physics0602147/ms.tex
1: \documentclass[twocolumn,showpacs,pre,amsmath,amssymb,floatfix]{revtex4}
2: %\documentclass[preprint,showpacs,pre,amsmath,amssymb,floatfix]{revtex4}
3: 
4: \usepackage{graphicx}% Include figure files
5: \usepackage{dcolumn} % Align table columns on decimal point
6: \usepackage{bm}      % bold math
7: \usepackage{latexsym}
8: \usepackage{color}
9: 
10: % BIBLIOGRAPHY %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
11: %\usepackage{natbib}
12: %\bibliographystyle{phfl}
13: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
14: 
15: % COLOR %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
16: \def\note#1{{\textcolor{red}{\bf [#1]}}}   % note
17: \def\add#1{#1}                             % addition
18: \def\ad#1{#1}                              % addition
19: %\def\add#1{{\textcolor{blue}{#1}}}         % addition
20: %\def\ad#1{{\textcolor{magenta}{#1}}}       % addition
21: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
22: 
23: %-My commands--------------------------------------------
24: \newcommand{\vomega}{\mbox{\boldmath $ \omega $}}
25: \newcommand{\vmu}{\mbox{\boldmath $ \mu $}}
26: %--------------------------------------------------------
27: 
28: %\topmargin -3pt
29: \begin{document}
30: 
31: \title{Magnetohydrodynamic activity inside a sphere}
32: 
33: \author{Pablo D. Mininni$^1$ and David C. Montgomery$^2$}
34: \affiliation{$^1$ National Center for Atmospheric 
35: Research, P.O. Box 3000, Boulder, Colorado 80307}
36: \affiliation{$^2$ Dept. of Physics and Astronomy,
37: Dartmouth College, Hanover, NH 03755} 
38: 
39: \date{\today}
40: 
41: \begin{abstract}
42: \add{We present a computational method to solve the magnetohydrodynamic 
43: equations in spherical geometry. The technique is fully nonlinear and 
44: wholly spectral, and uses an expansion basis that is adapted to the 
45: geometry: Chandrasekhar-Kendall vector eigenfunctions of the curl. The 
46: resulting lower spatial resolution is somewhat offset by being able to 
47: build all the boundary conditions into each of the orthogonal expansion 
48: functions and by the disappearance of any difficulties caused by 
49: singularities at the center of the sphere. The results reported here 
50: are for mechanically and magnetically isolated spheres, although different 
51: boundary conditions could be studied by adapting the same method. The 
52: intent is to be able to study the nonlinear dynamical evolution of those 
53: aspects that are peculiar to the spherical geometry at only moderate 
54: Reynolds numbers. The code is parallelized, and will preserve to high 
55: accuracy the ideal magnetohydrodynamic (MHD) invariants of the system 
56: (global energy, magnetic helicity, cross helicity). Examples of results 
57: for selective decay and mechanically-driven dynamo simulations are 
58: discussed. In the dynamo cases, spontaneous flips of the dipole 
59: orientation are observed.}
60: \end{abstract}
61: 
62: \pacs{47.11.-j; 47.11.Kb; 91.25.Cw; 95.30.Qd}
63: \maketitle
64: 
65: \section{\label{sec:intro}INTRODUCTION}
66: 
67: Magnetohydrodynamic ``dynamo'' processes are those in which the motions 
68: of an electrically conducting fluid amplify and maintain a finite magnetic 
69: field, starting from an arbitrarily small one. They have long been of 
70: interest for geophysics and astrophysics 
71: \cite{Moffatt,Glatzmaier95,Dikpati99,Kono02,Nandy02,Mininni04}, and have 
72: more recently become of interest with regard to laboratory attempts to 
73: generate dynamo magnetic fields in liquid metals 
74: \cite{Gailitis01,Steiglitz01,Noguchi02,Petrelis03,Sisan03,Spence05}.
75: The relevant theoretical and computational literature is vast, and 
76: extensive reviews have recently been given (see e.g. Refs. 
77: \cite{Roberts01,Kono02,Brandenburg05}).
78: 
79: In our own work, we have lately been studying dynamo processes numerically, 
80: using turbulent three-dimensional magnetohydrodynamic (hereafter, ``MHD'') 
81: codes of the familiar Orszag-Patterson pseudospectral variety 
82: \cite{Ponty05,Mininni05a,Mininni05b,Alexakis05,Mininni05c}. Such codes 
83: treat homogeneous turbulence efficiently, but are mainly useful in 
84: situations involving spatially periodic boundary conditions. Particularly 
85: for the case of planetary dynamos and laboratory experiments, restrictions 
86: to periodic boundary conditions are a severe limitation. Essential 
87: ingredients, such as rotation, global angular momentum, and the 
88: interfaces between conducting and non-conducting regions are not 
89: readily included. It is to be expected that all of these play a 
90: role in the physical situations of interest, and give rise to 
91: qualitatively new physical processes not accessible with periodic 
92: boundaries.
