198lApJS ... 46. .

211G

The Astrophysical Journal Supplement Series, 46:211-238, 1981 June

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

DYNAMICALLY CONSISTENT NONLINEAR DYNAMOS DRIVEN BY CONVECTION

IN A ROTATING SPHERICAL SHELL

Peter A. Gilman and Jack Miller

High Altitude Observatory, National Center for Atmospheric Research1
Received 1980 September 12; accepted 1980 December 30

ABSTRACT
We present extensive calculations from a convectively driven hydromagnetic dynamo for a rotating
spherical shell of fluid and compare the results with observations of the solar dynamo. What
distinguishes this dynamo model from previous models applied to the Sun is that the motion fields
used are themselves solutions to the nonlinear equations of momentum, thermodynamics, and mass
continuity, albeit for a Boussinesq fluid. The full feedbacks of the induced magnetic fields on these
motions are included. The motions take the form of a time-evolving spectrum of convection patterns
driven by uniform heating from below, which in turn drive a differential rotation whose outer
boundary amplitude and profile are similar to that of the equatorial acceleration of the Sun.
Despite the similarity of the calculated differential rotation to the observed one, we find the
calculated dynamo behaves much differently than the Sun. In general, global magnetic field reversals
are absent, as is equatorial migration of the toroidal magnetic field. There is no preferred symmetry
of the induced magnetic fields about the equator. The primary reason the model dynamo behaves so
differently from the Sun is that the model helicity is perhaps three orders of magnitude larger than
has been previously assumed to obtain the correct magnetic field reversal patterns in “a —w”
dynamos applied to the Sun. As a result, stretching of poloidal fields into toroidal fields by
differential rotation in our model is a relatively minor process, in contrast to its dominant role in
cl co dynamos. The helicity seems unlikely to be greatly reduced when new physics, such as
compressibility, is introduced. We suspect instead that the small-scale interactions between velocity
and magnetic fields on the Sun allow solar fields to escape much of this helicity.
There are at least two additional effects we find with this model which should be of general
interest in dynamo theory. One of these is that ohmic dissipation rates are substantially increased by
the presence of time dependence in the motion field. The other is that even very weak magnetic fields
can have a strong feedback on the inducing motions by causing cumulative unstable phase changes in
their time history. This effect introduces substantial randomness into the response of the motions and
argues against the existence of long time delays of fixed lag in the reaction of magnetic fields on the
motion.
Previous dynamo models of a — co type applied to the Sun owe much of their success to the ability
to independently choose the magnitude and profiles of helicity or “regeneration action,” on the one
hand, and differential rotation, on the other, without regard to compatability with the laws of fluid
dynamics. We have so far been unable to find any hydrodynamical solution for coupled global
convection and differential rotation containing the combinations of helicity and differential rotation
which have been successful in simulating the solar dynamo.
Subject headings: convection— hydromagnetics— Sun: interior

I. INTRODUCTION AND MOTIVATION
Over the past 15 years, there has been a large increase
in theoretical studies of hydromagnetic dynamos, much
of the effort having been applied to the solar dynamo
problem. The most successful solar dynamo models have
been of the “a —w” type, in which toroidal magnetic
field is produced from poloidal field due to stretching by
1
The National Center for Atmospheric Research is sponsored
by the National Science Foundation.
211
differential rotation, while the poloidal field is regener-
ated from toroidal field by the so-called “a-effect”, a
being a parameter or function which measures the
amount of twisting and lifting of toroidal field lines,
caused by turbulent convection influenced by rotation.
Although in the last decade there has been a veritable
explosion of quantitative calculations of a —co dyna-
mos, the basic ideas used in the solar context can be
traced back at least to the pioneering work of Parker
(1955 a,b) and Babcock (1961). We make no attempt

