / Planetary and Space Science 55 (2007) 547568 555
material (Pollack et al., 1996; Inaba et al., 2003; Alibert
et al., 2005).
Adams et al. (1989) have already pointed out, by
numerical integrations, that spiral density waves can be
excited on a reasonable timescale even if Toomres Q
parameter is greater than 1. Numerical simulations have
shown that nonaxisymmetric disturbances, which grow
as multiarmed spirals, become unstable for Qt1:5
(cf. Eq. (30)) (Papaloizou and Savonije, 1991; Tomley
et al., 1991, 1994). It has already been found in simulations
that the stability number Q of Toomre in relaxed
equilibrium disks does not fall below a critical value,
which lies about Q
crit
= 222:5 (e.g., Tomley et al., 1991,
1994). However, no adequate explanation of the latter has
been presented.
The fact that selfgravitating, differentially rotating
(dO=dra0 or 2O=k41, respectively) stellar disks can
exhibit strong nonaxisymmetric responses even when the
axisymmetric stability criterion Q = 1 is fullled, was most
convincingly demonstrated already by Julian and Toomre
(1966), concerning the gravitational effect of any single
orbiting mass concentration (such as gas lumps in
galaxies). These forced spiral waves are not to be confused,
of course, with Lin and Shus fully selfconsistent density
wave proposal explored in the present theory. Similar
result for selfgravitating, differentially rotating gas disk
was obtained by Goldreich and LyndenBell (1965). In
Toomre (1981) this amplication was discussed in terms of
swingmechanism, very reminiscent of the way I reach
the nonaxisymmetric stability criterion (30). Lau and
Bertin (1978, p. 509) have claried the problem by
considering the motion of a uid element: the density
response that is in phase with the potential minimum is
found to exceed, by an amount proportional to both dO=dr
and m, the corresponding response due to an axisymmetric
eld of equal strength. The modied SafronovToomre
stability criterion given by Eq. (30) is obviously only an
approximation, as it assumes spiral arm pitch angles c that
are relatively small, [ tan c[t1, and neglects (among other
things) the thickness of the disk. Nevertheless, numerical
experiments suggest that the formulation is approximately
correct (Morozov, 1981; Khoperskov et al., 2003; Liverts
et al., 2003; Griv and Gedalin, 2005).
A slightly improved SafronovToomre criterion was
obtained by Morozov by including a weak destabilizing
effect resulting from spatial inhomogeneity of the disk
(Morozov, 1980) and a weak stabilizing effect resulting
from the small but nite thickness of the disk (Morozov,
1981). Morozov (1981) has also shown that the effects of a
nite thickness and spatial inhomogeneity practically
cancel each other, at least for the parameters of the stellar
disk of our own Galaxy in the solar neighborhood.
ARTICLE IN PRESS
(a) (b)
Fig. 3. A schematic model of a Jeansunstable disk: (a) the SafronovToomre unstable disk (Qo1) and (b) the SafronovToomre stable disk (Qo1 but
Qt2O=k, and in highly attened protostellar disks 2O=k  2). In a differentially rotating disk (2O=k41), nonaxisymmetric, that is, spiral modes of
collective oscillations are more unstable than axisymmetric (radial) ones.
(a)
(b) (c)
Fig. 4. (a) Gravitationally unstable density waves with m = 1 arm in the (r; j)plane, (b) density waves with m = 2 arms, and (c) density waves with m = 3
arms. The lled circles represent the maxima of the perturbed density of Jeans waves, which are unstable to both radial and spiral perturbations. The
distribution of the surface density along the spiral arms is not uniform, but describes a sequence of maxima (protoplanets).
E. Griv / Planetary and Space Science 55 (2007) 547568 556
The growth rate of the instability is relatively high:
Io
+

2pGs
0
(k
2
+
=[k[)
_
and in general Io
+
O, that is, the
instability develops rapidly on a dynamical timescale (on a
time of 23 disk rotations, or t10
4
yr in the protostellar
disk). An important feature of the instability under
consideration is the fact that it is almost aperiodic
([Ro
+
=Io
+
[51) in a rotating frame. From Eq. (21), the
growth rate of the instability has a maximum at the
wavelength l
crit
 2c
2
s
=Gs
0
. At the boundary of instability,
that is, Q  1, l
crit
 2p
2
Gs
0
=k
2
(324)ph. It means that
of all harmonics of initial gravity perturbation, one
perturbation with l
crit
 10h, the associated number of
spiral arms m
crit
, and the pitch angle c
crit
will be formed
asymptotically in time of a single rotation. Because l
crit
bh,
the approximation of an innitesimally thin disk used
throughout the theory does not fail.
In the another, opposite to (23) frequency range,
[o
+
[
3
4pGs
0
O(m=r[kL[)5[o
J
[
3
, (32)
that is, [o
+
[5O, Eq. (20) has another root equal to
o
+3
 4pGs
0
O
o
2
J
m
r[k[L
. (33)
The root (33) describes the gradient (L
1
a0) branch of
oscillations. As is seen, the gradient perturbations are stable
and are independent of the stability of Jeans modes. These
lowfrequency ([o
+3
[5O), stable (Io
+3
= 0) oscillations are
obviously not important in dynamics of protostellar disks.