93: 
94: This present paper represents our attempt to begin a study of these 
95: processes by introducing, and displaying some results from, a 
96: computational method that is adapted to the geometry of isolated 
97: spheres. Our goal is not to reach realistic geophysical parameter 
98: regimes (out of the question for the foreseeable future, in any case), 
99: but rather to isolate and study those new physical processes that appear 
100: in this geometrically more realistic setting. \add{Our results are not 
101: to be compared with the ambitious geo-dynamo computation of Glatzmaier 
102: and Roberts (e.g., Refs. [2,21]) who include far more physical effects} 
103: \ad{and degrees of freedom} \add{than we have at this point. We will 
104: consider it not to be a limitation on our work if we are unable to reach 
105: realistic Reynolds numbers, so long as the spectra of the dynamo 
106: processes we do identify are well-resolved by the number of expansion 
107: functions we are able to retain. We think of these efforts as studies 
108: of dynamo behaviors exhibited by the MHD equations in spherical geometry 
109: without regard to their present applicability to geophysical or laboratory 
110: situations. It appears not to be necessary to reach presently-existing 
111: ranges of Reynolds numbers or magnetic Prandtl numbers in order to see 
112: interesting spontaneously-generated magnetic field behavior.}
113: 
114: \add{The numerical technique is entirely spectral, using an expansion 
115: basis that is specifically adapted to spherical geometry: the 
116: Chandrasekhar-Kendall (hereafter, ``C-K'') vector eigenfunctions of 
117: the curl \cite{Chandrasekhar57,Turner83,Yoshida91,Yoshida92,Cantarella00} 
118: (by construction, C-K functions are also eigenfunctions of the Helmholtz 
119: wave equation; see e.g. \cite{Mueller}). These functions form a complete 
120: orthogonal set under the boundary conditions we will choose. They can 
121: be normalized. Their completeness has been shown by Yoshida for the 
122: cylinder \cite{Yoshida92} and by Cantarella et al \cite{Cantarella00} for 
123: the sphere. The cylindrical version was used some years ago for studying 
124: processes believed to be involved in the disruptions of fusion confinement 
125: devices \cite{Montgomery78,Shan91,Shan93a,Shan93b,Shan94}. Their advantages 
126: lie in their natural geometrical relation to the specific geometries in 
127: which they are employed (all the boundary conditions can be built into 
128: each expansion function) and in certain desirable numerical properties, 
129: which include the following. The MHD equations involve several solenoidal 
130: fields of which several curls are taken. Taking the curl of a C-K function 
131: simply gives the function back again, times a multiplicative constant. 
132: This means that no numerical spatial differentiations are required, with 
133: their attendant complication of the expressions and loss of accuracy. The 
134: code preserves the ideal MHD invariants very accurately over long times 
135: (total energy, magnetic helicity, cross helicity), which is one of the 
136: few accuracy checks available to a strongly nonlinear code for which 
137: non-trivial analytical solutions are scarce. In addition, the ideal 
138: invariants are readily exhibited as simple algebraic quadratic sums 
139: involving the expansion coefficients, so no numerical integrations are 
140: required to evaluate them. Going along with these advantages is a severe 
141: disadvantage associated with the lack of a fast transform for the 
142: spherical Bessel functions involved, which means that the nonlinear terms 
143: lead to convolution sums which grow rapidly with the number of expansion 
144: functions retained in the Galerkin approximation. This limits us to 
145: mechanical and magnetic Reynolds numbers of the order of a very few 
146: thousand, and makes it unlikely that we can ever reach planetary parameters 
147: without modeling the small scales of the fluid motions.}
148: 
149: \add{But the absence of any coordinate singularities (e.g., $r=0$ in 
150: spherical polar coordinates) to worry about anywhere is a considerable 
151: advantage, as we noted previously in solving the two-dimensional 
152: Navier-Stokes equation inside a circle with non-ideal boundaries 
153: \cite{Li96a,Li97}. The code also can be readily parallelized and there 
154: are no potential aliasing problems.}
155: 
156: \add{The situation we wish to study is that of an electrically conducting 
157: fluid inside a rigid spherical boundary that isolates the fluid, 
158: mechanically and magnetically, from everything outside it. The boundary 
159: is regarded as a spherical shell that is perfectly conducting (so 
160: magnetic fields cannot penetrate the region outside) and is coated on 
161: the inside with a thin layer of insulating dielectric so that the 
162: electrical current density cannot penetrate the shell. The shell is 
163: regarded as mechanically impenetrable, so that the normal component of 
164: the fluid velocity vanishes there. We do not employ no-slip boundary 
165: conditions on the velocity field, but rather choose the vanishing of the 
166: normal component of the vorticity at the spherical boundary as the second 
167: boundary condition on the velocity field. This condition is implied by, 
168: but does not imply, no-slip boundary conditions on the velocity field, 
169: and thereby avoids certain mathematical procedures that sometimes attend 
170: the attempts at imposing no-slip boundary conditions.} \ad{These four 
171: boundary conditions for the fields in the external walls, plus regularity 
172: of each component of the velocity and magnetic fields at the origin, might 
173: be thought of as a total of ten boundary conditions for the system. The 
174: above mentioned boundary condition for the vorticity in the wall is easy 
175: to implement in the present formulation. Other boundary conditions (e.g. 
176: no-slip boundary conditions) can be studied with our method using a 
177: penalty method close to the walls (as done for example in Ref. 