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

198lApJS ... 46. .211G
212 GILMAN AND MILLER
here to review this large subject. Work prior to 1976 is
described well in the IA U Symposium 71 proceedings,
particularly in articles therein by Stix (1976) and Krause
(1976). More recent work is summarized by Stix (1981).
In some solar dynamo models, particularly those of
Yoshimura (1972, 1975, 1978û,Z>), the origin of a has
been explicitly ascribed to global or “giant celT’ convec-
tion. In most others the identification of a with a
specific scale of solar convection has been less definite.
In all cases there has been very little guidance from
theory, and none from observations, on what magnitude
to choose for a. In general, a has been chosen essentially
independently of the magnitude and profile of differen-
tial rotation, although in reality the two must be
intimately related through the fluid dynamics laws
governing the motion. What we report in this paper are
dynamo calculations in which all of the induction effects
arise from the same, self-consistent solutions of the
equations for convection in a rotating spherical shell.
Thus, the differential rotation in the model is driven by
nonlinear interaction with the convection in the model,
and both together provide all of the velocities which
induce magnetic fields via the MHD induction equation.
Starting from a hydrodynamic solution in which the
surface differential rotation takes the form of an equa-
tional acceleration similar in profile and magnitude to
the real Sun, we find that the resulting dynamo behaves
in most respects much differently than the Sun and than
previous a —to dynamo models applied to the Sun. The
reasons why this is so raise serious questions about solar
dynamo theory, which may require rethinking of many
concepts previously presumed to be valid for the solar
case.
We are not alone in questioning the validity of a —to
dynamo theory applied to the Sun. For a number of
years Piddington (1975, 1976) has been arguing against
solar a —to dynamos basically on grounds that proper
account has not been taken of the solar field being
largely confined to isolated magnetic flux tubes. (We
must state, however, that we do not agree with many of
Piddington’s arguments.) More recently Layzer, Rosner,
and Doyle (1979) have questioned many of the assump-
tions inherent in a —co dynamo theory applied to the
Sun, and Golub et ai (1981) have argued that much of
the emerging magnetic flux seen on the Sun does not
appear to be well connected to the dynamo that drives
the obvious features of the solar cycle. Galloway and
Weiss (1981) have advanced various arguments deriving
from the theory of convection in a magnetic field to
question whether the main solar dynamo can be con-
tained in the convection zone. Parker (1979) has argued
that magnetic buoyancy of the flux tubes may keep all
but the deepest layers of the convection zone from
acting as a dynamo. After presenting our own results we
will comment further on connections to these other
concerns. The common thread in these arguments is the
© American Astronomical Society • Provided by the NASA Astrophysics Data System
Vol. 46
role played by magnetic flux tubes. Our results also
point to reconsideration of this role.
Still another problem with assuming that the solar
dynamo (and differential rotation, for that matter) is
driven by global convection is that no one has been able
to clearly observe any giant cells on the Sun. Recently
Howard and LaBonte (1980) have placed an upper limit
on giant cell velocities of a few ms-1 per longitudinal
wavenumber from Mt. Wilson Dopplergrams. Gilman
and Glatzmaier (1980) have questioned these low limits,
and they are being revised upward somewhat (LaBonte,
Howard, and Gilman 1981), but they still are no more
than 13 m s-1 per wavenumber, which is smaller than
predicted from model calculations. It is clear, however,
that better observations are needed before great confi-
dence can be placed in these limits.
Despite the observational difficulties, there remain
strong theoretical reasons for expecting giant cells to be
present and to be the main driver for the solar differen-
tial rotation (Simon and Weiss 1968; Vickers 1971;
Gough et al 1976; Roxburgh and Tavakol 1979; Busse
1970, 1973; Dumey 1970, 1971; Gilman 1972, 1975,
1976, 1978, 1979). Much of the theoretical work on the
origins of solar differential rotation has been reviewed
recently in Gilman (1980«, b).
II. FORMULATION
Our main goal is to find out whether a dynamically
consistent combination of global convection and equa-
torial acceleration gives hydromagnetic dynamo behav-
ior similar to that of the Sun. In order to reach this goal
we have generalized our previously existing code for
nonlinear convection in a rotating spherical shell to
include electromagnetic induction effects in a complete
and physically consistent manner. In the resulting model,
we are able to solve simultaneously for the evolving
velocity and magnetic field and include the full feed-
backs of the magnetic field on the motion. The physics
of this generalized model is admittedly still much sim-
pler than the real Sun—for example, we deal with a
Boussinesq fluid, rather than a compressible one—but it
is more general and realistic than in previous models,
which either ignore the hydrodynamics all together (all
kinematic dynamos) or base the chosen motions on
solutions to the hydrodynamic equations which are even
more severely approximated than ours. For example,
Yoshimura (1972) uses motions found for a shallow
layer, in the limit of low rotation, in the hydrostatic
approximation, and without regard for whether the con-
vective velocities he uses could couple properly with the
differential rotation he assumes. In our calculations
none of these approximations are made, and the convec-
tion present is the motion which maintains the differen-
tial rotation that is present.
Although in our model all of the induction effects of
global scale motions are explicitly calculated, we are still
198lApJS ... 46. .211G
No. 2, 1981 DYNAMICALLY CONSISTENT NONLINEAR DYNAMOS
forced to represent the influence of scales of motion we
cannot resolve by diffusion coefficients for temperature,
momentum, and magnetic flux. Thus, our model retains
some aspects of a “mean field” model, like the a2 and
a —w dynamos. In this first calculation, we have also
found it necessary to allow the diffusion coefficient for
magnetic flux to be somewhat smaller than those for
momentum and heat in order to get dynamo action at
all. This assumption has no strong physical justification
but allows us in a simple way to raise the internally
determined magnetic Reynolds number to the point
where dynamo amplification of the magnetic field sets
in. This procedure is implicit in most kinematic dynamo
calculations, since raising the magnetic Reynolds num-
ber in such studies while keeping the same motion field
could be interpreted as lowering the magnetic field
diffusion rate while keeping the viscous diffusion rate
fixed in the imphed calculation of the motion field.
In previous calculations, we have found (Gilman 1972,
1977, 1978, 1979; Gilman and Foukal 1979) that in
order to obtain equatorial acceleration of substantial
amplitude, such as that which the Sun has, the influence
of rotation upon the convective motions must be strong
and the convecting layer must be rather deep, at least
one-third of the radius. Solutions for weak rotational
influence, such as those assumed by Yoshimura (1972)
and Busse (1970, 1973), lead to equatorial deceleration
when the flow reaches finite amphtude. Solutions for
shallow layers lead to an angular velocity which
increases toward the poles except when close to the
equator. Thus, these are unacceptable hydrodynamic
solutions to start from for the Sim. Yoshimura (1972,
1975) is able to use solutions in the limit of low rotation
because the differential rotation he takes is assumed and
not computed as a dynamical consequence of the global
convection he derives. The consequences of such incon-
sistencies are discussed again later.
Almost all of the hydrodynamic solutions we use here
are for a single choice of the governing dimensionless
parameters: Rayleigh number R = 2A\X104, Taylor
number T= 105, Prandtl number P— 1 (see the Appen-
dix for definitions) for a convection zone depth of 40%
of the outer radius, which is as deep a solar convection
zone as seems possible. This particular case was chosen
primarily because its average profile of differential rota-
tion with latitude seemed to be the best fit to the solar
case, as discussed in Gilman (1979). In any case, it is a
reasonable place to start. The precise numbers are not
important; other solutions we tried with neighboring
parameter values gave very similar results.
The hydrodynamic model we generalized is structured
in such a way that each dependent variable (three
velocity components, temperature, and pressure) is repre-
sented by a Fourier series in longitude, including longi-
tudinal wavenumbers between zero and some maximum
value. The amphtude coefficients of the terms in these
© American Astronomical Society • Provided by the NASA Astrophysics Data System
213
series are then computed on a set of finite difference
grids in the meridian (latitude-radius) plane. These grids
are staggered in space for different variables to optimize
efficiency and retain energy conservation in nonlinear
terms (details are given in the Appendix and in Gilman
1975, 1977, based upon earlier work by Wilhams 1969).
In generalizing the code to do magnetohydrodynamic
calculations, we are able to evaluate the three compo-
nents of magnetic field on the same grids as the corre-
sponding components of velocity field. Most of the extra
terms involving magnetic fields have their analogs in
ordinary hydrodynamic terms, which were therefore
taken as templates. The full magnetohydrodynamic
equations we actually solved are set down in the Ap-
pendix for the interested reader.
To obtain an ordinary hydrodynamic solution, the
three velocity components and temperature are marched
forward in time by solving the three equations of motion
plus the thermodynamic equation. The pressure is then
found at the new time step by solving a Poisson-
Helmholtz-type equation, which is obtained by taking
the divergence of the vector equation of motion and
invoking the condition that the fluid is incompressible (a
different solution technique must be used in the fully
compressible case). This allows conservation of mass as
well as momentum to be satisfied.
In the MHD case, there are three magnetic field
components corresponding to the three velocity compo-
nents, which we can obtain from the three components
of the equation for electromagnetic induction. However,
there is no additional electromagnetic variable corre-
sponding to the pressure. Nevertheless, the vector mag-
netic field must remain divergence-free (no magnetic
monopoles), so the three components must actually
satisfy four equations. In practice, therefore, we actually
solve three of the equations and then check to see how
closely the fourth is satisfied. If we were to solve the
three components of the induction equations, then the
divergence-free condition would be good only to trunca-
tion error. For dynamo studies we regard this as inade-
quate, so, in general, we solve two of the induction
equations for two of the magnetic field components and
then find the third from VB=0. (Before finite dif-
ferencing is introduced, specifying V *B=0 as an initial
condition is enough to ensure it remains so for all
subsequent times.) For one particular combination of
equations (described in the Appendix) plus all perfectly
conducting boundaries we are able to prove that all four
equations are satisfied exactly, i.e., to machine round off
in the actual computations. This powerful condition was
used as a check in the programmed equations, since it
leads to a number of work integrals which must cancel
to the same accuracy (see Appendix for details).
For physical reasons we chose in the actual computa-
tions to use a different boundary condition on the
magnetic field at the top, namely, to allow the field to
198lApJS ... 46. .211G
214 GILMAN AND MILLER
become radial there. This is largely true in the solar
photosphere, probably as a result of the action of small-
scale convective motions in maintaining magnetic flux
tubes at points of horizontal convergence. We beheve it
to be a better boundary condition for the solar problem
than the more traditional one in dynamo theory of
fitting the interior solution to a potential field. It is also
much easier to apply than the potential field condition
in a calculation such as ours. The radial field assump-
tion was first used in solar dynamo theory, we believe,
by Yoshimura (1975). With the radial field boundary
conditions we also monitored the total magnetic flux
crossing the boundary, to be sure it was zero to machine
round off. We did do a few calculations with both radial
field and perfectly conducting boundary conditions and
found that none of the major results we are presenting
here were sensitive to which components of the electro-
magnetic induction equations were actually solved, and
which were used as checks, given that V * Æ=0 was
satisfied.
In doing the hydrodynamic calculations reported in
previous publications, we found it necessary to filter the
solutions near the poles in order to prevent computa-
tional instability. Our filters were relatively crude, in
that we simply suppressed the solution for a given
longitudinal wavenumber m poleward of a specified
latitude. The higher the wavenumber, the lower the
latitude. Were we to do the same thing in the MHD
case, we would, in general, produce spurious magnetic
monopoles at the latitude where the filtering takes place
because, in general, V * B would no longer vanish there.
Better filtering schemes could undoubtedly be devised
(and a spherical harmonic expansion could avoid this
problem), but we chose instead for simplicity to put a
wall at latitude 75° N and S. This wall is a perfect
electrical conductor, as well as a thermal insulator. It
changes the hydrodynamic behavior in low and middle
latitudes virtually not at all. We may assume with good
confidence that dynamo solutions for this slightly trun-
cated sphere will reasonably characterize solutions for
the full sphere.
We also experimented with the spatial resolution and
the number of longitudinal wavenumbers m retained in
the Fourier expansion, before settling on a choice of
15, a grid with 3° latitude spacing and 12 radial
intervals. From previous hydrodynamical calculations,
e.g., Gilman (1979), this resolution was quite adequate
for resolving all the convective motions in the parameter
ranges of greatest interest for the Sun. This resolution
has proved to be somewhat less optimum for resolving
the magnetic fields generated because smaller scales are,
relatively speaking, more important. However, it repre-
sents about the finest resolution we could afford for
systematic study, even with a substantial allotment of
computer time on the NCAR CRAY 1A computer.
These are expensive calculations, requiring approxi-
© American Astronomical Society • Provided by the NASA Astrophysics Data System
Vol. 46
mately 10 s of CRAY CPU time for each step. We did
perform a few runs at still higher resolution, some with a
2° latitude grid and 20 radial intervals, and others with
up to 23 longitudinal wavenumbers. These tests con-
firmed the trends already seen at lower resolution and
resulted in very similar statistical properties of the solu-
tions, so we are satisfied that our primary resolution
choices are adequate for the results we obtained. Be-
cause of the time dependence of the solutions, we found
it difficult to devise more precise tests of accuracy.
We also went to the other extreme and examined a
number of cases in which only two wavenumbers, say
m=0, 5, were retained, or at most only five or six
wavenumbers. These results were considerably less
satisfactory because the time variations in the hydrody-
namic solutions were greatly amplified because of the
rather severe truncation. This obviously also influenced
the dynamo characteristics and led us to believe that it
would be difficult to generalize from these (admittedly
economical) solutions to the less truncated case. There-
fore, we focused on calculations with many modes pre-
sent.
III. TYPES OF NUMERICAL EXPERIMENTS PERFORMED
In general, we started up our dynamo solutions in
stages. First, we perturbed a state of rest (solid rotation)
with random numbers in the temperature field. From
these initial perturbations a spectrum of convection
develops. The Reynolds stresses in this convection begin
to redistribute angular momentum, generating differen-
tial rotation. During this stage, no magnetic field is
present. It generally takes ~2000 time steps for the
convection and equatorial acceleration to become fully
established. Once that happens, the flow field continues
to evolve in a somewhat random way, with different
longitudinal wavenumbers peaking in the convection
spectrum at different times. The amplitude of an indi-
vidual wavenumber can vary by a factor of 3 or 4 in a
few hundred time steps (corresponding to a month or
two of solar time). The spectrum changes on a time scale
comparable to the turnover time for the whole convec-
ting layer, although some dominant patterns can persist
for several turnover times. In addition to amplitude
variations, the rate at which each wavenumber rotates
about the axis varies with time on the same time scale
by several percent. Thus, different modes will grow,
decay, and move through each other.
The differential rotation pattern also evolves with
time on a somewhat longer time scale but always retains
its equatorial acceleration. Its fluctuations in amphtude
are more like 10% (relative to a uniformly rotating
reference frame), corresponding to changes in absolute
rotation rate of ~2%-3%.
Once the differential rotation and convection are fully
established, we introduce a small-seed magnetic field,
usually a purely toroidal field (a0 in the notation of the
198lApJS ... 46. .211G
No. 2, 1981 DYNAMICALLY CONSISTENT NONLINEAR DYNAMOS
Appendix) so that is automatically satisfied.
Experiments were performed with two kinds of initial
toroidal fields: a field of random numbers, and a profile
smoothly varying in latitude and in radius which was
purely antisymmetric about the equator. We had ex-
pected that dynamo action might organize the random
number pattern into something more regular, but that
did not happen. On the contrary, the smooth antisym-
metric pattern was soon broken up into more random
pieces.
To examine as fully as possible how the dynamo
behaved, we devised several long runs of the model,
which we describe below. All of these were started from
the same hydrodynamic solution so as to make compari-
sons easier. The principal parameter varied was the
magnetic Prandtl number Q—^/k, where, as defined in
the Appendix, tj is the magnetic diffusivity and k is the
thermometric conductivity. There is no externally de-
fined magnetic Reynolds number in our case, as op-
posed to the kinematic dynamo problem in which the
motion field is specified and usually steady. We find
that Q must be less than a certain value for sustained
amplification of the magnetic field to occur.
To examine the influence of feedback of the induced
magnetic field on the motion, as well as the role played
by time variations in the motion, we found it useful to
perform experiments in which either the feedback was
arbitrarily switched off, or the motion field was frozen
at a certain point in time. In this latter case we revert to
the classical kinematic dynamo problem, but with the
motions closely approximating real solutions to the
equations of motion. In addition, we also ran a “control”
case in parallel with the dynamo cases for which the
magnetic field was never introduced.
IV. RESULTS
a) Magnetic Energy History
The first question to be answered is whether we get a
dynamo when a seed magnetic field is added to the
hydrodynamic solution. Figure 1 shows one measure of
this, namely, the total magnetic energy as a function of
time, obtained for three different magnetic Prandtl num-
bers Q (0.4, 0.2, 0.1) after inserting a small-amplitude,
purely toroidal magnetic field a0 at time step 2800. In
each case full feedbacks are present, and the motion
field is allowed to continue to evolve. The threshold for
dynamo action appears to be in the neighborhood of
ß=0.2, since its time trace is nearly flat after an initial
rise. Clearly, ß=0.4 is below the dynamo threshold, and
ß=0.1 is above it. The initial steep rise in all the cases is
due to magnetic field being induced in all the nonzero
wavenumbers. As we will see below, the spectrum be-
comes virtually “full” by about step 3200. Thereafter, it
is a question of whether ohmic dissipation or induction
© American Astronomical Society • Provided by the NASA Astrophysics Data System
215
4 a 5
R = 2.4lx|0 I I0 P=l
Fig. 1.—Initial time history of total magnetic energy for three
dynamo runs, starting from the same initial conditions but with
different magnetic Prandtl number Q.
win. Since each time step is only 2 X10 _4 of the thermal
diffusion time ¿/2/k, the long-term trend in the magnetic
energy becomes established in only a small fraction of
an ohmic decay time for the shell. For example, for
ß=0.2 the ohmic decay time is five dimensionless units
(Q~l in our dimensionless time units), and yet the
ß = 0.2 curve levels out within ~500 time steps, which
is only ~0.1 time units. The reason equilibrium is
reached so fast is that magnetic energy is cascaded out
to the smallest scales present very efficiently, namely, in
one or two convective turnover times (Frisch 1977;
Frisch, Sulem, and Nelkin 1978), which is only a few
hundred time steps in our model. The ohmic decay time
at these small scales is similar, so magnetic energy is
easily dissipated as quickly as it arrives from larger
scales.
One might have supposed that the leveling out of the
magnetic energy trace for ß=0.2 was due to feedback
of the magnetic field on the motion. If so, this would
have been quite surprising, because the total kinetic
energy in these units was ~2X 104 units, or almost 104
times greater than the magnetic energy. Two tests we
performed confirmed it was not this feedback. In the
first we simply shut off the feedback, while still allowing
the motion to evolve, and obtained virtually identical
time traces out to about step 4500. In the second, we
redid the full feedback calculation but with an initial
magnetic field one-tenth as large. The magnetic energy
trace was exactly 100 times smaller.
With further tests, we discovered the time evolution
of the motion field contributes a great deal to the
216 GILMAN AND MILLER Vol. 46
4 5
R = 2.4lx|0 T = I0 P= l