According to Morozov (1985), analogous oscillation
branches, with frequencies proportional to the gradients of
the undisturbed equilibrium quantities, occur in the
terrestrial atmosphere (internal gravity waves), in inhomo
geneous plasmas (drift waves), in the terrestrial oceans
(Rossby waves), and in other inhomogeneous systems.
6. Turbulent heating
The Jeans instability of a disk will in fact be suppressed
because of the efcient turbulent heating of the medium
during the growth of that same instability: in the course of
developing the instability, the speed of sound (temperature)
grows, and the disk approaches the boundary of the
instability. Gravitational instabilities convert potential
energy into kinetic energy of random motions. Indeed, by
taking into account that terms o exp(ik
r
r imj io
+
t) do
not contribute (since they cancel out if they are averaged),
from Eqs. (9) and (16) one obtains that the perturbed
potential energy U is a negative quantity:
U =
s
1
F
+
1
2
_ _
=
[k[[
~
F[
2
2pG
e
2Io
+
t
, (34)
where ) denotes the time average over the fast
oscillations,
s
0
= s) =
1
T
_
T
0
sdt; s
1
) = F
1
) = 0,
and Tb2p=o is the characteristic time of the quasilinear
relaxation, i.e., the time during which the oscillations
inuence the equilibrium disk state, and F
+
1
is the complex
conjugate potential. Clearly as gravitational instability
develops the amplitudes [U[ of the unstable Fourier
harmonics of the perturbation will grow, setting energy
free to heat the medium. This effect will in turn diminish
the growth rate of the instability (see Eq. (21) above) and
ultimately cause it to become saturated. Fresh waves must
be continually created to maintain the density wave
pattern.
Following Morozov (1978), let us obtain mathematical
expressions describing the heating process. For simplicity,
the gas is supposed to obey a polytropic law
P = P
0
(s=s
0
)
g
, (35)
where the twodimensional adiabatic index g can be
mapped to a threedimensional adiabatic index in the
lowfrequency (static) limit; in protostellar disks
g = 1:522. Eq. (35) can be expanded as far as terms
quadratic in the amplitude of the perturbed density
P = P
0
g
P
0
s
0
ds
g(g 1)
2
P
0
s
2
0
ds
2
, (36)
where ds = s
1
2
s
2
and 51. Next, the result can be
averaged in the sense indicated above:
P P
0
) = dP(t) =
g(g 1)P
0
s
2
0
[ ~ s[
2
e
2Io
+
t
, (37)
where the growth rate Io
+
(t) is determined from Eq. (21)
by the relation
Io
+
= [k
2
2pGs
0
(k
2
+
=[k[) g(P
0
dP(t))k
2
+
=s
0
]
1=2
.
(38)
Eqs. (37) and (38) completely describes the heating process.
Of course, the system of Eqs. (37)(38) is correct only in the
linear approximation used throughout the theory,
[s
1
=s
0
[51, [P
1
=P
0
[51, and [F
1
=F
0
[51. The linear wave
considered in the present study is growing on the timescale
of about one cycle rotation period (O
1
) and in one cycle
rotation period it will reach nonlinear amplitudes. Strictly
speaking, this theory of weak turbulence describes only the
tendency of the disk to be heated by Jeansunstable density
waves and shows the direction of the disks evolution. One
understands that we still have to develop the theory of
strong turbulence.
Taking into account the fact that c
2
s
o P P
0
), from
Eq. (37) one obtains that during the linear (quasilinear)
stage of the instability the effective temperature c
2
s
grows
with time according to the law
c
2
s
o t. (39)
Of course, this is not really temperature, it is ordered
motion, i.e., coherent mechanical oscillations of the gas
element in response to the uctuating elds.
It is obvious that the uncooled disk manages to keep its
local stability parameter close to the critical value, Q  2
ARTICLE IN PRESS
E. Griv / Planetary and Space Science 55 (2007) 547568 557
(Griv et al., 1999, 2002; Griv and Gedalin, 2003). In this
case, once the differentially rotating disk has been heated
to values Q  2 (or c
s
 2c
T
, respectively) by gravitation
ally unstable density waves, no further unstable waves can
be sustained by virtue of the Jeans instabilityunless some
cooling mechanism is available leading to Toomres Q
value under approximately 2 or to the value of c
s
smaller
than approximately 2c
T
, respectively. To repeat myself, it
has been found in simulations that the stability number Q
of Toomre in relaxed equilibrium disks does not fall below
a critical value, which lies about Q
crit
= 222:5 (Tomley
et al., 1991, 1994). However, no adequate explanation of
the latter has been presented.
The development of gravitational instabilities would lead
to a selfregulation process. Namely, if the differentially
rotating disk is initially cold (QoQ
crit
), then gravitational
instabilities would repeatedly heat it according to the law
(39) on the dynamical timescale, bringing it toward
stability. On the other hand, if the disk is initially hot
enough (QXQ
crit
), then radiative cooling is going to bring
the value of Toomres stability parameter Q down toward
an unstable conguration. A similar approach has been
suggested in computer simulations (Lodato and Rice, 2004,
p. 630).