178: \cite{Shan94}) and will be considered in the future. However, we want 
179: to point out that the present election of boundary conditions also 
180: avoids some problems present in hydrodynamic incompressible flows 
181: when no-slip conditions are imposed (see e.g. \cite{Kress00,Gallavotti} 
182: for a detailed discussion). Since this topic is beyond the aim of our 
183: present study, we will use in the following these simple boundary 
184: conditions and leave the study of other options for future work.}
185: 
186: \add{An outline of the paper follows. Section \ref{sec:equations} 
187: introduces the expansion basis and shows how the nonlinear partial 
188: differential equations can be reduced to a set of ordinary differential 
189: equations, first order in the time, for the expansion coefficients. We 
190: include mention of other possible boundary conditions that it is 
191: intended to introduce in the future, then settle on the one just 
192: described for these runs. The numerical method is discussed in Section 
193: \ref{sec:numerical}. Sections \ref{sec:application1} and 
194: \ref{sec:application2} display our first applications of the code to 
195: some simple MHD problems that are considered interesting and that 
196: are peculiar to spherical geometry. We simultaneously explore 
197: the possibilities for some relative new flow visualization diagnostics 
198: when exhibiting our results \cite{vapor}; these are described when 
199: they appear. Section VI summarizes the results and discusses possible 
200: future applications of the method.}
201: 
202: \section{\label{sec:equations}THE SPECTRAL DECOMPOSITION}
203: 
204: C-K functions are constructed from a solution of the scalar Helmholtz 
205: equation:
206: \begin{equation}
207: \left( \nabla^2 + \lambda^2 \right) \psi = 0 ,
208: \label{eq:helmholtz}
209: \end{equation}
210: where we are referring to spherical polar coordinates $(r,\theta,\phi)$ 
211: and $\lambda$ is an eigenvalue that will eventually be determined by 
212: boundary conditions. Vector eigenfunctions of the curl appropriate to 
213: spherical geometry may be constructed according to the \add{following} 
214: recipe \add{from each solution to Eq. (\ref{eq:helmholtz})}:
215: \begin{equation}
216: {\bf J} = \lambda \nabla \times {\bf r} \psi + \nabla \times \left( 
217:     \nabla \times {\bf r} \psi \right) .
218: \end{equation}
219: From this, it may readily be verified that 
220: $\nabla \times {\bf J} = \lambda  {\bf J}$. Thus any single ${\bf J}$ 
221: is a ``force-free'' or ``Bernoulli'' field, though the sum of two or 
222: more of them is not. The relevant scalar $\psi$ is
223: \begin{equation}
224: \psi_{qlm} = C_{ql} \, j_l(|\lambda_{ql}| r) Y_{lm} (\theta,\phi) .
225: \label{eq:psi}
226: \end{equation}
227: Here, $C_{ql}$ is a normalization constant, $j_l$ is a spherical Bessel 
228: function of order $l$, and $Y_{lm}$ is the normalized spherical harmonic 
229: expressed in terms of the polar angle $\theta$ and the azimuthal 
230: (longitudinal) angle $\phi$. The number $C_{ql}$ is chosen to make 
231: the volume integral of ${\bf J}_{qlm} \cdot {\bf J}_{qlm}^*$ over the 
232: computational domain equal to unity (the asterisk denotes complex 
233: conjugate).
234: 
235: The integer $l$ is $1,2,3,\dots$ and $m$ runs in integer steps from $-l$ 
236: to $l$. An infinite sequence of values of $\lambda_{ql}$, labeled by 
237: $q=1,2,3,\dots$ and $q=-1,-2, -3,\dots$ \add{will be determined by a 
238: radial boundary condition momentarily to be invoked} on ${\bf J}$ for 
239: a given $l$.
240: 
241: \add{The boundary conditions that will be invoked, consistently with the 
242: physical description of the region and its boundaries that were given in 
243: Sec. \ref{sec:intro} refer to the magnetic field ${\bf B}$ and the 
244: velocity field ${\bf v}$, both solenoidal, and their curls. The curl of 
245: ${\bf v}$ is $\vomega$, the vorticity, and the curl of ${\bf B}$ is 
246: ${\bf j}$, the electric current density, in the dimensionless units that 
247: remain to be described.  We shall demand that the normal (radial) 
248: components of all four vector fields vanish at the boundary $r=R$. Since 
249: these four fields are to be expanded in the ${\bf J}$'s, making the 
250: radial component of each ${\bf J}$ vanish at $r=R$ will guarantee that 
251: the boundary conditions will be satisfied for any superposition of them. 
252: This vanishing of the normal components of the ${\bf J}$ determines the 
253: allowed values of $\lambda_{ql}$, positive and negative, by the locations 
254: of the zeros of the spherical Bessel functions. Then the explicit 
255: expression for the normalization constant $C_{ql}$ is}
256: \begin{equation}
257: C_{ql} =  \left|\lambda_{ql} \, j_{l+1}(|\lambda_{ql}|R) \right|^{-1} 
258:     \left[ l(l+1) R^3 \right]^{-1/2} .
259: \label{eq:normalization}
260: \end{equation}
261: Positive or negative $q$ is to be associated with positive or negative 
262: $\lambda_{ql}$ according to $\lambda_{-q,l} = - \lambda_{ql}$. For a 
263: variety of radial boundary conditions, \add{including the one just chosen,} 
264: the ${\bf J}_{qlm}$ functions corresponding to differing indices $q$, $l$, 
265: or $m$ can be shown to be orthogonal. Boundary conditions at two different 
266: radii can be imposed if spherical Neumann functions are also permitted in 
267: $\psi$.