Fig. 3
Fig. 2.—Initial time traces of magnetic energy for three cases with magnetic Prandtl number 2=0.2. The solid curve is the “standard”
case with the motion evolving and full feedbacks present. The curve with long dashes is for a kinematic dynamo calculation in which the
motion field is frozen at step 2800, when the magnetic field is first introduced. For the curve with short dashes the motion field is frozen at
time step 4540, whereupon the dynamo calculation becomes a kinematic one beyond that point.
Fig. 3.—Initial time history for two dynamo runs with magnetic Prandtl number Q increased to 2 — 0.4. The solid curve is the same
calculation as in Fig. 1 with evolution of the motion and full feedbacks, while the dashed curve is for a case in which the motion field is
frozen at time step 2800, so that subsequent dynamo is a kinematic one.
dissipation of magnetic fields. Figure 2 shows a longer shows neither long-term growth or decay. There are
time trace for 0=0.2, on which are also plotted the shorter episodes of both, and we interpret these to mean
magnetic energy when the motion is frozen, at step that there are some sequences of motion patterns which
2800, and at step 4540. In each case, the magnetic are slightly more favorable for dynamo action than
energy grows exponentially after an initial period of others, but neither are not sustained indefinitely. The
adjustment. Thus, if the motion were frozen, as in a g=0.1 case eventually shows a leveling off of magnetic
kinematic dynamo, the threshold for dynamo action energy, at a level about a factor of 6 or so below the
would be at a much higher value of Q. Figure 3 shows kinetic energy present. Thus, the solution approaches,
that the field grows even for 2=0.4, which strongly
decayed when the motion was allowed to evolve. Judg- R = 2.4I xlO4 T=I05 P=l
ing by the change in slope of the magnetic energy curves
between g = 0.2 and 0.4, the frozen motion threshold is
somewhere near g=0.5. Since the motion fields are the
same, this means the magnetic Reynolds number re-
quired to get a dynamo is only ~40% of what is needed
when the motion evolves. Since a reasonable RMS veloc-
ity is ~20 units, this means a magnetic Reynolds num-
ber of ~40, rather than 100.
What is happening in the fluid is that, with the
motion evolving, the magnetic field is never able to get
into the optimum configuration for growth. It starts to,
and then the motion changes to another pattern which
produces some cancellation of old field. If we were to
lengthen the evolution or turnover time for the convec-
tion (without reducing the velocity amplitude), then the
effect would be diminished. This is a process which has
not, to our knowledge, been noticed before in dynamo
theory, presumably because the motion is virtually al-
ways held steady.
The g=0.2 and 0.1 runs were carried out beyond
Fig. 4.—Total time history of magnetic energy for the two
time step 11,000 in order to generate long-term statis- principal dynamo runs with motion evolving and full feedbacks,
tics. The magnetic energy time traces for these full runs with magnetic Prandtl number g=0.2 (resulting in weak magnetic
are shown in Figure 4. We can see that the g=0.2 case fields) and 2—0.1 (resulting in strong magnetic fields).

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

198lApJS ... 46. .211G
No. 2, 1981 DYNAMICALLY CONSISTENT NONLINEAR DYNAMOS
but does not achieve, equipartition. We demonstrate
later, when examining the feedbacks of magnetic field
on the motion, that this leveling is due to such a
feedback. Presumably all solutions with 0.1<g<0.2
would show time traces falling between these two. We
do not present calculations for ß<0.1 because we found
serious difficulties with spatial resolution.
Given that we find dynamo solutions for ß=0.2,
what kind of dynamo is it? In particular, is there evi-
dence of magnetic field reversals? Since the motion field
is evolving in time, the magnetic field pattern will change
in complex ways, and whether time field reversals are
present might be obscured. But if we examine the solu-
tions for which we have frozen the motion field, we
always find the magnetic field pattern grows in time
without reversals or migration of the field toward the
equator or the poles. When the motion is changing with
time, we occasionally see superficial evidence of migra-
tion in the toroidal field aQ toward the equator near the
outer boundary, but it is a weak effect, not always there,
and not present deep in the convecting shell where
the field is strongest. These effects were seen only in the
ß=0.2 case, and when the motion field was frozen, the
migration stopped. We return to the question of field
reversals at several later points.
b) Spectral Characteristics
Subsequent to the introduction of a seed magnetic
field at time step 2800, the full MHD solutions can
evolve in different ways from each other and from the
parallel control case in which no magnetic field has been
introduced. Figure 5 shows sample average kinetic en-
R = 2.41 x IO4 T = I05
Fig. 5.—Sample average spectra of total kinetic energy (upper curves) and radial heat transport fraction (lower curves) as functions of
longitudinal wavenumber m. Three cases are shown: on the left, a “control” case, in which no magnetic field is present; in the middle, the
case with weak magnetic field and magnetic Prandtl number ß=0.2; and on the right, the case with strong magnetic field and 0 = 0.1.
© American Astronomical Society • Provided by the NASA Astrophysics Data System
217
ergy and heat flux spectra as functions of longitudinal
wavenumber m for three cases. These are the control
case with no magnetic field, the case with 0 = 0.2, for
which from Figure 4 the magnetic field remains weak,
and the case with ß=0.1, for which it becomes strong.
To facilitate comparison, these averages are obtained
over the same time period of 2860 steps, 4500 steps after
the introduction of the magnetic field. This length of
time average corresponds to ~10 convective turnover
times. Although the spectrum actually runs through the
range —15 = már +15, we have folded the spectra about
m=0, so the energy in each nonzero m is the sum of that
in the positive and negative wavenumber parts. This fact
is important for interpreting the magnetic spectra which
follow.
The general shape of the kinetic energy and heat flux
spectra in Figure 5 are similar for three cases. In each
case, m=0, representing differential rotation plus a small
meridional circulation has the most kinetic energy. The
convection amplitude peaks at some intermediate wave-
number and drops off by at least one order of magni-
tude by m= 15. The greatest heat flux is carried by the
wavenumbers with greatest kinetic energy, but very little
heat is carried by the meridional circulation represented
in wavenumber zero.
The most obvious differences in the spectra are that
different longitudinal wavenumbers dominate for the
different cases. For ß=0.2, it is /w=2 and 5, while for
ß=0.1, it is m = 3 and its first two harmonics, m = 6 and
9. Examination of the time history of the spectra reveals
that in both these cases the wavenumber combinations
listed above dominate through the whole time period,
while in the no field case the dominant wavenumbers
P = I STEPS 7300-10160 AVERAGED
218 GILMAN AND MILLER
are changing through the same period, resulting in a
smoother spectrum. Thus, we can see that even a weak
magnetic field can induce different convection patterns
to dominate, since at time step 2800 the three solutions
were identical. We examine this further when we look at
how the magnetic field reacts back on the flow.
The magnetic energy spectrum averaged over the same
time period is shown in Figure 6. By contrast with the
kinetic energy spectrum in Figure 5, it is very smooth
and falls off with m much more slowly; its peak occurs
at m—\. This curve represents a virtually universal
shape seen in all our calculations, no matter what value
of Q is used or what convection modes happen to
dominate. Starting from a purely toroidal field, it takes
less than 103 time steps to establish this spectrum, as
illustrated in Figure 7, for 0=0.2. At step 2800, m=A
happended to be the convective wavenumber of largest
amplitude, so magnetic energy grows fastest in m=4
and its overtones 8 and 12. But further interactions with
other wavenumbers rather quickly fill in the inter-
mediate modes, so that by step 3600 the instantaneous
magnetic energy spectrum is rather similar to, but some-
what smaller in amphtude than that in Figure 6. Through
this whole time sequence m=4 has remained the domi-
nant convective mode. Yet no peak in magnetic energy
remains at m=4 by step 3600, so efficient is the mixing
process.
The magnetic energy for the ß=0.1 case was still
rising somewhat through the time period 7300-10160.
R = 2.4IXI04,T=I05,P=I,
STEPS 7300-10160 AVERAGED
Vol.46
We ran the calculations further and observed a sudden
change in the dominant convective mode from m = 3 to
m=4, and by time step 11,000 the magnetic spectrum
appeared to be no longer growing. While the dominant
convective mode has changed, the magnetic energy spec-
trum has simply amplified for all wavenumbers, by
about a factor of 2. This is quite general behavior, seen
also in the 0=0.2 case, through all the rises and falls of
total magnetic energy seen in Figure 4. By time step
11,000, the magnetic energy at highest m for ß=0.1
actually exceeds the kinetic energy there. We expect that
trend would continue in still higher m if they were
included. The relatively sudden shifts from one domi-
nant convective mode to another after persistence of one
pattern for several turnover times is also common, both
with and without magnetic field present.
The development of the magnetic energy spectrum
with time, its greater breadth than the kinetic energy
spectrum, and the fact that at high wavenumbers the
magnetic energy exceeds the kinetic energy, while the
opposite is true at low wavenumbers, are all qualita-
tively similar to behavior seen by Pouquet and Patterson
(1978) in three-dimensional numerical simulations of
MHD turbulence. In both cases, magnetic energy is
quickly cascaded out to high wavenumbers and then
“backfilled” into intermediate scales. However, they see
in addition amplification of kinetic energy in high wave-
numbers by feedbacks from the magnetic field. That
effect, if present, is very small in our calculation. We
R = 2.41 x |04 T = I05 P=l Q = 0.2

LONGITUDINAL WAVE NUMBER, m

Fig. 6 Fig. 7
Fig. 6.—Spectrum of total magnetic energy for the same cases as given in Fig. 5
Fig. 7.—Time development of the magnetic energy spectrum for 2=0.2 for the first 800 time steps, starting from an initial purely
toroidal magnetic field (represented by the spike in the spectrum at m=0 for early times).

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

o

No. 2, 1981 DYNAMICALLY CONSISTENT NONLINEAR DYNAMOS 219

might see it if we had included much higher wavenum-
'"D bers, because extrapolating the energy curves to high m
a
in Figures 5 and 6 suggests the magnetic energy would
co
CTi become much larger than the kinetic energy, providing
an energy source at these scales more powerful than
buoyancy.
A breakdown of the kinetic and magnetic energies by
component is shown in Figure 8. (Upper case letters
U,V,W,A,B,C denote the total energy in the corre-
sponding velocity and magnetic field variables u,v,w, a,
b, c defined in the Appendix.) These kinetic energy
profiles are very typical of the spherical shell convection
solutions studied in earlier papers using this model.
East-west flow and east-west magnetic field pre-
dominate at low wavenumber, as should be expected
from the fact that both must satisfy divergence-free
conditions. Differential rotation energy is about a factor
of 100 larger than the meridional circulation, as a result
of the strong rotational influence on the motion. At high
wavenumbers, latitudinal and radial motion and field
comprise a greater fraction of the totals, which is also to
be expected. The ratio of total energies in each compo-
nent when summed over all wavenumbers is about the
same for velocity and magnetic fields. Thus, the more
flow in a particular direction, the more the magnetic
field is stretched out in the same direction. We see also
that most of the drop in magnetic energy with m is
contained in the east-west (a) component of the field.
By contrast, the energy spectra for b and c are much
flatter, with a broad shallow maximum in intermediate
wavenumbers, followed by a slow decay with m.
One of the most distinctive features of the magnetic
energy spectrum in Figure 6 and 8 is the sharp drop in
energy in all components in wavenumber 0 as compared
to wavenumber 1. This feature in fact is produced
simply by folding the spectrum about wavenumber 0.
Figure 9 shows the same spectrum unfolded, so negative
as well as positive wavenumbers are plotted. Viewed in
this way, more magnetic field is induced in ra=0 than
any other single m, but only in line with the monotonie,
nearly linear increase from higher positive and negative
wavenumbers.
In mean field dynamo theory, a special place is
accorded to axisymmetric (ra=0) toroidal and poloidal
fields, and all other fields are lumped together as per-
turbation departures. It is obvious from Figure 9, how-
ever, that the axisymmetric part of the field is simply the
peak of a broad spectrum. The field A0, in particular,

R = 2.4I x IO4 TSI05 P = l Q = 0.1 AVERAGE OVER STEPS 7300-10160

Fig. 8.—Spectra of kinetic and magnetic energies for the 0=0.1 case in Fig. 5, separated by component of motion and magnetic field.
The letters U,V,W denote the kinetic energy in the east-west, north-south, and radial motion fields, respectively, while A,B,C denote the
energies in the corresponding components of magnetic field.

