You are on page 1of 53

198lApJS...45..

335G

The Astrophysical Journal Supplement Series, 45:335-349, 1981 February


© 1981. The American Astronomical Society. All rights reserved. Printed in U.S.A.

COMPRESSIBLE CONVECTION IN A ROTATING SPHERICAL SHELL. I. ANELASTIC EQUATIONS

Peter A. Gilman and Gary A. Glatzmaier


High Altitude Observatory, National Center for Atmospheric Research1
Received 1980 April 17; accepted 1980 July 23

ABSTRACT
We derive anelastic equations for convection of a compressible fluid in a deep rotating spherical
shell. Our motive is to develop equations the solution of which can help us understand what role the
large density variation present in the solar convection zone plays in the maintenance of the solar
differential rotation through angular momentum transports by global scale convection. As such, the
model equations represent a generalization of a Boussinesq system that has been studied extensively
with the solar differential rotation problem in mind.
The equations are derived by a formal expansion in powers of a small parameter which is a
measure of the departure of the fluid from an adiabatic atmosphere. The resulting equations are
expected to apply best in the deep part of a convection zone, where such departures are expected to
be smallest. Several aspects of the fluid physics are treated in simplified ways. A constant molecular
weight is assumed, and small-scale transport of heat and momentum are represented by linear
diffusion, whose transport coefficients are at most functions of radius. The reference atmosphere
about which the fluid is perturbed is assumed to be fixed in time and to be a function only of radius.
We discuss similarities and differences between our equations and those of several other authors
derived for related physical problems.
Subject headings: convection— Sun: interior— Sun: rotation

I. INTRODUCTION AND MOTIVATION


Beginning in the early 1970s, one of us (P.A.G.) developed a nonlinear numerical model for simulating convection in
a rotating spherical shell. The motivation for development of the model was the desire to understand the origins of the
solar differential rotation in the interaction of rotation with global convection, for a broad range of conditions. The
model has now been extensively used to study this problem. Results have been reported and reviewed in several
publications by Gilman (1975, 1976, 1977, 1978, 1979) and Gilman and Foukal (1979). Through these studies, we have
gained considerable insight into the roles played by the varying angle between rotation and gravity with latitude, strong
versus weak rotation constraints on the motion, finite amplitude Reynolds stresses and heat transport, finite shell
depth, velocity and temperature boundary conditions, and symmetries about the equator. All of these effects have been
studied in the context of so called Boussinesq theory, in which variations of fluid density are suppressed in the model
except where they are coupled with gravity to produce buoyancy forces. Thus, the spherical shell convection model has
been developed for a stratified liquid shell heated from inside, rather than for a shell of gas with a large density
decrease between its bottom and top, such as in the real solar convection zone. We need now to examine what role this
density variation plays in processes that maintain differential rotation.
In order to focus on the influence of density stratification on the global properties of convection and differential
rotation, we wish to introduce this effect in such a way as to minimize the number of additional physical processes
present in the model. Therefore, instead of first considering a fully compressible system in which sound waves and
global pulsations are present, we will employ the so-called anelastic approximation, discussed by Ogura and Phillips
(1962), Gough (1969), Latour et al (1976), Kovshov (1978), and others. This approximation retains the inertial and
thermodynamic influences of a basic vertical density variation, while filtering out short-period, acoustic oscillations.
This assumes the flow then disturbed does not approach the speed of sound and that the filtered oscillations are not
important in the maintenance of differential rotation. The former can be tested a posteriori from the model
calculations. The latter is reasonable for the Sun, because the characteristic period of global acoustic oscillations is one
hour or less, while the rotation period is nearly one month, so that Coriolis forces have too little time to act to produce
Reynolds stresses that transport angular momentum. We discuss the anelastic approximation further in § III. Various
authors have used it in somewhat different forms, none of which is quite the same as our own, as we shall explain.
‘The National Center for Atmospheric Research is sponsored by the National Science Foundation.

335

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

336 GILMAN AND GLATZMAIER Vol. 45


In addition, other aspects of the model will be kept simple for the present. In particular, we will ignore partial
ionization, and deal with a fluid of constant molecular weight. This is consistent with focusing on global, deep,
convective motions when applying the model to the Sun. The model is incapable of resolving granule and supergranule
scale motions at the top of the convection zone, for which partial ionization is important. In the same spirit, the effects
of all small scales of motions must be parametrized as turbulent diffusion of heat and momentum. As in the Boussinesq
model, we will represent these processes by linear diffusion (but allow the scalar diffusion coefficient to vary with
radius). We recognize this is a crude approximation to a complex process, and expect to replace it with more
sophisticated, nonlinear formulations in later versions of the model. Such formulations are a whole subject worthy of
deep study in themselves, which we do not attempt to undertake here. For recent, exploratory efforts in this direction
for solar and astrophysical problems, see Gough (1979) and Dumey and Spruit (1979), in which rotational effects on
turbulent transports are estimated. Many approaches are possible, including borrowing from a number of formulations
used in geophysical fluid models. In our opinion, much further work needs to be done before such parametrizations are
heavily relied upon in global convection-circulation models.
Another way of looking at the fluid system we shall define and analyze is as an analog rather than an approximation
to the real Sun or another star. This analog comprises compressible, heavily stratified fluid in a deep spherical shell
which rotates and is heated from below. The anelastic perturbation equations define a Newtonian fluid. The dynamics
and circulation of such a system are physically well defined, not relying on ad hoc assumptions concerning processes
not explicitly represented. The results obtained from such a model are clearly still of interest for the Sun and stars,
because the physics of the two systems have so much in common.

II. NONLINEAR EQUATIONS


Before introducing our version of the anelastic approximation, we first write down the general equations for fluid
flow in a rotating spherical coordinate system. We employ a system in which X represents longitude, <j> latitude, and r
radius. The velocities in these three coordinates directions are defined as w, o, and w, respectively. The symbol p is
taken to be the fluid pressure, p its density, 0 its local temperature, and s its specific entropy. In addition, v is the
kinematic viscosity and k is the thermometric diffusivity, both of which may be taken as constants, functions of
position, or functions of the velocity field. In the present formulation, we will assume they are predetermined functions
of radius only. Gravity g(r) is assumed to be a central gravity, and self-gravitation of the fluid region is ignored. The
centrifugal force due to the basic rotation of the coordinate system is assumed to be very small compared to Newtonian
gravity, so that potential surfaces are essentially spherical. This assumption eliminates Eddington-Sweet currents from
consideration. For the Sun these would be extremely weak compared to the convective motions we are modeling.
Using the above definitions, the equation of motion for east-west flow may be written as

= + 20
PIT
K
8i +“1T
9i “iff
rcos<f> d\ p(
r
\ +——rksin^Oü—p(2flcos<¡>+
rcos<j> I — W
r /

1 9 1 9
2 (pKOCOS<í>)+^(pKHT2)j + PÍ\» (1)
reos# 9X (P« )+ rcos<i> 9<|>

in which F is the viscous force per unit mass, which we discuss in more detail below. The north-south equation of
motion is given as
9o , 9p _ 1 OE. u
P 97+ü9í “ p|2ß +
r 0<(> rcos<f>

1 8 1 8
(puv) + (po2cos$) +-^(prW) +pi*> (2)
rcos<> 9X rcos<j> 9<i>

and the radial equation of motion :

+w
97 = ■ 97 "pg+p(2öcos,i,+7 ) M+pl7

1 9/ -y, 1 9/
r^T(pMw)-l T -Zt(pvwcos<j>)-\—- -r-( pr w ) + pFr. (3)
rcos<í> 9X rcos<j> d<f> ’ r2 9rvr '

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

No. 2, 1981 CONVECTION IN A SPHERICAL SHELL 337


In equations (1), (2), and (3) above, the nonlinear velocity advection terms are written in momentum flux divergence
form, using the equation for continuity of mass (7) below. On the left-hand sides of (1), (2), and (3), the time derivative
of momentum density is split into velocity and density derivatives with time, in preparation for the anelastic
approximation to be made later.
The viscous forces Fx, F^, and Fr are evaluated in detail from the divergence of the viscous stress tensor in spherical
coordinates. Since we assume the fluid is Newtonian, this tensor is proportional to the rate of strain tensor. The
appropriate forms and manipulations for spherical polar coordinates are given for example in the appendix of
Batchelor (1967). The end result for the viscous forces is easily transformed to our coordinates. These final forms are
written to separate parts which arise for an incompressible liquid with constant kinematic viscosity (denoted by angular
brackets) from terms present only when the basic density field or viscosity varies with position or the fluid is
compressible. This is so we may easily distinguish between the Boussinesq equations for spherical shell covection which
have already been solved numerically, and the more general compressible equation. The result of all the above is

3 . , 92 1 32w
Fx = p{ ■gX + ^(«COS<i.) +
H r! -(Ï rw Ï)-
3<í> [r2cos<i> dr2 V rcos<f> drdX

4 1 3A / 3m u 1 3>v \ 1 3 , ,
3V rcos<¡> d\ \ 3/* r rcos<j> d\ } p '

3w \ /4. ^ \ i 1 du l_d_ v
+ — +(tan<|>)wH ——
r L 3<|>
2 7
cos<t> dX p 3<|> (f> )

_! .3^ _|_ w _ ûtan<#> _ A


+ (4)
rcos <¡>\ reos <j> dX r r 3

3/3« , \ , 32« 3« . , 1 3w \ 1 3 , .
v I^+f
2 2
r cos <i> r drd<l> r gr2 ) 3 /• d4> \ dr “ + 73^)p3;(p'')

I/I \+i 2 1 3u 1 3u 1 3 . .
P + (tan<i>)w (5)
r\rd<t> r 3/p3</ *'' [r cos<i. 3<i) r2cos2 </> 3A j p 3X ^

^_ / 1 3 / 3«\ , 1 32w , 1 3 ./ 3w 3 , J
F ++
'“r ÂÏÏ 3? r 9Ä ) âï h 4 âï “ 3?( )
1
* db tJdw A\ 1 3 , , , 1 dv _ û J_ 3>v 3 .
3r r r d<f>

1 I du u 3w
+- (6)
rcos<í>\ 3r r rcos<¡> dX

In equations (4),^ (5), and (6) above, the angular bracket terms are components of the vector — V X (v X v) (in
which v=wX+ü^>+mt), which is the vector invariant form of the viscous force for an incompressible, constant
viscosity fluid. This is the most convenient form for numerical integrations using the approach we have followed in the
Boussinesq case and will follow again. The symbol A = v*v, the three dimensional divergence of the velocity field.
In the same notation we write down the mass continuity equation as

■ly+ ^^-(r 2
pw)H ~r^T[(C0S£)P0]" 1
~Tïït(pm)=0- (7)
3í r2 dry ’ reos </> 3<#>LV ’ 2 rcos<j> 3ÀV ’

Assuming a perfect gas of constant composition, the equation of state is given by

p=pRifi, (8)

in which Ä* is the gas constant per unit mass.

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

338 GILMAN AND GLATZMAIER Vol. 45


The nonlinear thermodynamic equation expresses the total rate of change of internal energy ( p Ct [ dö/dt ]) as the
sum of the rate of change due to deformation ( H—pà ) and the rate of change due to thermal diffusion (D).

/ 90 u 90 u 90 90
pCv W = H-pA+D. (9)
\ 9i rcos</> 3X r 9<i> 3r

The variable Cv is the specific heat at constant volume (constant for a perfect gas), and (—pA) is the rate internal
energy changes due to compression and expansion. The quantity H is the rate internal energy increases due to viscous
dissipation, given by

H
=2Pv[eueiJ-^à2y (10)

The variable eiJ is the symmetric rate-of-strain tensor, whose components, in our notation, are

e
_ 3w _ J_ i w e
_ 1 du w _vtan<f> 9
rr 0r > r 0^ r > \\ fQQgÿ r r

_ cos^> 3/w\ 1 dv _ 1 dw r d Í u\
+ eXr_ +
e**- 2r 9^1 cos</>/ 2rcos<i> 9X’ 2rcös^ 9\ 2 9r 17T

_ r 8 /ü \ i 1 8w
er,l,
~2dr\~rJ+2rd^' (11)

The variable D is the negative of the divergence of the diffusive energy flux which, in general, is due to small-scale
(eddy) turbulence, radiation, and conduction. For the deep interior of the solar convection zone, small-scale turbulence
makes the largest contribution, followed by radiation, with conduction being negligible. For our model we assume the
total diffusive energy flux is due to the small-scale eddies and that the entire convection zone is superadiabatic.
Therefore, the diffusive flux should be proportional to the superadiabatic temperature gradient | V 0 —(V 0)AdI- Then
D takes the form

Z)=v{pC„/c[v0-(V0)AD]},

= pCpK ±±( , 1 d / , , 1 d2fl


r2 3r v dr / r2cos<f> d<l> / r2cos2</> d\2

90 9 , x , 1 90 30 9
+c 8r 8r (pic) + 2 2 (pk) (12)
8<|> d</} r cos <l> 9X

in which K is the eddy thermometric diffusivity, (V0)ad ~ —g=GM/r2, and Cp is the specific heat at
constant pressure (constant for a perfect gas). The constant G is the gravitational constant, and M is the mass
contained below the convection zone.
We will also need the thermodynamic equation (9) written in terms of specific entropy s, which, to within an
arbitrary constant, is defined as

s=Cp]nO—R*\np. (13)

This form of the thermodynamic equation is given by

ds v ds . ds\_
+ H+D. (14)
rcos<¡> 8\ 7 d<t>+Wdr)~

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

No. 2, 1981 CONVECTION IN A SPHERICAL SHELL 339


III. NONLINEAR ANELASTIC EQUATIONS

a) Anelastic Approximation and the Reference Atmosphere


The so-called anelastic approximation has been derived in slightly different forms by different authors, depending
upon the physical problem which motivated the analysis (see, e.g., Ogura and Phillips 1962; Gough 1969; Latour et al
1976). The basic assumptions made are these: All thermodynamic variable perturbations are small compared to their
reference or mean atmosphere values at the same level in the fluid. The time scale of the motion is reasonably
estimated by the buoyancy time (inverse Brunt-Vaisala frequency) characteristic of a rising convective element or a
gravity wave, and this time is similar to the advection time for the motions. A consequence of these assumptions is that
the motions must be small compared to the speed of sound in the fluid.
Given the above assumptions, a formal scale analysis of the equations is usually made. All dependent variables are
expanded in a power series in a small parameter representing the magnitude of departures of the thermodynamic
variables from the reference state, and only terms up to first order in the expansion are retained. As a result, the
equations become linear in all perturbation thermodynamic variables, but remain nonlinear in products of velocity
variables. Quadratic nonlinear products of velocity with thermodynamic variables drop out in the mass continuity and
momentum equations, but are retained in the thermodynamic equation. In addition, the time derivative of density in
the mass continuity equation (eq. [7] above) drops out. Dropping of the time derivative of density is what filters out
sound waves.
An issue arises as to the proper reference or mean atmosphere against which the thermodynamic variables are
linearized. Ogura and Phillips (1962), motivated by experience with convection in the Earth’s atmosphere, chose as
their reference state an atmosphere of adiabatic stratification. This is motivated by the observed fact that in dry
atmospheric convection, the mean lapse rate in the convecting region is very close to adiabatic. Gough (1969) on the
other hand, allows for a more general presumably superadiabatic atmospheric stratification, but still assumes it is a
function of radial distance only. Still more recently, Latour et al (1976) have allowed the horizontally averaged
atmosphere to change with time as the convection modifies the density distribution. This appears to be particularly
important if the potential temperature or specific entropy difference across the convecting layer is not a small fraction
of its average value, or if regions of partial ionization must be taken into account.
For the solar convection zone, mixing length arguments (see, e.g., Gough and Weiss 1976) indicate that only near the
top of the zone where granules and supergranules are formed is the stratification likely to be strongly superadiabatic.
Underneath, for the great bulk of the convection zone depth, it is estimated the temperature gradient departs no more
than about one part in 104 from the adiabatic value. Therefore, for purposes of modeling global convection filling this
lower region, it seems reasonable to use a fixed reference atmosphere which is nearly adiabatic and a function of radius
only. This is the procedure we adopt. The granule and supergranule layer above then becomes for global convection a
boundary layer interacting with the global convection layer below. This is discussed further in § IV.
The mean atmosphere is a function of radius and time and can be obtained by averaging over a spherical surface (for
given values of r and t). The reference atmosphere and the mean atmosphere are the same only before the
perturbations begin, i.e., while there exists a superadiabatic, unstable equilibrium with no large-scale motions. Since the
object of our model is to describe giant cell convection, the small scale turbulence will not be resolved, and thus will be
defined as part of the reference atmosphere. Therefore, before the onset of large scale convection, the total energy flux
is the reference state diffusive flux due to small-scale turbulence. For a solar model, this energy flux would be set equal
to the solar luminosity divided by (4t7t2).
With the onset of convective instability, perturbations develop and grow into large-scale convective motions which
transport the energy more efficiently. The total energy flux is now the sum of the reference-state diffusive flux, the
perturbation diffusive flux (which is inward), and the perturbation convective flux (outward). The total mean diffusive
flux (reference-state plus perturbation diffusive flux) is then less than it was before the onset of convection. The
convective flux reduces to steepness of the temperature gradient and therefore reduces (V ~ V Ad)> where v =
Bin 0/91np. The more efficient the convection, the smaller the value ( V ~ V Ad) becomes.
The model atmosphere is the sum of the reference atmosphere (a function of radius only) and the perturbations
(functions of position and time). After the onset of convection, the mean atmosphere is no longer the same as the
reference atmosphere because spherically averaged, nonlinear perturbations are present. However, the induced
departures from the reference atmosphere are no larger than the convective perturbations themselves.

b) Scaling
Each dependent variable,/, will be expressed as the sum of the reference state variable,/(r), and the perturbation,
ffr, (f), \, t). Since the uncertainty in k and p is great, k1 and p1 will be set equal to zero. Since we are neglecting
self-gravitation, the gravitational field is unperturbed.

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

340 GILMAN AND GLATZMAIER Vol. 45


We will let e be a measure of the relative departure of the reference atmosphere from the adiabatic atmosphere. Since
this departure is what drives the global convection, e should then also be a measure of the relative magnitude of the
perturbations.

(15)
P

This is the expansion parameter we will use in devising the anelastic equations. It is essentially the same one used
originally by Ogura and Phillips (1962) for a much simpler physical system. For the Sun, except near the very top of
the convection zone, e< 10 ~4 (Gough and Weiss 1976), so the departures of the reference atmosphere and the
perturbations are small indeed.
The spatial scale of variation of the reference variables is of the order of the pressure scale height which, for our
model, is of the order of the depth of the convection zone. The spatial scale for the perturbations depends on the size of
the convective cells. Since fluid elements moving in cells which reach from the bottom to the top of the convecting
region can release the most stored potential energy for a given density departure from their surroundings, we expect
such deep cells to be a prominent part of the total convection spectrum. Consequently we choose a length scale d equal
to the depth of convection zone for this analysis.
The scales for the density, p0, temperature, 0O, gravitational field, g0, thermometric diffusivity, k0, and kinematic
viscosity, v0, are chosen to be their respective values at the top of the convection zone. The entropy scale, s0 will be Cp.
For convenience we define

d
s
5 = -(V = -A(di\ (16)
AD ) =-~
/J +go/C
CD \ dr jo
''o L 5A)0 '

where I is the local pressure-scale height and the subscript 0 means the value of the variable at the top of the
convection zone.
Assuming our motions are at most moderately nonlinear, and in order to easily compare our results to the
Boussinesq case, we choose the advection time scale to be the (eddy) thermal diffusion time. Thus,

(17)

Consequently, the velocity scale is


w0=K0/d. (18)

Pressure perturbations extract energy from the radial motions to drive the horizontal motions. Therefore, the
perturbation pressure term and the advection term in the equation of motion must be of the same order of magnitude.
This implies we should define a pressure scale ps according to

2 PoK04
eps=p0w0 =—r- (19)

Note that ps is not the pressure at the top of the convection zone p0.
This scaling scheme gives e another physical interpretation. From equation (19),

_
£= í I characteristic Mach number|2. (20)
(Ps/Po)

In other words, small e also implies velocities small compared to a characteristic sound speed (a/Po)1/2* Om* time
scale is also, in general, somewhat larger than the buoyancy time for a parcel in the slightly superadiabatic fluid.
Some convenient and conventional dimensionless numbers will be used which follow from our scaling. The Rayleigh
number takes the form

4
-go +go/Cp d
egpä3
R= (21)
(wo) (OoWo)

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

No. 2,1981 CONVECTION IN A SPHERICAL SHELL 341


The Prandtl number is
p=v
o/«o- (22)
The Taylor number is

(23)
"o2 '
The Froude number is
F
=Ko2/(eCp60d2). (24)

The values R, P, and T all appear in the Boussinesq case, but F is a new number peculiar to the compressible case.
Note that the product FPR=(y— \)d/(yIo) = (god/CP0o) and is determined by the reference state. Here y is the ratio
of specific heats CPCy. We will also define an aspect ratio
ß=(r0-d)/d, (25)

where r0 is the radius of the sphere.