268: 
269: \add{Given} the completeness of the C-K functions so defined, they are then 
270: appropriate for expanding various vector fields of incompressible MHD: 
271: the magnetic field, fluid velocity, vorticity, electric current density, 
272: and vector potential in the Coulomb gauge. \add{These are considered to 
273: be the necessary set of quantities to be expanded when studying spherical 
274: MHD and dynamos in the class of problems studied here.} The ones of 
275: slowest spatial variation (smaller $|q|$, low $l$ and $|m|$) contain the 
276: dipole and low-order multipole components of the fields and provide a 
277: natural ordering of the spectral contributions from various spatial scales, 
278: \add{starting with those of the largest scales.}
279: 
280: The MHD equations to be solved are, in familiar dimensionless 
281: (``Alfv\'enic'') units, \add{common in MHD turbulence theory,} 
282: an equation of motion for the fluid velocity 
283: ${\bf v}$,
284: \begin{equation}
285: \frac{\partial {\bf v}}{\partial t} = {\bf v} \times \vomega + 
286:     {\bf j} \times {\bf B} - \nabla \left({\mathcal P} + \frac{v^2}{2} 
287:     \right) + \nu \nabla^2 {\bf v} + {\bf f} , 
288: \label{eq:momentum}
289: \end{equation}
290: and the induction equation for advancing the magnetic field ${\bf B}$,
291: \begin{equation}
292: \frac{\partial {\bf B}}{\partial t} = \nabla \times \left( {\bf v} 
293:     \times {\bf B} \right) + \eta \nabla^2 {\bf B} ,
294: \label{eq:induction}
295: \end{equation}
296: 
297: \add{In these units, ${\bf v}$ may be considered to be normalized by 
298: the rms value of the velocity field ($U$, say), and the magnetic field 
299: in units that, using the square root of the mass density, converts a magnetic 
300: field to an Alfv\'en speed based upon the same rms velocity. In Eqs. 
301: (\ref{eq:momentum}) and (\ref{eq:induction}), lengths are measured in 
302: units of the spherical radius $R$ (the radius $R$ will be 1 in the 
303: dimensionless units) and times in units of $R/U$.} ${\mathcal P}$ is 
304: the dimensionless ratio of pressure to mass density, with the mass 
305: density assumed to be spatially uniform. A forcing function ${\bf f}$ 
306: has been written on the right hand side of Eq. (\ref{eq:momentum}) to 
307: represent any externally applied mechanical force that can be chosen 
308: to mimic a variety of physical effects, a common convention in 
309: dynamo-motivated computations \add{in periodic boundary conditions}. 
310: Both ${\bf B}$ and ${\bf v}$ and their curls are solenoidal. \ad{Note 
311: that in the incompressible case, nonlinear coupling between modes is 
312: provided by the nonlinearities in both the equation of motion and the 
313: induction equation. Here, we are keeping all of these. In the context 
314: of planetary dynamos, other physical effects sometimes justify dropping 
315: the advection term in the equation of motion. Then other numerical 
316: methods (see e.g. Ref. \cite{Ivers03}) for linearised MHD can be used.}
317: 
318: An evolution equation for vorticity can be obtained by taking the curl of 
319: Eq. (\ref{eq:momentum}):
320: \begin{equation}
321: \frac{\partial \vomega}{\partial t} = \nabla \times \left( 
322:     {\bf v} \times \vomega \right) +  \nabla \times \left( 
323:     {\bf j} \times {\bf B} \right) + \nu \nabla^2 \vomega + 
324:     \nabla \times{\bf f} .
325: \label{eq:vorticity}
326: \end{equation}
327: 
328: \add{Given the} appropriate boundary conditions, the vorticity $\vomega$ 
329: can be used to determine the velocity ${\bf v}$. The dimensionless viscosity 
330: $\nu$ and magnetic diffusivity $\eta$ can be interpreted, respectively, as 
331: reciprocals of kinetic and magnetic Reynolds numbers based on $U$, $R$, and 
332: the laboratory (dimensional) values of kinematic viscosity and magnetic 
333: diffusivity.
334: 
335: The numerical scheme is to solve Eqs. (\ref{eq:momentum}) or 
336: (\ref{eq:vorticity}) and (\ref{eq:induction}) by representing ${\bf v}$ 
337: and ${\bf B}$ as the Galerkin expansions,
338: \begin{eqnarray}
339: {\bf v}({\bf r},t) &=& \sum_{qlm} \frac{\xi^v_{qlm}(t)}{\lambda_{ql}}
340:     {\bf J}_{qlm}({\bf r}) \;\;\;\; (r \le R) \label{eq:expanv} \\
341: {\bf B}({\bf r},t) &=& \sum_{qlm} \xi^B_{qlm}(t) \, {\bf J}_{qlm}({\bf r}) .
342: \label{eq:expanb}
343: \end{eqnarray}
344: Here, the unknowns are the \add{now} complex, time-dependent expansion 
345: coefficients $\xi^v_{qlm}$ and $\xi^B_{qlm}$. The vorticity and current 
346: density are given by the same series, each term multiplied by the 
347: appropriate value of $\lambda_{ql}$.
