2.

4 Accretion disks
2.4.1 Governing equations – radial structure

• So far, we have considered accretion flows that, by construction, have zero net angular
momentum, which has allowed us to neglect the e↵ects of rotation as the gas flows in
towards our accreting object. However, this situation is unlikely to be encountered in
reality: any real astrophysical flow will have at least some angular momentum.
• Consider an initially spherical shell of gas around our mass M with a small but non-zero
angular momentum. If the gas in this shell conserves angular momentum as it flows
in towards M , then its rotational velocity will increase as vrot / r 1 . Inflow parallel
to the axis of rotation will occur in much the same fashion as it would if the gas were
not rotating, but inflow perpendicular to the axis of rotation will feel an increasing
degree of rotational support, and hence will proceed more slowly. Eventually, the
rotational velocity will reach the Keplerian velocity, and collapse in this direction will
stop entirely. The gas will therefore settle into a flattened rotating accretion disk.
• This process will also occur in flows that do not conserve angular momentum, unless
the process responsible for transferring angular momentum away is extremely efficient.
This is difficult to arrange, particularly once the inflow onto M becomes supersonic,
and so we expect accretion disks to be a common feature of accreting astrophysical
systems.
• It is therefore important to understand how accretion works in a disk geometry. In
our initial study of disk accretion, we will make a couple of important simplifying
assumptions. We will assume that the disk is axisymmetric, and that it is thin, with
a scale height h ⌧ Rdisk , the radius of the disk.
• In place of the usual 3D spatial mass density ⇢, we work in terms of the mass surface
density ⌃, defined as Z 1
⌃(R, t) ⌘ ⇢(R, z, t)dz, (153)
1
where R and z are the radial and vertical coordinates in our cylindrical coordinate
system. The velocity of the flow can be written as
~v = vR~êR + v ~ê , (154)
where ~êR and ~ê are unit vectors in the radial and angular directions, respectively. Our
assumption of axisymmetry then implies that the radial and tangential velocities vR
and v are functions of R but not of .
• To further simplify our treatment of the problem, we assume that the mass of the gas
in the disk is much less than the mass of the central object, so that we can ignore the
self-gravity of the disk, and that the motion of the gas is nearly Keplerian, so that
vR ⌧ v and r
GM
v ⇠ . (155)
R
• Having made these assumptions, it is straightforward to write down the continuity
equation in our cylindrical coordinate system:
@⌃ 1 @
+ (R⌃vR ) = 0. (156)
@t R @R
• The momentum equation requires a little more thought. The velocity divergence of
gas in an axisymmetric disk that is rotating in a perfectly Keplerian fashion can easily
be shown to be zero. In our 2D cylindrical setup, we have
1 @ 1 @v
r · ~v = (RvR ) + . (157)
R @R R
If the gas flow is perfectly Keplerian, then vR = 0, v = (GM/R)1/2 , and so both terms
in this expression vanish. In reality, of course, the flow will not be perfectly Keplerian
(since in that case there would be no accretion), but if the deviation from Keplerian
rotation is small, then the velocity divergence will also be small. We therefore expect
the dominant source of viscosity to be the shear viscosity acting between fluid elements
orbiting at slightly di↵erent radii within the disk.
• A convenient way to explore the e↵ects of this shear viscosity is through its e↵ects
on the angular momentum of the rotating gas. In the absence of viscosity, angular
momentum is conserved within the disk, since the only force acting is the gravity of
the central object, which exerts no torque upon the gas. Therefore, in the absence of
viscosity, the angular momentum per unit area of the disk obeys a simple continuity
equation:
@
⌃R2 ⌦ + r · ⌃R2 ⌦~v = 0. (158)
@t
Axisymmetry implies that the component of the divergence term is zero, and so we
can rewrite this equation as
@ 1 @
⌃R2 ⌦ + ⌃R3 ⌦vR = 0, (159)
@t R @R
or alternatively
@ @
R ⌃R2 ⌦ + ⌃R3 ⌦vR = 0. (160)
@t @R
• If the gas in the disk is rotating in Keplerian fashion, or indeed in any form of di↵eren-
tial rotation such that the angular velocity ⌦ varies with radius, then the di↵erence
in angular velocities results in a viscous torque acting on the gas. If we consider two
thin annuli of material on either side of some surface of constant R in the disk and with
vertical thickness dz, then it is easy to show that the outer annulus exerts a viscous
force
d⌦
Fvisc = 2⇡Rdz ⇥ µR (161)
dR
Writing this in terms of the kinematic viscosity ⌫ and integrating in the vertical direc-
tion yields
d⌦
Fvisc = 2⇡⌫⌃R2 . (162)
dR
 Finally, as this force is acting at a distance R from the centre of rotation, it is easy to
see that it will exert a torque
d⌦
T = 2⇡⌫⌃R3 . (163)
dR
Note that if ⌦(R) decreases with radius, as in a Keplerian disk, then T is negative –
the viscosity acts to slow down the rotation of the inner parts of the disk and speed
up the rotation of the outer parts, resulting in a net outward transport of angular
momentum.