The independent variables are therefore scaled as follows:

r^>(d)r,

(26)

The dependent variables are scaled such that

P-*Po(P+eP)>

0^>0o(O+e0),

_ (Poko2\,=+, PoSod, = ,
-^T )(p ep)--px-(P+ep),

S->Cp(s + £S),

fsdV/2
PR )

egodV/2
-PR)

egod\l/2
•GîiM
2
g->go )

(27)

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

342 GILMAN AND GLATZMAIER Vol. 45


where the nondimensional reference state variables, /, and the nondimensional perturbations, /, are of order unity. In
addition, we use the solution 2ü = (egQTP/Rd)x/2.

c) The Full Nondimensional Equations


With the above scaling, the six nonlinear, coupled differential equations take the following nondimensional forms.

i) Continuity Equation (from eq. [7])

0=e3/2 + :
a7 7"l;tr2M,(P+ep)í + 7^¿t(p+^)ÜCOS<í>] + ^¿[(p+^)M]- (2g)

ii) Equations of State (from eqs. [5] and [13])

(5+e5)=ln(ö+eÖ)— -^-^-ln(^+e/?). (30)

Note, in equation (30), the constant arising from the scales of 0 and p has been dropped since the equations involve
only derivatives of entropy.

iii) Radial equation of Motion (from eqs. [3] and [6])

s(p+ep)^-+s2w^ = -3~(p+ep)-PR(p + ep)^l^f-]]

(u2 -bo2)
+ £r1/2P(p + £p)t/cos<i>+e(p + ep)
r

{10 10
rcos<f>0\ [(P + £P)7MH;1J ^ rcos<f>
7 77LV 7 77 í(p + ep)t»vcos<í>]
0£LV 7 J

+ (31)
7 7 [(P+cP)r2>v2] } +eP(p+ep)Fr.

iv) East- West Equation of Motion (from eqs. [ 1 ] and [4])

e(p+ep)|y +e2«^ = - +er> Jp(p+ep)(t>sm«i>-»vcos<i>)

+e(p+ep)( “)( v tan <$> — w —~r 4cLV[(p+£p)w


7 2]J H “T -4r [(p+ep)«t?cos<#>]
rcos<J>0\ rcos<i> 0<i>LV 7 J

+ -^-^[(p+ep)«wr2]}+ei>(p+ep)Fx. (32)

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

No. 2,1981 CONVECTION IN A SPHERICAL SHELL 343


v) North-South Equation of Motion (from eqs. [2] and [5])

, s 9t> 2 _ e dp (wü + w2tan<í>)


,(p+,p)¥+et,-57--;:5ï — er1/2P(p + ep)wsin<í> —e(p + ep)

1 1 2
-£{ [(p + ep)«o] +
rcos<i> 9À rcos<¡> d<j> [(p+ep )o cos<i)]

+
^Jr [(P+eP)r2t,M;] ] +eP
(P+eP)Ff (33)

In the above equations, Fr, FXf and F^ have the same form as equations (4), (5), and (6), but with v and p replaced by v
and (p+ep), recalling that p and v are not functions of <i> or \.

vi) Thermodynamic Equation (entropy form from eq. [14])

ds eu ds t ev ds , 9 /_ , X 3/2 /2
e1//2(p + ep)(0 + e0) E H JJT +w—(s + es 7) = e FPH+e' D, (34)
dt rcos<t> d\ r 9<i> 9rv

in which

//= 2( p+ep ) p ( e,7 e,7 - y A2 ), (35)

and

, e 9 / d0\ , e d20
D=(p+ep)«U|;[r2|:(ö+ei>) H 2: I cosió-— I
r cos<i> 9<|> \ r cos <j> dX2
2 2

e2k dO dp e2k dO dp
+ 0^(0 + e0)^[(p
dr + ep)/c] + —— 2 ^ + 2 2
r d<j> d<j> r cos <¡> dX

-FPR(±±£f j-r[(j>+ep)K]. (36)

d) Dimensionless Reference Atmosphere Equations


The dimensionless equations for the reference atmosphere can now be obtained by setting all perturbations equal to
zero. These equations are:

(37)
Y

s=]n6 (38)

(39)

f2fr(r2PKT)=0, (40)

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

344 GILMAN AND GLATZMAIER Vol. 45


where the reference state, superadiabatic temperature-gradient profile IJr) is defined as

d0_ LÊ.
r=- + FPR (41)
e dr e dr '

Note that T is unity at the top of the convection zone, so —(e0o/d)T is the actual superadiabatic temperature
gradient.
The solutions to the reference state equations within the anelastic approximation are found by expanding all
reference state variables/in series of the form +e/(,) +e2/(2) + • • •. Then, to the lowest order in e, the reference
state thermodynamic variables are adiabatic. Consequently, equations (37) and (39), together with the adiabatic
temperature gradient

(42)

describe a polytropic reference state. The solution is

FPR(ß+lf
0=[l-FPÄ(ß+l)] + (43)

P=Ön, (44)

P= (45)

where n is the polytropic index equal to 1/(y_ l)=3/2, the product FPR is equal to [(eVo '' — \ )ß]/(ß+1), and the
parameter Np is defined as the number of density e-folds across the convection zone.
Note that the thermodynamic equation (40) is of order e3/2 since the difference between the actual temperature
gradient and an adiabatic gradient must be accounted for. Once the functional form of ic(r) is specified in equation
(40), T(r) can be calculated. Alternatively, if T(r) were specified, ic(r) could be calculated.

e) Dimensionless Anelastic Perturbation Equations


To find the anelastic perturbation equations, we expand all perturbation dependent variables / in series of the form
/—/(0) +e/(1) +e2/(2) + • • •. We then retain the terms of lowest order in e in each of the governing equations. This will
result in only/(0) terms being retained, so the system is complete. We then drop the superscript (0).
When we allow e->0, we assume that the Rayleigh number R and the Froude number F, both of which contain e,
remain finite.
For convenience, we also define the anelastic divergence V of a vector/* as

i
+ cos, , +
reos# d\ ’ rcos<¡> ¿(^
d<¡> í)
pr

(46)
p dr

Then the dimensionless perturbation equations take the following forms.

i) Continuity Equation, to Order e,//2


w
A , —
o=v¿ ' v=A+ dp (47)
p dr

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

No. 2,1981 CONVECTION IN A SPHERICAL SHELL 345


ii) Equation of State, to Order e

(48)
P P 9 ’

(49)

iii) Radial Equation of Motion, to Order e

dw
~a7 ~^~PR-p(^r )2+ r'/^cos <*>+ - V „•( wv)

-UP A/ 1
d2w , 1 8 Í dw
\ r cos</>drV 3X /
2
r cos <¡> dX2
2 2
r2cos<i> lC°S d<¡)

4 _ 3A 2/ dw A \ d(vp)
(50)
3V dr p \ 3r 3 / dr

iv) East- West Equation of Motion, to Order e

du 1 dp
+ r1//2P(t>sin<í>—vvcos<i>)
Jt prcos<¡> 3A

+ — (t?tan<i>—w) — V a#(wv)

_3_ 1 _3_ , 1 92M 1 02w


+P v 2
(ucos<f>)
d<¡> r cos<f> d<f> ^ 9r2 rcos<i> 9r3X

4 y 9A 1 / 8u m 1 W
(51)
3 reos# 3X P l 9r r rcosQ 9X jdr^^V

v) North-South Equation of Motion, to Order e

dv (H^ + w2tan<f>)
-Va-(v\)
Jt pr d<j> r

du I 9jro) \
+P -( COS<t>
d<¡> \
9<í> l 3A ^ 3r2 /
i 4 3A J_ dv _ v . 1 (hv
(52)
3 r d<f> p dr r r 3<i>

To derive the correct form of the perturbation entropy equation, we must take note of the fact that by our
assumptions, the reference state entropy gradient ds/dr is of order c, since the departures from an adiabatic
atmosphere are small. Then, using equation (41), we derive the following equation.

vi) Thermodynamic Equation (Entropy Form), to Order e3/2

9s
+v*Vs -WpY=2FFpv I <?,7e,7 - yj + v(p¡cV0). (53)
9/

A more convenient form of the dimensionless thermodynamic equation (for both the demonstration of energy
conservation and the actual numerical computation) can be obtained by substituting equations (47), (48), and (49) into

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

346 GILMAN AND GLATZMAIER Vol. 45


(53). Then

S |y=wr+EP/?(^)2^-va*(<?y)+|vV/>+2FI>^e,7e,7-y) + ^v(p<cV0). (54)

Equations (47)-(52) plus equation (54) thus constitute the anelastic rotating spherical shell equations we propose to
solve in subsequent papers.

IV. BOUNDARY CONDITIONS


The primary goal of developing equations for compressible convection in a rotating spherical shell is to model the
deep, nearly adiabatic part of the solar convection zone. Sitting on top of this region in the real Sun is the granule and
supergranule layer, which for our purposes must be regarded as a turbulent boundary layer. It is not clear how much
and what kind of influence this boundary layer has on the dynamics of global convection and differential rotation
below. This should be studied in the future, but for the present we will apply stress-free velocity boundary conditions
at the top, as well as at the bottom of the convecting region. This implies

8m 8ü
=0, (55)
9r 87
at the top and bottom.
In the future we must also examine the role of penetration of global convection modes, both into the top boundary
layer mentioned above, and into the convectively stable region below. But for the present, we ignore this effect, and set
h>=0 (56)

at top and bottom. When we consider penetration at the bottom, we can apply boundary conditions (55) and (56) at a
depth well below the region of unstable stratification beyond the effective region of penetration, which can be
determined by experimentation with the model.
A variety of boundary conditions on temperature are also possible. If we ignore penetration of global modes above
and below, it is probably most reasonable to assume a constant (eddy) diffusive energy flux at the bottom boundary.
Then, to leading order in e, the perturbation temperature gradient vanishes there, i.e.,

90/9r=O. (57)

At the top, we again have the granule and supergranule layer to consider. We might consider this layer to be an
efficient heat exchanger between the global convection below and space above, in which case it is plausible to match
the small scale turbulent flux at the top with a blackbody flux at the effective temperature. This results in the following
(nondimensional) condition on the perturbation temperature and its gradient at the top, i.e.,

0_ / cpPoko \ 8^
l 4o6¿d I 8/* ’

Here a is the Stephan-Boltzmann constant. However, since /c0æe1/2 (see Eq. [27]), a constant temperature top
boundary, i.e. 0=0 at the top, is implied to leading order in e. Also, the coefficient in parentheses in equation (58) is of
the order of 10-3 when typical solar values are used. For the Boussinesq case in Gilman (1978), we have already
examined the effect on the rotational dynamics of several linear combinations of perturbation temperature and
temperature gradient at the boundaries.

V. ENERGETICS
It can be shown that the anelastic perturbation equations derived above are energetically consistent, in the same
sense as discussed by Gough (1969). That is, a total energy equation (kinematic plus internal plus potential) derived
from the original governing equations, to which our anelastic scaling and expansion is applied, will yield the same
result as when an anelastic total energy equation is formed from equations (47)-(52) and (54). The result is very similar
to equation (5.9) of Gough (1969) except for scaling and change of variables.
One difference is that since our reference atmosphere is fixed in time, no correction to it due to dynamical pressure
effects is included. But this is also consistent with our scaling. Such effects do come in when determining the form of
our spherically averaged or mean departure from the reference atmosphere.

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

No. 2,1981 CONVECTION IN A SPHERICAL SHELL 347


Certain energy integrals can be derived from the system of equations (47)-(54), which are useful for interpretation.
If we take the scalar product of the velocity with the equations of motion (50)-(52) in vector form, and integrate over
the entire mass of the spherical shell, assuming no fluid enters or leaves the top or bottom, we get, to leading order in e

_8_ (ß+\f
jp—^-dT = — jwp dT—PR j pwdT+ j pyFdr, (59)
dt

in which dr=r2cos <(> drd<l>dX.


We can obtain a perturbation thermodynamic-energy integral from the perturbation entropy equation (54):

•(0+1)
IT dT= (60)
dt Jí^F~
r r JfwpdT+PRÍ-P
T J T r
l J pwdr+ (lidT+ fp^=,wdT+ ¡Ddr,
JT JT r JT
in which

H=2Ppv —j,

and

ß=|;V(pKVÖ).

The first integral on the right in equation (60) represents the energy extracted from the thermal field by pressure
forces doing work on the fluid in equation (59). Similarly, the second integral on the right in equation (60) may be
interpreted as work done by the buoyancy forces in extracting energy from the thermal field and adding it to kinetic
energy of the fluid through the second integral on the right in equation (59). If we apply stress-free velocity boundary
conditions at the top and bottom of the fluid, then it can be shown that

j pvFdT= — j Hdr, (61)

so that all the work done by the motion against the viscous force appears as internal heating.
The fourth integral on the right in equation (60) actually vanishes. This can be seen as follows. If we integrate the
anelastic continuity equation (47) over an entire spherical surface, and denote this operation by angular brackets, then
we are left with

¿(^<^»=0. (62)

Since by equation (56), w=0 at the top and bottom of the shell, obviously (w) =0 there, too. Integrating equation (62)
in radius implies ( w ) =0 everywhere. But since T, ¡5, and F are all independent of longitude and latitude,

JpywdT= Jpj(w) dT=0. (63)

Thus, while the term wT is locally important (even dominant, especially in the linear case) in determining changes in
the perturbation thermal field, its contributions cancel out when integrated over a full spherical surface.
Finally, the integral fTD dr in. (60) integrates to the top and bottom boundaries as

pK dO
(64)

If we make use of all the above results, the equation for total energy, which is the sum of equations (59) and (60),
reduces to
_ 0s 1_ p/c dO p/c dO
da- da, (65)
p
y+2pV*V “f ■/0 bottom TYr

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

348 GILMAN AND GLATZMAIER Vol. 45


showing that under our assumptions, the only way total perturbation energy can change is through a net flux of heat
across the boundaries into the fluid due to perturbation-temperature gradients on the boundaries. Otherwise, the
system conserves total perturbation energy.

VI. DISCUSSION
The new physical effects we are adding to the problem of compressible convection in a star are rotation and
spherical geometry. Both are necessary to addressing the problem of global circulation. Our perturbation expansion
and the resulting anelastic equations are somewhat more restrictive than those in Gough (1969) or Latour et al (1976),
because our departures from the adiabatic atmosphere are kept small while theirs need not be. Actually, we have
returned to the approach formalized by Ogura and Phillips (1962), but we have considered more physical effects than
they did—specifically rotation, spherical geometry, and diffusion of both heat and momentum.
The formulation of Latour et al. (1976), including the capability for large changes in the reference atmosphere,
allows those authors to start from an atmosphere in (unstable) radiative equilibrium that may be quite far from
adiabatic. Ours assumes some effect of motions is already present, which has already brought the stratification close to
the adiabatic. A possible disadvantage of their formulation is that it increases the risk that the predicted velocities
approach the speed of sound rendering the anelastic approximation less valid. This is an implication of our equation
(20).
Since our reference atmosphere does not change, our perturbation thermodynamic equation is simpler than in
Latour et al. (1976). In effect, we treat changes in the mean atmosphere as perturbations to the same level of accuracy
as local changes. This mean could be taken over a complete spherical surface, or, more appropriately in the rotating
case, as an average in longitude, so that global changes in the latitude dependence remain explicit. We could derive a
separate prediction equation for these changes by averaging the perturbation thermodynamic equation (53) or (54) over
a spherical surface, but within our approximations there is no need to calculate these quantities separately. Changes in
the mean atmosphere which is defined by the longitudinal average would be predicted normally from equations for
longitudinal wavenumber m=0 in a Fourier or spherical harmonic expansion.
Latour et al. (1976) and Toomre et al. (1976) include specifically the effects of radiative diffusion in their
calculations. By implication, they assume this process is more important in transporting heat than turbulent diffusion
at scales smaller than they resolve. Assuming turbulent diffusion coefficients for heat and momentum are similar in
magnitude for a given scale of motion imphes they are generally dealing with effective Prandtl numbers substantially
less than unity. This may well be appropriate for modeling the thin convection zones of A stars, or even solar granules
and supergranules. In our own formulation, we have not explicitly restricted the Prandtl numbers yet, but for the
global motions we wish to model, it seems unlikely the turbulent transport rates for heat and momentum for all the
unresolved motions are likely to differ much from each other, so Prandtl numbers near unity seem most reasonable. If
our model could resolve spatial scales small enough that radiative processes could actually compete with small scale
turbulence, then perhaps Prandtl numbers much less than unity would be appropriate, as Massaguer and Zahn (1980)
have argued. However, this would require much greater resolution in a global model than can be afforded in the near
future.
Formally, the most attractive first problem to address, using our anelastic rotating spherical shell equations, would
be the linear convective-stability problem, equations for which would be defined by simply suppressing all nonlinear
products in equations (47)-(54) above. We do this in Paper II, which follows. But in physical terms, this begs the
question of the source of the small-scale turbulence implied in our reference state. It could have come from the global
convection we are now trying to represent explicitly, or it could have come from a “primordial” stratification
associated with a radial gradient in (unstable) radiative equilibrium. Thinking of our linear system as an essentially
Newtonian analog rather than as a physically realizable approximation to the real Sun (or another star) is useful for
coping with this somewhat philosophical difficulty. The nonlinear system does not really suffer from this point—rather,
we can argue that the parametrization of unresolved scales is simply quite crude. Also, solving the linear stability
problem will give us much useful guidance for the nonlinear case.
It does not seem meaningful to use our system for very supercritical calculations for which sharp boundary layers
would presumably form, since these would be boundary layers based on the parametrizations assumed. In that case,
our scaling also begins to break down because the eddy diffusion time d2/K0 becomes much longer than the buoyancy
or turnover time. The eddy diffusion represents all scales smaller than those explicitly calculated. There is no reason to
expect a large deficit in the turbulent energy throughout the convection zone at scales slightly smaller than are resolved
explicitly. Thus we have no reason to assume a large difference in these two time scales. At this point it would be better
to simply replace our crude parametrization with a better, presumably nonlinear one.
Our treatment of the reference atmosphere is in many respects similar to that of Belvedere and Paterno (1977)
among others. However, they choose to parametrize all the effects of rotation upon convection into a single physical

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

No. 2,1981 CONVECTION IN A SPHERICAL SHELL 349


quantity, namely a heat-transport coefficient which is a function of latitude. This effect is treated as a perturbation on
a reference state which is independent of latitude, as is ours. In solving the convection problem, we choose to confine
rotational influences to those motions we explicitly calculate, and so we leave out latitude dependence of small-scale
transport coefficients. Nevertheless, there is little doubt that rotation, as well as nonlinear interactions, shearing
instabilities and probably other effects too, will influence these small-scale transports and should ultimately be taken
into account. Dumey and Spruit (1979) and Gough (1979) focus primarily on rotational effects and estimate turbulent
velocities by local convective instability theory with Coriolis forces included. Dumey and Spruit intend their
parametrization to represent all motions except for differential rotation and axisymmetric meriodional circulation.
Therefore, rotational effects must be prominent. Our approach is different, in that we will calculate the influence of
rotation for all global compressible convective modes exactly without resorting to a local approximation, but with a
cruder representation of small-scale turbulence. We believe that ultimately a combined approach would be ideal, in
which the global convection is calculated explicitly, but rotation and nonlinear transfers are taken into account in the
scales of motions we do not calculate. Even in that case, centrifugal effects due to the rotation of the system could still
be ignored, as we have done above, compared to Coriolis effects.

We are pleased to acknowledge several useful discussions with Jean-Paul Zahn while he was a visitor to HAO in the
summer of 1979. Among other things, he called our attention to an error in one form of the thermodynamic equation
we were using. We also wish to thank David H. Hathaway for reviewing the manuscript.

REFERENCES
Batchelor, G. K. 1967, An Introduction to Fluid Mechanics (Cam- Gough, D. O. 1969, /. Atmos. Sei., 26, 448.
bridge: Cambridge University Press). . 1979, Proceedings of the Workshop in Solar Rotation,
Belvedere, G., and Paterno, L. 1977, Solar Phys., 54, 289. Catania, Osservatorio Astrofísico di Catania, Pubblicazione No.
Dumey, B. R. and Spruit, H. C. 1979, Ap. J., 234, 1067. 162, p. 337.
Gilman, P. A. 1975, J. Atmos. Sei., 32, 1331. Gough, D. O., and Weiss, N. O. 1976, M.N.R.A.S., 176, 589.
. 1976, in IAU Symposium 71, Basic Mechanisms of Solar Kovshov, V. I. 1978, Sov. Astrom. Aj., 22, 288.
Activity, ed. V. Bumba and J. Kleezek (Dordrecht: Reidel), p. Latour, J., Spiegel, E. A., Toomre, J., and Zahn, J.-P. 1976, Ap. J.,
207. 207, 233.
. 1977, Geophys. Ap. Fluid Dyn., 8, 93. Massaguer, J. M., and Zahn, J.-P. 1980, Astr. /Ip., 87, 315.
. 1978, Geophys. Ap. Fluid Dyn., 11, 157. Ogura, Y., and Phillips, N. A. 1962, J. Atmos. Sei., 19, 173.
. 1979, Ap. J., 231, 284. Toomre, J., Zahn, J.-P., Latour, J., and Spiegel, E. A. 1976, Ap. J.,
Gilman, P. A., and Foukal, P. V. 1979, Ap. J., 229, 1179. 207, 545.