348: 
349: Next, we may substitute Eqs. (\ref{eq:expanv}) and (\ref{eq:expanb}) 
350: into Eqs. (\ref{eq:induction}) and (\ref{eq:vorticity}), say, and take 
351: inner products one at a time with the functions ${\bf J}_{q'l'm'}$. 
352: Using their orthonormality, the dynamical equations become
353: \begin{eqnarray}
354: \frac{\partial \xi^v_i}{\partial t} &=& \sum_{j,k} \left( A^i_{jk} 
355:     \xi^v_j \xi^v_k + B^i_{jk} \xi^B_j \xi^B_k \right) - \nu \lambda_i^2 
356:     \xi^v_i + \xi^f_i 
357: \label{eq:CK1} \\
358: \frac{\partial \xi^B_i}{\partial t} &=& \sum_{j,k} C^i_{jk} \xi^v_j \xi^B_k
359:     - \eta \lambda_i^2 \xi^B_i .
360: \label{eq:CK2}
361: \end{eqnarray}
362: 
363: The indices $i,j,k$ are regarded as shorthand; each one of them represents 
364: the triple of numbers $q,l,m$ necessary to identify a single member of the 
365: family of the C-K functions ${\bf J}_{qlm}$. The sum is over all the values 
366: retained in the Galerkin expansion.
367: 
368: The nonlinearities, both in the original partial differential equations and 
369: in the ordinary differential equations (\ref{eq:CK1}) and (\ref{eq:CK2}), 
370: are quadratic. The kinematic coupling coefficients (which do not contain 
371: the $\xi$) $A^i_{jk}$, $B^i_{jk}$, and $C^i_{jk}$, are numerical integrals 
372: of considerable complexity. Their evaluation and storage as a table is one 
373: of the most demanding parts of the computation, and some features of that 
374: evaluation are described in Section \ref{sec:numerical}. On the right hand 
375: side of Eq. (\ref{eq:CK1}), we have represented the forcing function 
376: ${\bf f}$, if any, by its expansion over the retained C-K functions. The 
377: coefficients $\xi^f$, if non-zero, are to be regarded as known, possibly 
378: time-dependent functions that may excite the velocity field, and 
379: may stand for various mechanical processes.
380: 
381: The coupling coefficients in Eqs. (\ref{eq:CK1}) and (\ref{eq:CK2}) are 
382: reducible to
383: \begin{equation}
384: A^i_{jk} = C^i_{jk} = \frac{\lambda_i}{\lambda_j} I^i_{jk} \, , 
385: \;\;\;\;\;\; B^i_{jk} = \lambda_i \lambda_j I^i_{jk} \, ,
386: \end{equation}
387: where
388: \begin{equation}
389: I^i_{jk} = \int {\bf J}^*_i \cdot \left( {\bf J}_j \times {\bf J}_k 
390:     \right) {\textrm d}^3 x .
391: \label{eq:integral}
392: \end{equation}
393: 
394: \section{\label{sec:numerical}NUMERICAL METHOD}
395: 
396: \begin{figure}
397: \includegraphics[width=9cm]{fig1.eps}
398: \caption{Speed-up of the code in two Linux clusters in parallel 
399:     simulations with $\max\{q\} = \max\{l\} = 7$. The dotted line 
400:     indicates the ideal scaling.}
401: \label{fig:speedup}
402: \end{figure}
403: 
404: Equations (\ref{eq:CK1}) and (\ref{eq:CK2}) are solved numerically with 
405: double precision in a sphere of radius $R=1$. The expansion coefficients 
406: $\xi$ are in general complex, and since the fields are real, the 
407: coefficients satisfy the condition $\xi_{q,l,-m} = (-1)^m \xi^*_{qlm}$. 
408: As a result, only the coefficients for non-negative values of $m$ are 
409: stored and evolved in time.
410: 
411: Before the simulation is started, for a given resolution in $q$ and $l$ 
412: (and all possible values of $m$) all required values of the normalization 
413: coefficients $C_{ql}$ are computed using Eq. (\ref{eq:normalization}) 
414: and stored. The values of $\lambda_{ql}$ are computed numerically as 
415: the roots of the spherical Bessel functions using a combination of 
416: bisection and Newton-Raphson \cite{Press}. Finally, the coupling 
417: coefficients $I^i_{jk}$ are computed and stored.
418: 
419: The coupling coefficients are complex, and from Eq. (\ref{eq:integral}) 
420: satisfy the relation $I^i_{jk} = - I^i_{kj}$. The integral in Eq. 
421: (\ref{eq:integral}) is separable in spherical coordinates. In the $\phi$ 
422: direction, the integral reduces to the condition $m_k = m_i-m_j$; in 
423: any other case the coupling coefficients $I^i_{jk}$ are zero. The 
424: radial integral reduces to seven integrals involving three spherical 
425: Bessel functions or their derivatives, and the integral in the polar 
426: angle reduces to seven integrals on three Legendre functions and their 
427: derivatives. Radial integrals are computed numerically with high 
428: precision using Gauss-Legendre quadratures, while integrals in $\theta$ 
429: are computed using Gauss-Jacobi quadratures \cite{Press}. Due to 
430: symmetry properties of the Legendre functions, all coupling 
431: coefficients with $l_i+l_j+l_k+m_i+m_j+m_k$ even are purely real, 
432: while the remaining coefficients are purely imaginary, another 
433: property that can be used to save memory.