Because the typical relaxation time, that is, the time
required to stabilize the initially unstable disk tO
1
, the
disk fragments when t)Ot1, where t) is an effective
cooling time dened as the average internal energy of the
model divided by average cooling rate (see, e.g., Gammie,
2001). Thus, I believe to have obtained a theoretical
interpretation of Johnson and Gammies (2003)
experimental result: pure gaseous disks with t)Ot1
fragment rapidly on the dynamical timescale and thus do
not exist.
A cooled disk could generate transient Jeansunstable
density waves even without an external trigger. Here I refer
to papers by Miller et al. (1970), Quirk (1972), Sellwood
and Carlberg (1984), Carlberg (1987), Griv and Chiueh
(1998), Gammie (2001), and Johnson and Gammie (2003)
as evidence for how this could come about. Note especially
the work by Quirk (1972), who shown that the azimuthal
gravitational forces produced by the spiral arms are
necessary to maintain the longterm recurrent wave
structure of the galaxy in the gas clouds component,
which was dissipative through inelastic cloudcloud colli
sions. As cooling process always exists in the protostellar
disks, the Jeans instability can be considered to be a long
term generating mechanism for unstable density waves,
thereby leading to recurrent shortlived O
1
10
4
yr spiral
or spiralring patterns in the observed protostellar and
circumstellar disks.
7. Angular momentum transfer
Let us now turn to the question of how to account for
the concentration of angular momentum in the planets and
of mass in the star. The collective torque per unit area
exerted by the gravity perturbations on the disk is
(1=r)(dG=dr) =
_
dj(r =F
1
)s
1
) or
1
r
dG
dr
=
_
2p
0
s
1
(r; j
/
)
qF
1
(r; j
/
)
qj
/
dj
/
_ _
. (40)
Using Eq. (17), in terms of the Fourier components
dened in Eq. (1), G =
o
m=1
G
m
, from Eq. (40) one nds
1
r
dG
m
dr
 8p
m
2
rL
s
0
O
k
2
Io
+
F
1
F
+
1
if Io
+
40, (41)
or (1=r)(dG
m
=dr) = 0 if Io
+
p0, and F
1
F
+
1
= [
~
F[
2
exp
(2Io
+
t). Eq. (41) is correct only in the main domain of
the system under study between the inner and outer
Lindblad resonances. A special analysis of the solution
near corotation (o
+
= 0) and Lindblad (o
+
k = 0)
resonances is required. Thus, the points r
ILR
and r
OLR
in
which o
+
k = 0 are called the points of inner and outer
Lindblad resonances. They play an important role in the
theory: the solution of spiral type given by Eq. (2) rapidly
oscillating in the radial direction ([k
r
[rb1) lies between r
ILR
and r
OLR
. Outside the resonances, ror
ILR
and r4r
OLR
, the
solution decreases exponentially (Lin and Lau, 1979).
A special analysis of the solution near spatially limited
corotation and Lindblad resonances is required. Reso
nances of a higher order, o
+
lk = 0 and l = 2; 3; . . . ; are
dynamically of less importance (Lin and Shu, 1966; Lin
et al., 1969; Shu, 1970). To emphasize it again, the present
analysis is restricted to consideration of the main part of a
disk between the Lindblad resonances. The spatially
limited waveparticle resonances have been investigated
by LyndenBell and Kalnajs (1972), Goldreich and
Tremaine (1979, 1980), MeyerVernet and Sicardy (1987),
and Griv et al. (2000).
Four physical conclusions can be deduced from Eq. (41).
(a) The distribution of the angular momentum of a disk
will be changed under the action of only the nonaxisym
metric forces o m. The latter is obvious: axially symmetric
motions of a system produce no gravitational couplings
between the inner parts and the outer parts. (b) The
distribution of the angular momentum will be changed
with time only under the action of growing, that is,
gravitationally (Jeans) unstable perturbations (Io
+
40).
(c) Growing spiral perturbations can transfer angular
momentum only in spatially inhomogeneous disks
((d=dr)(s
0
Ok
2
)a0). This result is also anticipated, be
cause in homogeneous disks the angular velocity of spiral
perturbations O
p
= 0 (see Eq. (25) above), and, therefore,
there is no exchange of angular momentum in the wavegas
system. And nally (d) because in selfgravitating disks in
equilibrium Lo0, (1=r)(dG
m
=dr)40: an applied gravita
tional torque increases the angular momentum of the given
gas element and thus leads to motion of the gas element at
a larger radius and thus tends to decrease O. (As is known,
O(r) in a selfgravitating disk is a decreasing function of r,
whereas the angular momentum of a unit mass, Or
2
, is an
increasing function of r.) This takes place in the main
domain of the disk between the Lindblad resonances where
ARTICLE IN PRESS
E. Griv / Planetary and Space Science 55 (2007) 547568 558
spiral density waves are selfexcited via a nonresonant
waveuid interaction. This in turn cannot be done for all
masses because the total orbital momentum must remain
constant. As a result, a group of inner particles with radii
ror
ILR
moves inward and a group of particles with radii
r
ILR
oror
OLR
moves outward. The secular evolution of
protostellar disks, therefore, proceeds in the direction of
increasing central mass concentration in the baryonic
material and of extending of outer portions (see Fig. 5).
As a result, the bulk of angular momentum is transferred
radially outward, whereas an inner gaseous medium moves
inward, losing a large part of the angular momentum.