Peter A. Gilman and Gary A. Glatzmaier: High Altitude Observatory, National Center for Atmospheric Research,
P. O. Box 3000, Boulder, CO 80307

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

The Astrophysical Journal Supplement Series, 45:351-380, 1981 February


© 1981. The American Astronomical Society. All rights reserved. Printed in U.S.A.

COMPRESSIBLE CONVECTION IN A ROTATING SPHERICAL SHELL.


II. A LINEAR ANELASTIC MODEL

Gary A. Glatzmaier and Peter A. Gilman


High Altitude Observatory, National Center for Atmospheric Research1
Received 1980 April 17; accepted 1980 July 23

ABSTRACT
We study the onset of convection for a compressible fluid in a rotating spherical shell via linear
anelastic fluid equations for a depth of 40% of the radius, constant kinematic viscosity and
thermometric diffusivity, Taylor numbers up to 105, and density stratifications up to seven e-folds
across the zone. The perturbations are expanded in spherical harmonics, and the radially dependent
equations are solved with a Newton-Raphson relaxation method.
The most unstable modes are single cells extending from the bottom to the top of the convection
zone. As the density stratification and rotation rate increase, the horizontal dimension of these cells
decreases, while their prograde longitudinal phase velocity and the enhancement of the velocity near
the top of the zone increase. Cylindrical cells arranged symmetrically about the equator develop
parallel to the rotation axis as the rotation rate increases, but for large stratifications, their axes bend
toward the pole at midlatitude instead of intersecting the outer surface, as they do for small
stratifications. The buoyancy force does negative work near the top of the zone, while the pressure
force does a net positive amount of work for large stratifications. Helicity profiles and convective
heat flux profiles for the most unstable modes do not change significantly as the density stratification
increases, although distinct differences develop for modes that are not the most unstable. The radial
momentum flux for the most unstable modes at high rotation rates is inward for large stratifications
instead of outward as it is for small stratifications; this may have a significant effect on differential
rotation and magnetic field generation in nonlinear, compressible models.
Subject headings: convection— Sun: interior— Sun: rotation

I. INTRODUCTION of slow rotation, thin shell, hydrostatic equilibrium in


Having developed nonlinear, anelastic equations for a the radial direction and with a parameterized differen-
stratified, rotating, spherical shell in Paper I (Gilman tial rotation, by Gilman (1975) for a rotating spherical
and Glatzmaier 1981), we begin our investigation of shell of finite depth, and by Busse and Cuong (1977) for
compressible, global convection with a linear stability a spherical shell in the limits of rapid rotation and thin
analysis in order to predict and interpret the major shell. Dumey (1968, 1970) developed a quasi-linear,
effects of compressibility and stratification on cell struc- Boussinesq, rotating, spherical shell model and solved it
ture and on the maintenance of differential rotation. as an initial value problem neglecting the fluctuating
Previous studies have been made that account for rota- self-interactions. Nonlinear Boussinesq models have been
tion in spherical geometry; others account for com- studied as initial value problems by Weir (1974) for an
pressibility in plane-parallel geometry. We present a axisymmetric rotating sphere, by Gilman (1976, 1977,
self-consistent hydrodynamic model that accounts for 1978a, b, 1979) for a rotating spherical shell, and by
rotation and compressibility in spherical geometry. Soward (1977) for a sphere in the limit of rapid rotation.
Most models that account for spherical geometry and Cuong (1979) studied Boussinesq convection in a rotat-
rotation employ the Boussinesq approximation; that is, ing sphere as a linear stability problem and as a mildly
the fluid is assumed incompressible except in the nonlinear problem with interaction between the axisym-
buoyancy term in the equation of motion. Linear metric mode and one nonaxisymmetric mode.
Boussinesq models are described by Chandrasekhar However, as discussed by Spiegel and Veronis (1960),
(1961) for an axisymmetric rotating sphere, by Roberts the Boussinesq approximation is applicable only when
(1965, 1968) and Busse (1970) for a sphere in the limit “the vertical dimension of the fluid is much less than
of rapid rotation, by Yoshimura and Kato (1971) and any scale height.” Since the solar convection zone spans
Yoshimura (1971, 1972) for a spherical shell in the limit many density scale heights, the Boussinesq approxima-
tion is applicable only in the very lower part of the zone.
’Sponsored by the National Science Foundation. Compressible convection in a stratified medium has
351

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

352 GLATZMAIER AND GILMAN Vol. 45


been studied mainly via linear, nonrotating, plane- cally in Paper III (Glatzmaier and Gilman 1981). In
parallel models with polytropic reference states. Lamb subsequent papers we shall investigate the effects of
(1910), Skumanich (1955), Unno (1957), Bohm and different boundary conditions, different viscosities and
Richter (1960), Bohm (1963), Spiegel and Unno (1962), conductivities, and different convection zone depths. We
and Spiegel (1964) calculated the convective growth shall also solve for induced-mean second-order quanti-
rates for an inviscid fluid. The onset of steady-state ties such as differential rotation and meridional circu-
convection with constant dynamic viscosity and conduc- lation.
tivity has been studied by Unno, Kato, and Makita
(1960), Spiegel (1965), and Gough et al. (1976). Kato
and Unno (1960) and Heard (1973) investigated the II. REFERENCE STATE
combined effects of rotation and compressibility for a The anelastic fluid equations for a rotating spherical
steady-state, plane-parallel polytrope; Kato and Unno shell have been derived in Paper I by expressing each
assumed an isothermal reference state, while Heard as- dependent variable as a sum of a reference state variable
sumed rapid rotation. Graham (1975) solved a nonlin- and a perturbation. The ratio of the latter to the former
ear, two-dimensional, nonrotating, compressible model is of the order of the maximum (V ~ Vad) of the model
with a polytropic reference state; and Graham and convection zone which was shown in Paper I to be of the
Moore (1978) investigated the effects of different viscos- same order as the Mach number of the maximum con-
ities and conductivities on a nonrotating, plane-parallel, vective velocity and assumed to be ~ 10 ~4 in our model.
stratified medium via a linear stability analysis. The formal scale analysis resulted in the anelastic per-
Another approach to solving the compressible prob- turbation equations with coefficients that depend on the
lem involves the use of the anelastic approximation as chosen reference state.
discussed in Paper I. Toomre et al. ( 1976) and Massaguer As in Paper I, we assume a perfect gas, a central
and Zahn (1980) have developed nonlinear, anelastic gravitational field (ignoring self-gravitation), and an
models; these models are nonrotating and plane-parallel eddy diffusive heat flux proportional to the superadia-
with specified horizontal planforms. An anelastic liquid, batic temperature gradient, i.e., the entropy gradient.
nonrotating, plane-parallel model has been used by Jarvis The nondimensiqnal reference state variables, pressure
and McKenzie (1980); it includes the effects of com- /?, temperature 0, density p, and entropy s, are time
pressibility, but only in the limit of small density stratifi- independent and functions of radius r only. The refer-
cations and infinite Prandtl number. ence state is determined by specifying the radial depen-
A few models that have been developed include dence of the eddy thermometric diffusivity ic(r) and two
spherical geometry, rotation, and compressibility, but nondimensional parameters: ß, which is the ratio of the
these rely heavily on parameterization of convective heat bottom radius to the depth of the convection zone, and
flux and momentum flux. Belvedere and Paterno (1977) Np, which is the number of density ¿-folds across the
developed such a model for axisymmetric flow in the zone.
limit of large viscosity and slow rotation to explain The perturbation equations described in this paper
differential rotation. Turbulent diffusion of heat is as- allow the kinematic viscosity v and the thermometric
sumed to be a function of latitude; this results in diffusivity K to be functions of radius, but the solutions
a meridional circulation which drives the equatorial presented in this paper were obtained by assuming v and
acceleration. A similar approach has been used by k constant and equal. This assumption is in fairly good
Vandakurov (1975a, b). agreement with solar mixing-length models. Since both v
In this paper we study the effects of compressibility and k represent eddy diffusivities, they are roughly the
and stratification in a deep, rotating, spherical shell via product (eddy velocity) (mean free path); the product of
the linear, anelastic fluid equations. The reference states the convective velocity and the mixing length in solar
we use are discussed in § II. We describe the perturba- mixing-length models (see Baker and Temesvary 1966)
tion equations in § III and the numerical solution tech- varies slowly throughout the convective zone. Conse-
nique in § IV. Section V begins with a summary of the quently, the Prandtl number P=v/k—\.
major differences we find between Boussinesq and com- Mixing-length theories predict the bottom of the solar
pressible convection in rotating spherical shells. We then convection zone to be ~ 20% of the solar radius below
describe and interpret these differences in terms of the photosphere (see, e.g., Gough and Weiss 1976), but
stability, phase propagation, mode structure, kinetic en- this prediction depends on the chosen ratio of the mix-
ergy balance, helidty, momentum flux, and heat flux. ing length to the pressure scale height and on the chosen
We also compare our results with those of many of the hydrogen, helium, and metal abundances (see, e.g., Baker
models mentioned above. 1963). Rhodes, Ulrich, and Simon (1977) argue that,
In order to better understand the physics of the phase based on the k—co plane analysis of nonradial solar
propagation, we define a simple analog to the com- /?-mode oscillations, the depth of the solar convection
pressible, rotating, spherical shell and solve it analyti- zone is greater than 25% and could be as much as 50%

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

No. 2, 1981 LINEAR ANELASTIC SHELL MODEL 353


of the solar radius, but the observed surface abundance The nondimensional reference state density p, tem-
ratio of Li/Be places an upper limit of — 40% on the perature 0, and (V-VAd) are plotted versus radius
depth of the zone (see, e.g., Weymann and Sears 1965). in Figure 1 for the five density stratifications. Solar
In addition, Gilman (1979) has found in his nonlinear mixing-length models (see, e.g., Baker and Temesvary
Boussinesq calculations that the polar vortex, which is 1966) produce (V~VAd) profiles similar to those for
unobserved on the Sun but present in his calculations Np—5 and 7.
for a depth of 20%, is not present for a depth of 40%.
Therefore, for this paper we have chosen a depth of 40% III. PERTURBATION EQUATIONS
of the radius of the top boundary, i.e., ß=\.5. As a
result, when solar values are assumed for the rotation a) Separation of variables
frequency and the outer radius, the Taylor number T is The stability problem is solved by examining the
of the order of 105. onset of stationary convection under the assumption
In order to see how the density stratification affects that the perturbations are small enough to drop the
the dynamics of the convection, we examine five differ- nonlinear terms in the equations derived in Paper I.
ent density stratifications corresponding to iV^lO“2, Stationary convection here means in the global sense;
1, 3, 5, and 7. The value 10 ”2 is used to compare our that is, the total kinetic energy of the convection zone
results with previous Boussinesq calculations. An Np remains constant. The resulting nondimensional, linear,
between 5 and 7 is typical for the Sun since we are anelastic perturbation equations of state, mass, momen-
assuming that the top boundary of our model corre- tum, and energy are:
sponds to a surface — 3% of the solar radius below the
£ = £ + !., !=l Y~1\^. (2a)
photosphere.
The entropy difference A5 (scaled by CP) across the p p e e Y IP’
convection zone is a measure of the departure from an
adiabatic stratification and is equal to V *v— — — ÚR (2b)
pit’

ß+\ 1 d0_ 9v
dr ^r+T]/2P\Xz
Jr
Jß dr JßR t)e dr 97 P \ r ) p
: P dpv
+ Pv v(w)-vx VX v +
= -{v-VADXy{\-e-N>), (1) p dr

X 2 +
where y is the ratio of specific heats and V is the (£-"M!7-7 7!^
logarithmic derivative of the temperature with respect to
the pressure. The entropy difference plays the same role 3w u 1 8w
+ (2C)
in a compressible atmosphere in which diffusion is via 87 r rcostp d\
small eddies that the temperature difference A0 plays in
a Boussinesq atmosphere in which diffusion is via con- 0 !y=rw+iv(pKV0). (2d)
duction.

r/r0 r/r0 r/r0


2
Fig. 1.—Reference density, temperature, and (v — Vad) plotted vs. radius for density e-folds Np = 10 , 1, 3, 5, and 7. The density and
temperature are scaled by their values at the top of the zone and (V — Vad) is assumed to be 10 4 at the top.

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

354 GLATZMAIER AND GILMAN Vol. 45


As in Paper I, the spherical coordinates are longitude X, dwr
latitude <p, and radius r; the corresponding components u(r,cp,\)= 2 wr+
/, m dr
of the velocity v are w, v, and w. Also, Sl = Qz- The
thermodynamic perturbation variables are pressure p,
density p, temperature 0, and specific entropy s. The 1 arr , zmwr
Taylor number T, the Rayleigh number R, the Prandtl X 7 • (5b)
cos<p 3X 9<p
number P, and the Froude number P have been defined
in Paper I.
The function is defined as (\/p)dp/dr. In addition,
The spatial and temporal dependences of the per-
the v/ m are orthogonal with respect to integration over
turbations are separated employing a Fourier expansion
radians, so
of the time dependence and a spherical harmonic
expansion ( Y/” [ <p, X ]) of the spatial dependence. Conse-
quently, the onset of stationary convection is char- f \y\2dQ= 2 /(/+1)
acterized by Imco^O. For m=£0, the Rew in general l, m
does not vanish, so the phase of the convection pattern
propagates in longitude with respect to the rotating
l{l+\)\wr\2 +r2 (7+ÄP)^
coordinate system. The onset of stationary convection in
the axisymmetric (ra = 0) case requires the frequency co
to vanish. dw,"
The simple equation of state allows a straightforward + r22|z;
i 'ym i 2
(6)
dr
spherical harmonic expansion of the spatial amphtudes
of the thermodynamic perturbations.
The radially dependent amphtudes 0/”, p?, p™, W¡m,
e(r,<p,X)=^er(r)Yr(<p,X); (3a) and Z/” are, in general, complex with the property
mj Note the spherical harmonics also
have this property. Consequently, the spatial amphtudes
p(^<pA)= (3b) of the perturbations, for a given m, have the property
m,l

p(r,<P,\)=2p7(r)Yr(<P,\). (3c) wm*(r,(P,\)='2l(l+l)Wr*Ylm*


m, / i

=
However, the anelastic mass continuity equation re- 2 l(l+ l)W,,~mY/~m =w~m(r,(p,X).
quires a more complicated expansion for the compo- I
nents of velocity. Since the mass flux (pv) is solenoidal, (7)
it can be expanded in basic poloidal and toroidal
vectors (see Roberts and Stix 1972), as is normally Therefore, to obtain real spatial amphtudes for a given
done for the velocity when the fluid is Boussinesq (see |m|, the m and —m complex amphtudes are added
Chandrasekhar 1961). With the following expansions together.
for the radial components of the velocity and vorticity,

w(r,qt»)A)= 2 l(l+l)Wr(r)Yr(<p,\), b) Radial Equations


m,l
In order to arrive at equations that depend only on
(4a) the radial coordinate r, the radial component of the curl
of the equation of motion (2c) and the radial component
( V X v(r, <p, A))r = 2/(/+1 )Z/m(O^/m(<P A), of the curl of the curl of the equation of motion are
mj formed. The (L2p) term is eliminated from the latter
(4b) equation by first employing the equation of state (eq.
we get the following expansions for the latitudinal and [2a]) to get the density perturbation in terms of the
longitudinal components of velocity: pressure and temperature perturbations, and then get-
ting {L2p) in terms of the velocity by subtracting the
radial derivative of the radial component of the equa-
dwr tion of motion from the divergence of the equation of
v(r,<p,\)= 2 '+
l,m (AJ"'''■ * motion. Here [—L2/r2] is the angular part of the Lapla-
cian operator such that L2Y[m =1(1+ l)T/m. The per-
97J Z,m 317” } turbation equations thus reduce to an infinite set of
X (5a)
3<p cos<p 3\ J’ coupled, first-order, complex differential equations in

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

No. 2,1981 LINEAR ANELASTIC SHELL MODEL 355


terms of the following complex functions of radius. d) Boundary Conditions
For this paper we choose to apply stress-free boundary
wr, er, conditions at the top and bottom of the convection
zone. Therefore, at the boundaries, 9/3r(t>/r)=0 and
8/9r(«/r)=0, which implies that 3/9r(v Xv), =0.
xr
Since

dW,m ddi
DW,m = Der = ¿(v Xv)r= 2 /(/+ \)DZrYr, (10a)
dr ’ dr /, m
dxr /tym DZ? dY¡m
Dxr = Dzr=-~~.
1 (8)
dr ’ dr 9r (7)=,2
/, m d(p cosqp 3X , (10b)

As can be seen in equations (A4)-(A11), the Coriolis 97,"' 97


force couples the equations for a given / with equations ' +nzr »(10c)
dr (”)=,2
/, m cos <p 9X dcp
for (/—l) and (/+1). But the equations are uncoupled
with respect to the longitudinal wavenumber m. Dumey
(1968) found this type of coupling for a Boussinesq fluid each DZfm and X/” vanish at the boundaries. We also
in a rotating spherical shell. The amplitudes are coupled choose the boundaries to be impermeable, so w=0, and
in the following way. therefore, by equation (4a), each W™ =0 at the
boundaries. In addition, we apply constant temperature
boundary conditions, so 0=0, and therefore, by equa-
tion (3a), each 0™=0 at the boundaries. Note that
different boundary conditions will be considered in a
future paper. Finally, an arbitrary amplitude must be
specified to solve this set of coupled linear equations.
We set DW™ = — 1 at the top boundary for one value
of /.
The axisymmetric case (m=0) again requires special
attention. Note first that the chosen spherical harmonic
expansion for automatically conserves total angu-
(9) lar momentum in the direction of Ö because the contri-
bution relative to the rotating frame,
For the case of no rotation (T=0), there are only six
coupled, first-order, differential equations in terms of
the six amplitudes in the first column of display (9). Lz = fpurcosydr (ii)
There is no radial vorticity and no / or m coupling; all
amplitudes are real. vanishes term by term (see eq. [5b]) since

c) Symmetry with Respect to the Equator J


f Yimœscpd(pd\=8li0Ôm0. (12)
4ir
A convenient symmetry with respect to the equator
exists because of the above coupling scheme and the fact However, for m=0, the /— 1 term of the longitudinal
that (/— I m I) is the number of nodes in Y™ between the velocity includes a y0° term from dY^/dcp. Conse-
poles. If the coupling scheme begins with JF/™n, i.e., quently, for the axisymmetric symmetric solution,
l
min = \m\ for m=£0, then w, u, 0, p, p, and (VXv)^
peak at the equator, while v, (VXv)r, and (VXv)A, L 4
(j)'/2J*+1pZ°r4dr.
z= (13)
vanish there. The opposite is true if the coupling scheme
begins with Z/h . The former case is called a symmetric
solution since tfiere is no flow across the equator; the Dumey (1968) found this same condition in his study of
latter case is an antisymmetric solution. This same sym- axisymmetric Boussinesq convection. Since the radial
metry property has been used by Roberts (1968). component of vorticity (for m=0, 1= 1) is proportional
For the axisymmetric case (m=0) l^n = 1 because y0° to sincp, equation (13) represents an arbitrary solid
is a constant, and so its contribution is included in the rotation with respect to the rotating frame. Therefore,
reference state variables. Therefore, for m=0, a symmet-
ric coupling scheme begins with Z® while an antisym-
metric scheme begins with IF,0.