• The net torque acting on an annulus with radius R and width dR is the di↵erence
between the torque at the inner and outer edges of the annulus, which we will denote
as dT . The surface area of the annulus is 2⇡RdR, and so the net torque per unit area
of the annulus is therefore
dT 1 @T
= . (164)
2⇡RdR 2⇡R @R
• With the e↵ects of this viscous torque included, our equation for the evolution of the
angular momentum per unit area of the disk becomes
@ @ 1 @T
R ⌃R2 ⌦ + ⌃uR R3 ⌦ = . (165)
@t @R 2⇡ @R
Note that a similar expression can be derived directly from the Navier-Stokes equation
for the gas in the disk. However, the derivation is complicated by the fact that we
cannot necessarily assume that the viscosity in the disk is independent of radius.

• If we assume that ⌦ does not vary significantly with time (which is a valid approxi-
mation provided that the accretion time tacc = M/Ṁ is much longer than the orbital
period of the disk), then we can combine our mass conservation and angular momen-
tum conservation equations (Eqs. 156 and 165, respectively) to yield the following
expression: "⇢ #
1
@⌃ 1 @ @ @T
R = R2 ⌦ . (166)
@t 2⇡ @R @R @R
If the rotational velocity of the disk is given by the Keplerian value
✓ ◆1/2
GM
⌦= , (167)
R3

then it follows that

d⌦ 3⌦
= , (168)
dR 2R
T = 3⇡⌫⌃R2 ⌦, (169)
@T @
= 3⇡R3/2 ⌦ ⌫⌃R1/2 . (170)
@R @R
 • We can therefore write the time evolution of the disk surface density as

@⌃ 3 @ @
= R1/2 ⌫⌃R1/2 . (171)
@t R @R @R
Given this, the radial velocity then follows straightforwardly from the equation of mass
conservation:
3 @
vR = 1/2
⌫⌃R1/2 . (172)
R ⌃ @R
Finally, the mass flux through the disk at a radius R is simply
Ṁ (R) = 2⇡R⌃|vR |, (173)
@
= 6⇡R1/2 ⌫⌃R1/2 . (174)
@R
• Comparing our expression for the mass flux with that for the evolution of the surface
density, we see that
@⌃ 1 @ Ṁ
= . (175)
@t 2⇡R @R
We therefore see that if our accretion disk is in a steady state, with @⌃/@t = 0, then
the mass flux through the disk is the same at all radii.

2.4.2 Governing equations – vertical structure

• It is common to assume that the gas in the disk is in hydrostatic equilibrium in the
vertical direction. This is a reasonable assumption if the timescale to come to equilib-
rium – typically, the dynamical timescale of the gas – is much shorter than the time
that it takes for gas to flow through the disk and onto the central object.
• In hydrostatic equilibrium, we know that
d 1 dp d⇢
= = c2s . (176)
dz ⇢ dz dz