7
LyndenBell and Kalnajs (1972, p. 6) proved that the
gravitational torques can only communicate angular
momentum outward if the spirals trail. LyndenBell and
Kalnajs as well as Goldreich and Tremaine (1979, 1980)
and MeyerVernet and Sicardy (1987) considered, however,
only resonant waveparticle interactions at spatially
limited resonances (in Eq. (17), o
+
k = 0) in gravitation
ally stable disksatellite systems (Io
+
= 0). The outward
transfer of orbital momentum allows the central parts of
disks to contract and, thus, allows the star to contract
without breaking up, and the remnant disk of gas is the
reservoir for forming planets.
Eq. (41) derived in this paper to describe the wavegas
exchange of angular momentum for unstable protostellar
disks is a new result, and one has to realize that this is the
focus of the paper. At the qualitative level, the inuence of
viscosity on a rotating mass of gas was well understood by
1920s (LyndenBell and Pringle, 1974; Pringle, 1981).
(Following Pringle, 1981, by viscosity I mean the mechan
ism, whatever it is, which enables angular momentum to be
transferred and energy to be dissipated.) In general, the
dissipative processes act to spread the disk out, allowing
the inner parts to move in and necessitating, through
conservation of angular momentum, the outer parts to
move out (Pringle, 1981, p. 140). LyndenBell and Kalnajs
(1972), by considering the evolution of at galaxies, also
suggested that selfgravitating disks want to transfer their
angular momentum outwards. Numerical experiments
have been demonstrated that gravitational instabilities
produce growing nonaxisymmetric density waves and
associated gravitational torques, which are potent agents
of angular momentum transport (Anthony and Carlberg,
1988; Tomley et al., 1991; Laughlin and Ro zyczka, 1996;
Yorke and Bodenheimer, 1999; Gammie, 2001; Pickett
et al., 2003). Eq. (41) shows that although the angular
momentum exchange between gas and the growing wave
is of second order in the perturbation potential F
1
,
it is obtained by taking inner products of rstorder
solutions. Thus, going to nonlinear (quasilinear) approx
imation of the theory is unnecessary to derive the
nonresonant waveuid angular momentum exchange
(cf. LyndenBell and Kalnajs, 1972, p. 10).
According to Eq. (41), the angular momentum transfer
efciency of nonaxisymmetric Jeansunstable density waves
depends on their spatial and temporal form. Let us next
evaluate the collective torque for a realistic model of the
disk. In accordance with the theory, the fastest growing
mode with m\1, k
+
= k
crit
, and Io
+
O is considered.
Taking into account that 8pm
2
F
1
F
+
1
F
2
0
(an astrophysicist
might well consider a perturbation with F
1
=F
0
of
1
10
to be
quite small) and F
0
r
2
O
2
=2, where F
0
is the basic
potential, from Eq. (41) one obtains [(1=r)(dG
m
=
dr)[s
0
r
3
O
2
=4[L[. The angular momentum of the given
gaseous element P = s
0
r
2
O. Then the characteristic time of
the angular momentum redistribution is tP=[(1=r)
(dG
m
=dr)[(4[L[=r)O
1
O
1
. Thus, already in the rst
disk revolution, in, say, about 10
4
yr, the unstable
protostellar gas disk sees a large portion of its angular
momentum transferred outward, and mass transferred
both inward and outward (Fig. 5). I suggest that the
gravitational instability studied here can give rise to
torques that can help to clear disks around stars on
timescales of 10
5
210
6
yr, in accord with astronomical
requirements (Taylor, 1992, Section 2.9.2; Bally et al.,
1998; SiliciaAguilar et al., 2005; Greaves, 2005; Roberge
et al., 2005).
I must emphasize, however, that here the stability of an
unbounded, isolated, twodimensional system is treated, by
applying the local WKB approximation. Thus the numer
ical results given above should be regarded merely as
approximate estimates, in order of magnitude.
Closing, Paczynski (1978) considered a model of a thin
disk which rotates around a central compact object that is
marginally stable to selfgravity. His model followed the
standard accretiondisk theory by Shakura (1973), Shakura
and Sunyaev (1973), and LyndenBell and Pringle (1974).
These standard models have understood angular
momentum transport in that they assumed that it is due
ARTICLE IN PRESS
0 0.5 1 1.5
0
0.5
1
1.5
2
r, radius
0
,
s
u
r
f
a
c
e
d
e
n
s
i
t
y
t =0
t =
Fig. 5. Surface mass density of a disk before (t = 0) and after (t = o)
initially Jeansunstable gravity disturbances have increased.
7
Additional emission of angular momentum occurs at the inner
Lindblad resonance and absorption of angular momentum occurs at the
outer Lindblad resonance and at corotation (LyndenBell and Kalnajs,
1972).
E. Griv / Planetary and Space Science 55 (2007) 547568 559
to turbulent stresses. Accordingly, turbulent viscosity
driven by local disk instabilities was assumed to transfer
angular momentum outwards and makes the accretion of
matter onto the central object possible. There is one
important difference between the model presented here and
the models by Paczynski (1978), Shakura (1973), Shakura
and Sunyaev (1973), and LyndenBell and Pringle (1974),
namely in the present model the physical cause for the
turbulence is included selfconsistently by considering a
selfconsistent system of the gasdynamic and Poisson
equations. Paczynski (1978) also considered the vertical
structure, which is a less simplied approach than
assuming an innitesimally thin disk, but this is not
required for an understanding of angular momentum
transport.