434: 
435: Once tables containing all these values are stored, the evolution of 
436: the system reduces to solving the set of ordinary differential 
437: equations defined by Eqs. (\ref{eq:CK1}) and (\ref{eq:CK2}). These 
438: equations are evolved using a Runge-Kutta method of fourth order 
439: \cite{Canuto}. The MHD equations have three quadratic ideal invariants: 
440: the total energy $E$, the magnetic helicity $H$, and the cross helicity 
441: $K$. In spectral space the invariants can be computed as
442: \begin{eqnarray}
443: E &=& \frac{1}{2} \sum_i \left( \frac{|\xi^v_i|^2}{\lambda_i^2} + 
444:     | \xi^B_i|^2 \right) , \\
445: H &=& \frac{1}{2} \sum_i \frac{|\xi^B_i|^2}{\lambda_i} , \\
446: K &=& \frac{1}{2} \sum_i \frac{\xi^v_i \xi^{B*}_i}{\lambda_i} .
447: \end{eqnarray}
448: The conservation of these quantities up to the numerical precision 
449: serves as a test of the code and have been verified in simulations 
450: with $\nu=\eta=0$. In a simulation with $\max\{q\} = \max\{l\} = 5$, 
451: the invariants were conserved up to the sixth decimal place after 
452: 200 turnover times. The turnover time is defined as $T=R/U$.
453: 
454: The system is evolved entirely in spectral space and all global 
455: quantities are also computed spectrally. To obtain representations 
456: of the fields in configuration space, Eqs. (\ref{eq:expanv}) and 
457: (\ref{eq:expanb}) are used.
458: 
459: The reciprocal of the smallest $|\lambda|$ may be identified with the 
460: largest length scale in the dynamics allowed by the boundary conditions, 
461: and the reciprocal of the largest $|\lambda|$ retained may be considered 
462: to be the smallest resolvable spatial scale. In a typical computation 
463: (see e.g. Section \ref{sec:application2}), these numbers may be 
464: $\approx 4.59$ and $\approx 41.3$, respectively. The minimum and maximum 
465: wavenumbers in our previous 3D periodic dynamo computations (e.g. 
466: in a $256^3$ dealiased simulation using the $2/3$-rule \cite{Canuto}) 
467: have typically been 1 and 85 in dimensionless units, by comparison.
468: \add{The fact that the maximum to minimum ratio of allowed wavenumbers 
469: is more than $9$ times greater for the $256^3$ code shows that far 
470: less spatial resolution appears to be available in the C-K code as 
471: presently run. If ``degrees of freedom'' are taken as a formal measure 
472: of the resolution, then its $1.6\times10^4$ makes the C-K code roughly 
473: comparable with the $7.8\times10^4$ available to a de-aliased periodic 
474: 3D code that is of resolution $64^3$. That is not, however, the whole 
475: story, since very many Fourier coefficients would be needed to represent 
476: accurately any of the C-K functions which are, individually, all 
477: consistent global physical states of the system obeying all the 
478: boundary conditions. We find that there are many situations that can be 
479: computed with the C-K code that are nontrivial, highly nonlinear, and 
480: that are well-resolved as indicated by the presence of an unmistakable 
481: dissipation range for their $\lambda$-spectra; these include situations 
482: exhibiting a variety of dynamo behavior}.
483: 
484: Besides the quadratic ideal invariants and the reconstruction of 
485: the field components, two vector quantities will be of interest. The 
486: angular momentum ${\bf L}$ is defined as
487: \begin{equation}
488: {\bf L} = \int {\bf r} \times {\bf v} \, {\textrm d}^3 x \, ,
489: \end{equation}
490: where a unity mass density is assumed, and the magnetic dipole moment 
491: $\vmu$ is given by
492: \begin{equation}
493: \vmu = \frac{1}{2} \int {\bf r} \times {\bf j} \, {\textrm d}^3 x \, .
494: \end{equation}
495: In terms of the C-K functions, these two quantities become
496: {\setlength\arraycolsep{2pt}
497: \begin{eqnarray}
498: {\bf L} &=& 4 R^3 \sqrt{\frac{\pi}{3}} \sum_q C_{q,1} 
499:     \frac{j_1'(|\lambda_{q,1}| R)}{|\lambda_{q,1}|} \left[ 
500:     -\xi^v_{q,1,0} \hat{z} + \right. \nonumber \\
501: && {} \left. + \sqrt{2} {\textrm Re} \left(\xi^v_{q,1,1}\right) \hat{x} - 
502:     \sqrt{2} {\textrm Im} \left(\xi^v_{q,1,1}\right) \hat{y} \right] , \\
503: \vmu &=& 2 R^3 \sqrt{\frac{\pi}{3}} \sum_q C_{q,1} |\lambda_{q,1}| 
504:     j_1'(|\lambda_{q,1}| R) \left[ -\xi^B_{q,1,0} \hat{z} + \right. 