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

356 GLATZMAIER AND GILMAN Vol. 45


must vanish in order to have a zero total angular 140 points in a meridian plane. Then the number of
momentum relative to the rotating frame. Note that this harmonics is increased by two, thus increasing the num-
complication does not arise in the axisymmetric anti- ber of complex equations by eight, and the entire proce-
symmetric case. dure is repeated. New values of w, 0, and (V X v),. are
To apply this additional condition we first solve for computed at the same 140 meridian points. When the
DZ® using the radial component of the curl of the average relative change in each of w, 0, and (vxv)r
equation of motion (eq. [All]) and the boundary condi- drops below 5X10-3, and when the maximum relative
tions on DZ® and W®. The result is change drops below 5X10-2 for each of these func-
tions, the series of spherical harmonics is truncated. This
dZ° 3 1. convergence criteria does not have to be quite this
DZ°X = (14) stringent if one is only interested in the eigenvalues, i.e.,
dr v 15
the Rayleigh number and the frequency, which converge
Then we solve for Z^r) by integrating equation (14) much faster than the eigenfunctions. However, we are
from the bottom of the convection zone to a radius r. also interested in the detailed structure of the modes.
Inserting this solution into equation (13) and setting the The number of spherical harmonics required to describe
integral equal to zero results in the following boundary the perturbations increases with density stratification
condition: and rotation rate because these factors tend to com-
plicate the mode structure. The largest number of
fß+]pr4[ (W2/v)dr dr spherical harmonics used in our calculations was 22,
TV/iJß K V corresponding to 88 coupled complex differential equa-
tions.
Since there is no /-coupling in the nonrotating case
(T=0), the system reduces to six equations which are
(15) independent of time and the longitudinal wave number
m, and are real. Consequently, only the first level of
Consequently, for a symmetric solution with m=0, one convergence is required because the Rayleigh number
equation {dDZ^/dr— •••) and two boundary condi- serves as the eigenvalue and a single / specifies the
tions (DZ® =0 at r=ß, ß+\) are eliminated, while the problem.
new boundary condition (eq. [15]) is added. In the axisymmetric case («2=0), as in the nonrotat-
ing case, the frequency vanishes, the equations are real,
IV. NUMERICAL TECHNIQUE and the Rayleigh number serves as the eigenvalue. But
The solution to the reference state equations provides unlike the nonrotating case, the Coriohs forces produce
the coefficients (functions of r) for the perturbation /-coupling. Consequently, the second level of conver-
equations. Since these equations are uncoupled with gence is not needed, but the third level is still necessary
respect to the longitudinal wavenumber m, we are inter- to determine a valid truncation of the spherical harmonic
ested in the symmetric and antisymmetric solutions on series.
the stability boundary for each m. The Newton-Raphson relaxation method uses 37 ra-
Three levels of convergence are employed in the pro- dial mesh points with a finer mesh near the top of the
cess of solving this double eigenvalue stability problem. zone. The nonrotating solutions easily converge and are
The first level is a Newton-Raphson relaxation method not very sensitive to the initial guesses for the dependent
that solves a set of «-coupled, first-order, complex dif- variables. These solutions can then be used as initial
ferential equations with «-complex dependent variables guesses for the next higher Taylor number and so on, for
and «+1 complex boundary conditions for given the same wavenumber and symmetry. This bootstrap
Rayleigh (R), Taylor (T), and Prandtl (P) numbers. method works particularly well for wavenumbers larger
The eigenvalue, unless m=0 or T=0, is the complex than the critical «2, i.e., the m corresponding to the
frequency co. The second level of convergence is a smallest Rayleigh number for a given value of T, be-
Newton-Raphson method that finds the Rayleigh num- cause the structure of the most unstable modes changes
ber that yields a vanishing imaginary frequency. Its relatively smoothly as the rotation rate increases. But
iteration terminates when the ratio of the imaginary when m is less than the critical «2, the structure of the
frequency to the real frequency drops below 10 4. The most unstable mode can change drastically for small
third level of convergence finds the minimum number of changes in T or «2. Consequently, the solution for
spherical harmonics required to accurately represent the («2—1) or («2+1) of the opposite symmetry but same
actual solution with a truncated series. After finding a Taylor number must also be tried as an initial guess in
solution on the stability boundary for a given number of order to find the most unstable solution for the given T
spherical harmonics (starting with l—\m\ or l=\ if and «2. Care must also be taken to converge on the
m—0), the values of w, 0, and (V X v)r are computed at fundamental radial mode since higher radial harmonics,

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

No. 2, 1981 LINEAR ANELASTIC SHELL MODEL 357


i.e., multiple cells in radius, exist at higher Rayleigh Rayleigh numbers, when evaluated in the center of the
numbers. Consequently, solving for the most unstable zone, are essentially the same for a stratified gas and a
solution with a Newton-Raphson relaxation method, Boussinesq liquid, the stratified gas is relatively more
especially for large T, large Np, and small m, requires stable because the Rayleigh numbers are much larger
good initial guesses for the dependent variables, the when evaluated in the region of maximum convective
Rayleigh number, and the frequency. velocity. As in Boussinesq convection, single cells ex-
Although Dumey (1968, 1970) also studies convection tending from the bottom to the top of the zone are more
in a rotating spherical shell by expanding in spherical unstable than multiple cells in radius for the parameter
harmonics, his numerical solution technique is signifi- range we have investigated. But the critical wavenum-
cantly different than ours. Following Chandrasekhar’s bers, associated with the most unstable modes at large
(1961) method, Dumey expands his radially dependent rotation rates, are much larger than those for the
functions in a combination of spherical Bessel functions Boussinesq case; and the stability curves are much shal-
that satisfies the boundary conditions. Then, with initial lower. This suggests that larger longitudinal wavenum-
estimates for the functions at the two boundaries, he bers and a larger number of wavenumbers will be re-
integrates outward from the bottom and inward from quired in nonlinear, compressible calculations. The
the top. A generalized Newton-Raphson method is then longitudinal phase velocities of the convective patterns
used to find solutions that match in the center of the are prograde at all rotation rates; in the Boussinesq case,
zone. For the axisymmetric, steady-state, rotating prob- the modes propagate in the retrograde sense at low
lem Dumey (1968) sets the Rayleigh number to 1500, rotation rates or for high wavenumbers. Also the magni-
which is about twice the critical Rayleigh number for tudes of the phase velocities are much larger than those
r=0. He begins by using /=8-12, instead of starting for the Boussinesq case. This suggests that the nonlinear
with /= 1, and continues to increase the range until the patterns will propagate faster and only in the prograde
relative difference in his radial amplitudes between sense.
successive iterations is less 20%. His largest Taylor num- Differences in the mode structure also exist between
ber is 502, for which he uses 15 spherical harmonics; for the compressible and Boussinesq cases. Velocity is very
T>502 convection is completely inhibited by rotation enhanced in the upper part of the convection zone for a
(for R= 1500). For the time dependent nonaxisymmetric constant kinematic viscosity. This contradicts the con-
problem Dumey (1970) sets R= 1500 and 77=4, and clusions Graham and Moore (1978) make based on their
integrates in time using only /=8-12 and only 5 radial compressible models. Cylindrical cells parallel to the
modes. The main difference between our numerical rotation axis develop at high rotation rates as they do in
technique and Durness is that we search for the most the Boussinesq case; but unlike the Boussinesq case,
unstable modes, each corresponding to a given T, m, their axes bend toward the pole at midlatitude instead
and Np, by solving a linear eigenvalue problem, whereas of intersecting the outer surface. We find, as did
Dumey searches for solutions for a given Rayleigh Massaguer and Zahn (1980), a reversal of the density
number above critical by solving a quasi-linear initial perturbation near the top of the zone which results in
value problem. Also we require more stringent conver- negative buoyancy work there; this never occurs in
gence criteria and find solutions for much higher Taylor Boussinesq convection. Also instead of zero net work,
numbers (up to 105); consequently, we must use a larger pressure does a net positive amount of work in strati-
number of spherical harmonics. fied, compressible convection. Although helicity profiles
Cuong (1979) uses a spherical harmonic expansion in for a stratified gas and a Boussinesq liquid are com-
his linear stability analysis of a rotating sphere. He pletely opposite at moderately low wavenumbers, they
expands his radial dependent functions in sines and are similar for the most unstable modes. But the peak
cosines and solves for the expansion coefficients via a helicity in a stratified gas (for constant v and k) occurs
Newton-Raphson iteration method. The sets of equa- near the top boundary of the zone instead of in the
tions that Dumey and Cuong solve are different from center, as it does for a Boussinesq liquid.
ours since they use the Boussinesq approximation, and Second-order differences are also suggested and will
we use the anelastic approximation. be tested in future papers. For example, the mean
differential rotation in a nonlinear system is partially
determined by the momentum flux profiles. For large
V. RESULTS AND INTERPRETATION
density stratifications, the radial momentum flux at high
rotation rates is inward instead of outward as it is in the
a) Summary Boussinesq case. Since differential rotation and helicity
Some of the major differences and similarities we are the two important ingredients in stellar dynamo
have found between very stratified, compressible con- theories, compressible dynamos could be quite different
vection and Boussinesq liquid convection in a rotating, than the Boussinesq dynamos. Also the mean tempera-
spherical shell are the following. Although the critical ture gradient in a nonlinear system is partially de-

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

358 GLATZMAIER AND GILMAN


termined by the convective heat flux profiles. Although T for the five different values of Np. But the critical
the latitudinal convective heat flux profiles for a strati- wave numbers mcñi increase dramatically with density
fied gas and a Boussinesq liquid are completely opposite stratification at high rotation rates and with rotation
at moderate rotation rates, die respective radial convec- rate for large stratifications, as illustrated in Figure 2
tive heat flux profiles are very similar and dominate, and summarized in Table 1. This effect was also found
especially for large stratifications, over the latitudinal by Kato and Unno (1960) for an isothermal, plane-
heat transport. parallel model. However, Heard (1973) found for a
We discuss in detail below the differences and similar- rapidly rotating plane-parallel polytrope only a slight
ities between Boussinesq and compressible convection in increase in the critical wave number, with increasing
a rotating spherical shell, concentrating on the most stratification for a constant v model and a slight de-
unstable modes since they are more likely to char- crease for a constant pv model. Figure 2 also shows how
acterize what would happen in a nonlinear system and mcnt increases slightly with increasing T for Np ^0, as
on Taylor number 105 since it approximates the solar has been observed by Gilman (1975) and Cuong (1979)
case. for Boussinesq convection. In the nonrotating case, mcrii
increases slightly as Np increases; this has also been
b) Stability and Phase Propagation observed by Unno et al (1960), Gough et al (1976), and
Graham and Moore (1978) for a plane-parallel poly-
The onset of stationary convection for a given mode trope. Also notice how much shallower the stability
occurs at the minimum superadiabatic temperature curves in Figure 2 (left) are for large stratifications.
gradient that is capable of sustaining convective per- The real frequencies associated with the stability solu-
turbations by balancing the work done by the pressure, tions are plotted on the right-hand side of Figure 2 and
buoyancy, and viscous forces. This stability boundary is are scaled, as in Paper I, by the thermal diffusion time.
characterized by a certain value of the Rayleigh number Notice, for large stratifications, how they are all pro-
which is the ratio of the product of the viscous and grade; i.e., values for Rew>0, are much larger than
thermal diffusion times to the square of the buoyancy those for small stratifications and remain large with
time. Since the net work done by all the forces vanishes, increasing m. The plot in the center of Figure 3 shows
the total kinetic energy remains constant on the stability how the critical frequencies cocrit increase with rotation
boundary. This requires the frequency to vanish in the rate, for a given Ap, and asymptotically increase to
axisymmetric (m—0) case but not necessarily in the T-dependent limits as the density stratification in-
nonaxisymmetric case. When m^O, the rate of change creases. Also notice that when the critical frequencies
of the total kinetic energy is proportional to the imagin- are scaled by the rotation frequency, i.e., when those
ary part of the frequency which must vanish, but the plotted in Figure 3 are multiplied by 27,-1/2, they are
real part of the frequency need not vanish. A nonzero almost independent of ß, as they are in Paper III. The
real frequency represents the rate the nonaxisymmetric critical phase velocities vP d are plotted on the right side
cellular pattern propagates in longitude with respect to of Figure 3. Physical explanations for these properties of
the rotating frame. Since the perturbations are propor- Rayleigh numbers, wavenumbers, and frequencies are
tional to ez(mX_a)r), the phase velocity, in radians per discussed in § W and in Paper III.
time, is Solutions with multiple cell structure in radius also
exist but at higher Rayleigh numbers and with larger
Re
V/> = (16) critical wave numbers. For example, the value of /crit for
m
r=0 and Np = 10-2 is 4 for single cells, 8 for double
A positive value of Re co represents prograde phase cells, and 13 for triple cells. The critical Rayleigh num-
propagation, while a negative Re co represents retrograde bers for double and triple cells are, respectively, 16 and
propagation. 83 times that for single cells. For T—0 and Np =7, the
The Rayleigh numbers Rci evaluated in the center of value of /crit is 6 for single cells, 9 for double cells, and
the convection zone and associated with the onset of 13 for triple cells, and the critical Rayleigh numbers for
convection for single cells in radius r are plotted versus double and triple cells are, respectively, 12 and 50 times
longitudinal wavenumber m in Figure 2 (left) for six that for single cells.
different Taylor numbers T, five different density Although the critical Rayleigh numbers plotted in
stratifications Np, and both the symmetric and antisym- Figure 3 do not vary significantly with density stratifica-
metric solutions. The critical Rayleigh numbers Rcrit, tion, the actual stability of the convection zone does.
i.e., the smallest Rayleigh numbers for a given T and Np, The Rayleigh number, as defined in Paper I, is
are circled in each case. Notice how the critical Rayleigh
numbers increase with rotation rate but are relatively
insensitive to the density stratification; this is also evi- g0^[(^7¿r)0+g0/CP]
(17)
dent on the left in Figure 3 where Rcúx is plotted versus 0oVoKo

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

RAYLEIGH NUMBER
SYMMETRIC ANTISYMMETRIC
I I I I I I ■ I ■ I M M M ■ M M~

0 2 4 6 8 10 12 14 16 18 20 220 2 4 6 8 10 12 14 16 18 20 22
m m
Fig. 2.—Rayleigh numbers Rct on the left and frequencies co on the right plotted vs. wavenumber m for the symmetric and antisymmetric
modes, for Taylor numbers r=0, 102,103, 104, 3X104, and 105, and for density e-folds Np = 10 _2, 1, 3, 5, and 7. The Rayleigh numbers are
evaluated in the center of the zone, and the frequencies are scaled by the thermal diffusion time.

© American Astronomical Society • Provided by the NASA Astrophysics Data System


FREQUENCY
SYMMETRIC ANTISYMMETRIC
HO i I i M I i I i I r i i i i i i-i i -| i
100
90 Vier2
80
70
60 I04
50
40 / 3XI04
30 // I05
20
10
O
-10
HO M I I I I'I I I I T I M I I M I I I
100
90 iVi
80
70
60
50
40
30
20
10
O
-10 » 1 » I I I I I I I I I I I I I I I I I I
110 i ri i " i r u m 'u -i i i i i i i i i i Ï r i--r i T ( nr i i-1 i i i i i i i r i
100
90
80
70
60
50
40
30
20
10
O
-10 i i i i ■ i ■ i » i » i i i i i i i i i » ,i 1 i 1 i 1 -L I ,,i. I i I i I i I i .1 i I I
HO
100
90
80
70
60
50
40
30
20
10
O
-10 i L_i I j I i I i I i 1 i I i I i I i I i
110
100
90
80
70
60
50
40
30
20
10
O
-10
2 4 6 8 10 12 14 16 18 20 22 0 2 4 6 8 10 12 14 16 18 20 22
m m
Fig. 2.—Continued

© American Astronomical Provided by the NASA Astrophysics Data System


198lApJS...45..335G

LINEAR ANELASTIC SHELL MODEL 361


SYMMETRIC MODES

Fig. 3.—Critical Rayleigh numbers Rcrit on the left, critical frequencies wcrit in the center, and
2
critical phase velocities vpcncn( —^crit/wcrit
on the right plotted vs. Taylor number T for the symmetric modes, for density e-folds Np = 10 ~ , 1,3, 5, and 7.

and is based on the values of the variables at the top of thickness with weak density stratification.” In our model
the convection zone. However, since we are dealing with gravity decreases slightly with radius, but the super-
stratified zones, the local Rayleigh number is actually a adiabatic temperature gradient increases very rapidly
function of radius. The problem of how to define the with radius for large stratifications (see Fig. 1); and k
Rayleigh number in a stratified zone has been discussed are assumed constant in this paper. As a result, our local
by Unno et al (1960), Spiegel (1965), Graham (1975), Rayleigh number increases with radius for large stratifi-
Gough et al (1976), and Graham and Moore (1978) in cations. In addition, the radial velocity peaks in the
relation to the nonrotating, plane-parallel polytrope. In upper part of the zone for large stratifications (see
their models the superadiabatic temperature gradient, discussion in § Vc). Therefore, we also conclude, based
the gravitational field, and the dynamic viscosity (vp) on the same type of reasoning put forth by Gough et al
and conductivity (Cpicp) are assumed constant. As a (1976), that larger stratifications are more stable. In
result, their local Rayleigh number decreases with height addition, the buoyancy force does negative work in the
above the bottom (faster for larger stratifications). In upper part of stratified zones (see discussion in § Ne)
addition, since the kinematic diffusivities are relatively which adds to the stability of large stratifications. One
large near the top, the vertical velocity peaks in the could argue that an average Rayleigh number over the
lower part of the zone (more so for large stratifications). convection zone, using, for example, the radial velocity
Therefore, Gough et al (1976) conclude that since the as a weighting factor, would be a more representative
vertical velocity peaks in the region where the local ratio of diffusion time to buoyancy time, but the actual
Rayleigh number is large, “a layer with strong density definition used is arbitrary. We find that large stratifica-
stratification is more stable than a layer of the same tions have larger (smaller) critical Rayleigh numbers

TABLE 1
Critical Wavenumbers mcrit for the 3Symmetric
and Antisymmetric Modes

Nn 0 102 103 104 3 X 104 105


10' 4,3 4.3 5,5 4.5 5,6 6,8
1. 4.3 4.4 5.4 5.6 6,8 8,10
3. 5.4 5.4 6.5 7,8 9,10 12,13
5. 5.4 6.5 7.6 9,10 12,13 16,17
7. 6.5 6,5 7.7 10,11 13,15 18,20
3
Values for the symmetric and antisymmetric modes are given preceding
and following the comma respectively.

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

362 GLATZMAIER AND GILMAN Vol. 45


when evaluated at the top (bottom) of the zone com- for an odd m. Consider now solutions for m>4. Since
pared to small stratifications, but when evaluated in the l>m for a given m, and since the Rayleigh number
center of the zone the critical Rayleigh numbers are monotonically increases with / for />4, the most unsta-
comparable for all stratifications, as discussed previ- ble symmetric solution occurs for /=m, while the most
ously. Unno et al (1960) also found that the critical unstable antisymmetric solution occurs for /=w+l.
Rayleigh numbers, when evaluated in the center of the Consequently, the most unstable solution is symmetric
zone, are essentially independent of stratification. for all m>4. In addition, the antisymmetric Rayleigh
Therefore, we have chosen to plot in Figure 2 the number for m is identical to the symmetric Rayleigh
Rayleigh number evaluated in the center of the zone: number for (m+1) when 7=0 because exactly the same
set of radial equations is solved, beginning with /=w +1.
£±I Stability in the rotating case depends on all l>m.
*ct ( Pct^ct ) ^> (18)
The spherical harmonics with large (l—m) peak near
the poles, while those with small {l—m) peak near the
where pct and 0ct are the reference density and tempera- equator. These peaks become more pronounced and
ture at the center of the zone scaled by their respective localized as / increases. The Y¡m for which l~mcrh
values at the top. contribute the most to the perturbations when m<mcrit.
Stability in the nonrotating case (T=0) depends only Consequently, when the velocity peaks near the
on the / of the spherical harmonic that describes the poles, and the mode is called a polar mode; when
perturbations. Consequently, all wavenumbers w</crit m^mcúi the velocity peaks near the equator, and the
can represent modes having the same value of Rcrit> mode is called an equatorial mode. Polar and equatorial
although their horizontal structure and symmetry with modes were first noticed for the sphere by Roberts
respect to the equator (which is an arbitrary great circle (1968) and for the spherical shell by Gilman (1975) and
for 7=0) depend on (/crit — m). As an example, consider have been later discussed by Busse and Cuong (1977).
the nonrotating case with one density e-fold across the Roberts compares polar modes to a Benard layer per-
zone. The Rayleigh number Rct is plotted versus the / pendicular to il and g, and equatorial modes to a
number on the left in Figure 4. The critical Rayleigh Benard layer parallel to il and perpendicular to g. The
number is 635.8 with /=4, while the next smallest transition from polar to equatorial mode is very definite
Rayleigh number is 676.3 with /=3. The right side of for small stratifications because mcrit is small and the
Figure 4 shows the corresponding plots of Rayleigh valley in the stability curve is small (see Fig. 2). But for
number Rci against longitudinal wavenumber m for the large stratifications mcrit is large, and there are a larger
symmetric and antisymmetric cases. Note that a solution number of /=values associated with the valley of the
is symmetric (antisymmetric) if (/crit —m) is even (odd). stability curve that make significant contributions.
Consider first solutions for m<4. The most unstable Therefore, for large Np, various intermediate mode tran-
symmetric solution occurs for /=4 if m is even and for sitions occur as m increases from 0 to mcrit. The sym-
/=3 if m is odd. On the other hand, the most unstable metric Np=! plot in Figure 2 shows two examples of
antisymmetric solution occurs for / = 3 if m is even and transitions that occur as a result of the intersection of
for /=4 if m is odd. Consequently, the most unstable different stability curves; the broken lines represent the
solution is symmetric for an even m and antisymmetric continuation of the respective stability curves. These

N/» =l, T =0

Fig. 4.—Rayleigh number Rct plotted vs. / on the left and m on the right for = 1,7=0. The symmetric modes, on the right, are
indicated by dots, the antisymmetric modes by crosses.