8. Turbulent diffusion
Since one of the reasons for the Jeans instability is the
inhomogeneity of the disk, oscillations developing in
consequence of the instability cannot die away until the
inhomogeneity is completely destroyed. In other words,
such an instability must lead to turbulent diffusion: during
the evolution of unstable disks mass must be transported
radially both inward and outward while angular momen
tum must be transferred outward.
It is quite generally accepted that the accretion of
material onto rotating astrophysical disks is inefcient if
the viscosity in the disk is determined by the classical
transport coefcients. The presence of developed turbu
lence is usually postulated to explain the observed features
of disks rotating about galactic centers, black holes,
neutron stars, protostars, and Saturns rings. Shakura
(1973) and Shakura and Sunyaev (1973) proposed that
turbulence might enhance the transport, and so introduced
a turbulent or eddy viscosity:
n
eddy
aHc
s
, (42)
where a is dimensionless number (less than unity because
turbulent eddies would most likely be subsonic and no
larger than the scale height H = 2h of the disk), and the
sound speed c
s
evaluated at the disk midplane where most
of the mass is concentrated. The source of this anomalous
turbulence has been highly controversial, and the para
meter a summarizes the uncertainties related to the sources
of anomalous viscosity. See Balbus and Hawley (1998) and
Stone et al. (2000) for thorough reviews.
In the spirit of Morozov and Khoperskov (1990), let us
showin a speculative mannerthat in gravitating sys
tems the anomalous turbulent diffusion arising by the
Jeans instability may exceed the ordinary microscopic
(molecular) diffusion substantially. Unlike Morozov and
Khoperskov (1990), who considered the gradient convec
tivetype instability (similar to the internal gravitational
waves in the atmosphere of the Earth, the Rossby waves in
the terrestrial oceans, or drift waves in plasmas) in the
nonselfgravitating limit, the gravitational instability is
studied.
The gravitational instability is most likely to be the
mechanism which does produce the effective disk diffusion.
The selfsustained turbulence with outward angular mo
mentum transport that may arise as a result of instability is
related to stochastic motions of uid elements. By
considering a plasma, Kadomtsev (1965) has estimated
the diffusion coefcient D
tur
from the exponential damping
factor determining its instability:
D
tur
Io
+
k
2
min
, (43)
where k
min
represents the minimum value of the wave
number (the maximum value of the wavelength, or the
principal scale) for which one still has an instability, and
[Io
+
=Ro
+
[b1. Morozov and Khoperskov (1990) have
emphasized that Eq. (43) is valid if Io
+
b[Ro
+
[ (otherwise,
the nature of the estimate changes). Expression (43) given
in Kadomtsevs review for the turbulent viscosity
resulting from the development of drift or dissipative
instabilities is quite universal (Horton, 1984). Indeed, from
dimensional considerations, turbulent diffusivity is of the
form
D
tur
t
1
l
2
pl
,
where t is the time of the correlations disappearance that is
reasonable to choose I
1
o
+
, and l
pl
is the characteristic
scale of turbulent uctuations in the direction normal to k
(Branover et al., 1999). The turbulence scale in the disk
plane l
pl
can be evaluated as follows (to order of
magnitude): l
pl
k
1
pl
, where the dimension of uctuations
k
1
pl
is obtained from the condition of maximum growth
rate. Then we will have a good estimate (43).
According to the results of the stability analysis
described above, in an unstable disk Io
+
O, k
min
1
2
h
pGs
0
=c
2
s
, and c
s
pGs
0
=k. Substituting these quantities
into Eq. (43), one obtains
D
tur
4Oh
2
p
2
G
2
s
2
0
O
3
. (44)
In Eq. (44), the requirement of the hydrostatic equilibrium
in the z direction for a selfgravitating slab model was
adopted, which gives
h
c
s
4pGr
0
_
c
2
s
2pGs
0
and r
0
s
0
=2h is the volume mass density in the midplane.
A similar result, D
tur
G
2
s
2
0
=O
3
, was already discussed by
Lin and Pringle (1987) and Takeda and Ida (2001) by using
a simple dimensional analysis. Daisaka et al. (2001) have
found numerically that in planetary rings in the presence of
the growing spiral wakes, the effective viscosity n
eff
is given
as n
eff
= CG
2
s
2
0
=O
3
, where the nondimensional correction
factor C = 6220 (cf. Eq. (44)). As was pointed out to the
author by the rst referee of the paper, the main result,
ARTICLE IN PRESS
E. Griv / Planetary and Space Science 55 (2007) 547568 560
Eq. (44)the derivation of the expression for the effective
turbulent viscositycan be obtained by dimensional
analysis without the formalism elaborated in the text.
Clearly, the turbulent viscosity is, in order of magnitude,
simply the square of the JeansToomre length scale
l
crit
= 2c
2
s
=GS
0
, where c
s
pGS
0
=k, for selfgravitating
disks times the angular velocity O of local disk rotation.
In the fully nonlinear regime, the justication provided by
the quasilinear analysis cannot be considered much better
than the dimensional analysis just stated above.