505:     \nonumber \\
506: && {} \left. + \sqrt{2} {\textrm Re} \left(\xi^B_{q,1,1}\right) \hat{x} - 
507:     \sqrt{2} {\textrm Im} \left(\xi^B_{q,1,1}\right) \hat{y} \right] .
508: \end{eqnarray}}
509: 
510: As previously mentioned, in the code a sphere with $R=1$ is used. 
511: Note that only modes with $l=1$ and $m=0, \pm1$ give a contribution to 
512: the angular momentum and the dipole moment. With the boundary conditions 
513: considered in this work, the angular momentum is not a conserved quantity 
514: unless $\nu = 0$. Other choices of the boundary conditions can lead to a 
515: conservation of the angular momentum even in the non-ideal case. Those 
516: boundary conditions apply to the case when the sphere of magnetofluid is 
517: isolated from torques, but we will defer consideration of that situation 
518: to a later paper.
519: 
520: The code is written in Fortran 90 and parallelized using MPI. Since 
521: most of the computing time is spent in the sums in Eqs. (\ref{eq:CK1}) 
522: and (\ref{eq:CK2}), the parallelization is done as follows. Each 
523: processor contains a complete copy of the expansion coefficients 
524: $\xi$, but only a portion of the coupling coefficients $I^i_{jk}$. 
525: The array $I^i_{jk}$ is distributed in $q$ if the number of processors 
526: is smaller or equal than $2q$, and distributed in $q,l,m$ in any other 
527: case. Each processor computes the sums in Eqs. (\ref{eq:CK1}) and 
528: (\ref{eq:CK2}) locally for the corresponding values of $q,l,m$, and 
529: after each iteration of the Runge-Kutta method the coefficients 
530: $\xi$ are synchronized between all processors. The required 
531: communication is minimal and the scaling of the code as the number 
532: of processors is increased is close to ideal.
533: 
534: Figure \ref{fig:speedup} shows the speed-up vs. the number of 
535: processors in two different Linux clusters, in a simulation using 
536: $\max\{q\} = \max\{l\} = 7$. The clusters differ in the network 
537: configuration. The speed-up is defined as the time required to do 
538: one time step in $N$ processors divided by the time required in 
539: one processor. The code shows ideal scaling up to $N\approx 2 \max\{q\}$. 
540: A drop is then observed and is related to the change in the parallel 
541: distribution of the array $I^i_{jk}$. However, after this drop a 
542: linear scaling is again recovered.
543: 
544: \section{\label{sec:application1}SELECTIVE DECAY}
545: 
546: \begin{figure}
547: \includegraphics[width=9cm]{fig2.eps}
548: \caption{Total energy (dotted line), magnetic energy (solid line), and 
549:     kinetic energy (dashed line) as a function of time in run II. At 
550:     late times the system is dominated by magnetic energy. \add{Since the 
551:     only dynamic evolution possible for ${\bf B}$ requires a velocity 
552:     and since the kinetic energy has essentially disappeared, the system 
553:     has ``frozen'' into a nearly purely magnetic state, which can only 
554:     slowly resistively decay hereafter.}}
555: \label{fig:decay_ener}
556: \end{figure}
557: 
558: \begin{figure}
559: \includegraphics[width=9cm]{fig3.eps}
560: \caption{Relative magnetic helicity as a function of time for runs I 
561:     (solid line) and II (dashed line). The relative helicity at late 
562:     times increases with the Reynolds number. Note the maximum 
563:     possible value for the relative helicity is 
564:     $\min^{-1}\{|\lambda|\} \approx 0.22$.}
565: \label{fig:decay_hel}
566: \end{figure}
567: 
568: \add{``Selective decay'' refers to a frequently studied MHD turbulence 
569: situation in which a system with initial magnetic helicity present 
570: evolves with a  rapid decay of total energy relative to the magnetic 
571: helicity. This situation has been studied previously in periodic 
572: boundary conditions and it is of interest to see if the behavior 
573: is affected by spherical geometry.} The late-time state is a quasi-steady 
574: configuration in which the remaining energy is nearly all magnetic 
575: and is condensed into the longest wavelength modes allowed by the 
576: boundary conditions. Selective decay has been extensively studied in 
577: periodic boundary conditions (see, e.g., \cite{Matthaeus80,Ting86,Kinney95}). 
578: In Ref. \cite{Mininni05d} it was found that given isotropic initial  
579: conditions in a periodic box, the final state of the magnetic field 
580: corresponds to an Arn'old-Beltrami-Childress (ABC) field at the largest 
581: possible scale with $A$, $B$, and $C$ equal. It is therefore of interest 
582: to test selective decay in a sphere, and to observe the geometry 
583: of the magnetic and velocity fields in the late-time state.
584: 
585: Two runs were done, the first (run I) with $\nu = \eta = 1\times 10^{-2}$, 
586: and the second (run II) with $\nu = \eta = 6\times 10^{-3}$. Run I was 
587: done with $\max \{q\} = \max \{l\} = 7$, and run II with 
588: $\max \{q\} = \max \{l\} = 9$. In both simulations, no external 
589: force was applied, and the system was allowed to evolve for a 
590: long time (20 initial large scale turnover times).