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

No. 2, 1981 LINEAR ANELASTIC SHELL MODEL 363


transitions are reflected in the corresponding frequency bottom is essentially 1 for the case of constant (pv) and
curves as “jumps” from the characteristic frequency of asymptotically approaches 3.5 as Np increases for con-
one mode structure to that of the other. In addition, for stant v. We find much larger ratios of the horizontal
large values of Np and T, à large number of spherical velocity at the top to that at the bottom in our inviscid
harmonics are required to describe low wavenumber (¿^O), rotating, polytropic model of Paper III because
polar modes because the series begins with /=m<mcrit there is no viscous resistance. However, like the viscous
and terminates with an />wcrit. For this reason, many polytropic models, the ratio approaches a limit with
of the low wavenumber solutions were not calculated for increasing stratification because, as the density stratifi-
large Np and T. cation increases, the density scale height asymptotically
The solutions for iVp = 10 ~2 have been compared to approaches a limiting radial profile (see Paper III).
Gilman’s Boussinesq calculations for the same depth Therefore, since conservation of mass requires the diver-
(unpublished). When the anelastic Rayleigh numbers are gence of the velocity to equal the vertical velocity di-
adjusted to coincide with the definition of the vided by the local density scale height, the velocity
Boussinesq Rayleigh numbers (see Jeffreys 1930) by profile asymptotically approaches a limit as Np increases
using ks/Cp instead of «A0 (see eq. [1]), they agree to for a polytropic model. Graham and Moore (1978) also
within ±3%. Similarly, the frequencies agree to within look at a constant ic model and find that the ratio of the
— ±5%. We consider this good agreement, considering horizontal velocity at the top to that near the bottom is
the difference in the two methods. Using the technique only ~ 3, with only slight differences between the cases
described in his 1975 paper, Gilman solved an initial of constant (pv) and constant (v). They therefore state
value problem on a staggered grid in radius and latitude, that “the horizontal velocity at the upper surface was
while this compressible problem has been solved as an always found to be comparable with the velocity
eigenvalue problem with a spherical harmonic expan- throughout the bulk of the fluid, in spite of the great
sion. variations of the density”. Although their constant /c
model allows large variations in pressure, the maximum
ratio of the density at the bottom of the superadiabatic
c) Mode Structure zone to that at the top is 2.5, about one density e-fold.
We wish to understand how the structure of the linear We get similar results in our present constant v and k
modes varies with rotation rate, density stratification, model for Np = 1 and 3. However, for Np =1, the ratio of
and wavenumber. The structure and its variations will the peak horizontal velocity at the top to that at the
help explain the stability and propagation properties bottom (for the most unstable modes) is of the order of
described above. Velocity, mass flux, thermodynamic 10 at low rotation rates and of the order of 100 at high
perturbations, and vorticity are first-order perturbation rotation rates, because the cell center moves toward the
quantities which are constructed directly from the linear top as the value of m increases. Velocity profiles for the
solutions. We also are interested in work densities, helic- most unstable symmetric modes are plotted in constant
ity, momentum flux, and heat flux. These are second- latitude (4°) surfaces in Figure 5 for five different
order perturbation quantities also constructed from the density stratifications; the nonrotating case (T= 0) is on
linear solutions. The profiles of both the first- and the left, the rapidly rotating case (r=105) is on the
second-order quantities depict only the initial tendencies right. The three-dimensional flow is illustrated with
at the onset of convection, but, based on the results of arrows representing the component of velocity in the
the linear and nonlinear Boussinesq calculations plotted surface and contours representing the velocity
(Gilman, 1975, 1977), these linear solutions may serve as normal to the surface. Notice how the velocity becomes
qualitative predictions of the nonlinear solutions. enhanced near the top of the zone as the stratification
The enhancement of the velocity near the top of a increases and as the wave number increases for large
stratified zone has been observed by several authors. values of Np.
Kato and Unno (1960) and Heard (1973) show for their This w-dependence of the local velocity enhancement
plane-parallel, polytropic models how the mean vertical has been observed by several authors. Spiegel (1965) and
velocity peaks in the upper part of the zone for constant Heard (1973) found for their viscous polytropic models
v and in the lower part for constant ¡5 in the latter case that the vertical velocity peak moves toward the top of
the velocity is damped in the upper part of the zone the zone as the wavenumber increases. This agrees with
because v is large there. Graham and Moore (1978) the results of the inviscid, nonrotating polytropic models
consider the effects of different viscosities and conduc- of Skumanich (1955), Bohm and Richter (1960), Bohm
tivities on convection in a nonrotating, plane-parallel (1963), and with the rotating model of Paper III. There-
compressible fluid. For a constant dynamic conductivity fore, as m increases, the horizontal velocity at the top
(Cppic), i.e., a polytrope, with density stratifications as increases relative to that at the bottom because of
large as ptop/pbot = 1000, i.e., Npæl, they find that the conservation of mass, and the center of the cell moves
ratio of the horizontal velocity at the top to that at the toward the top of the zone.

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

VELOCITY AT 4° LATITUDE
ROTATION

Fig. 5.—Velocity profiles in constant latitude surfaces (4° latitude) for the most unstable symmetric modes corresponding to Np = 10 “2,
1, 3, 5, and 7. T=0 on the left and 105 on the right. The arrows represent the component of velocity in the plotted surface, while the contours
represent the velocity normal to the surface; solid contours mean flow away from the equator, i.e., o>0; broken contours mean flow toward
the equator, i.,e., t?<0.

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

LINEAR ANELASTIC SHELL MODEL 365


In addition, we find that the center of the cell moves However, at low latitudes the Coriohs force acts mainly
toward the bottom of the zone as the value of T increases, on the radial velocity of the east-west rohs; so rising and
for a given Np and m, in an attempt to satisfy the sinking fluid flows inefficiently along indirect paths.
Taylor-Proudman theorem (discussed later) by reducing Therefore, the most unstable modes for w<mcrit are
the velocity gradients produced by the surface curvature. polar. When m>mcrit, north-south rohs dominate. At
This effect is apparent in Figure 5 for Np —10 '2, in low latitudes the Coriohs force acts nearly perpendicu-
which case mcúx does not change much as T increases larly to the axis of the cells and is offset by radial and
from 0 to 105. The center of the cell also moves toward longitudinal pressure gradients; so rising and sinking
the bottom as Np increases, for a given T and m, in fluid near the equator moves mainly in the radial direc-
order to reduce the gradient of the velocity divergence tion optimizing the buoyancy work. But at high latitudes
and therefore the viscous damping (see Fig. 1 and eqs. the small longitudinal dimension results in large velocity
[2b] and [2c]). This effect can be seen in Figure 5 for shears in longitude and therefore large viscous stresses
r=0, in which case mcrit does not change much as Np which damp the convective motions. Thus, the most
increases from 10 ~2 to 7. unstable modes for m>mcrit are equatorial.
In the nonrotating case, convection in a spherical Consider the symmetric, equatorial modes. At low
shell can be thought of as a combination of north-south rotation rates the north-south rohs become hehces. The
rolls and east-west rolls which are periodic in longitude. five lines in Figure 6 for T= 102 mark the axes of the
This is apparent on the left side of Figure 5; the arrows north-south hehces for the five stratifications. Hehces
illustrate the periodic north-south roll structure, while develop because the Coriohs forces tend to expand the
the contours illustrate the periodic east-west roll struc- cross sections of counterclockwise north-south rohs; as a
ture. Since the solutions in Figure 5 are symmetric, there result, latitudinal pressure gradients force the fluid in
is no flow across the equator. Notice how these two the counterclockwise cells toward the equator, in the
structures are 90° out of phase and so share the same direction of increasing longitudinal dimension. Clock-
radial flow. The latitudinal structure depends on (/—m); wise circulation produces Coriohs forces that tend to
north-south rolls dominate if m — l, and only east-west contract the cellular cross section, and so there is flow
rolls exist if /rz=0. As discussed in the previous section, toward the pole. Because of conservation of mass, fluid
stability in the nonrotating case depends only on the / of at low latitudes spills over from the equatorward hehces
the spherical harmonic which describes the hori- to the poleward hehces, and vice versa at high latitudes,
zontal structure; note that the effective horizontal producing between these alternating hehces radial hehces
wave number is (/[/+1])1/2. However, as discussed by
Chandrasekhar (1961), the most unstable modes prefer
to have comparable horizontal and radial dimensions. If
the radial dimension were much smaller than the hori-
zontal, very large buoyancy forces, which work only on
the radial flow, would be required to drive the fluid
against the viscous forces which work everywhere. If the
horizontal dimension were much smaller than the radial,
V2v, and therefore the viscous forces would be very
large; again large buoyancy forces would be required.
Thus, the required Rayleigh numbers for the onset of
convection (see Fig. 4) have a minimum for a certain
value of /, i.e., /crit. Notice, for ^<1 and m=/, a
longitudinal dimension (through the middle of the zone
and at 45° latitude) equal to the depth of the zone
requires /crit to be 4 which agrees with Figure 4.
The flow patterns in the rotating case are distorted by
the Coriohs forces. When w<wcrit, east-west rolls
dominate. As discussed by Gilman (1975), the longitudi-
nal pressure gradient, which is proportional to m, offers
little resistance to the Coriolis forces which curve the
fluid trajectories in the planes perpendicular to the
rotation axis. Therefore, at high latitudes the Coriohs
force acts mainly on the horizontal velocity of the
east-west rolls and not on the radial velocity; so rising Fig. 6.—Axis of helices plotted in radius and latitude for
and sinking fluid tends to flow in the direction of the Np = \0~2, 1,3, 5, and 7. The axes that parallel the surface {broken
buoyancy force optimizing the buoyancy work ( — pgw). lines) are for T= 102, while those that are cylindrical at low
latitudes (solid lines) are for T = 105.

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

366 GLATZMAIER AND GILMAN Vol. 45


2
a a 10 ~ and 1. But because of the curvature of the outer
boundary, rising fluid is forced toward the equator and
sinking fluid toward the pole (see the plots in Figs. 5
and 8 for T=105, Np = \0~2, and 1). Therefore, the
meridional velocity profile at high rotation rates, as
illustrated on the right in Figure 9 for Np = 10 _2, T= 105,
is considerably different from the east-west roll profile
at low rotation rates as illustrated on the left of Figure 9
for Np = 10 ~2, T=103. Fluid in the counterclockwise
cells still spirals toward the equatorial plane, but now it
is because the radial dimension increases because of the
NORTH-SOUTH ROLL
surface curvature. Similarly, fluid in the clockwise cells
Fig. 7.— Schematic streamlines representing helices for a spirals away from the equatorial plane. A schematic
north-south roll on the left and for a cylindrical cell on the right. streamline is depicted on the right in Figure 7.
For high rotation rates and large stratifications, the
of rising clockwise and sinking counterclockwise vortices. Taylor-Proudman theorem does not apply because the
A schematic stream-line is depicted on the left in Fig- other torques in equation (19) become significant. As a
ure 7. result, the axes of the cells bend toward the pole at
For high rotation rates (T^IO4) and small density midlatitude (see Fig. 6 for T=105, Np=3,5,T). The
stratifications (Np&0) the Coriolis forces, which are reason for this can be understood by considering the
perpendicular to the rotation axis, overwhelm the com- vorticity equation (19). Recall that the surfaces of con-
ponents of buoyancy forces parallel to the rotation axis, stant reference density parallel the boundary surface
producing cylindrical cells with axes parallel to Ö. This and that the density scale height drops off
effect has been discussed by Roberts (1968) and Busse rapidly near the top of the zone. In addition, the radial
(1970) for a Boussinesq fluid in terms of the Taylor- velocity w tends to be smaller at midlatitude compared
Proudman theorem (see Proudman 1916; Taylor 1921). to low latitudes, while the longitudinal gradients of the
The linear, anelastic, vorticity equation (for constant v), pressure and density tend to be larger because of the
i.e., the curl of equation (2c), is useful for understanding smaller longitudinal dimension. As a result, the pressure
under what conditions the Taylor-Proudman theorem and buoyancy torques in equation (19), which are mainly
applies. in the latitudinal direction for large values of m,
dominate near the top of the zone above midlatitude,
a(vxv) hp % , ^ x generating vorticity in latitude instead of parallel to the
7^ = -zrfX V p+^rx Vp + (2fí*v)v rotation axis. However, the viscous torque is also signifi-
dt p p
cant in this region because of the smaller longitudinal
dimension, i.e., larger (1 /cos2 <p)(32v/3X2), and smaller
+ 2ñ/ipw + ¿'[v 2(V X v)+V X (/ip2?)].
density scale height, i.e., larger V #v. Consequently, for
(19) large stratifications at high rotation rates, the cells are
cylindrical at low latitudes and weak north-south rolls at
Here B is the second expression in equation (2c) that is and above midlatitude. But low m modes will probably
enclosed in brackets. The terms on the right side of dominate at high latitudes in a nonlinear model.
equation (19) are, from left to right, the pressure, the There is a simpler way of understanding why these
buoyancy, the stretching, the compressibility, and the cells bend at high latitudes in stratified mediums. If
viscous torques per moment of inertia. Note that every- equation (2c) were multiplied by p before taking its curl,
thing is dimensional in equation (19) and hp <0. The the (dimensional) rotation term would be
Taylor-Proudman theorem assumes steady-state, uni-
form density, and no viscosity; therefore, the vorticity
equation reduces to (2O»v)v=0. As a result, the cells (20-v)(pv)=2ßp|^+2ßv|^.
are cylindrical with axes parallel to the rotation axis.
For small stratifications at high rotation rates, our solu-
tions have small values of <o and hp, and the buoyancy If it is assumed that the other terms in the equation are
and viscous torques tend to cancel; so the velocity does negligible, the Taylor-Proudman theorem would apply
not vary significantly in the direction of Q. The axes of at low latitudes because the reference density p is fairly
the cylindrical cells parallel the rotation axis and inter- constant in the £ direction; so 3v/3zæO. But above
sect the outer boundary at ~ 40° latitude (see Fig. 6 for midlatitude (l/p)(3p/3z)^/ip, which is very negative
r=105, A^IO-2, 1). This is also illustrated in the near the top of the zone; so (l/vz)(3vz/3z) is very
constant radius (r=0.98 r0) plots in Figure 8 for Np = positive, which accounts for the bend in the cell axis.

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

No. 2, 1981 LINEAR ANELASnC SHELL MODEL 367


VELOCITY AT r»0.98 r0

90° 0°
Fig. 8.—Velocity profiles in constant radius surfaces (r=0.98r0) for the most unstable symmetric modes at 105, for Np = 10 -2, 1, 3,
5, and 7. The north pole is in the center of the figure, and the equator is the perimeter. Solid contours represent rising fluid, broken contours
sinking fluid.

The constant latitude cross sections of these cells density stratification (see the top row in Fig. 10). For
become tilted in longitude at high rotation rates. This this case, the relative pressure perturbation (p/p) in
can be understood by examining the thermodynamic equation (2a) is small compared to the relative density
perturbations. The temperature, pressure, and density (p/p) and temperature (0/0) perturbations; so the tem-
perturbations for the most unstable symmetric modes perature and density perturbations tend to be 180° out
are plotted in Figure 10 in the equatorial plane at of phase. Notice how rising and sinking fluid have
T=105 for the five density stratifications. First, con- higher and lower temperatures and lower and higher
sider almost incompressible fluid associated with a small densities, respectively. At low rotation rates (see Fig. 4

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

368 GLATZMAIER AND GILMAN

V T = I05
m=6
- N V
w
. . . x ^ ^ ^ ^gX^
' ' N ‘ X ^ >w v,
. > ^ N ^ N \>
x ^ ^ 'N. N. N^\
,y ' ' N >» :N»XN \>\
•.\W\\x\\% \v*X\
' ' ' , rr I X \\X \\
' ' ' ,
''s,'
" " ^ ^// Vx\x\\V\
x\> x\x\X
X\^x\x\x\s\
x\X ^
* V-. >> X
‘ x >•
\ x -* A;
y* v\
3
x -
T= I0 \'
m=5 v *'
‘ * Tl LÜ
■ §

RADIUS
Fîg. 9.—Velocity in constant longitude surfaces at longitude of maximum upflow. These are the most unstable symmetric modes for
AT = 10 at T—10^ on the left and r= 105 on the right.

of Gilman 1976) the horizontal flow is down the pres- Cuong attempts to explain this phenomenon by arguing
sure gradient in order to offset the horizontal viscous that the phase velocity (co/ra) increases with radial
force; as a result, the buoyancy driven radial flow is distance r. However, in his linear stability calculations
against the pressure gradient. But as the rotation rate (and in ours) both w and m are constants; w is an
increases, the pressure perturbations become larger and eigenvalue and /w is an input parameter. Consequently,
peak in the centers of the cells (see Fig. 10) in order to the phase velocity is a constant, for a given solution.
offset the Coriolis forces; however, notice how the radial Now consider highly compressible fluid associated
flow is still against the pressure gradient (more so near with a large density stratification (see the bottom rows
the bottom of the zone because of the 1 /r1 gravity), and of Fig. 10). Note that the vectors plotted in Figure 10
the horizontal flow is still down the pressure gradient. are mass flux (pv). The plots of the density perturbation
Also, the pressure contours are not uniformly spaced. clearly illustrate a reversal of the density perturbation
The gradient is larger in the regions where the buoyancy near the top of the zone. This density perturba-
force has a component in the direction of the Coriolis tion reversal has previously been observed in the
force, and smaller in the regions where the buoyancy plane-parallel, nonrotating, anelastic calculations by
force has a component in the direction of the negative Massaguer and Zahn (1980). As a result, the buoyancy
pressure gradient. As a result, for small stratifications force does negative work near the top of the zone in
and high rotation rates, pressure contours are tilted in regions where heavy fluid rises and light fluid sinks. We
the prograde direction. Since the buoyancy forces are have found that this effect is also present between cells
smaller in the upper part of the zone because of the when there are multiple cells in radius, so it is not totally
smaller gravity, the flow more closely parallels the pres- due to the temperature boundary condition, as sug-
sure contours there. Notice how the prograde tilt is gested by Massaguer and Zahn (1980). They argue that
evident for iVp = 10_2 and 1 in the constant latitude the vanishing 0 at the top boundary forces p to have the
plots of Figure 10 and the constant radius plots of same sign as p near the top because of equation (2a).
Figure 8. Also notice how the pressure gradients are small near
This prograde tilt has also been observed for a the top because, as a result of the density profile, the
Boussinesq fluid by Gilman (1976) and Cuong (1979). buoyancy forces there aid in offsetting the Coriolis

Fig. 10.
2
— Thermodynamic perturbations plotted in the equatorial plane for the most unstable symmetric modes at T=105 for
Np = 10 “ , 1, 3, 5, and 7. Temperature is on the left, pressure in the center, and density on the right. Arrows represent mass flux; solid and
broken contours represent positive and negative thermodynamic perturbations, respectively. Contour levels are arbitrarily chosen for the
different perturbations.