So, n
tur
4Oh
2
. Because c
s
Oh, this is of the same form
as the standard Shakura (1973) and Shakura and Sunyaev
(1973) a prescription of accretion disks (42), and it suggests
that a should be of order unity in a system (see also
Papaloizou and Lin, 1995; Lin and Papaloizou, 1996).
Let us now estimate the order of magnitude of D
tur
in the
protostellar gaseous (the atomic hydrogen) disk. Since the
velocity dispersion c
s
of particles of protostellar disks
cannot be determined observationally, we shall regard the
disk as marginally stable in the disk plane, that is,
c
s
pGs
0
=k10
3
cms
1
(we take the mass of the disk
M
d
 M
yr
1
.
9. Numerical experiments
9.1. The model
In this section, numerical experiments that simulate the
nonlinear development of gravitational instabilities are
used to test the validities of the theory. Selfgravitating
collisionless disks, represented by a distribution of Nb1
point masses in their own gravity eld, were set up
according to initial conditions. The evolution of the disks
were calculated for several rotation periods, until the disks
approached a statistical quasisteady state, in the sense that
certain azimuthally averaged quantities tended toward
constancy. To repeat myself, Anthony and Carlberg (1988),
Papaloizou and Savonije (1991), Tomley et al. (1991),
Adams and Benz (1992), Laughlin and Bodenheimer
(1994), Laughlin and Ro zyczka (1996), Yorke and
Bodenheimer (1999), Pickett et al. (2000), Gammie
(2001), Durisen et al. (2003), and other have used
numerical simulations to study the nonlinear behavior of
disks around young stars. Important results have been
obtained regarding the enhanced transport of angular
momentum and mass by gravitational instabilities. Cassen
et al. (1981), Pickett et al. (2003), Lodato and Rice (2004),
Mayer et al. (2002), and other have examined numerically
the fragmentation of the protostellar disk into protopla
nets. It has been stated that stable gas disks favor
formation of planetesimals by the accumulation of solid
material; unstable disks allow the possibility of direct
condensation of gaseous protoplanets. The aim of the
simulations presented here is just to illustrate the analytical
results obtained in Sections 5 and 7 above.
Tomley et al. (1991) have pointed out that the use of a
collisionless particle code to study a hydrodynamic system
requires justication. The gravitationally (Jeans) unstable
disks are characterized by collective interactions that act
over distances greater than the disk scale height, and it is
the description of such interactions for which the particle
code is well suited. Although the code cannot be used to
study acoustic phenomena, it is dissipationless. This is an
advantage over hydrodynamical codes, in which numeri
cally induced viscosity can affect the propagation of excited
spiral modes, and, therefore, the strength and distribution
of calculated torques. On the other hand, Griv et al. (2000)
have already demonstrated that there exist no dominant
instabilities in the collisionless Jeansunstable system that
have no counterparts for a gaseous system (see Tomley
et al., 1991, p. 531 for a discussion).
The traditional point of view is that, unlike in gaseous
systems, shock waves in almost collisionless particulate
systems are impossible. In fact, it has already been
proved that in particulate rotating systems the existence
of the socalled collisionless shock waves, with character
istic dimensions much smaller than the particle mean free
path, is possible (e.g., Friedman et al., 1981). This
possibility can serve as an effective mechanism of energy
transfer in gravitating collisionless systems. The physical
situation is very close to the condition in a collisionless
plasma described by Sagdeev (1966) where, in the presence
of the magnetic eld (cf. the regular rotation of gravitating
systems), a shock wave can be formed with the thickness of
the front much smaller than the mean free path of plasma
particles. Generally, similarities between selfgravitating
systems and ordinary plasmas arise from the common
longrange o 1=r
2
nature of the basic forces, whereas
differences arise from the opposite signs of these forces
(Lin and Bertin, 1984).
Thus, one concludes that the collisionless particle code
can be used to simulate Jeansunstable gas disks. Appar
ently, Cassen et al. (1981) and Anthony and Carlberg
(1988) were the rst to simulate the planet formation
ARTICLE IN PRESS
E. Griv / Planetary and Space Science 55 (2007) 547568 561
in gravitationally unstable gaseous disks by Nbody
experiments.
A disk of identical particles is investigated by direct
integration of Newtons equations of motion. The numer
ical procedure is rst to seek stationary solutions of the
Boltzmann kinetic equation in the selfconsistent eld
approximation for the equilibrium parameters of the
system, and then to determine the stability of those
solutions to spontaneous gravity perturbations. The
rapidly rotating model disk is considered. As a solution
of a timeindependent collisionless Boltzmann equation, at
the start of the Nbody integration, our simulation
initializes the particles on a set of concentric rings with a
circular velocity V
rot
of disk rotation in the equatorial
plane; the system is isolated. Then the position of each
particle is slightly perturbed by applying a random number
generator. The Maxwelliandistributed random velocities v
are added to the initial circular velocities V
rot
, and
[v[5[V
rot
[; the number of particles N = 120; 000. In the
numerical experiments using the particle code, the initial
temperature of the disk (the initial sound speed c
s
) is
represented by an imposed rootmeansquare random
velocity dispersion (Lin and Lau, 1979, p. 112; Binney
and Tremaine, 1987, p. 363; Tomley et al., 1991, p. 532).