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

220 GILMAN AND MILLER
4 5
R = ^4IXI0 J=I0 IP=|,Q=0.I,
AVERAGE OVER STEPS 7300-10180
Fig. 9.—Total and component magnetic energy spectra for
Q=0.1 case unfolded so that negative and positive longitudinal
wavenumbers are both depicted.
contains typically no more than ~5% of the total
magnetic energy. If this were true on the Sun, it seems
unlikely the Sun would be able to sustain the Hale
sunspot polarity law so well, since all of the polarity
information is contained in the axisymmetric part. To
get the toroidal field to be a larger fraction of the total
requires the magnetic spectrum fall off more steeply
with m.
In addition, on the Sun there is strong evidence the
axisymmetric field is antisymmetric about the equator.
Sunspot pairs have opposite leading and following
polarities in north and south hemispheres, and the north
and south polar fields are of opposite polarity virtually
all the time. In our numerical experiments, we de-
liberately tried to force the antisymmetric magnetic
structure to dominate by starting our calculations from
an initial toroidal field which was purely antisymmetric
about the equator. Because of the influence of rotation,
the convection and differential rotation in the calcula-
tion are predominately symmetric about the equator. If
they were purely so, then all of the magnetic field
components would remain antisymmetric. However, even
a small amount of flow (~10% of the total kinetic
energy) of the opposite symmetry about the equator is
enough to induce both symmetries in the magnetic field,
and after a few thousand time steps no preferred sym-
metry about the equator can be found in the magnetic
field. This question of how the Sun chooses a particular
symmetry remains vexing. Some a —dynamo solutions
do favor antisymmetric fields, such as those with angu-
lar velocity increasing with depth in Roberts (1972) and
Roberts and Stix (1972), but others, particularly those
© American Astronomical Society • Provided by the NASA Astrophysics Data System
Vol. 46
which have most of the dynamo action in higher lati-
tudes, such as in Belvedere, Paterno, and Stix (1980),
show little preference. The long series of solar dynamo
models by Yoshimura (1972, 1975, 1978a, 6, 1979)
sidestep this problem by simply assuming the correct
symmetry.
c) Maintenance of Magnetic Energy
The induction equations which predict the evolution
of the magnetic field are laid out in the Appendix (eqs.
[6]-[8] and [21]-[23]). The ohmic diffusion terms (all
those multiplied by the magnetic Prandtl number ß), in
general, produce only decay of the field. The induction
terms (all terms involving products of velocity and mag-
netic field variables) which can sustain the field can be
broken into two types, which we call shear terms and
transport terms. In each component of the induction
equation, shear terms are those which involve a velocity
in the same direction as the field component being
induced, coupled with magnetic fields in the two per-
pendicular directions. For example, in equations (6) and
(21) for induction of the east-west field a, the shear
terms are all those containing w, the east-west velocity.
All these terms are “shear terms” because variations in
latitude or radius of u shear the fields b and c, respec-
tively, into the east-west direction to induce an a field.
Transport terms are all terms which involve products of
the magnetic field component induced in that equation
with velocities in the two perpendicular directions. Thus,
in equations (6) and (21), the transport terms are all
those involving products of a with either v or w. These
terms are said to “transport” field because field is
simply moved from place to place without being sheared.
From the finite difference spectral equations (21)-(23),
it is clear that both processes convert magnetic field of
one longitudinal wavenumber to another, through non-
linear interactions. The question then is which transport
and which shear processes are most important in main-
taining the magnetic energy spectra seen in Figures 6-9.
To find out, we calculate the rate at which magnetic
energy of each field component at each wavenumber is
produced by each of the shear and transport terms in
the equations. Formally, this is done by first multiplying
each induction equation by the complex conjugate of the
magnetic field variable predicted by that equation, and
then averaging all the equations to a common grid in the
meridian plane (as described at the end of the Appen-
dix). Finally, each equation is summed over all the
points in the grid. Figure 10 shows the magnetic energy
maintenance rates for each wavenumber, summed over
the three field components for the same ß=0.2, ß = 0.1
cases as in Figures 5 and 6. For comparison with Figure
9, we have left the energy maintenance spectrum un-
folded. The curve marked “total” in each case gives the
sum of shear and transport processes, which in a long-
term average of a statistically stationary solution is
No. 2, 1981 DYNAMICALLY CONSISTENT NONLINEAR DYNAMOS 221
4 5 s
R = 2.4I x IO4 T = I05 P= l R = 2.4IX|0 T = I0 P=l Q O.I

LONGITUDINAL WAVE NUMBER, m

Fig. 10 Fig. 11
Fig. 10.—Rates (dimensionless) at which magnetic field associated with each longitudinal wavenumber is maintained by transport and
shear processes defined in the text. Sample averages are for same time period as in Figs. 5 and 6.
Fig. 11.—Magnetic energy maintenance rates from Fig. 10, separated by component of field

approximately the negative of the magnetic energy dis-
sipation rate.
We see that for both 0=0.2 and 0=0.1 cases, the
large magnetic energy at low wavenumbers is main-
tained primarily by shearing of magnetic field. Trans-
port is very small at low m but is slightly larger than
shearing at high m. This is evidence that the convection
takes magnetic field of a given component and wave-
number and packs it into smaller scales or higher wave-
number.
We break down the magnetic energy maintenance by
field component in Figure 11 for the 0=0.1 case. Here
we see that most of the shearing of low wavenumbers
goes into maintaining the large east-west field By
contrast, the a field at high m is maintained entirely
by transport, which means packing into cell boundaries.
By comparison, maintenance of the b and c fields is
predominately by shearing at all m. The strength of this
shearing generally increases with m to counteract the
larger dissipation due to diffusion across smaller longi-
tudinal scales.
We should expect large shearing of b and c fields by u
motions at low m because, from Figure 8, there is a lot
of kinetic energy in the m field at low m, compared to
either v or w. This is especially true for m=0, which
contains the differential rotation. The resulting large
values of am at low m are transported around by v and
w fields of intermediate m to produce higher wavenum-
ber a fields. This is an effective process, because from
Figure 8, v and w energies peak at intermediate m. From
Figures 8 and 9, at all ma fields are larger than b and c
fileds. Therefore, the shearing produced by intermediate
scale v and w flows on a m are primarily responsible for
maintaining the bm and cm fields. From Figure 11, some
clumping of the radial field is occurring at cell
boundaries, but this is clearly a weaker process, due to
the large amplitude in a. The resulting spectra for b and
c are flatter than for a, because while a drops with
increasing m, v and w increase, out to rather high m.
Finally, since from equation (24) we must satisfy V • Bm
=0, where =am\ + hwÖ+cmr, even a small amount
of induced bm and cm at low m requires a large am,
which is consistent with the result.
In the so-called “a—œ” dynamos, the toroidal field
(our a0) is maintained primarily by shearing of the
poloidal field b0, c0 by the differential rotation u0. The
poloidal field is maintained by the “a effect”, which is
really the sum of our shear and transport processes that
contribute to m = 0, i.e., the sum of the m=0 points on
the Bm and the Cm plots in Figure 11. The a effect is
assumed to contribute little to the maintenance of the
toroidal field in such models. How true is this in our
model? Figure 12 shows the fraction of the total mainte-
nance rate for toroidal magnetic energy A0 and poloidal

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

o
222 GILMAN AND MILLER
l_
0
a Thus, our dynamo is more like an “a2” dynamo, in
0 I 2 3 4 5 6 7 8 9 10 II 12 13 14 15
LONGITUDINAL WAVE NUMBER,m
Fig. 12.—Fractions of the total maintenance of the axisym-
metric toroidal and poloidal magnetic field which are contributed
from induction in each longitudinal wavenumber, for 2=0.2
{upper figure) and 2=0.1 {lower figure) cases. These are sample
averages for the same time period as in Figs. 5 and 6. Contribution
to the maintenance of toroidal field by differential rotation stretch-
ing out the poloidal field is given by the m=0 dot on the toroidal
field maintenance trace. Note that for both Q=0.\ and 0.2, it is
~ 10% of the total.
magnetic energy B0 and C0, for both Q=0.2 and g=0.1,
as functions of longitudinal wavenumber m (spectrum
now folded about m=0). In each case, the maintenance
rate is the sum of that due to shear and transport. We
see that in each case only ~10% of toroidal field
maintenance comes from the shearing of poloidal field
out into the east-west direction due to differential rota-
tion (the m—0 points on the toroidal field maintenance
curves). About 90% comes from similar action due to
periodic east-west motions um shearing periodic radial
and north-south fields cm and bm, together with trans-
port of periodic east-west field am by periodic north-
south and radial motions vm and wm. This combination
is the a effect.
The largest contributions are in general from the
longitudinal wavenumbers of largest velocity amplitude.
Thus, m = 3 contributes the most for g=0.1 to toroidal
field maintenance due to periodic motions, followed by
its harmonics m = 6, 9, and 12. For g=0.2, it is m = 5
which is the biggest contributor. If one splits + and —
wavenumbers, the contribution of shearing of the
poloidal field by differential rotation to the maintenance
of toroidal field is larger than every single nonzero
wavenumber except ±3 or ±5 in the g=0.1 and
g=0.2 cases, respectively, but it is not nearly large
© American Astronomical Society • Provided by the NASA Astrophysics Data System
Vol. 46
enough to say it dominates, as in the a —co dynamos.
which the lifting and twisting due to helical motions is
primarily responsible for both poloidal and toroidal
fields. From Figure 12, it is clear that the periodic
motions dominate in determining the poloidal field, just
as they are assumed to do in the a — w dynamos.
As a further test to demonstrate the reduced role
differential rotation plays in the maintenance of the
toroidal field compared to an a —co dynamo, we per-
formed dynamo calculations for a case in which the
Rayleigh number was raised to the point that equational
acceleration was replaced by deceleration. The resulting
dynamo was similar in virtually all respects to that
displayed here.
It is obvious that in order for our dynamo to act like
an a —to dynamo, and therefore (hopefully) more like
the Sun, the contributions of convective, m^0 veloci-
ties to the induction of toroidal field must be substan-
tially reduced, compared to the m = 0 contribution from
differential rotation. This means that either the convec-
tive velocities needed to drive differential rotation must
be greatly reduced relative to the differential rotation
they drive or that the magnetic field, perhaps because on
the Sun it is in isolated flux tubes, somehow escapes
most of the helicity of the motion, while still
feeling the full impact of the m=0 differential rotation.
With respect to the first possibility, we have so far been
unable to find examples in which the differential rota-
tion driven is much larger than the convection which
drives it. The case we chose here is near the maximum
possible. With respect to the second possibility, pre-
sumably what has to be addressed is the detailed inter-
action between a flux tube and the fluid flow around it.
d) Typical Solution Patterns
The structure and evolution of the magnetic field
patterns in our dynamo solutions is quite complex, even
though the magnetic energy spectrum of Figures 6-9
look rather simple. The reason, of course, is that the
spectrum is broad, so many scales of field contribute to
the total pattern. We show here only a very small
sample, to illustrate. Figure 13 displays typical patterns
of radial motions in the global convection and of the
radial magnetic field at the same time and the same
radial level, namely, near the outer boundary. These
solutions are for g=0.2, at step 8240. Clearly, the
north-south oriented convective rolls, centered on the
equator, are the dominant radial velocity pattern. This is
very typical of hydrodynamic solutions strongly in-
fluenced by rotation. The corresponding radial magnetic
field patterns look quite different. They are smaller in
horizontal scale, which we should expect from the flat
spectrum for radial field, more patchy, and much more
strung out diagonally—suggesting deformation by dif-
ferential rotation. Figure 13 shows a little of that effect
also on the motion fields themselves.
O