© American Astronomical Society • Provided by the NASA Astrophysics Data System


DENSITY
ROTATION
PRESSURE
ROTATION

Fig. 10
TEMPERATURE
ROTATION

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

370 GLATZMAIER AND GILMAN Vol. 45


forces. In the bulk of the zone, pressure contours are broader as Np increases because the difference in hori-
essentially in phase with the flow patterns, while the zontal dimension between consecutive values of m is less
phase of the temperature contours is slightly ahead, and at large wavenumbers. That is, the effects discussed in
the phase of the density contours is slightly behind. the previous paragraph are less sensitive to m when m is
Consequently, the majority of the rising or sinking fluid large.
is still hotter or cooler and less or more dense than the The type and magnitude of the phase propagation
reference state; but there are regions throughout the also varies significantly with density stratification. Equa-
zone where the local buoyancy work density (—pgw) is tion (2c) shows how a net force, composed of the
negative and the radial convective heat flux (JpBw) is pressure gradient, the buoyancy, the Coriolis, and the
inward. In addition, the regions where the buoyancy viscous forces, locally accelerates the fluid, but a more
force has a component in the direction of the Coriolis convenient equation for explaining phase propagation is
force, and therefore the regions of larger pressure gradi- the vorticity equation (19). An easy way to understand
ents, have shifted from their positions in the low stratifi- what the torques in this equation represent is to consider
cation case. Thus, the pressure contours and the flow a rotating, symmetric, equatorial mode with north-south
patterns in the equatorial plane are now slightly tilted in or cylindrical cells. Each cell is made up of fluid col-
the retrograde direction, as illustrated for Np=5 and 7 umns revolving about the axis of the cell. At low lati-
in the constant latitude plots of Figure 10 and the tudes the major component of the vorticity (VXv) is
constant radius plots of Figure 8. This tilt is much larger either parallel or antiparallel to 0, and it is locally
for the more stable, low wavenumber modes because the changing with time, in the rotating frame, as a result of
horizontal dimension is larger. the torques in equation (19). Since we have Fourier-
analyzed the perturbations, the frequency of the phase
d) Interpretation of Stability and Phase propagation is equal to the vorticity generation rate
Propagation divided by the vorticity:
As discussed in § V7>, the critical wavenumber in-
creases with T and Np. The tilting effect at large T is
responsible for the density stratification dependence of " = '(^fI)/(vxv). (20)
the critical wavenumber. Notice how at7T=105,m=6is
the critical wavenumber for Np —10 “2, whereas, for
Np—1, the critical wavenumber is m =18 (see Fig. 2). When only the real parts of the numerator and de-
The reason for this is that the very tilted flow patterns at nominator in equation (20) are considered, they must be
lower wavenumbers for Np =7, T= 105 are very ineffi- evaluated 90° out of phase. Since the peak vorticity
cient convective heat transport mechanisms. The smaller generally occurs in the centers of the cells, the torques
horizontal dimension at m = 18 results in less tilting and contribute more to a> if they peak 90° out of phase with
therefore is more efficient because rising and sinking the vorticity, i.e., in the regions of strong radial flow.
fluid moves nearly parallel to the buoyancy force opti- Profiles of the latitudinal vorticity and its generation
mizing the buoyancy work. For Np ^0, mcrit increases rate in the equatorial plane are illustrated on the left of
slightly as T increases because larger longitudinal pres- Figure 11 for a retrograde frequency and on the right of
sure gradients, which are proportional to m, are re- Figure 11 for a prograde frequency. Notice how the
quired to offset the Coriolis forces. For T=0, mcrit prograde (retrograde) sides of the vortices are becoming
increases slightly as Np increases because the majority of stronger for prograde (retrograde) propagation; i.e.,
the mass flux is distributed over a smaller radial dimen- negative (positive) vorticity is generated in the regions of
sion; so /crit increases, decreasing the horizontal dimen- upflow for prograde (retrograde) propagation.
sion of the most unstable mode and therefore keeping The pressure torque in equation (19) exists because
the effective aspect ratio close to unity which is most there is a pressure gradient perpendicular to the refer-
efficient for energy conversion. This has also been dis- ence state density gradient. This term is more important
cussed by Kato and Unno (1960). at large T because then the longitudinally pressure
The dependence of the stability curves on T and Np gradient (see Fig. 10) has negative (positive) peaks in
(Fig. 2) can also be explained based on previous argu- regions of upflow (downflow). Since the upper side of a
ments. The critical Rayleigh number, for given Np and rising fluid column has a smaller density because of the
m, increases with T because the Coriolis force tends to density stratification, the longitudinal pressure gradient
produce inefficient flow patterns, but the amount of tends to shear rising fluid columns, generating vorticity
increase is much smaller at higher wavenumbers because parallel to fí. Similarly, the longitudinal pressure gradi-
the Coriolis forces are less effective because of the larger ent tends to shear sinking fluid columns, generating
viscous forces. As a result, the stability curves for differ- vorticity antiparallel to £2. Therefore, since rising (sink-
ent values of T converge as m increases. In addition, the ing) fluid columns are on the retrograde (prograde) side
valley of the stability curve for a given T becomes of composite cells having vorticity parallel to fí, the

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

No. 2, 1981 LINEAR ANELASTIC SHELL MODEL 371

RETROGRADE PROPAGATION PROGRADE PROPAGATION


ROTATION ROTATION

Fig. 11.—Latitudinal components of 3vorticity (top row) and vorticity generation rate (bottom row) plotted in the equatorial plane for the
most unstable symmetric modes at 10 for = 10 ~2 on the left and = 103 on the right. Solid and broken contours represent vorticity
toward the north and south poles respectively; arrows represent velocity in the equatorial plane.

pressure torque, in the case of large T, contributes to The stretching torque exists because a velocity gradi-
retrograde phase propagation. At low rotation rates ent parallel to O changes the moment of inertia of the
(r<102) the longitudinal pressure gradient tends to fluid columns. As discussed previously, as T increases,
peak in the center of the cells, so in these cases the the meridional flow changes from periodic east-west
pressure torque contributes little to phase propagation. rolls at low rotation rates to equatorward (poleward)
The buoyancy torque exists because there is a density flow in rising (sinking) fluid at high rotation rates (only
gradient perpendicular to the gravitational field. Usually at low latitudes for large stratifications). Notice in the
rising (sinking) fluid columns have negative (positive) left-hand plot of Figure 9 that, for rising fluid, the
density perturbations (see Fig. 10), so the longitudinal gradient in the Ö direction of the velocity in the Ö
density gradient tends to have a positive (negative) peak direction is positive, whereas it is negative in the right-
in regions where the vorticity is parallel (antiparallel) to hand plot of Figure 9. Consequently, at low rotation
Í2. That is, the vorticity and its buoyancy generation rates rising fluid columns are stretched, and at high
rate tend to be in phase, so the buoyancy torque con- rotation rates sinking fluid columns are stretched. The
tributes little to phase propagation. However, for large stretching causes the fluid columns to contract in the
density stratifications and rotation rates, as discussed plane normal to the rotation axis which produces
previously, the density perturbations are almost in phase tangential Coriolis forces that torque the contracting
with the pressure perturbations, and therefore the longi- columns in the direction of Ö. Similarly, expanding fluid
tudinal density gradient (see the lower right of Fig. 10) columns are torqued in the direction antiparallel to O.
has negative (positive) peaks in regions of upflow Therefore, since rising (sinking) fluid columns are on the
(downflow). In these cases, a rising fluid column has a retrograde (prograde) side of composite cells having
lower density on its prograde side, so the longitudinally vorticity parallel to Ö, the stretching torque contributes
dependent, radial buoyancy force tends to shear rising to retrograde phase propagation at low rotation rates
fluid columns, generating vorticity antiparallel to ß. and, for small stratifications, contributes to prograde
Similarly, the radial buoyancy force tends to shear sink- phase propagation at high rotation rates. This high
ing fluid columns, generating vorticity parallel to £2. rotation stretching effect has been discussed by Hide
Therefore, the buoyancy torque, in the case of large T (1966) for an incompressible rotating fluid shell. How-
and Np, contributes to prograde phase propagation. ever, at high rotation rates for large density stratifica-

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

372 GLATZMAIER AND GILMAN Vol. 45


tions, the axis of the cell bends toward the pole at Cuong attempts to explain them based on a geometrical
midlatitude, so in these cases the stretching torque con- observation made by Hide (1966) for an incompressible
tributes to prograde phase propagation at low latitudes fluid. That is, if the fluid motion were mainly perpendic-
and to retrograde propagation at middle and higher ular to the axis of rotation, a fluid column parallel to the
latitudes. axis of rotation would lengthen as it moves away from
The compressibility torque exists because the radial the rotation axis if inside a cylinder tangent to the inner
density gradient causes rising (sinking) fluid columns to boundary and coaxial with the rotation axis; the oppo-
expand (compress) and therefore changes the moment of site is true if the column were outside this tangent
inertia of the columns. As a fluid column rises toward cylinder. As a result, the phase of cells inside the tangent
the upper surface, it expands in the plane normal to the cylinder would propagate in the retrograde sense and
rotation axis. The resulting Coriolis forces torque the that outside in the prograde sense. Cuong argues that
expanding column in the opposite direction of Í2, gener- retrograde frequencies exist at low rotation rates be-
ating vorticity antiparallel to Q. Similarly, vorticity cause the major part of a convective column is inside the
parallel to Ö is generated in sinking fluid columns. tangent cylinder. However, for the most unstable modes,
Consequently, the compressibility torque always con- it is evident from his plots (and ours) that most of the
tributes to prograde phase propagation. This effect is convection takes place outside this tangent cylinder. In
discussed in more detail in Paper III. addition, the less unstable polar modes do not assume
The viscous torque exists because there is diffusion of this column structure because the buoyancy forces would
vorticity; an additional term is also present as a result of be almost perpendicular to the convective flow.
the density stratification and is important for large For Np>\ the phase propagation for all but the very
stratifications near the top of the zone. The viscous high wavenumber modes is prograde (see Fig. 2). The
torque tends to weaken local vorticity peaks via diffu- stretching torque contributes to retrograde propagation
sion in the bulk of the zone and spins up the less dense for r<103 and to prograde propagation for T> 104,
fluid near the top of the zone; therefore, it is generally except at high latitudes for large stratifications, but it is
— 180° out of phase with vorticity in the bulk of the generally smaller than the other torques for Np>3. The
zone and in phase near the top. As a result, it contrib- prograde contribution of the compressibility torque in-
utes little to phase propagation and is cancelled mainly creases with rotation rate and density stratification. The
by the buoyancy torque for small stratifications and by pressure torque makes a retrograde contribution. As m
a combination of the buoyancy and pressure torques for increases, for small Np, the pressure torque manages to
large stratifications. decrease the prograde frequency and eventually makes
The relative importance of the torques on the right it retrograde. For large Np, the buoyancy torque, which
side of equation (19) varies with the rotation rate, den- makes a prograde contribution and is also proportional
sity stratification, wave number, and position. The to m, offsets the retrograde contribution made by the
stretching and compressibility torques are obviously pressure torque, so the frequency becomes nearly inde-
proportional to Tx/1. The pressure and compressibility pendent of m (see Fig. 2 and Paper III).
torques are inversely proportional to the local density
scale height. At high wavenumbers the longitudinal pres- e) Kinetic energy balance
sure and density gradients are larger than the latitudinal Large-scale kinetic energy density is locally time de-
gradients, so the pressure and buoyancy torques are pendent because of the phase propagation, but the aver-
approximately proportional to m. In addition, since the age in longitude is time independent for these linear
cell structure varies with latitude, so do the torques. stability solutions. The real part of the dot product of v*
For very little density stratification (Np «0), the pres- and the equation of motion (eq. [2c]) is the time and
sure and compressibility torques are generally small, and longitude independent linear anelastic kinetic energy
the buoyancy and viscous torques are, respectively, in equation (for constant v)\
phase and 180° out of phase with vorticity and tend to
cancel. Therefore, the stretching torque determines the
phase propagation, which is retrograde for T<102 and |¿(ípM2)=-Real(v**v/>)
generally prograde for T> 103 (see Fig. 2). The reason
for this, as discussed previously, is the change in the —gReal(pw*) + pií Real(v*#A+/ípv*-B),
meridional velocity profile as T increases. As the wave- (21)
number increases for large T, the pressure torque (which
is proportional to m) contributes more and more to where A and B are the expressions enclosed in brackets
retrograde propagation, so the frequencies in Figure 2 in equation (2c). Note everything is dimensional in
decrease and eventually become negative (retrograde). equation (21), and both sides of this equation vanish for
Retrograde frequencies have also been observed for a solutions on the stability boundary. The terms on the
Boussinesq fluid by Gilman (1975) and Cuong (1979). right side are, respectively, the average pressure,

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

No. 2, 1981 LINEAR ANELASTIC SHELL MODEL 373


buoyancy, and viscous work densities per time. The result in left-handed helices in the northern hemisphere
Coriolis force does no work. The average pressure work (see Fig. 7). Similarly, right-handed helices develop in
is positive near the bottom and the top of the zone the southern hemisphere. Consequently, the latitudinal
because the fluid flows down the pressure gradient, and contribution to helicity [t>(V X v)^] tends to be negative
negative in the middle of the zone because there the in the northern hemisphere and positive in the southern
fluid flows against the pressure gradient. In addition, the hemisphere. This is the major contribution to helicity
pressure perturbation does work on expanding and con- because of the large latitudinal component of vorticity.
tracting fluid. The buoyancy force, for the most part, In addition, the spill-over effect produces left-handed
converts gravitational potential energy into large-scale radial helices between the latitudinal helices in the
kinetic energy, so the buoyancy work is mainly positive. northern hemisphere. Consequently, the radial contribu-
That is, in general, less dense fluid rises, while denser tion to helicity [w(V X v)r] also tends to be negative in
fluid sinks. However, as discussed previously, for Np >0 the northern and positive in the southern hemisphere.
there is negative work done by buoyancy near the top of The longitudinal contribution to helicity [w(V X v)x] is
the zone. The viscous force generally converts large-scale the smallest contribution and is, on the average, negative
kinetic energy into small-scale turbulent energy, so the near the bottom of the zone and positive near the top.
viscous work is negative. All work densities peak in the However, the total helicity for the most unstable modes
equatorial plane, for the most unstable symmetric modes, is negative everywhere in the northern hemisphere and
because of the large perturbations there. positive in the southern hemisphere for low rotation
The work densities, averaged in longitude and lati- rates.
tude, are plotted against radius in Figure 12 for the most At high rotation rates and small stratifications the
unstable modes at T=0 (on the left) and T—105 (on the helices intersect the outer boundary. Fluid spiraling
right) for the five density stratifications. They are nor- toward the equatorial plane also drifts away from the
malized so the average viscous work density over the rotation axis because of the surface curvature. In addi-
entire zone is —1. Notice how the negative buoyancy tion, since the surface curvature forces rising (sinking)
work near the top increases with density stratification fluid toward the equator (pole), there exists a compo-
and is offset by large positive pressure work. These plots nent of vorticity perpendicular to the rotation axis in the
also illustrate how the convective activity moves toward direction of the fluid drift. Consequently, positive helic-
the bottom of the zone as T increases for small stratifi- ity exists near the top boundary. As illustrated on the
cations and toward the top as m increases for large left in Figure 13, helicity, averaged in longitude, for the
stratifications. most unstable modes is negative in the northern hemi-
The total pressure work done in the convection zone, sphere except near the top surface at low latitudes; the
opposite is true in the southern hemisphere.
Stix (1976) pointed out that the helicity profiles for
—Real j (y**Vp)dr= —Real jhppw*dr, (22)
Yoshimura’s (1972) model of a rotating Boussinesq
sphere contradicted the helicity profiles (unpublished)
vanishes in the incompressible case (hp=0) but is posi- for Gilman’s (1972) model of a rotating, Boussinesq
tive and comparable to, although less than, the total annulus. Gilman’s results are very similar to ours for the
buoyancy work when Np is large. This is also discussed most unstable modes with small density stratification,
by Massaguer and Zahn (1980). Physically, a rising while Yoshimura found negative (positive) helicity in the
(sinking) fluid parcel in the upper part of the zone, upper (lower) part of the zone in the northern hemi-
where the expansion (compression) rate is greatest, gains sphere. Yoshimura’s model applies only for slow rota-
expansion (compression) kinetic energy while experienc- tion and for a convection zone depth much smaller than
ing a positive (negative) pressure perturbation. the horizontal scale and the radius of the sphere. We
find average helicity profiles at low rotation rates and
very small wavenumbers that are similar to those of
/) Helicity Yoshimura, but these are not the most unstable modes.
Helicity (v V X v) is of considerable interest because it On the other hand, both types of average helicity profiles
is the basis of “cyclonic turbulence” which is a key have appeared in nonlinear Boussinesq calculations for
factor in mean field solar dynamo theories (see, e.g., Stix a rotating spherical shell (Gilman 1977; see his Figure
1976; Moffatt 1978). Physically, helical fluid motion 23); the type that appears depends on the specified
twists toroidal magnetic field lines producing poloidal Rayleigh number.
magnetic fields. Our model cannot resolve small scale At high rotation rates and large stratifications, both
turbulence but does provide profiles of helicity due to the velocity and the vorticity peak near the top boundary,
large scale convective motions. so helicity also peaks there. The north-south cell struc-
As discussed previously, at low rotation rates the ture, opposed to the cylindrical cell structure, at midlati-
combination of spherical geometry and Coriolis forces tude fails to produce positive helicity near the top

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

T=0 T=I05
3
2
1
lyicr2 o
-I
-2
-3
3
2
1

VI 0

-2
-3
3
2
1

V3 0

-I
-2
-3
3
2
1

V5 0

-I
-2
-3
3
2

V7 0

-I
-2
-3
0.60 0.65 0.70 0.75 0.80 0.85 0.90 0.95 1.00 0.65 0.70 0.75 0.80 0.85 0.90 0.95 1.00
r/r0 r/r0
Fig. 12.—Average buoyancy (B), pressure (P), and viscous (V) work densities per time plotted vs. radius for the most unstable
symmetric modes. 7=0 on the left and 105 on the right, for Np = 10 -2, 1, 3, 5, and 7. The work densities are scaled so the average viscous
work density (over the entire zone) is — 1.

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

LINEAR ANELASTIC SHELL MODEL 375


HELICITY AVERAGED IN LONGITUDE

RADIUS
Fig. 13.—Helicity in the northern hemisphere averaged in longitude and plotted in radius and latitude for the most unstable symmetric
modes at r=105 and for A^ = 10-2, 1, 3, 5, and 7. Broken contours represent negative (left-handed) helicity; solid contours represent
positive (right-handed) helicity.

boundary. Consequently, the longitudinally averaged tensor (—V(pwv)) is instrumental in the determina-
helicity is negative in the northern hemisphere and tion of the initial tendency of the mean longitudinal
positive in the southern hemisphere and peaks near the momentum density [p(3(«)/3/)]asa function of radius
top boundary for the most unstable, symmetric modes and latitude. In this paper we examine the momentum
with large stratifications, as illustrated on the right in flux in radius and latitude for different stratifications in
Figure 13. But for equatorial modes with m<mcrit there order to understand how the most unstable modes might
is more latitudinal structure, so larger positive longitudi- produce differential rotation in a time-dependent non-
nal contributions to helicity exist near the top of the linear model.
zone. As a result, average helicity is positive (negative) First, consider the radial momentum flux (puw) for
everywhere in the northern (southern) hemisphere—just the most unstable symmetric modes. For small density
the opposite of the profiles for the most unstable modes. stratifications at low rotation rates the flow from
It is also interesting to note that, for large stratifications, the counterclockwise north-south rolls spills over to
double cells in radius have positive (negative) average the adjacent clockwise rolls at low latitudes. As a result,
helicity in the upper (lower) part of the zone in the the radial momentum flux directed inward, i.e., pww<0,
northern hemisphere; but again these are not the most at low latitudes is much stronger than that directed
unstable modes. outward, i.e., puw>0; (see the velocity vector plots of
Fig. 11). At high latitudes the flow spills over from the
g) Momentum Flux clockwise to the counterclockwise rolls; thus, the radial
Differential rotation is of major interest because it is momentum flux tends to be directed outward, but has a
the only measurable large scale global motion that can much smaller magnitude because the velocities are small
be used to calibrate giant cell models at present. Dop- compared to their magnitudes at low latitudes. At high
pler observations (see, e.g., Howard and Harvey, 1970) rotation rates, for small stratifications, the prograde tilt
provide a measure of the surface differential rotation in associated with the cylindrical cells produces a stronger
latitude, while the k—co diagrams of observed solar outward radial momentum flux (see the mass flux vector
/?-mode oscillations can potentially provide a measure of plots of Fig. 10 for Np — \0~2 and 1). Gilman (1975)
the radial differential rotation (see Deubner, Ulrich, and found this same effect for the Boussinesq case. But when
Rhodes 1979). In addition, differential rotation is a key the stratification is large, the radial momentum flux is
factor in mean field solar dynamo theories; differential mainly directed inward at low latitudes for all T because
rotation shears poloidal magnetic field lines, producing of the spill-over effect and the retrograde tilt (see Fig. 10
toroidal magnetic fields. The second-order calculations for Np=5 and 7). Therefore, the average in longitude of
in a future paper will show how the longitudinal compo- the radial momentum flux, except for small stratifica-
nent of the convergence of the mean momentum flux tions at high rotation rates, is directed inward at low

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

376 GLATZMAIER AND GILMAN Vol. 45

Fig. 14.—Latitudinal momentum flux averaged in longitude (top row) and radial momentum flux averaged in longitude (bottom row)
plotted in radius and latitude for the most unstable symmetric modes at T= 105 and Np = 10 “2, 1, 3, 5, and 7. Broken contours represent
latitudinal flux toward the equator (top row) and inward radial flux (bottom row). For a given Np, the values of the contour levels are the
same for the latitudinal and radial fluxes.