The acceleration of the ith particle is
a
i
=
N
jai
(r
j
r
i
)
(r
2
ij
r
2
cut
)
3=2
. (46)
In Eq. (46), r
cut
is the socalled cutoff radius. This
softening of the gravitational potential is a device used
in Nbody simulations to avoid numerical difculties at
very close but rare encounters. As is known, the gravita
tional softening which is introduced in the model can be
interpreted physically: each of our particles has a nite
extentthey are in fact Plummer spheres with a scale size
of r
cut
(Athanassoula and Sellwood, 1986).
The initial surface density of the disk
s = s(0)(1 r
2
=R
2
)
3=2
,
the initial squared angular velocity
O
2
(r) = (3p
2
=4R)Gs(0)[1 (
3
4
)(r
2
=R
2
)],
the initial thickness h = 0:02R, the initial radial dispersion
of random velocities c
r
= 0:7c
T
(or Toomres stability
parameter Q = 0:7, respectively), the initial azimuthal
dispersion c
j
= (k=2O)c
r
, and the initial vertical dispersion
c
z
= 0:2c
r
, where s(0) is the surface density at r = 0.
Finally, the angular velocity O was replaced by (Hohl,
1972)
O = O
2
1
rs(r)
q
qr
[s(r)c
2
r
(r)]
_ _
1=2
.
Slight corrections have been applied to the resultant
velocities and coordinates of the model particles so as to
ensure the equilibrium between the centrifugal and
gravitational forces, to preserve the position of the disk
center of gravity at the origin, and to include the weak
effect of the nite thickness of the disk to the gravitational
potential. Thus, the initial model is very near the dynamical
equilibrium for all radii (see Griv and Chiueh, 1998 for a
discussion). A time t = 1 was taken to correspond to a
single revolution of the initial disk. In the experiment the
simulation was performed up to a time t = 10. It should be
noted here that after about three rotations the picture is
practically stabilized and no signicant changes in gross
properties of the model over this time are observed. Any
difference between the results of simulations with and
without applying the socalled quiet starts procedure (Griv
and Chiueh, 1998) to select the very regular initial
coordinates of particles was not found. A few runs for
larger systems containing N = 150; 000 model particles and
for smaller systems containing N = 10; 000 ones were also
performed. It was found that the results obtained for those
systems are qualitatively indistinguishable: I did not detect
in the experiments any dependence of the type
ds o N
1=2
,
where ds is the amplitude of the density variations. The last
is clearly inconsistent with Toomres (1990) and Toomre
and Kalnajs (1991) hypothesis that the spirals observed in
simulations can be explained by the swingamplied
particle noise (kaleidoscope of chaotic arm features
which are responses to the random density irregularities
orbiting within the particulate disk). I advocate a way to
describe the rapidly evolving planar structures, such as
those reported in the simulations, in terms of true
instabilities of collective Jeanstype oscillations.
As is seen from Eq. (41), the angular momentum transfer
rate depends on equilibrium surface density s
0
(r), but not
on the particle size or number, so that the timescale of
evolution of a particulate disk is independent of the
number of particles N that is used to represent the disk, if N
is large enough to represent the spatial density wave
structure. Moreover, tests indicated that the results were
insensitive to changes in other parameters: the cutoff
parameter in the range r
cut
= (0:00120:01)R, the initial
velocity dispersion in the range c
r
= (0:220:8)c
T
, the
initial vertical velocity dispersion c
z
= (020:3)c
r
, and
the initial disk thickness in the range h = (020:075)R.
9.2. Results
Fig. 6 displays a series of snapshots from a simulation
run. As calculations show, during the rst rotation,
spontaneous unstable gravity perturbations break the
SafronovToomre unstable system (Qo1) into several
fragments. In agreement with the theoretical prediction (see
Section 5 above), in the linear regime at a time t  0:5 the
disk takes the form of a cartwheel (cf. Fig. 3a). The
simulations show that at times t = 0:220:7, the model
resembles the cartwheel pattern most. In agreement with
the theory, the ring can be considered as the crest of a high
amplitude m = 0 Jeansunstable density wave, while the
ARTICLE IN PRESS
E. Griv / Planetary and Space Science 55 (2007) 547568 562
spokes are probably nothing but density maxima of spiral
(ma0) density waves. A straightforward estimate shows
that a number of hot spots seen in Fig. 6 at times t =
0:320:5 is in fair agreement with a theoretical expectation
based on Eq. (31). Interestingly, the Gemini North 8meter
telescope nearinfrared image of the disk around the binary
star pair GG Tauri AB has revealed a somewhat similar
cartwheellike structure (http:www.gemini.edu/gallery,
Astronomical Images). See also Roddier et al. (1996,
Plate 23). At the end of the third rotation, a quasi
stationary system of a massive star and a pair of minor
hot Jupiter is seen. The manybody calculations pre
sented here have nothing to do whatsoever with Wetherills
(1980) and Wetherill and Cox (1984, 1985) ones who
considered planets grow by the gradual addition of matter,
following twobody collisions and coherence (the classic
coreaccretion model).