• We have already assumed that the gravitational potential of the system is dominated
by the central
p object. If we write the distance to object this from a point (R, z) in the
disk as r = R2 + z 2 , then it is easy to see that
d d dr d z
= = . (177)
dz dr dz dr r
The circular velocity for a test mass orbiting the central object at a distance r is defined
as
d
vc2 = r , (178)
dr
and so we can write the vertical component of the gravitational force as
d v2z
= c2 . (179)
dz r
 • If we now assume that the disk is thin, so that r ⇠ R and vc (r) ⇠ vc (R), we can write
this as
d v2z
= c2 . (180)
dz R
Substituting this into our equation of hydrostatic balance, we find that

d⇢ vc2
= zdz. (181)
⇢ R2 c2s

This has a solution of the form

✓ ◆
z2
⇢ / exp , (182)
2h2

where the scale height h is approximately

h cs
⇠ . (183)
R vc
Therefore, provided that the disk is cold, and that its rotation is supersonic, our
assumption that it is thin is well-justified, since in this case we will have h ⌧ R. Note
also that for a disk in approximately Keplerian rotation, ⌦ ' vc /R, and so cs ' h⌦.

2.4.3 The need for an anomalous viscosity

• We can use the equations derived in the previous section to estimate the typical rate
at which mass will flow through a Keplerian accretion disk. From Equation 172, we
see that to within an order of magnitude,
⌫
vR ⇠ . (184)
R
The typical accretion timescale is therefore

R R2
t⌫ ⇠ ⇠ , (185)
vR ⌫
and the mass flow rate is
Ṁ ⇠ 2⇡⌃⌫. (186)

• If the only source of viscosity in the disk is the standard molecular viscosity, then
⌫ ⇠ vT , where vT is the thermal velocity and is the mean free path. In this case,
we see immediately that the inward velocity of the flow will be very small,

vR ⇠ vT , (187)
R
as we know that for astrophysical fluids, we are almost always in the regime where
⌧ R.
 • Consider a protostellar accretion disk with a midplane temperature T ⇠ 1000 K, a
midplane density n ⇠ 1014 cm 3 , and a radius of 10 AU. In these conditions, we have
⇠ 10 cm, vT ⇠ 2 ⇥ 105 cm s 1 , and hence ⌫ ⇠ 2 ⇥ 106 cm2 s 1 . This implies that the
accretion timescale in such a disk will be t⌫ ⇠ 1022 s ⇠ 1015 yr if molecular viscosity
is the only source of viscosity acting to drive the inflow. This is vastly longer than
the lifetime of a protostellar accretion disk, and hence implies that in a disk governed
solely by molecular viscosity, little or no accretion will actually occur.
• A similar analysis applied to other accreting systems (e.g. X-ray binaries) gives similarly
long accretion timescales, and associated accretion rates that are much smaller than
those that are inferred from observations. We therefore see that there must be some
additional “anomalous” viscosity present in the disks, with a magnitude that is much
larger than the molecular viscosity, in order to explain the observed accretion rates.

2.4.4 The ↵-disk prescription

• The question of what physical process or processes are responsible for the anomalous
viscosity in accretion disks is a difficult one to answer, and the issue is not yet entirely
settled. However, it was realised by Shakura & Sunyaev in the mid-1970s that we
can understand a great deal about the behaviour of real astrophysical accretion disks
without understanding the source of the viscosity.
• If the source of the anomalous viscosity is some form of turbulent process, as seems
likely given the very high Reynolds number of the flow, then we expect on dimensional
grounds that ⌫ ⇠ vturb Lturb , where vturb is the characteristic turbulent velocity and
Lturb is the size of the largest turbulent eddy. If the turbulence was highly supersonic,
then it would rapidly dissipate energy in shocks, and so in practice, we expect that
vturb ⇠ cs . Furthermore, if the turbulence is approximately isotropic, then the size of
the largest eddies will not exceed the scale height of the disk, i.e. Lturb ⇠ h.
• These considerations led Shakura & Sunyaev’s to propose the following form for the
kinematic viscosity:
⌫ = ↵hcs . (188)
↵ here is a measure of our uncertainty – we expect on general physical grounds that
↵  1, but it could in principle by very much smaller. Nevertheless, this ↵-disk
prescription allows one to solve for the physical structure of the accretion disk, the
mass flow rate, etc. in terms of only this single unknown parameter.
• For a thin disk, we have already seen that
h cs
⇠ , (189)
R vc
and so we can also write our prescription for ⌫ as
R
⌫ = ↵c2s , (190)
vc
= ↵c2s ⌦ 1 , (191)
 where the second line follows if we assume that the rotation is Keplerian (or approxi-
mately so).