latitudes. The mean radial momentum flux profiles for the most unstable symmetric modes. At low rotation
the most unstable modes at T= 105 for the five stratifi- rates stronger latitudinal momentum flux near the top
cations are displayed in the lower part of Figure 14. For (bottom) of the zone is directed toward the equator
large stratifications the inward mean momentum flux (pole) because the fluid in the counterclockwise (clock-
increases with radius in the lower part of the zone and wise) rolls spirals toward the equator (pole). At high
decreases with radius in the upper part; that is, the rotation rates for small stratifications the latitudinal
mean radial flux of longitudinal momentum converges momentum flux near the bottom of the zone is also
in the lower part and diverges in the upper part of the directed toward the equator at low latitudes because of
zone. Based on just the radial flux, this suggests that the prograde tilt and the meridional flow (see the upper
angular velocity in a nonlinear model will decrease with left of Fig. 14), and the flux peaks along a line parallel
radius. The radial momentum flux profile for small to the axis of the cylindrical cells. The constant radius
stratifications with T> 104 suggests just the opposite; plots of Figure 8 illustrate the equatorial transport of
this was also observed by Gilman (1975) for linear longitudinal momentum, i.e., puv<0. Again the same
Boussinesq calculations. effect was found in Gilman’s Boussinesq calculations
However, the angular velocity profile also depends on (1975). However, the flux profiles for large stratifica-
the latitudinal momentum flux (pwu); again consider tions at high rotations rates (see upper right of Fig. 14)

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

No. 2, 1981 LINEAR ANELASTIC SHELL MODEL 377


resemble those at low rotation rates because of the perturbation convective heat flux. Second-order calcula-
north-south roll structure above midlatitude. Note for tions will show how the convergence of the mean
the most unstable symmetric modes that the changes in convective enthalpy flux (—V(c/>p0v)) is instrumen-
the radial and latitudinal momentum flux profiles as Np tal in the determination of the initial tendency of the
increases for 7= 105 (Fig. 14) are similar to the respec- mean enthalpy density perturbation [p0(d(s )/dt)] as a
tive changes as T decreases for Np =0, shown in Gilman function of radius and latitude. Therefore, in order to
(1976). In the upper part of the zone, the equatorward understand how mean entropy and temperature
mean momentum flux increases with latitude at low perturbation gradients might develop in a time-
latitudes and decreases with latitude at higher latitudes; dependent nonlinear model, we examine in this section
that is, the mean equatorward flux of longitudinal the convective heat flux, i.e., convective enthalpy flux, in
momentum converges at low latitudes and diverges at radius and latitude.
high latitudes. Based on just the latitudinal flux, this First, consider the radial heat flux (pOw) for the most
suggests that angular velocity in a nonlinear model will unstable symmetric modes. Since hotter fluid tends to
decrease with latitude in the upper part of the zone. rise while cooler fluid sinks, the radial heat flux is
The differential rotation that develops in a nonlinear mainly outward, i.e., p0w>O, although at high rotation
model depends on the sum of the convergences of the rates small regions of inward radial heat flux exist
radial and latitudinal momentum fluxes as well as the because of the tilt in the cells and the shift in the
influence of axisymmetric meridional circulation (see temperature profile (see Fig. 10). However, the mean
Gilman 1977, 1978a, b). For small stratifications at high convective radial heat flux is outward and increases with
rotation rates both convergences are positive in the radius in the lower part of the zone and decreases with
upper part of the zone at low latitudes, and the radial radius in the upper part; that is, the mean radial flux
momentum flux peaks at the equator; therefore, equa- diverges in the lower part and converges in the upper
torial acceleration, as observed on the Sun, is predicted part of the zone. This suggests that mean radial
for Rayleigh numbers close to the critical value. It is not perturbation diffusive heat flux in a nonlinear model
obvious from these momentum flux profiles whether the will be inward and therefore will tend to cancel the
angular velocity will increase with radius because, al- outward mean radial convective heat flux. Another way
though the radial flux profile suggests it will, the equa- of looking at this is that the mean entropy gradient will
torward latitudinal flux increases with depth in the tend to become less negative in the bulk of the zone, so
upper part of the zone. In Gilman’s (1977) nonlinear the mean temperature gradient will become even closer
Boussinesq calculations, angular velocity decreases with to an adiabatic gradient.
latitude and depth for small Rayleigh numbers, while Now consider the latitudinal heat flux (pOv) for the
just the opposite occurs for large Rayleigh numbers. It is most unstable symmetric modes. Because of the periodic
known that mean field kinematic dynamo models re- east-west roll structure for small stratifications at low
quire an angular velocity that increases with depth to rotation rates, hot rising (cool sinking) fluid in the lower
maintain a solar-like magnetic cycle when helicity is part of the zone has a component of velocity toward the
negative (positive) in the northern (southern) hemi- equator (pole), while in the upper part it has a compo-
sphere (see, e.g., Stix 1976; Moffatt 1978). However, nent toward the pole (equator); therefore, the mean
Gilman (1977, 1978 a) never found a case in his nonlin- latitudinal heat flux is toward the equator in the lower
ear Boussinesq calculations for which angular velocity part of the zone and toward the pole in the upper part.
increases with depth but decreases with latitude. For At higher rotation rates the average latitudinal heat flux
large stratifications at high rotation rates, the conver- is toward the equator everywhere and peaks near the
gence of the mean latitudinal flux is positive in the middle of the zone because the curved outer boundary
upper part of the zone at low latitudes, while the conver- forces rising (sinking) fluid toward the equator (pole).
gence of the mean radial flux is negative; the magni- The mean latitudinal heat flux profiles for large stratifi-
tudes of these fluxes are comparable (see Fig. 14, right). cations are similar to those for small stratifications, but
Also it is not clear from these first-order solutions how the peak in the upper part of the zone is much stronger
the Coriolis force acting on the induced meridional because the velocity and temperature perturbations peak
circulation will contribute to the induced differential near the top of the zone. Also the switch-over from
rotation/ Therefore, a second-order calculation is re- convective heat transport away from the equator to
quired to determine the mean differential rotation at the transport toward the equator occurs at a higher rotation
onset of convection in a compressible atmosphere. rate in the compressible case. The convergence of the
mean latitudinal heat flux suggests that in a nonlinear
h) Heat Flux model the mean latitudinal perturbation diffusive heat
In a nonlinear anelastic model the mean perturbation flux will be poleward (equatorward) in the lower (upper)
diffusive heat flux, which is driven by the gradient of the part of the zone for low rotation rates and poleward
mean entropy perturbation, tends to balance the mean everywhere at high rotation rates. That is, the mean

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

378 GLATZMAIER AND GILMAN Vol. 45


entropy and temperature at the surface will tend to be order solutions predict a mean temperature that is larger
smaller (larger) at the equator for low (high) rotation near the equator. However, certain more stable modes
rates. for large stratifications have radial convective heat flux
Again it is the sum of the convergences that de- profiles that are more uniformly distributed in latitude.
termine the mean entropy profile; the pressure, If these become important in a nonlinear calculation, a
buoyancy, and viscous work densities also help to main- more uniform mean surface temperature may result as
tain it. For the most unstable modes, the radial heat flux observed on the Sun.
is ~ 3 times larger than the latitudinal heat flux for low
stratifications and — 10 times larger for large stratifica- We wish to thank David H. Hathaway for reviewing
tions. Consequently, at high rotation rates, these first- the manuscript.

APPENDIX
We use the following definition for the spherical harmonics:

1/2 (l—sin2<jp)|m|/2 d|m|+/(sin2<p—l/


2/+1 (/~M)! „irnX
y/m(9,\)=<T \m\+l (Al)
Att (/+|m|)! 2ll\ d(sm(p)\

where <r=(— l)w for m>0 and 1 for m<0. Recall that <p is latitude, \ is longitude. We also define the inverse scale
heights

j 1 dp _ 1 dpv _ 1 dpK
hp hp hk (A2)
~pdr’ -pv^-’ ~Jt~dir9

and the coefficient

(/+H)(/-|m|) 1/2
(A3)
(2/-l)(2/+l)

The radial dependence of the perturbation equations reduce to the following set of nondimensional, coupled,
first-order, complex differential equations.

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

No. 2,1981 LINEAR ANELASTIC SHELL MODEL 379


2(3-/(/+l)) ,/wrV2
+
* /(/+1) p\Xi

¡2yFPR(l+ßf | 2/., /(/+!) , _ yFPR (l+ß\:


(y— l)6r4 r2 r2

Í yFPR ( \±1\2(1K , 4 \ | 2KK | 4^


(y-l)tf V i" > \ r2 ' r3 r2 r3

l(l+l)yFPR(l+ß)
(1+^) / _4 \
+ (A9)
(y~l)Ô >4 3Aí>)

/('+!) dh„ dh 2hp 6


+ + -r+hl-2hllhfl
dr dr

i /(/+!) \2 | ¿/(/+l)to | /wrl/2 / yFPR(1+)8)2 \


+ 2 2 2
\ r / ^r P ¿^r [ (y—l)ör |

+ ii^L-iiÎÂ(IM)V4p + î)(4( + i)J,

j-1/2
yFPR I l+ß )2 , , , 2-/(/-l) (/— 1) 1
+ Ä„ + -
(y-l)<9 (/+ l)r WTV) AZ'-
a

p/2 yFPR i 1 +ß \2 1+3 a


/+l.'n I -7m
+hn +
(y-l)^" l r I /+i r,+1

r'/2 (/-iK,
ÿ /(/+i)

f/(/+i)(^+i)2 ,WjFpr _/ .zj..,...,/. 4 \


{ ÏW + —[2"(*. + 7) + '''('+l)('',-3''.)

+ ■W,"
""7"/2r(1 + WÏ) ('■»+ 7 ) ) - ^ +
f)

[ iuFP imTl/,2r2 iur2


?(2-/(/+l)) + DWr (A10)
l * 1(7+1) P~

American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

380 GLATZMAIER AND GILMAN Vol. 45


7
iœFP Pvr2 I 4\1 Í iuFPvr2 )

iuFPT'/2r2 (/-l)û/,, iuFPT'/2r2 a/+


/+i1, m
/(/+1) W-l /+1 -/+1>

_ \ r/2ial+u Í rl/2«/+2,.
DWr+ Z)^,1+2
dr v(l+2)(l+\)] { v (1+2)

yJ/2
+■
/+1 lai+\,K
■ W”
(/+2 j/lp (/+2)/- + /! >+
/+2\ ' r) (/+2)(/+l)
(All)
[l+2\ (/+3)(/+2) 1 4
l + 2,m
w;1+2
1/+ f)'^ (/+ l)r 1+2

imT]/1 /(/+3) HQ
[h. + ^DZ^.
?(/+l)(/+2) r2 ï'P W+l

Note that equation (A9) is the radial component of the curl of the curl of the equation of motion, equation (A10) is the
thermodynamic equation, and equation (Al 1) is the radial component of the curl of the equation of motion.

REFERENCES
Baker, N. 1963, The Depth of the Outer Convection Zone in Hide, R. 1966, Phil. Trans. Roy. Soc. London, A, 259, 615.
Main-Sequence Stars (New York: NASA Institute for Space Howard, R., and Harvey, J. 1970, Solar Phys., 12, 23.
Studies). Jarvis, G. T., and McKenzie, D. P. 1980, J. Fluid Mech., 96, 515.
Baker, N., and Temesvary, S. 1966, Tables of Convective Stellar Jeffreys, H. 1930, Proc. Cam. Phil. Soc., 26, 170.
Envelope Models New York: NASA Institute for Space Studies. Kato, S., and Unno, W. 1960, Pub. Astr. Soc. Japan, 12, 427.
Belvedere, G., and Paterno, L. 1977, Solar Phys., 54, 289. Lamb, H. 1910, Proc. Roy. Soc. London, /l, 34, 551.
Bohm, K. H. 1963, Ap. J., 137, 881. Massaguer, J. M., and Zahn, J. P. 1980, Astr. /Ip., 87, 315.
Bohm, K. H., and Richter, E. 1960, Zs. f Ap., 50, 79. Moffatt, H. K. 1978, Magnetic Field Generation in Electrically
Busse, F. H. 1970, J. Fluid Mech., 44, 441. Conducting Fluids (Cambridge: Cambridge University Press).
Busse, F. H., and Cuong, P. G. 1977, Geophys. Ap. Fluid Dyn., 8, Proudman, J. 1916, Proc. Roy. Soc. London, A, 92, 408.
17. Rhodes, E. J., Ulrich, R. K.,' and Simon, G. W. 1977, Ap. J., 218,
Chandrasekhar, S. 1961, Hydrodynamic and Hydromagnetic Stabil- 901.
ity (London): Oxford University Press). Roberts, P. H. 1965, Ap. J., 141, 240.
Cuong, P. G. 1979, Ph.D. thesis, University of California, Los . 1968, Phil. Trans. Roy. Soc. London, A, 263, 93.
Angeles. Roberts, P. H., and Stix, M. Í972, Ast. Ap., 18, 453.
Deubner, E. L., Ulrich, R. K., and Rhodes, E. J. 1979, Ast. Ap., 72, Skumanich, A. 1955, Ap. J., 121, 408.
177. Soward, A. M. 1977, Geophys. Ap. Fluid Dyn., 9, 19.
Dumey, B. 1968, J. Atmos. Sei., 25, 771. Spiegel, E. A. 1964, Ap.J., 139, 959.
. 1970, Ap.J., 161, 1115. . 1965, Ap. J., 141, 1068.
Gilman, P. A. 1972, Solar Phys., 27, 3. Spiegel, E. A., and Unno, W. 1962, Pub. Astr. Soc. Japan, 14, 28.
. 1975, y. Atmos. Sei., 32, 1331. Spiegel, E. A., and Veronis, G. 1960, Ap. J. 131, 442.
. 1976, in ÍAU Symposium 71, Basic Mechanisms of Solar Stix, M. 1976, in IAU Symposium 71, Basic Mechanics of Solar
Activity, ed. V. Bumba and J. Kleczek (Dordrecht: Reidel), p. Activity, ed. V. Bumba and J. Kleczek (Dordrecht: Reidel), p.
207. 367.
. 1977, Geophys. Ap. Fluid Dyn., 8, 93. Taylor, G. I. 1921, Proc. Roy. Soc. London A, 100, 114.
. \91Sa, Geophys. Ap. Fluid Dyn., 11, 157. Toomre, J., Zahn, J. P., Latour, J., and Spiegel, E. A. 1976, Ap. J.,
. 1978/>, Geophys. Ap. Fluid Dyn., 11, 181. 207, 545.
. 1979, Ap. J., 231, 284. Unno, W. 1957, Ap.J., 126, 259.
Gilman, P. A., and Glatzmaier, G. A. 1981, 4/?. J. Suppl., 45, 335 Unno, W., Kato, S., and Makita, M. 1960, Pub. Astr. Soc. Japan,
(Paper I). 12, 192.
Glatzmaier, G. A., and Gilman, P. A. 1981, 4/?. J. Suppl., 45, 381 Vandakurov, Y. V. 1975¿z, Solar Phys., 40, 3.
(Paper III). . 1975/7, Solar Phys., 45, 501.
Gough, D. O., Moore, D. R., Spiegel, E. A., and Weiss, N. O. 1976, Weir, A. D. 1974, Ph.D. thesis, University of Cambridge.
Ap.J., 206, 536. Weymann, R., and Sears, R. L. 1965, Ap. J., 142, 174.
Gough, D. O., and Weiss, N. O. 1976, M. N. R. A. S., 176, 589. Yoshimura, H. 1971, Solar Phys., 18, 417.
Graham, E. 1975, J. Fluid Mech., 70, 689. . 1972, Ap. J., 178, 863.
Graham, E., and Moore, D.R. 1978, M. N. R. A. S. 183, 617. Yoshimura, H., and Kato, S. 1971, Pub. Astr. Soc. Japan, 23, 57.
Heard, W. B. 1973, Ap. J., 186, 1065.

Peter A. Gilman and Gary A. Glatzmaier: High Altitude Observatory, National Center for Atmospheric Research,
P. O. Box 3000, Boulder, CO 80307

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

The Astrophysical Journal Supplement Series, 45:381-388, 1981 February


© 1981. The American Astronomical Society. All rights reserved. Printed in U.S.A.

COMPRESSIBLE CONVECTION IN A ROTATING SPHERICAL SHELL.


III. ANALYTIC MODEL FOR COMPRESSIBLE VORTICITY WAVES

Gary A. Glatzmaier and Peter A. Gilman


High Altitude Observatory, National Center for Atmospheric Research1
Received 1980 April 17; accepted 1980 July 23

ABSTRACT
A simple analog to compressible convection in a rotating spherical shell is described and solved
analytically to enable us to understand the basic physics of the prograde phase velocities that resulted
from the numerical calculations of Paper II. The analog is for an inviscid, adiabatically stratified,
rotating, equatorial annulus of gas for which a form of potential vorticity is conserved. Linear
perturbations in this system take the form of vorticity waves which propagate prograde relative to the
rotating reference frame as long as the density decreases outward. Their frequencies depend on the
longitudinal wavenumber and density stratification in the same way as the convective modes of
Paper II. Simple physical arguments explain these dependences.
Subject headings: convection— Sun: interior— Sun: rotation

I. introduction little density stratification, i.e., depth of zone is much


Stability solutions to the linear anelastic fluid equa- less than the reference state density scale height; he also
tions for a rotating, compressible spherical shell are concludes that all waves propagate in the prograde
discussed in Glatzmaier and Gilman (1981), hereafter sense. Dumey and Skumanich (1968) study nonradial
referred to as Paper II. Paper II illustrates how cellular oscillations of an inviscid, polytropic sphere in the limit
patterns propagate in the direction of rotation and how of slow rotation. They calculate what the polytropic
the corresponding prograde frequencies depend on the index must be for marginal stability and also find pro-
Taylor number 7, the longitudinal wavenumber m, and grade frequencies.
the number of density e-folds Np across the convection
zone. II. THE MODEL
In this paper we develop an extremely simple model
for which a form of the potential vorticity theorem a) Assumptions
holds. The resulting compressible vorticity waves are
The most unstable modes of a rotating spherical shell
then used to explain the general properties of the pro-
manifest themselves as north-south rolls which peak in
grade frequencies found for the rotating spherical shell
the equatorial plane. This cellular structure is described
model (Paper II). Hide (1966) has studied Rossby waves
by Gilman (1975) for a Boussinesq fluid shell of sub-
in rotating spherical shells with fluid columns parallel to
stantial depth, and in Paper II for a compressible fluid.
the axis of rotation. Prograde vorticity or Rossby waves
We adopt the simplified geometry depicted in Figure
exist in his model because the surface curvature forces
la, a thin annulus concentric with the axis of rotation,
the columns outside the cylinder tangent to the inner
which is meant to simulate the equatorial latitudes of a
sphere equator to shrink as they move away from the
spherical shell. The radius of the annulus is assumed
rotation axis and stretch as they move toward it. How-
large enough compared to the depth d that we may
ever, Hide has assumed a homogeneous, incompressible
employ a local Cartesian coordinate system. Longitude
fluid, which is unrealistic for stellar models. We also
now increases in the x- direction. The rotational
consider fluid columns parallel to the rotation axis, but
frequency ß and the gravitational field g are assumed
in a highly stratified, compressible fluid. Vorticity waves
constant and in the y- and negative ¿-directions, respec-
exist in our model because the fluid columns expand as
tively. The problem is assumed two-dimensional in the
they move away from the rotation axis into lower ambi-
xz-plane; the bottom of the zone is at z=0, the top at
ent fluid density and compress as they move toward the
z=d. As in Paper II, the reference state temperature and
axis and higher density. Gibbons (1980) considers a
density are assumed functions of radius only (z in this
similar compressible problem but in the limit of very
case), the ratio of specific heats is taken to be 5/3, and
impermeable bounding surfaces are assumed. However,
1
Sponsored by the National Science Foundation. for this model we neglect the diffusion of both momen-
381

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

382 GLATZMAIER AND GILMAN Vol. 45

a
î y
^ A

/i.'- —
\
H-z H(z)
f

HOI
d
(a) (b)
Fig. 1.—Rotating annulus geometry for (û) ( dH/dz)=0 and ( Z?) (dH/dz)=/=0

tum and temperature. Convection and gravity waves are Therefore, as z0 oo, Np-*0 (the Boussinesq case), and
filtered out of this problem by assuming that both the as z0->d, Np^> oo.
reference state and the perturbations are adiabatic.
Acoustic waves have been filtered out via the anelastic
approximation. Thus, the only time dependence that can c) Linear Anelastic Perturbation Equations
arise is from rotational waves. With the above assumptions, the anelastic perturba-
tion equations of state, mass, energy, and momentum
b) Reference State Equations become:
With the above assumptions the reference state is
Z _ £ i il (4a)
described by
P- P" 0 ’
p=R*pO, (la) V(pv)=0, (4b)