LATITUDE

RADIAL MAGNETIC FIELD c (0 = 0.2)

LATITUDE

Fig. 13.—Computer generated contours of radial motion w (upper figure) and radial magnetic field c (lower figure) near the outer
boundary of the spherical shell, at time step 8240 with 0 = 0.2. Solid contours denote outward motion or field, dashed contours inward
motion or field.

Despite the pronounced small-scale structure in the
magnetic field, there appears to be a global ordering as
well. There are clearly two main crescent shaped bands
of magnetic “activity”, one which crosses the equator in
a band between ~60° and 150° longitude, and the
other between ~270° and 330°. These regions occur in
longitude bands where the radial motions, both up and
down, are most intense.
Figure 14 shows solutions for radial velocity and
magnetic field for 0.1. One can see the same sort of
global scale ordering here, too. One can also see that
reduction of the magnetic Prandtl number by a factor of
2 has clearly resulted in even finer structure in the
magnetic field. In fact, the model resolution is really
being pushed to its limit here.
Plots of the total horizontal magnetic field, and the
toroidal field, which we have not attempted to show
here, also show much fine structure and patchiness,
despite the very broad dominant spacial scales in the
velocities which induce them.

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

' 1913lApJS . . . 46 . . 211G

224 GILMAN AND MILLER Vol. 46

RADIAL MOTION w

RADIAL FIELD c (Q = O.I)

0 20 40 60 80 I00 I20 I40 I60 I80 200 220 240 260 280 300 320 340 360
LONGITUDE
Fig. 14.—Same as Fig. 13, but for 0=0.1

e) Feedbacks of the Magnetic Field on the Motion
One of the most important reasons for studying non-
linear dynamos is to find out just how the induced
magnetic fields feed back on the motions which induced
them. We expect such feedbacks to be important in
determining the amphtudes of the solar cycle. Yoshimura
(19786, 1979) has gone further and invoked ad hoc
highly parametrized forms of these feedbacks, including
a time delay in the reaction they produce, to explain
such features as variations in the amplitude of the
envelope of the cycle.
In our own calculations, we have found two kinds of
feedback, one of which depends upon the magnetic
energy reaching a significant fraction of the kinetic
energy of the motion, while the other can be produced
by much weaker magnetic fields, although the effect on
the fluid dynamics in a certain sense is not weak at all.
We display this latter property first.
As we have already demonstrated, dynamo solutions
for 0=0.2 produce a weak magnetic field, of total
energy only a few parts in 104 of the kinetic energy
which induced it, while ß=0.1 gives solutions in which
the magnetic energy is as large as 20% of the total

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

o
C\]
- No. 2, 1981 DYNAMICALLY CONSISTENT NONLINEAR DYNAMOS
kinetic energy. One way to measure the effect these
^ magnetic fields are having on the motion is to plot a
Í time trace of total kinetic energy in each case against the
S time trace of kinetic energy of the parallel solution we
obtained when no magnetic field is present. The re-
sulting time tracks through “phase space” of the two
pairs of solutions are illustrated in Figure 15, for Q—^2
(left) and Q=0A (right), respectively. If the solution
with magnetic field present in each plot felt no feedback
on the motion, then the kinetic energy time trace would
be identical in the two cases, and the result would be a
straight line of unit slope. What we find instead is that
the two solutions in each case do produce such a straight
line from step 2800 when the seed magnetic field was
introduced, to about step 4500. After that time histories
of both solutions with magnetic field present diverge
away from the parallel solution without magnetic field,
resulting in a randomly meandering phase history curve.
The time dependence of the solutions with magnetic
field becomes uncorrelated with that the solution without
magnetic field. Since Figures 15 (left) and (right) also
bear little resemblence to each other, the two solutions
with weak (ß=0.2) and strong (ß=0.1) magnetic fields
are also uncorrelated. One can make similar phase dia-
grams for individual convective modes with the same
result, but with a larger range of amphtude variations.
The same thing occurs in the differential rotation, il-
lustrated in Figure 16 (left) and (right). Here the time
scale for variations is obviously longer (fewer turns in
225
the phase history curve for the same total time span)
than for the total kinetic energy, reflecting the character-
istic “spin up” time for angular momentum changes,
compared to the shorter convective turnover time. This
point was discussed in more detail in Gilman (1978).
What all this means is that even small magnetic fields
can produce small amphtude and phase changes in the
velocity patterns, which are unstable to further change,
resulting in two hydrodynamic solutions, initially identi-
cal, diverging away from each other. Presumably, any
other small force, or even random noise, introduced into
the solution, will have similar effects. This effect does
not arise fundamentally from such limitations as lack of
resolution, but rather from the fact that the fluid dy-
namics itself is unstable to small fluctuations. This prob-
lem has been recognized for almost two decades in
dynamic meteorology (see, e.g., Leith 1978) as providing
fundamental limitations on our ability to make accurate
predictions of global weather patterns.
If one takes a weak magnetic field solution, such as
the one with ß=0.2, and compares the maintenance of
kinetic energy of the flow compared to the parallel case
without magnetic field, one can find substantial numeri-
cal differences. However, these can not be accounted for
directly by the fact that in only one case is work done
against the electromagnetic body force. Calculation of
this work term shows it is too small by perhaps three
orders of magnitude. Instead, the phase histories of the
two solutions are sufficiently different, due to the un-

R = 2.41 x IO4 T = I05 P=l

1.4 1.5 1.6 1.7 1.8 1.9 2.0 2.1 2.2 2.31.4 1.5 1.6 1.7 1.8 1.9 2.0 2.1 2.2 2.3 2.4 2.5
4
TKE WITH MAGNETIC FIELD (X I0 )
Fig. 15.—Phase diagram for the total kinetic energy (TKE) changes with time of two pairs of solutions. The ordinate in each case is the
total kinetic energy from the control solution with no magnetic field present, while the abscissa is the total kinetic energy for (2=0.2 {left)
and 0 = 0.1 (right) solutions with magnetic field present. The numbered dots represent time in hundreds of time steps, and the arrows show
the direction of time advancing. The point of each figure is that the total kinetic energy time histories with and without magnetic field are
uncorrelated with each other, resulting in the random wandering of the curve through phase space. The two solutions with magnetic field are
also uncorrelated.

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

226 GILMAN AND MILLER Vol. 46

R = 2.4IXI04 T = I05 P=l

U0KE WITH MAGNETIC FIELD (X I04)

Fig. 16.—Same as Fig. 15 but for the differential rotation kinetic energy t/0 KE

stable growth of phase differences, that different convec- dissipation curve fluctuates about zero—total kinetic
tive modes dominate in the solution, and different energy is rising when his curve is above zero, falling
amounts of buoyancy work are done. when it is below. One complete rise and fall is a measure
Given this unstable, random character to the feed- of the convective turnover time of a few hundred time
back from weak magnetic fields, it is hard to see how the steps. We see that as the work done against the jxB
feedback could be accurately described as having a force becomes significant, the average value of the
fixed, substantial time delay, as Yoshimura (19786) has buoyancy work dissipation rises above zero to com-
assumed, or even giving a deterministic response at all. pensate. But the turnover time remains about the same,
And even though for ß—0.2 the induced magnetic fields
are quite weak, the feedback on the time history of the R = 2.4lxl04 T = I05 P=l Q=O.I
inducing motion field is prompt, rendering it quickly
uncorrelated with the no field case. We believe this
characteristic to be a very general one for dynamos with
feedbacks. This will make the task of understanding the
effects of such feedbacks in a naturally occurring dy-
namo like the Sun that much more difficult and subtle.
Turning to solutions with ß=0.1, we now examine
the nature of the feedback when the magnetic field is
strong. As we have already illustrated in Figures 15 and
16, the time histories of the solutions with strong field
and no field become uncorrelated with each other after a
few thousand time steps. Thus, the feedback is clearly
not in the form of a “drag”, which reduces each convec-
tive motion by a specified fraction, at least on a time
scale of a few convective turnover times. To see the total
effect of the feedback, we have plotted in Figure 17 the
work done by buoyancy in maintaining the motions,
reduced by the total viscous dissipation. The result is the
sharply fluctuating solid curve. The dashed curve is the Fig. 17.—Time history of work done by buoyancy in maintain-
total work done against the electromagnetic body force, ing the convection minus the viscous dissipation {solid curve),
over the same time period, which is always negative. We compared to that for the work done against the electromagnetic
body force for the case when the magnetic field becomes strong
see that before the field gets strong, the buoyancy- (0=0.1).