• Substituting this value for ⌫ into our expression for Ṁ , we find that

Ṁ = 2⇡⌃↵c2s ⌦ 1 . (192)
3/2
For a Keplerian disk, ⌦ / R , and so we see that

Ṁ / ⌃↵c2s R3/2 . (193)

If the accretion flow through the disk is steady, then Ṁ must be the same at any
radius, and so we therefore expect that

⌃↵c2s R3/2 = constant. (194)

In the simple case where ↵ does not vary strongly with radius, then we can simplify
this expression further, to
⌃T / R 3/2 , (195)
where we have made use of the fact that c2s / T .

• To make further progress, we need to determine how the disk temperature T varies as
a function of radius. Solving for T in the general case is a complicated problem, as
realistic accretion disks are often optically thick over a wide range of radii. However,
there are a few simple cases that we can look at that are nevertheless informative.

• Let us start by considering the case of an isothermal disk. This is a reasonable ap-
proximation if the cooling rate of the disk is a very steep function of temperature, as
can happen in disks where the cooling is dominated by H opacity or by H2 cooling
(e.g. in population III protostellar accretion disks). In this case, Equation 195 shows
us that
⌃ / R 3/2 . (196)
Moreover, since
cs
h/ / T 1/2 R3/2 , (197)
⌦
we see that an isothermal ↵-disk flares strongly, and that its mean density
⌃
⇢¯ = (198)
h
falls o↵ very steeply with radius, as ⇢¯ / R 3 .

• Another simple case is an optically thin disk with an opacity  that does not depend on
temperature, whose heating is dominated by the radiation from the central accreting
object. In this case, the heating rate of the disk varies with radius as

/ R 2. (199)
 If the disk is in equilibrium, the cooling rate must vary with radius in the same fashion.
However, we also know that the cooling rate will vary with the temperature as ⇤ / T 4 .
If we assume the density dependence of the heating and cooling rates is the same, it
then follows from the fact that = ⇤ that
T 4 / R 2, (200)
T / R 1/2 . (201)
From this, it then also follows that
⌃ / R 1, (202)
h / R5/4 , (203)
⇢¯ / R 9/4 . (204)

• Finally, consider the case where the dissipation of energy by viscosity in the disk is the
main heating source. In this case, one can show that4
3/4
T /R , (205)
and hence we find that
⌃ / R 3/4 , (206)
h / R9/8 , (207)
⇢¯ / R 15/8 . (208)

• These few simple examples barely scratch the surface of the subject, but show how
powerful the ↵-disk prescription can be. This deceptively simple model for the disk
viscosity, combined with a thermal model for the gas, allows us to solve for the structure
of the disk in a wide variety of di↵erent physical scenarios. It is therefore not surprising
that the original paper by Shakura & Sunyaev has become one of the most highly cited
papers in the history of astrophysics.
• Nevertheless, the ↵-disk prescription still presents us with one major issue: what phys-
ical process is responsible for the anomalous viscosity? What sets the value of ↵? In
the next section, we will examine one important source of disk viscosity that appears
to be responsible for determining ↵ in a wide variety of di↵erent types of disk: the
magnetorotational instability.

2.4.5 Magnetorotational instability

• The magnetorotational instability (or MRI) is an instability that arises due to the
action of a weak magnetic field in a di↵erentially rotating disk. It grows rapidly, on
the orbital timescale, and once it becomes non-linear can drive turbulence in the disk.
The resulting turbulent viscosity is a leading candidate for the anomalous viscosity
required in order to explain the large accretion rates inferred for most real accretion
disks.
4
We will explore this in more detail in exercise sheet 4
 • The MRI was originally analyzed by Chandrasekhar in the 50’s, but was then neglected
and forgotten for many years, before being rediscovered by Balbus and Hawley in 1991.

• In Balbus and Hawley’s seminal 1991 paper5 , they studied the MRI by starting with
the fluid equations for a magnetized, di↵erentially rotating disk and applying the tools
of first-order perturbation theory. However, the resulting derivation of the conditions
for the instability is rather lengthy, and so for our purposes, we will look instead at a
much simpler but more physically intuitive model for how the instability operates.