ÊL = . (lb) £=5/3£, (4c)


dz -Pg,
P P
de A. 9v
(le) = - V p-pgz+2Qp\Xÿ. (4d)
dz Cp
Notice that although both the reference state and the
where we are using the same notation as in Paper IL The perturbations are adiabatic, thermodynamic perturba-
solution of these equations is a polytrope: tions do exist. The buoyancy term in the momentum
equation is usually omitted in the development of
(2a) Rossby waves, but here it is absolutely necessary be-
Cp
cause the curl of the Coriolis term vanishes in the
equatorial plane.
p=, pt>) The vorticity equation results from taking the curl of
(zo-d) the momentum equation and making the usual substitu-
tions with equations (2) and (4):
2gPo
P= e/2, (2c)
5(z0-dŸ/2
f=2ÛVv. (5)
where, for convenience, we have defined £=z0 —with
z0 being a constant (larger than the depth d) which Here f=(VXv)^ = 9t//9z —9w/9x, hp =d\np/dz, w
determines the density stratification. Notice that £ is is the vertical (z-directed) velocity, and u is the horizon-
proportional to the local density scale height. The num- tal (x-directed) velocity. From equation (5) we can show
ber of density e-folds Np across the zone is that within the linear approximation,

(3) (6)

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

No. 2,1981 VORTICITY WAVES IN A SPHERICAL SHELL 383


Therefore, the potential vorticity (f+2ß)/p is con- applying the boundary conditions, we get
served following a fluid particle.
It can also be shown that if the curvature of the outer
surface, as depicted in Figure \b, were taken into
account, equations (5) and (6) would have the following
approximate forms:

^=2û(Ap+fc„)W) (7a)
(9)

A/¿±2fí\ =0, (7b) The phase velocity u/m is positive and therefore pro-
dt\ Hp J grade, unless z0-»oo, or m=0, or m-»oo, in which
cases the frequency vanishes. To get retrograde waves
where hH = d\n.H/dz and H(z) is the height of a fluid with our assumptions requires changing the sign of both
column. These equations illustrate the additive effects of gravity and the radial density gradient. Of course, this
compressibility and stretching in the generation of nega- would have no stellar application but might be realized
tive (positive) vorticity in rising (sinking) fluid columns. in a laboratory experiment.
Simple geometrical arguments show that hp^hH for
Vp æ 1 ; hence, for more than one density e-fold, the
d) Solutions
density stratification should be more important than the
surface curvature in the production of vorticity waves. We have chosen the method of Frobenius (see, e.g.,
We ignore this surface curvature effect here. Compari- Hildebrand 1949) to solve equation (8), so no numerical
sons with the results of Paper II should therefore be difference scheme is required. The input parameters are
better when Np>\. z0 and m. The general solution for the vertical velocity
As in Paper II, we Fourier-analyze in longitude by amplitude is a linear combination of two-power series in
assuming the perturbations are proportional to el(kx~03t\ f, beginning with powers of 1 and —3/2, respectively.
where k is the wavenumber and is the frequency. In The boundary conditions determine the transcendental
order to compare this simple model to the spherical shell equation that is satisfied by various (real) eigenfrequen-
model we devise a relationship between the plane cies a). The largest eigenfrequency always results from a
wavenumber k and the longitudinal wavenumber m. Let single cell reaching from the lower to the upper
k equal m divided by the radius of the midpoint of the boundary. The smaller eigenfrequencies correspond to
spherical shell. We assume, as we did in Paper II, a multiple cells. The infinite summations are truncated
depth of 40% of the stellar radius; hence, k=m/2d. The when relative contributions made by additional terms
phase velocity, in radians per second relative to the become less than 10 ~6.
rotating frame, is \p —^/m. The eigenfrequency (scaled by 12) for single cells is
Substitution of equations (2b) and (4b) into (5) then plotted versus wavenumber for 10 different density
leads to the following second-order differential equation stratifications in Figure 2a. The corresponding phase
in terms of the (real) vertical velocity amplitude w(z), velocities are plotted in Figure 2b. Figure 3 shows some
typical plots of velocity, mass flux, vorticity, vorticity
d2w 1 dw m m generation rate, and perturbation pressure, density, and
+ w=0, (8) temperature in the xz-plane for different wavenumbers
de £2
and density stratifications. The phase of the vorticity
generation rate is 90° ahead of the phase of the vorticity
where we have employed the following scaling:
and thus illustrates the prograde (+x-directed) phase
propagation. The thermodynamic perturbations are
w—»(Q)co,
adiabatic; thus, they have the same sign and same
longitudinal dependence; but, like the reference state
thermodynamic variables, they have different radial de-
pendences. Notice that the flow is almost geostrophic in
z0^(d)z0.
that it nearly parallels the pressure contours with coun-
terclockwise circulation around the lows and clockwise
The boundary conditions require w to vanish at £=z0 around the highs. But the balance between pressure
and £=z0 —1. gradient and Coriolis forces implied by this flow pattern
Without solving equation (8) in detail we can demon- is not exact; the imbalance among these forces and
strate that all phase velocities are greater than or equal buoyancy (see eq. [4d]) results in the inertial acceleration
zero, and therefore all waves are prograde. Multiplying needed for prograde phase propagation. Notice also that
equation (8) by w, integrating from £=z0 to z0 — 1, and as a result of conservation of mass, the velocity becomes

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

384 GLATZMAIER AND GILMAN

Fig. 2.—Frequencies (a) and phase velocities (b), scaled by the rotation frequency, vs. the longitudinal wavenumber for density e-folds
1-10.

very large near the upper surface as the stratification a lower (higher) density on its prograde side, the longitu-
increases; consequently, the relative Coriolis forces dinally dependent, radial buoyancy force tends to shear
are stronger there. This force, when directed upward the fluid column, generating negative (positive) vorticity.
(+z-direction), is partially balanced by a downward But at the same time the upper side of a rising (sinking)
(—z-directed) buoyancy force, i.e., a positive density fluid column has a smaller density because of the den-
perturbation near the upper surface due to the perturba- sity stratification; therefore, the longitudinal pressure
tion density reversal there, and likewise by an upward gradient tends to shear the fluid column, generating
buoyancy force when the Coriolis force is directed positive (negative) vorticity. Therefore, the pressure and
downward. buoyancy torques tend to generate the opposite vortid-
ties.
e) Physics of Compressible Vorticity Waves To understand why the frequency of the waves varies
with wavenumber and density stratification, as il-
Consider a north-south roll that reaches from the lustrated in Figure 2a, recall that the frequency is
lower boundary to the upper boundary (a single cell) proportional to the vorticity generation rate divided by
composed of many fluid columns, parallel to the rota- the vorticity. Using equation (5) and the fact that
tion axis, all revolving about the axis of the roll. As a hpcc l/£, we have
particular column rises toward the upper surface, it
expands in the plane normal to the rotation axis. The w
resulting tangential Coriolis forces torque the expanding
fluid column in the opposite direction to fí, generating
negative vorticity relative to the rotating frame as pre-
dicted by equation (5) (note hp<0). Similarly, positive 3z 3x
relative vorticity is generated in sinking columns (see
Fig. 3a). Therefore, since rising fluid columns are on the When only the real parts of the numerator and de-
prograde side of negative vorticity rolls and on the nominator in equation (10) are considered, they must be
retrograde side of positive vorticity rolls, the cellular evaluated 90° out of phase because the vortidty and its
pattern should propagate with a prograde phase velocity generation rate are 90° out of phase. For convenience
as predicted by equation (9). we will consider the peak values of the numerator and
Only the above “compressibility” torque appears in denominator. Notice that the arbitrary velodty ampli-
the vorticity equation. There is no torque due to stretch- tude in this linear analysis cancels out of equation (10).
ing of the fluid column along its own axis because there Also notice that if a cell’s relative vortidty is parallel to
is no velocity gradient parallel to ß, and obviously there ß, (3m/3z) is positive and (dw/dx) is negative every-
is no viscous torque because the fluid is inviscid. In where within the cell, and vice versa if its relative
addition, since the fluid is adiabatic, the pressure torque vortidty is antiparallel to ß.
per moment of inertia [(hp/p)(dp/dx)] and the First, we examine how the frequency depends on the
buoyancy torque per moment of inertia [(g/p)(3p/0x)] longitudinal wavenumber for a given density stratifica-
exactly cancel (see eqs. [2] and [4]). The fact that these tion. For small wavenumbers, the horizontal dimension
two effects oppose each other can be seen qualitatively of a cell is greater than its vertical dimension. As a result
from Figure 3. Since a rising (sinking) fluid column has of conservation of mass, the peak u is greater than the

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

VELOCITY

/ S *■* ^9 —V - •\i •
^N\V f r'w'' 1 f / 'w^: t j f ^ w ' x t r, '•U‘
''Vi, ¡ i t î . 1 u ' , r t, ' i p t +1 v i ¿ , 11, ; u ' . *
> t 1 1 1 t 1 t .. /ll'
' ii ‘, r t, ;'¿iv
1 ; mi í Í * J' ' < 'll' - ’
^ i I 4 1 l V ' /li
— » »
' / / I I * v * < -. Í }. - '1 >-í 1 v ' •' '

m=5 N^ = l m = 15 N^=l
% /
t 9

m =5 ^=5 m = 15 N^ = 5

MASS FLUX

^^ X % #
\ I 4 / ✓^ ^^ ^ ^ ' ’ ' ' ^ — ^\ I 4 / f T '4 4 ' ' t { ^ ' w ' ' t î r '4 4 4 ' i Í r ' '41 ' ' t
t , 4‘ 11 ' , f f t '414 tit 11 ' , tf , li ', . r
'vu 1114 / " - ' 't i {} ' ' r ■ ' " ‘ * i ' ' ' v
t, 11 • , r t x ; i .4 111 ! ■ J. » f t, ; 111 . f
‘ 4 * * * » • ‘ .. , > f r t m X x,. •''*111''
'* i i
•< * / 4
i \ \
I
s/ ¿ i i
j i \ p * {ï * - ; ! ' '41 ' ■ ' ! ' * ' *4 ' ' ! Î ' '4 4
' ''

m=5 N^=i m = l5 ^=1

^ i Í '• '»X
■X.XV X I/ ' ' * f f Ss ~ ^N\ Í 4 iS* r 1
'V r( '• H ^.' w
'~r
Í t m
' ' ' ^ ’V w
'f
^\\ \ l t i**' y \ T t t t f ''^ \ \ \ l l *'
\ i I I i i f î t» î111,.
t t» 11 t • ' ' X l l l i I * îr'ü'rîîpMt-:’
• * i i l
''é i l
l
l
l X I
\ \ \ '4 4 ' ! ' -4 4 ' * ' ' 4
i ‘ '4 4
' '
#^ ✓ / i I \ — -•'*'** ' * * ' i I \ 'x 'x
*> é % X 'te
m=5 ^=5 m = !5 N. =5

Fig. 3a.—Velocity, mass flux, vorticity, and vorticity generation rate plotted in the xz-plane for various wavenumbers m and density
stratifications Np. In the vector plots, the arrow lengths are proportional to the vector magnitudes; in the contour plots, solid {broken) lines
represent the vector in the +y (—y )-direction.

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

386 GLATZMAIER AND GILMAN Vol. 45


PRESSURE PERTURBATION

m=5 N^l

m=5 IV5 m = l5 Np=5

TEMPERATURE PERTURBATION

m= 15 =I

m = l5 Np = 5

Fig. 3b.—Thermodynsimic perturbations plotted in the xz-plane for various wavenumbers m and density stratifications Np. Solid
{broken) lines represent positive {negative) perturbations.

peak w, and the vertical distance over which u changes cell is less than its vertical dimension and consequently
from zero to its peak value is less than the horizontal the peak (du/dz) is less than the peak (dw/dx). There-
distance over which w changes from zero to its peak fore, equation (10) is approximately
value (see Fig. 3a). Consequently, the peak (du/dz) is
greater than the peak (dw/dx). Therefore, equation (10)
is approximately

Since (3w/9x)peakoc mWpeak, 1 /m. Hence, as de-


(HU picted in Figure 2a, the frequency is approximately
proportional to m at low wavenumbers because the
Now, because of conservation of mass, the ratio of the vertical shear of the horizontal velocity is the major
peak horizontal velocity to the peak vertical velocity is contributor to vorticity and is approximately inversely
approximately proportional to the wavelength, so co~/w. proportional to m at large wavenumbers because the
For large wavenumbers, the horizontal dimension of a horizontal shear of the vertical velocity is the major

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

No. 2, 1981 VORTICITY WAVES IN A SPHERICAL SHELL 387


contributor to vortidty. Based on these arguments, the
frequency should peak near the wavenumber that results
in comparable horizontal and vertical velocity shears.
Equation (9) predicts a vanishing frequency for m=0
and for m —> oo. When m=0 there cannot be any vertical
velocity, so there is no vortidty generation and thus no
frequency. When /w-»oo, the vortidty becomes in-
finitely larger than its generation rate as a result of the
horizontal shear of w, and so the frequency vanishes.
Now we examine how the frequency depends on the
density stratification. When there is very little stratifica-
tion, the horizontal and vertical velodty shears are com-
parable for a longitudinal wavelength equal to twice the
depth of the zone. This occurs at m=6 when d is 40% of
the stellar radius. Consequently, the frequency should
peak at m = 6 for small stratifications (see Fig. 2a). But
as the stratification increases, for a given m, conserva-
tion of mass forces the peak vertical shear to increase
relative to the peak horizontal shear (see Fig. 3a). Fig. 4.—Frequencies, scaled by the rotation frequency, vs. the
Therefore, the frequency peaks at a larger wavenumber longitudinal wavenumber for density e-folds 1, 3, 5, and 7 from
for a larger stratification (see Fig. 2a) because the peak Paper II for Taylor number 104.
horizontal shear becomes comparable to the peak verti-
cal shear at a higher wavenumber.
Except for the cases of very little density stratification
As the density stratification increases, the local den- with low rotation rates or high wave numbers, the
sity scale height, which is proportional to £, decreases.
frequendes of the equatorial modes in the spherical shell
Therefore, equation (10) suggests that the frequency
model are prograde, and, as in the analytic model, the
should increase with density stratification. But notice
frequendes for single cells are ~50% greater than the
that although the ratio of the top density to the bottom corresponding frequendes for double cells. The frequen-
density continues to grow as Np increases, the local des of symmetric, single cells from Paper II for Taylor
adiabatic density scale height asymptotically approaches number 104 are plotted for four different stratifications
2/3(d—z). Therefore, since conservation of mass re- in Figure 4. Comparing this plot to the analytic frequen-
quires the divergence of the velodty to equal the vertical des in Figure 2 a illustrates the vorticity wave properties
velodty divided by the density scale height (see eq. [4b]), of the convective spherical shell modes.
the velodty profile and the propagation frequency re- These figures differ in several respects. The curves in
sulting from equation (8) asymptotically approach limits Figure 4 appear shifted to lower frequencies, relative to
as the density stratification increases (see Fig. 2a). Also their respective positions in Figure 2û, in proportion to
notice that since the velodty profile approaches a limit, the number of density e-folds They also drop off faster
the relative mass flux at the top surface decreases as Np with increasing wavenumber than do the curves in Fig-
increases (see Fig. 3a). ure 2a. These differences exist because the linear anelas-
Equations (9) and (10) predict a vanishing frequency tic vortidty equation for the convective spherical shell
for z0—>oo, i.e., a zero density stratification. When the has, in addition to the compressibility torque, a pressure
density is constant, rising fluid does not expand; hence, torque, a stretching torque, a buoyancy torque, and a
no vortidty is generated, and the frequency vanishes. viscous torque (see discussion in Paper II). The pressure
So far we have considered just single cells, but more and buoyancy torques do not cancel as they do in the
than one cell in the vertical direction is possible. For the adiabatic model because the peak density perturbations
same wavenumber and density stratification a multiple occur in the regions of peak radial flow while the peak
cell will have a larger relative vertical shear and about pressure perturbations occur in the centers of the cells
the same relative horizontal shear compared to a single (in order to offset the Coriolis forces). Consequently, the
cell. Consequently, the frequency for a multiple cell in pressure torque, which tends to produce retrograde phase
this model should be less than that for the correspond- propagation and is inversely proportional to the density
ing single cell, as is the case. scale height, causes the prograde frequency curves to be
shifted to lower frequencies in approximate proportion
III. COMPARISON WITH PAPER II to the number of density c-folds. Also, since the pressure
The general characteristics of the frequendes found torque is proportional to m, the curves in Figure 4 drop
for the numerical spherical shell model in Paper II are off faster with increasing m than they do in Figure 2a.
very similar to those of the analytic model in this paper. The stretching torque, except at low Taylor numbers for

© American Astronomical Society • Provided by the NASA Astrophysics Data System


198lApJS...45..335G

388 GLATZMAIER AND GILMAN


small stratifications, tends to generate prograde phase
propagation due to the surface curvature. The buoyancy
torque tends to strengthen local vorticity peaks by driv-
ing the convection, while the viscous torque tends to
weaken local vorticity peaks by the diffusion of vortic-
ity, but neither makes a significant contribution to the
phase propagation because the former is approximately
in phase with the vorticity, while the latter is ~ 180° out
of phase with the vorticity. Some differences also exist
at very low wavenumbers because the equatorial annulus
model is invalid at wavenumbers for which the most
unstable modes are polar modes.
Figure 5 shows how the peak frequencies of the
symmetric spherical shell for different Taylor numbers
and stratifications compare with those predicted by the
analytic model. The peak frequencies tend to increase
toward their corresponding analytic predications with
increasing Taylor number T because the pressure torque
does not explicitly depend on T, while the compressibil- Fig. 5.—Peak frequencies, scaled by the rotation frequency, vs.
ity and stretching torques are proportional to r1/2. Also Taylor number from Paper II compared to the peak frequencies
(broken lines) predicted by the analytic model for density c-folds 1,
the assumption of rolls aligned with the rotation axis is 3, 5, and 7.
least accurate for small T. Notice that the discrepancy at
large T is smallest for jVp = 1. This is probably because
the surface curvature, which we have ignored in the wavenumbers probably dominate giant cell stellar con-
analytic model, generates an additional amount of vection, and these have correspondingly smaller phase
vorticity via the stretching torque, which is comparable velocities relative to the basic rotation rate. Another
to that generated by the density stratification when factor is the depth of the zone which we assumed to be
Np^\. 40% of the stellar radius. A shallower zone would pro-
The fact that the prograde frequencies of the convec- duce a given frequency at a larger wavenumber and
tive spherical shell depend on the wavenumber and would result in a correspondingly smaller phase velocity.
density stratification in the same general way as do the In addition, if the angular velocity decreases with depth
frequencies of this analytic model means that the com- as predicted in nonlinear spherical shell calculations for
pressibility torque is the primary cause for the prograde Boussinesq convection (Gilman 1977, 1978, 1979), a
phase propagation in the rotating, stratified, spherical deep global convective mode should feel a lower ro-
shell. tation rate than the surface value. Consequently, the
period for a mode to propagate around the Sun would
IV. STELLAR IMPLICATIONS be longer, perhaps close to the observed surface equa-
The portion of the solar convection zone that contains torial rate. It is known that coronal holes, which may be
giant cells (below the granules and supergranules) prob- connected to such deep convective patterns, rotate at
ably has an Np >5. This simple model predicts a pro- about this rate (see, e.g., Kreiger 1977).
grade phase velocity, relative to the rotating frame, of
~0.18 fi (for w=l). This means that, if the Sun is We wish to thank David H. Hathaway for reviewing
observed from the Earth to rotate once every 27 days, a the manuscript. Papers I, II, and III are part of a
cellular pattern (m= 1) would be observed from the dissertation by G. A. G. submitted in partial fulfillment
Earth to propagate completely around the Sun (in the of the requirements for a Ph.D. in the Department of
direction of rotation) in ~23 days. However, larger Physics at the University of Colorado.

Dumey, B., and Skumanich, A. 1968, Ap. 152, 255. Hide, R. 1966, Phil. Trans. Roy. Soc. London, A, 259, 615.
Gibbons, M. P. 1980, J. Fluid Mech.y 96, 493. Hildebrand, F. B. 1949, Advanced Calculus for Engineers (New
Gilman, P. A. 1975, J. Atmos. Sei., 32, 1331. York: Prentice-Hall), p. 121.
. 1977, Geophys. Ap. Fluid Dyn., 8, 93. Kreiger, A. S. 1977, Coronal Holes and High Speed Wind Streams
. 1978, Geophys. Ap. Fluid Dyn., 11, 157. (Boulder: Colorado Associated University Press), p. 71.
. 1979, Ap. J., 231, 284.
Glatzmaier, G. A., and Gilman, P. A. \9S\, Ap. J. Suppl., 45, 351,
(Paper II).

Peter A. Gilman and Gary A. Glatzmaier: High Altitude Observatory, National Center for Atmospheric Research,
P.O. Box 3000, Boulder, CO 80307

© American Astronomical Society • Provided by the NASA Astrophysics Data System

You might also like