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

198lApJS ... 46. .211G

No. 2, 1981 DYNAMICALLY CONSISTENT NONLINEAR DYNAMOS 227

as does the typical amplitude of a convective surge. The
strength of the electromagnetic feedback is much more
regular in time than is the convection, indicating that
even though the magnetic energy is a subsantial fraction
of the kinetic energy, the time dependence of the con-
vection is still governed principally by nonelectromag-
netic processes.
If we examine the work done against the electromag-
netic body force in each longitudinal wavenumber, we
find that it is roughly proportional to the kinetic energy
present in that wavenumber. This is a result of the rather
flat spectrum of the magnetic field with m, which results
in a rather flat spectrum of body force. Thus, modes
with more than average kinetic energy are net sources of
magnetic energy, compared to the ohmic dissipation in
that wavenumber, while modes with below average
kinetic energy are net sinks. Therefore, there must be o TIME STEP (inhundreds)
nonlinear transfer of magnetic energy from “source” [ Meridional Circulation
Work Done by j Eddy Momentum Transports
wavenumbers to “sink” wavenumbers. I— Electromagnetic Body Force
Figure 18 presents a similar sort of plot for the
Fig. 18.—Time trace for the fraction of total work done to
maintenance of differential rotation, except that here we maintain the differential rotation by meridional circulation {solid
plot the fractions of total work done which can be curve), eddy momentum transports {dashed curve), and by electro-
attributed to each process. Here the rate of dissipation magnetic body force {dot-dashed curve), for same case as in Fig.
of differential rotation by viscous diffusion is, by defini- 17.
tion, — 1. The dashed trace fluctuating near +1 is the
work done by Reynolds stresses, transporting angular
momentum in latitude and radius, to maintain differen-
tial rotation. Work by axisymmetric meridional circula-
tion, which is generally a brake on differential rotation, R = 2.4I xI04,T=I05,P=I,Q=0.I,
is given by the much more highly fluctuating solid curve. AVERAGE OVER STEPS II000-H600
The work done against differential rotation by the elec-
tromagnetic body force is rather small, (<0.2), and one
can see only a small tendency for the Reynolds stresses
to rise to compensate for it.
Even though the mathematical form of the “Maxwell
stresses” providing the feedback is identical, except for
sign, with that of the Reynolds stresses driving differen-
tial rotation, their profiles with longitudinal wavenum-
ber m are much different. This is illustrated in Figure
19, which depicts the contribution of each wavenumber
averaged over the time interval 1100-11600 when the
magnetic field is largest. About two-thirds of all the
work done by Reynolds stresses in maintaining differen-
tial rotation comes from a single wavenumber, m=4,
which is the dominant convective mode in this time
period. By contrast, no wavenumber contributes more
than 2% of electromagnetic braking, compared to the
viscous dissipation, although the total braking from all
wavenumbers is —20%. This is clearly due to the rather
flat spectrum of magnetic energy and means the domi- Fig. 19.—Fraction of the total maintenance of differential
nant convective mode still proceeds rather unhindered rotation contributed by eddy Reynolds stresses or momentum
in maintaining whatever differential rotation it chooses. transports in each longitudinal wavenumber m {solid line segments)
compared with the fraction of braking of differential rotation
A good way to illustrate the total feedback effect and contributed by the electromagnetic Maxwell stresses associated
how this bounds the magnetic energy is to simply switch with each wavenumber. Sample average is over the last 600 steps of
off the feedback temporarily, or freeze the motion field. the run with Q—O.l when the magnetic field was largest.

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

198lApJS ... 46. .211G
228
4 5
R*2.4lx I0 T=I0 P=l
< braking by electromagnetic forces is quite enough to
LU
X
No Feedback
Feedback
10"
No Feedback "• /-,—__•/
Feedback
/
y /
I03
/ Motion Frozen
/ .***
101 103 105 107 109 III
TIME STEP (in hundreds)
Fig. 20.—Time traces of several energy statistics of the <2—0.1
dynamo solution, comparing the full dynamo solution with motion
evolving and feedbacks present, to cases when either the feedback
is shut off at time step 10,100, or the motion is frozen at that point.
Vertical scales on the left-hand side refer (reading down), respec-
tively, to the total radial heat flux through the layer (normalized to
its value at the inner boundary), the total kinetic energy TKE, and
the differential rotation kinetic energy U0 KE. Each shows a pair
of curves, with feedback (continuous) and without (dashed). The
vertical scale on the right-hand edge refers to the total magnetic
energy TME represented by the group of three curves near the
bottom of the figure.
Figure 20 illustrates what happens to several statistics of
the solution when this is done in the 2=0.1 solution at
time step 10,100. Near the bottom is plotted the total
magnetic energy for the three cases (amplitude scale on
the right). We can see that the growth of magnetic
energy, when either the motion is frozen or the feedback
is suppressed but the motion is allowed to evolve, is
immediate and strong. Initially the frozen motion case
grows more slowly because the motion has been frozen
at a somewhat lower level, determined by the feedback.
But eventually this case overtakes the no feedback case,
because the evolving motion causes more dissipation of
field, as we illustrated in Figure 2 and discussed in
From the upper plots in Figure 20, we see that with
feedback the total kinetic energy was falling as was that
of differential rotation, for the first several hundred time
steps. Shutting off the feedback largely arrested this fall
and produced a significant rise in differential rotation,
temporarily as large as 50% in energy. But in the top
curves, the total radial heat flux shows similar changes
© American Astronomical Society • Provided by the NASA Astrophysics Data System
GILMAN AND MILLER Vol.46
Q=O.I with time, until the two solutions have evolved away
from each other in phase by time step 11,200 or so.
It is clear from Figure 20 that even the rather modest
keep the magnetic field bounded, at a level somewhat
below equipartition of energy. Presumably, for g <0.1,
the amount of braking required would be larger, and the
magnetic energy would more closely approach equiparti-
\cP tion (or even exceed it). But this possibility can not be
studied without introducing substantially greater resolu-
tion into the model, which is extremely expensive in
terms of computing time, even on a computer as fast as
the CRAY 1A. Even at our present resolution, in order
m to establish what the feedback does to the long-term
I04 average kinetic energy spectrum would require much
longer computer runs than we have tried, perhaps 10-
100 times longer. This would be necessary to average
out the shorter term peaks when a given longitudinal
wavenumber dominates in the convection spectrum.
Judging by the magnetic energy level, and that of the
I03 electromagnetic body forces, the differences in long-term
113 average kinetic energy spectra with field and without are
not likely to be more than 10%-20%. We do not con-
sider such long calculations to establish such modest
differences to be worthwhile at present.
v. DISCUSSION
Our principal conclusion is that global convection
which drives the correct surface differential rotation
profile for the Sun does not give the right dynamo for
the Sun. Why not? It has already been amply dem-
onstrated, e.g., Stix (1976), that a-w dynamos can give
behavior much like the Sun. The induction equation for
a-co dynamos is a particular case of the mean field
dynamo equation (Stix 1976, Moffat 1978):
RR
— = VX(FX5+aB)-ijVX VX5, (1)
at
in which the velocity F is a pure differential rotation, a
is a parameter measuring the combination of lifting and
twisting of the magnetic field by motions not explicitly
resolved, and tj is the turbulent magnetic diffusivity. But
our model does not behave like an a-co dynamo. We
demonstrated this in terms of processes which maintain
magnetic energy in § IVc. Here we show that the reason
is that the effective a our model generates is much too
large. Figure 21 shows a typical profile of differential
rotation generated by the model, with contours nearly
parallel to the axis of rotation in low and middle lati-
tudes. It also illustrates the pattern of helicity, or scalar
product of velocity and vorticity, associated with the
convection. Now Steenbeck and Krause (1969) have
demonstrated that for kinematic dynamos with certain
No. 2, 1981 DYNAMICALLY CONSISTENT NONLINEAR DYNAMOS 229

Differential Rotation Helicity

Fig. 21.—Contours in the meridian (latitude-radius) plane of differential rotation linear velocity, and helicity (scalar product of velocity
and vorticity) for a typical time step in the dynamo solution. Continuous differential rotation contours denote dimensionless values greater
than the rotation of the coordinate system, dashed contours smaller speeds. Continuous contours of helicity denote positive values, dashed
contours negative values.

approximations a is proportional to helicity:
«=-1 F-vXFr,
in which the overbar denotes an average over small-scale
velocity fluctuations and r represents a correlation time
for the velocity field. This correlation time is no shorter
than the turnover time for the convection and could be
somewhat longer. In our solutions, it is certainly no
smaller than say 6X10-2 dimensionless units. From
Figure 21, a typical value for X K in our units is
~ 103. Thus, a(dimensionless)æ20. The quantity a has
dimensional units of velocity, and for the solutions we
have studied, one of our dimensional units is ~4.6 m
s-1, so our a(dimensional)~90 ms-1. From Stix (1976)
we see that in order to get the right reversal period, we
would need a<0.1 m s-1, or three orders of magnitude
smaller. For more persistent global convection, r could
be larger, and the discrepancy increased.
We concede that formula (2) may give an over esti-
mate of the true a since, because our dynamo is nonlin-
(2) ear, magnetic helicity can be generated and at least
partially cancel the “kinetic” helicity represented in
equation (2). Pouquet, Frisch, and Léorat (1976) have
observed this in some of their MHD turbulence calcula-
tions. We have not attempted to estimate the magnetic
helicity in our present model calculations.
One definition of an a-o) dynamo is that the magnetic
Reynolds number Ra associated with a is very small
compared to that associated with differential rotation,
R^. In our notation, suitable definitions are Ra =
a(dimensionless)/ß,Rœ=u0/Q, in which u0 is our di-
mensionless differential rotation linear velocity. So the
ratio Ra/R„æa/u0. In our model this is about unity,
so the stretching by differential rotation and the twisting
and lifting from a compete on virtually equal terms, and
we cease to have an “a-co” dynamo. So, as we said in
§ IVc, either the motions driving the differential rota-
tion must be greatly reduced or the magnetic field must

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

1-98lApJS ... 46 . . 211G
230 GILMAN AND MILLER
escape most of the helicity. With respect to the former
possibility, certainly a compressible model for convec-
tion which drives differential rotation must be explored.
But it seems unlikely even this change could produce a
drop of three orders of magnitude in the helicity. In fact,
we should expect the addition of a large density de-
crease between the bottom and the top of the convecting
layer to enhance helicity—such effects have been in-
voked by Steenbeck and Krause (1969). Furthermore,
such a reduction must come about without reducing the
Reynolds stress below the level needed for maintenance
of the observed differential rotation. This seems quite
difficult to achieve, because helicity and convergence of
momentum flux have the same dimensional units and
are intimately related in motions influenced by rotation.
In this regard, perhaps solutions for lower Prandtl
number P (lower viscosity) would be better, since lower
Reynolds stresses would be needed to overcome eddy
viscous diffusion of momentum. This possibility will be
explored further, even though physical justification for
the low P case is not apparent at the moment. This
might also improve the comparison with observations of
giant cells discussed in § I.
Also, even at a Prandtl number of 1, we need to
explore further solutions at higher Taylor number and
higher Rayleigh number (also lower viscosity) to see if
significantly smaller convection amphtudes in that part
of the parameter range can maintain the needed dif-
ferential rotation. Earlier, cruder calculations were not
especially encouraging on this point, but it is worth
further study. In our earher work, we had seen evidence
of the differential rotation amplitude being self-limiting,
due to feedback of the large shear flow on the convec-
tion which was driving it, at a point when differential
rotation and convection kinetic energies were nearly the
same magnitude. The hydrodynamic solution we used
here was chosen partly because it contained the largest
differential rotation (in the form of equatorial accelera-
tion) compared to convection of any we obtained.
Finally, even if the helicity and therefore a could be
reduced without losing the differential rotation, the
combination of helicity and differential rotation profiles
seen in Figure 21 would lead to migration of the toroidal
field toward the poles throughout the bulk of the con-
vection zone, opposite to the observed butterfly dia-
gram. Consequently, still other changes would have to
take place in the dynamics. Perhaps this will happen in
the compressible case, but early results of compressible
calculations by Glatzmaier and Gilman (1981) are not
particularly encouraging.
We are left with the possibility that the true interac-
tion between velocity and magnetic fields on the Sun has
not been well captured in any of the dynamo models so
far developed (including our own). If this is so, then
much basic theoretical and observational work needs to
be done before more progress in solving the solar dy-
© American Astronomical Society • Provided by the NASA Astrophysics Data System
Vol. 46
namo problem can be expected.
Given this somewhat discouraging point of view, why
have the Yoshimura dynamos, for example, apparently
worked so well for the Sun? We beheve the reason is
that the “regeneration action” or a that Yoshimura uses
is not dynamically consistent with the differential rota-
tion he assumes. Yoshimura (1972) takes some guidance
as to the magnitude of a from perturbation solutions to
the global convection problem. But these solutions are
found in the limit of weak rotational influence, and
hydrostatic balance. Hydrostatic balance filters out the
radial component of Coriohs force, which is important
in determining helicity and a. (Glatzmaier and Gilman
1981 discuss the generation of helicity in global convec-
tion further.) With weak rotational influence, the “re-
generation action” or helicity will necessarily be smaller
for a given convective velocity amphtude. This would be
acceptable if such motions weakly influenced by rota-
tion could actually drive a large-amplitude equatorial
acceleration such as the Sun has. Yoshimura did not
demonstrate that this would happen. In our nonlinear
convection calculations we have repeatedly found they
can not. When rotational influence is weak, whatever
differential rotation is generated is small compared to
the motions which drive it. Thus, if we were to scale up
the differential rotation to solar values, the giant cell
convection velocities would be much larger still—much
larger than the upper limits provided by observations
(Gilman 1980c). Consequently, their helicity or regenera-
tion action would also be much too large. But the
difficulty with low rotation solutions is even worse in
that, in general, it has proved impossible to find finite
amphtude solutions which predict equatorial accelera-
tion at all. They all produce rather weak equatorial
deceleration. This is discussed in Gilman (1980 c) and
previous publications cited therein. Yoshimura (1975)
has gotten around this problem by choosing differential
rotation and regeneration action profiles and especially
magnitudes independently of each other without regard
to dynamical consistency.
One might hope that with the addition of compressi-
bility to the nonlinear convection models these difficul-
ties could also be circumvented, but early calculations
by Glatzmaier and Gilman (1980) suggest that all the
ingredients which lead to these problems are still pres-
ent. Further study is needed.
All of the above arguments and complimentary ones
being made by others, several of which were cited in the
introduction, lead this writer to beheve we are much
further away from a final solution to the solar dynamo
problem than has been previously claimed to be the
case.
David Galloway and Michael Stix kindly read this
lengthy manuscript and responded with many useful
comments. We have modified the text to accomodate.
198lApJS ... 46. .211G