• Consider two neighbouring fluid elements, A and B, rotating around a mass M at

slightly di↵erent radii, rA and rB , in a di↵erentially rotating disk. Suppose that the
disk is magnetised and that the two fluid elements are linked by a magnetic field line
that is oriented primarily vertically in the disk.

• As the disk rotates, the two fluid elements will begin to move apart, owing to the
di↵erential rotation. If the rotational velocity decreases with increasing radius in the
disk, i.e. if
d⌦
< 0, (209)
dr
then the inner fluid element will rotate faster than the outer fluid element, and hence
will move ahead of it.

• Since the two fluid elements are linked by a magnetic field line, both will feel a magnetic
tension force as the fluid elements move apart. This force will act to slow down the
inner element A and speed up the outer element B. Consequently, both fluid elements
will change their angular momentum: A will lose angular momentum and B will gain
angular momentum.

• The response of the fluid elements to a change in angular momentum depends on how
the specific angular momentum varies with radius. If it increases with increasing radius,
then when fluid element A loses angular momentum, it will move inwards, to a smaller
orbit. Similarly, fluid element B’s gain in angular momentum will cause it to move
outwards, to a larger orbit. However, this motion will increase the distance between
the fluid elements and hence increase the magnetic tension force. It is therefore easy
to see that the situation is unstable: any initially tiny di↵erence in rA and rB will
rapidly amplify, until the simple picture sketched above breaks down.

• If the specific angular momentum of the disk decreases with increasing radius, then
the MRI does not operate. However, in this case the disk will violate the Rayleigh
stability criterion,
d(r2 ⌦)
> 0, (210)
dr
and will be unstable to purely hydrodynamical perturbations.
5
“A powerful local shear instability in weakly magnetized disks. I - Linear analysis.”, Balbus, S. A.,
Hawley, J. F., 1991, ApJ, 376, 214
• A more detailed analysis of the instability shows that in practice, the disk will be
susceptible to the MRI only if the magnetic field is “weak” in the sense that

Bz2 3
< 2 ⇢c2s , (211)
8⇡ ⇡
where Bz is the vertical component of the magnetic field, and only if the rotational
velocity of the disk satisfies
d
⌦2 < 0. (212)
dr
• Keplerian accretion disks always satisfy the second of these criteria, and hence are
MRI-unstable whenever the vertical field threading them is sufficiently weak. Note
that if the field is too strong, the magnetic tension between our hypothetical fluid
elements A and B is strong enough to prevent them from separating significantly,
thereby preventing the instability from developing.

• Although a simple linear analysis of the MRI breaks down once the perturbations be-
come large, numerical simulations show that the non-linear evolution of the instability
rapidly leads to turbulence. The MRI is therefore a simple but e↵ective way to drive
turbulence in a magnetised accretion disk.

• How do we then get from this to a prescription for the anomalous viscosity? For our
↵-disk model, the key feature of the viscosity is that it allows for the radial transport
of angular momentum through the disk. Random “turbulent” motions can also lead
to such an outward transport of angular momentum, by a process sometimes known
as turbulent viscosity, but which is better thought of as the turbulent di↵usion of
momentum.

• The characteristic length scale over which molecular viscosity transfers momentum is
the particle mean free path, . In contrast, the characteristic length scale over which
turbulent viscosity transfers momentum is the scale of the largest turbulent eddy, Lturb .
Since it is often the case that Lturb , turbulent viscosity provides a much more
e↵ective means of transporting angular momentum through the disk than molecular
viscosity.

• Numerical models of MRI-driven turbulence in accretion disks show that the resulting
large-scale behaviour of the disk is reasonably well described by the ↵-disk model, with
↵ ⇠ 0.01. Recalling our discussion of the disk accretion timescale, we find that in this
case
R2 R2 ⌦
t⌫ ⇠ ⇠ . (213)
↵hcs ↵c2s
Substituting in some plausible numbers for a protostellar accretion disk – T ⇠ 1000 K,
R ⇠ 10 AU, ⌦ ⇠ 0.2 yr 1 – we find that t⌫ ⇠ 4 ⇥ 109 ↵ 1 ⇠ 4 ⇥ 1011 s for ↵ = 0.01, i.e.
a timescale of around 104 yr. This is much more consistent with observations that the
timescale of 1015 yr that we obtained when we only considered molecular viscosity.