From the edgeon view pictured in Fig. 7, one can see
that a fully threedimensional disk develops immediately at
t  1:0. A straightforward estimate shows that a mean
height h of the disk above the plane corresponds to the
force balance between the gravitational attraction in the
plane and the pressure due to the velocity dispersion c
z
(i.e., temperature) in the zdirection:
h 
c
z
4pGr
0
_ 
c
2
s
2pGs
0
,
where r
0
 s
0
=2h is the volume mass density in the
midplane (Binney and Tremaine, 1987). Clearly, this
pressuresupported (in the zdirection) threedimensional
structure seen to form very rapidly on the timescale of a
single vertical oscillation, O
1
, with rather sharp edges.
After a time t  1, practically there is no change in the
edgeon structure of the system under study; it is still a
disk.
Fig. 6 gives a feeling for the evolutionary process, but in
order to trace and quantify the growth of instabilities in the
disk, it is necessary to compute Fourier decompositions of
the surface density distribution for various mode numbers.
In Fig. 8a the time evolution of the Fourier spectrum of the
spiral pattern shown in Fig. 6 is plotted. The complex
Fourier coefcients are determined from the summation
A(p; m) =
1
N
N
j=1
exp{i[mj
j
p ln(r
j
)]],
where (j
j
; r
j
) are the polar coordinates of the jth particle,
the pitch angle of an marmed logarithmic spiral c is given
by c = arctan(p=m), positive p corresponds to trailing
spirals and negative p to leading. Thus, the power spectrum
is constructed by employing a technique where one regards
the particle, labeled by j, as a discrete dfunction to
calculate the density Fourier component. In addition, the
azimuthal mode number m assumes discrete values
compatible with the azimuthal symmetry, whereas the
radial wavenumber p=r
0
assumes continuous values be
cause of the background ow shear (Athanassoula
and Sellwood, 1986; Sellwood and Athanassoula, 1986;
Laughlin and Ro zyczka, 1996; Chiueh and Tseng, 2000;
Griv and Gedalin, 2005).
This logarithmic spiral Fourier analysis shows that the
peak of the signal of the dominant Fourier mode moves
from negative p at times tt0:3 to positive p with increasing
time, reaches a maximum at t = 0:520:7, and eventually
decays as the pitch angle continues to increase. It is natural
to attribute the observed instability to the dynamical Jeans
instability so far discussed in the paper.
ARTICLE IN PRESS
0
1
2
t = 0.5
0
1
2
t = 1.0
0
1
2
t = 1.5
0
1
2
3
t = 2.0
0
1
2
3
t = 2.5
0
1.5
2.5
3.5
t = 3.0
Fig. 6. The time evolution (faceon view) of a SafronovToomre unstable disk (c
r
oc
T
, or Qo1, respectively). The time is normalized so that the time
t = 1 corresponds to a single revolution of the initial disk, and the rotation is taken to be counterclockwise. The model is violently unstable against
spontaneous, both axisymmetric and nonaxisymmetric gravity perturbations.
E. Griv / Planetary and Space Science 55 (2007) 547568 563
As is seen in Fig. 8b, the mass density of the disk is
effectively redistributed by the unstable waves on the
dynamical timescale. In agreement with the theory, the
surface density becomes more peaked as the wave energy
increases (cf. Fig. 5). Growing density waves (trailing spiral
arms) transfer the angular momentum outwards (Fig. 8c).
Summarizing, spiral density perturbations arising from
gravitational instabilities grow and exert torques which
redistribute both mass and angular momentum. In agree
ment with the theory, in the quasisteady state disk (after a
time tO
1
) a major fraction of the mass (eventually
residing in the star) retains only a minor fraction of the
angular momentum, most of the latter deposited in outer
portions of the system under study.
As is seen in Fig. 6, in the nonlinear state of instability
once a signicant fraction of the particles has been caught
up in clumps, the furious production of new gravitationally
bound objects is ceased, and the system evolves as a set of
point masses. After a long time t\3 the system reaches
dynamical stability, with only a handful of objects
remaining (cf. Mayer et al., 2002). The study of the
nonlinear state is beyond the scope of the present paper.
10. Summary
The dynamically cold (Qo1, or Top
2
G
2
s
2
0
=Rk
2
,
respectively) thin (c
s
5rO) protostellar disk is likely subject
to radial and azimuthal gravitational Jeanstype instabil
ities and might therefore be clumpy. The instability is
driven by a strong nonresonant interaction of the gravity
uctuations with the bulk of the particle population; and
the dynamics of Jeans perturbations can be characterized
as a nonresonant waveparticle interaction. The growth
rate of the instability has a maximum at the wavelength
l
crit

2c
2
s
Gs
0
and the mass of a protoplanet
M
prot
pl
2
crit
s
0
.
At the boundary of instability (Q  1), l
crit
 2p
2
Gs
0
=
k
2
10h. Protostellar disks are likely to be unstable to the
gravitational instability on scales 10h, and locally stable
to the gravitational instability on scales much less than 10h
and much greater than 10h.
For the parameters of a marginally stable protostellar
disk (R1000 AU, k = 2p=T
orb
10
10
s
1
, the total mass
of M
disk
M
, and c
s
c
T
), one obtains the following
surface density s
0
1 g cm
2
, sound speed c
s
5
10
3
cms
1
, disk temperature T10
K, typical wavelength
l
crit
10 AU, and typical mass of a protoplanet
M
prot
10
3
M
300M