No. 2,1981 DYNAMICALLY CONSISTENT NONLINEAR DYNAMOS 231

APPENDIX

EQUATIONS FOR A DYNAMO DRIVEN BY BOUSSINESQ CONVECTION IN A ROTATING

SPHERICAL SHELL
The basic system of equations we have solved is developed in outline form below, with annotations. Solutions to
these equations without magnetic fields have been discussed in previous papers cited in the text, as has much of the
reasoning behind the formulation we have chosen.
1. Coordinate system: A=longitude; <i>=latitude; r=radius.
2. Variables: m, u, velocities in longitude, latitude, and radius; p = mean density of Boussinesq fluid; a,b,c=
magnetic fields in longitude, latitude, and radius, divided by ^4vrp (so they have dimensional units of velocity);
6 — temperature departure from reference state; 7r=p/p, pressure perturbation/mean density.
3. Physical parameters: kinematic viscosity (constant); /c = thermometric diffusivity (constant); 17 = magnetic
diffusivity (constant); g0= gravity at outer boundary of shell (gravity ~r~2); £2 = rotation rate of coordinate system
(constant); depth of convection zone (constant); /; = radius of inner boundary of spherical shell; A0 = total
temperature difference between bottom and top of convection zone in the reference state; a = coefficient of volume
expansion of Boussinesq fluid (constant).
4. Scale factors to make variables dimensionless: w, v, w: n/d; a, b, c: n/d; r: d\ 6: A©; tt: (n/d)2.
5. Dimensionless parameters resulting from scaling: Rayleigh number R=g0aA®d3/Kv; Taylor number T=
4ü2d4/v2\ Prandtl number P—v/k\ magnetic Prandtl number Q=r]/^l radius of inner boundary =ri/d=ß.
6. Reference state: no motion, hydrostatic, conductive temperature gradient^ —ß(ßJr \)/r2 (scaled relative to A0).
7. Dimensionless continuous equations for spherical shell (written in form convenient for finite differencing).

a) Equations of motion:

Longitude:

du 1 877 — pj1'/! ]/2

w cos <£> + pt v sin <i>
dt rcos<¡> d\

8u 9 , x 1 a2w
+P
ö<i> [r2cos<i> - r dr ™1
rcos<¡> 3r9X

1 3 2, 1 3 , ^,13/ z2x
reos TTT«
3a H rcos<f> wt>cos4>)+
3<f> —-5-(ww
2
r 9r'
)

2 2 2
+ -(i)tan<i>, —w)X
1
3 / z +b +c
7TT-
r rcos<f> 3A \

2
H ^--^-(a¿>cos<í>)+ -^^-(acr
2 K )+-(c—¿>tan^)). (1)
rcos<¡> 3A rœs<t> a<j> r dr ' r

Latitude:

3o 1 9,7 d (du A^d2v 1 82w 1 d2rv

-Pr1/2«sin<i.+Pl 1
~Z~~ I -TT- cos ó IH -
3i r d<}> r2cos2 </> d<j>\dX ) 0X2 r drd<j> r 0r2

1 3 / x , 1 9/2z ^ . 1 9 , 2 , t/2tan<ÿ> vw 1 9 a2-\-b2+c2

rcos<¡>7 dX 7
reos <¡> —
d<j>x(o cos<i>)4-—-
J 2 —x
r dr
(rW)7 r r r d<¡> 2

1
+ 9 yab)+
/ L\ i 1 7 7t(6
9 /l2cos<M+
^ , ——(r
1 9 2
2l ^ , a2tan<i> , be •
fo)H
v (2)
rcosÿ 9 A rcos<j> 9<i> -2 d<}> /

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

198lApJS ... 46. .211G

232 GILMAN AND MILLER Vol. 46

Radius:

3w 9w ß+\
PR6+PTi/2ucos<t>
"37 +

1 3 1 92w , 1
3 dw 9 / x
2
+ 2 2 2 H ;
2 COS <t>
r cos<j> \ 9X / r cos <j> 9A r cos<¡> d<¡>

u2 +ü2
rcos<j> 9A■(
vWH;)J+T^T^:(t;vvcos<i>)+"7^:(r ^ ) + -
rcos<l> d<¡> d<¡>

3 (a2+b2+c2) i a , 1 3 ^ , 1 9 ^ 2 2^ û2+è2
1 rTracH ^ïï7(fcccos^)+_2Jä_(,'c ) r (3)
3r 2 rcos</> 3\ rcos<¡> 3<¡> r dr

b) Thermodynamic equation:

afl_;8(ft+l) 1 3 / 23g\ , 1 3 / a9Ö\, 1 32ô

w+
3i r2 P 3r T dry r
2 C S 2
co¡ 2<t> dX2
cos * 9<i> l ° ^ 9<i» J r cos

t> cos< > 7 (4)

—^“7
rcos<i> 9a v 7 ^¡r(
rcos<i> 9<i>v ^ í )"^ “7
2
r 9r
v 7

(Frictional and joule heating are ignored—consistent with Boussinesq approximation and reasonable bounds on
magnetic field strength.)

c) Mass continuity equation:

1 , 1 t)COS<
9 ,> + _, r1 w9 / 2 N .
(5)
rcos<í> 9X 7^T(
rcos<j> d<j> i ) ~7ö (r2 9r)“°-
v

J) Electromagnetic induction equations (written in form convenient for finite differencing):

Longitude:

9a 1 9 2/ 1 9 ,/ L\ tan<|> ^
Tt=^Yrr{uC~Wa)~~r{uC~Wa)~7^^C°^Va~ub)

3 1 9
^ , 9&
+ I_91/' x 1 92c
+Ô 3î(‘ C“f)-3X (6)
2
9<í> r cos<J> r 9r2 ra rcos^> 3r3X

Latitude:
19
9i> 1 3 / , , N , A , x
—- = —(va—ub) —r¿2(wb-vc)+ -(wb-vc)
3i rcos<j> d\K ' r dr y
2
' rK

d2b d ( da i\ 92c
+Ô 2 2 2 (7)
r cos (/} r %r ^ rdrd<f>

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

198lApJS ... 46. .211G

No. 2, 1981 DYNAMICALLY CONSISTENT NONLINEAR DYNAMOS 233

Radius:

3c _ 1 3 1
cos<J>(w¿>—t>c)- (uc—wa)
3r rcos<f> 3<i> rcos<i> 3\

1 32c 1 3 1 3
+Ô 2 2 2 2 2
cos<i>
r cos <¡> 3A r cos<j> dr r cos<¡> d<¡>

e) Continuity of magnetic flux:

1 da 1 9
+ Ocos<i>)+-^-^(r2c)=0. (9)
rcos<i> 3 À rcos<f> d<t>

8. Boundary conditions.
a) Velocities: (i) top, bottom (stress free, no penetration):

_9_ _9_
=0; (f)=0; -=0. (10)
3r 3r

ii) Polar boundaries (stress free, no penetration):

_3_
=0; o=0. (11)
d<l>

b) Temperature: (i) top: constant temperature, 0=0.

30
Ü) Bottom: constant heat flux, dr =0.

30
in) Polar boundaries: perfect insulators, — =0. (12)

c) Pressure: at all boundaries, derived from component of equation of motion normal to the boundary.

d) Magnetic field:
i) Top: perfect electrical conductor or radial field,

j^(ar)=0’ i;(br)=0’ c=0; (13a)

or

a,b=0, -^(r2c)=0. (13b)

ii) Bottom: perfect electrical conductor,

-j^(ar)=0, yr{br)=0, c=0. (14)

in) Polar boundaries: perfect electrical conductor,

^(acos<i>)=0, b=0, |^=0. (15)

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

198lApJS ... 46. .211G

234 GILMAN AND MILLER Vol. 46

9. Fourier analyzed, finite difference equations actually solved (differenced in a form to retain energy conservation
properties of continuous equations; diffusion terms are lagged one time step for computational stability).

a) Fourier expansion of any dependent variable F\

1 fm(<t>,r,t)eimX.
F=
m= — oo
b) Finite difference operators used, for any independent variable jc:

Sum operator: /x= | /(x+^ j|.

Difference operator: Sx/=^ /(*+^)-/(x-^)j.

Finite difference grid staggered in meridian plane, according to layout in Figure 22.

c) Equations of motion:

Longitude:
[1] P] [3]

S'û'm = - +f>T1/2tan<í>ümcos<í>'f>—.PT’1/2^^ r2wm

+P +
M ^(w"lCOS<#>)]} + ~r8jruJ~7Ú^imSrW" t-Lt
[7]

+ 2 ]—[imunum^n+S^utvm_ncos<t>)]-\dr(urnr2wm^n)+-^-unvm_ncos<i>',,-r^u„wm^nr2
rcos<i> reos <t>
[5]
im a a
+ 2 2rcos<i) ( n m-n +^m-n)

[9] [10] [11]

1 r 2
+- imanam_„ +S<l,(atbm^ncos<l>)+ \sr(a nr cm_n) + y ( </>)[• (16)
rcos<f>

Fig. 22.—Finite difference staggered grids in meridian plane on which solutions for the various components of velocity and magnetic
field, as well as temperature and pressure, are obtained. Grids for different variables are staggered with respect to each other for purposes of
economy of computation, as well as conservation of energy and other quadratic quantities by the nonlinear terms in the equations of motion,
thermodynamics, and electromagnetic induction.

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

198lApJS ... 46. .211G

No. 2,1981 DYNAMICALLY CONSISTENT NONLINEAR DYNAMOS 235

Latitude:
i [1] [2]
/2
8Ä = - 7r Vm ~PT' wmsm<f»<> + pj - 1
[imumcos+ 5r<f>vwm + -^5rr() [ ( 17)
rzcoszóL J
r r t-At

+
^ Ii _ reos 9L ^ t,--«cos * * )] - V—
r cos<¡>
cos
)

[6] [8]
r<j>
, ,, 2
7r «nMm-ntan<í ' -4ünW'm-
r* n''

[5]
+ 2 {- j-r +fe„-^n +cÂ_„)]

[12]
1
+- imatbm_„ +bn84, bm_„cos~^-8^ b„b„^n '
rcos<j>

[12] ^ [13] [14]

r 2 ,> r
+ ~r—ô r(b- r cm^ncos4> ’) + -\b
rL nc *_n + JanJJ
rzcosé

Radius:
[1] [4] [3]
(ß+1)
+ - 9- p/t0-r+prl/2ürcos<t>+p im8r(nim)-
r cos<i> /•2cos2 <P

1
+ ^ [cos <i) ( - 5rrom )]
rz2cos<t> t-At

+ 2 i -T^T['WrM«W’m-«+5<.(W-f«m-«''rCOS,í>)]
„=-<,0 [ r2cos<i>L
[7] [8]
r<j>)
2 r
^8r(w¿r Wm_n )+ ~unum—n +^~cos<Kom_„

[5]

+ 2 \-\8r{a„am_n+b^b^n+crncrm_n)
[15]

im ra
+ 'T~\ +5<>( C$ rb m-nCOS <^)]J + \ C„8r r2C m-n+8r ^
rzcos<¡)L rz ¿

- 7¿^ [cos</>a„am_„ + cos <£. bn bm _ fl r^] l. (18)

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

1981ApJS'. . . 46 . . 211G

236 GILMAN AND MILLER Yol. 46

d) Thermodynamic equation:
[4]
ß(ß+1)
±8r(rXem)+-T !--8,(cos<l>0,6m)--I2 ?^—0, m
r2cos</> r cos 2 </)± t—At
2 rv
r rw (19)
+ Í ( “ 7~TT +^(öm ^.„cos<#>)] - -jô
r r(0„ ,• m—r
n= — oo ^ ^

e) Mass continuity equation:

imum , 1
+ ü COS < , +
- f ) -TÔr(r2wm) = 0. (20)
reos# rcos<f>

/) Induction equations:

Longitude:

Vm=ô +
^(amCOS<í,)]} + 7^(ra'”)_ t—At
[9]
00
1 1
+ _2 |■^Sr[/■2(M;Cm_„-ä>m_^)]--|;^^^[cOS^(â*üm-«-M^m-n)]

[10] í16] [14] [11]

“nCm-n , sin^>t?n am_„ | tan<]>Kn^_„ (21)
r r2 rcos<l> r

Latitude:

Radius:

C ..
S'C^Ql-^-SXraJ-^ m + z¿ ~8<t>[COS<t>(8<f>Cm ~8rrbm] !
I rzcos<f> rzcos <J> r cos<¡> t — At

r
+ 2 2
r^[cos<í)(rh„
L v
w^_„-ro„rc*_„)]-iw(rM/cm_„-ra„rwm_„)l. (23)
n =-oo lr cos<¡> 'J y yj

g) Continuity of magnetic flux:

+
~^>8*(b”'COSri+^8r(r2c'n)=0- (24)
rcos<¡>

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

No. 2,1981 DYNAMICALLY CONSISTENT NONLINEAR DYNAMOS 237
10. Boundary conditions.

a) Velocities: i) (top, bottom) j =0, wm =0;

ii) (polar boundaries) 8^|j = 0, 8^vm =0, vm= 0.

b) Temperature: i) (top) ömr=0,

ii) (bottom) 8r6m =0,

iii) (polar boundaries) =0.

c) Magnetic field: i) (top) 8r{amr)=Q, 8r(bmr)=Q, cm=0; or amr, bmr=0, 8r(r2cm)=0

ii) (bottom) 8r(amr)=0, 8r(bmr)=0, cm = 0

iii) (polar boundaries) a mcos <¡> )=0, bm=0, 8<f>cm = 0

With the above finite difference forms, there are a large number of terms in different equations which cancel in pairs
or groups of three when the finite difference equivalents of kinetic, magnetic, and thermodynamic energy integrals are
derived and perfectly electrically conducting boundaries are assumed. These energy equations are obtained by
multiplying the prediction equations (16)-(19), (21)-(23), for the velocities, temperature, and magnetic field variables
by the complex conjugate of the respective variable being predicted and then averaging each equation to the same finite
difference grid, namely, the one on which u, a, 6,77 are calculated. For equations (17) and (22), for v and b,
respectively, the averaging operator is (cos <i>) “1 (cos <¡> )<i>, while for equations (18) and (23), for w and c, respectively,
it is r~2(r2 y. The thermodynamic equation must also be multiplied by the scale factor /?-1(/?+ l)PR. The resulting
finite difference integrands are then summed over the whole meridian plane, weighted by the finite difference element
of integration lirr2cos<¡> ArA<¡>. The terms in the equations which cancel in this way are denoted with the same
numbers in square brackets written over the terms. In total there are 16 pairs and one group of three. These represent a
powerful check on the accuracy of the computer code, since each pair separately must cancel to machine round off. In
addition, the nonlinear inertial terms in flux divergence form in the equations of motion cancel within each equation
when summed over all m.
In the actual solution procedure, for the hydrodynamic part, um,vm,wm,0ni are advanced in time using equations
(16)-(19). Then a new pressure 7rm is found from a Poisson-Helmholtz-type equation for each m obtained by taking
the divergence of the vector finite difference equations of motion (16)-(18) and assuming the velocity divergence at the
new time step vanishes. The pressure equation for each m is then solved by direct matrix inversion.
To advance the magnetic field, we use a somewhat different procedure for the m=0 (axisymmetric) variables than
for m=£0. For m—0, we advance a0 and c0 in time by solving equations (21) and (23), respectively. Then b0 is found by
solving the continuity equation (24). It can be shown that the fields obtained in this way satisfy the boundary condition
=0 at both polar boundaries, even though in solving equation (24), only one boundary condition may be applied to
b0 because it appears differentiated only once in latitude. This can be proven by applying the operator (1 /r2)8rr2( )
to equation (23), and then substituting from equation (24). For ra^O, we solve equations (22) and (23) for new values
of bm,cm, respectively, and then find the new am from equation (24). Equation (21) for am is then used as a check.
When the boundaries are assumed to be perfect electrical conductors, exact energy conservation is obtained for all the
nonlinear induction terms, with corresponding terms in the equations of motion, as marked by pairs of equal numbers.

There may be other finite difference representations which retain the same properties, but the one we have chosen
took the greatest advantage of the already existing computer code.

REFERENCES
Babcock, H. W. 1961, Ap. /., 133, 572. Busse, F. 1973, Astr. Ap., 28, 27.
Belvedere, G., Paterno, L, and Stix, M. 1980, Astr. Ap., 86, 40. Dumey, B. R. 1970, Ap. J., 161, 1115.
Busse, F. 1970, Ap. J., 159, 629. . 1971, Ap. /., 163, 353.

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

238 GILMAN AND MILLER
Frisch, V. 1977, in IAU Colloquium 38, Problems of Stellar Convec-
tion, ed. E. A. Spiegel and J. P. Zahn (New York: Springer-
Verlag), p. 325.
Frisch, V., Sulem, P. L., and Nelkin, M. 1978, J. Fluid Mech., 87,
719.
Galloway, D. J., and Weiss, N. O. 1981, Ap. J., 243, 945.
Gilman, P. A. 1972, Solar Phys., 27, 3.
. 1975, Atmos. Sei., 32, 1331.
. 1976, in IAU Symposium 71, Basic Mechanisms of Solar
Activity, ed. V. Bumba and J. Kleczek (Dordrecht: Reidel), p.
207.
. 1977, Geophys. Ap. Fluid Dyn., 8, 93.
. 1978, Geophys. Ap. Fluid Dyn., 11, 157.
. 1979, Ap.J., 231, 284.
. 1980 a, in press.
. 19806, Highlights of Astronomy (Dordrecht: Reidel), 5, 91.
. 1980c, Geophys. Ap. Fluid Dyn., 14, 252.
Gilman, P. A., and Foukal, P. V. 1979, Ap. J., 229, 1179.
Gilman, P. A., and Glatzmaier, G. A. 1980, Ap. J., 241, 793.
Glatzmaier, G. A., and Gilman, P. A. 1981, Ap. J. Suppl, in press.
Golub, L., Rosner, R., Vaiana, G. S., and Weiss, N. O. 1981, Ap.
J., 243, 309.
Gough, D. O., Moore, D. R., Spiegel, E. A., and Weiss, N. O. 1976,
Ap. J., 206, 536.
Howard, R., and LaBonte, B. J. 1980, Ap. J., 239, 738.
Krause, F. 1976, 'mlAU Symposium 71, Basic Mechanisms of Solar
Activity, ed. V. Bumba and J. Kleczek (Dordrecht: Reidel), p.
305.
LaBonte, B. J., Howard, R., and Gilman, P. A. 1981, in prepara-
tion.
Peter A. Gilman and Jack Miller: High Altitude Observatory, National Center for Atmospheric Research, Box
3000, Boulder, CO 80307
© American Astronomical Society • Provided by the NASA Astrophysics Data System
Layzer, D., Rosner, R., and Doyle, H. T. 1979, Ap. J., 229, 1126.
Leith, C. E. 1978, Ann. Rev. Fluid Mech., 10, 107.
Moffatt, H. K. 1978, Magnetic Field Generation in Electrically
Conducting Fluids (Cambridge: Cambridge University Press).
Parker, E. N. 1955a, Ap. J., 121, 491.
. 19556, Ap. J., 122, 293.
. 1979, Cosmical Magnetic Fields (Oxford: Clarenton).
Piddington, J. H. 1975, Ap. Space Sei., 38, 157.
. 1976, m IAU Symposium 71, Basic Mechanisms of Solar
Activity, ed. V. Bumba and J. Kleczek (Dordrecht: Reidel), p.
389.
Pouquet, A., Frisch, V., and Leorat, J. 1976, J. Fluid Mech., 77,
321.
Pouquet, A., and Patterson, G. S. 1978, J. Fluid Mech., 85, 305.
Roberts, P. H. 1972, Phil. Trans. Roy. Soc. London, A, 272, 663.
Roberts, P. H., and Stix, M. 1972, Astr. Ap., 18, 453.
Roxburgh, I. W., and Tavakol, R. K. 1979, Solar Phys., 61, 247.
Simon, G. W., and Weiss, N. O. 1968, Zs. Ap., 69, 435.
Steenbeck, M., and Krause, F. 1969, Astr. Nach., 291, 49.
Stix, M. 1976, 'm IAU Symposium 71, Basic Mechanisms of Solar
Activity, ed. V. Bumba and J. Kleczek (Dordrecht: Reidel), p.
361.
. 1981, Solar Phys., in press.
Vickers, G. T. 1971, Ap. J., 163, 363.
Wilhams, G. P. 1969, J. Fluid Mech., 37, 727.
Yoshimura, H. 1972, Ap. J., 178, 863.
. 1975, Ap. J. Suppl, 29, 467.
. 1978a, Ap. J., 220, 692.
. 19786, Ap. J., 226, 706.
. 1979, Ap. J., 221, 1047.