591: 
592: The non-vanishing initial expansion coefficients in both simulations are
593: \begin{eqnarray}
594: & \xi^v_{3,3,0} = \xi^v_{-3,3,0} = -u_0 , \;\;
595:     \xi^B_{3,3,0} = \frac{5}{3} \xi^B_{-3,3,0} = B_0 , \\
596: & \xi^v_{3,3,m} = \xi^v_{-3,3,m} = u_0(1+i) , \\
597: & \xi^B_{3,3,m} = \frac{5}{3} \xi^B_{-3,3,m} = B_0(1-i) ,
598: \end{eqnarray}
599: where $m$ runs from 1 to 3 and negative values of $m$ are given by 
600: $\xi_{q,l,-m} = (-1)^m \xi^*_{qlm}$. The amplitudes $u_0$ and $B_0$ were 
601: chosen to have initial kinetic and magnetic energies of order unity 
602: ($u_0 = 4$ and $B_0 = 0.4$). These initial conditions correspond to 
603: a small cross correlation between the velocity and magnetic fields, a 
604: non-helical velocity field at an intermediate scale, and a 
605: (non-maximally) helical magnetic field at the same scale. The 
606: initial angular momentum is zero and remains negligible during 
607: the complete simulation. The kinetic and magnetic Reynolds numbers, 
608: based on the length $R=1$ and the initial rms velocity, are 
609: respectively defined as
610: \begin{eqnarray}
611: R_V & = & \frac{R U}{\nu} , \\
612: R_M & = & \frac{R U}{\eta} .
613: \end{eqnarray}
614: In run I $R_V = R_M \approx 98$, and in run II $R_V = R_M \approx 165$.
615: 
616: \begin{figure}
617: \includegraphics[width=9cm]{fig4.eps}
618: \caption{Spectrum of kinetic energy as a function of $\lambda$ for run 
619:     II, at $t=0.1$ (solid line), $t=1.5$ (dotted line) and $t=3$ (dashed 
620:     line). \add{Note the appearance of a clear dissipation range.}}
621: \label{fig:decay_ksp}
622: \end{figure}
623: 
624: \begin{figure}
625: \includegraphics[width=9cm]{fig5.eps}
626: \caption{Spectrum of magnetic energy as a function of $\lambda$ for run 
627:     II, at $t=0.1$ (solid line), $t=1.5$ (dotted line) and $t=3$ (dashed 
628:     line). \add{A magnetic dissipation range becomes clearly visible.}}
629: \label{fig:decay_msp}
630: \end{figure}
631: 
632: Figure \ref{fig:decay_ener} shows the time history of the kinetic, 
633: magnetic, and total energies in run II. At late times, the kinetic 
634: energy is negligible and the system is dominated by magnetic energy. 
635: The magnetic helicity decays slowly compared to the energy, as 
636: indicated by the evolution of the relative helicity $H_M/E_M$ (Fig. 
637: \ref{fig:decay_hel}). As time evolves, the ratio $H_M/E_M$ increases 
638: until a steady state is reached. The final state reached by the 
639: system is not a ``Taylor state'', a state of maximum possible helicity 
640: for the given energy (the maximum possible value for the relative helicity 
641: is $\min^{-1}\{|\lambda|\} \approx 0.22$). In run I, this is because 
642: after $t\approx 5$ most of the kinetic energy has decayed, and as a 
643: result the spectral exchange between modes stopped. In run II the decay 
644: of the kinetic energy takes place at a later time, and as a result the 
645: final value of the relative helicity is larger than in run I. Larger final 
646: values of the relative helicity can be expected if the Reynolds numbers 
647: are further increased.
648: 
649: Since the C-K functions are eigenfunctions of the curl with 
650: eigenvalue $\lambda$, the Laplacian operators in the diffusion terms 
651: in Eqs. (\ref{eq:momentum}-\ref{eq:vorticity}) are proportional to 
652: $\lambda^2$. As a result, $|\lambda|$ plays in this case the role 
653: of the wavenumber $k$ in the Fourier base. To define the energy spectrum, 
654: we linearly bin the spectral space in shells of constant $|\lambda|$ and 
655: sum the power of all the coefficients in each shell, in analogy with 
656: the usual procedure in Fourier-based spectral methods. Figures 
657: \ref{fig:decay_ksp} and \ref{fig:decay_msp} show respectively the 
658: resulting kinetic and magnetic energy spectrum as a function of 
659: $|\lambda|$ in run II at three different times.
660: 
661: \begin{figure}
662: \includegraphics[width=9cm]{fig6.eps}
663: \caption{Amplitude of the coefficients $\xi^B$ in run II at $t=10$ 
664:     as a function of $q$ and $l$ (summed over all values of $m$). Most 
665:     of the magnetic energy is at the largest available scale, and an 
666:     asymmetry is observed between positive and negative values of $q$.}
667: \label{fig:decay_xi}
668: \end{figure}
669: 
670: At early times ($t=0.1$), the signature of the initial conditions 
671: in the spectrum can be easily recognized. Both spectra peak at 
672: $|\lambda| \approx 14$, corresponding to the non-vanishing initial 
673: perturbation at $q=3$ and $l=3$. As time evolves, the amplitude of the 
674: kinetic energy spectrum decays but the position of the peak remains 
675: approximately constant. On the other hand, the peak in the magnetic 
676: energy spectrum moves to smaller values of $|\lambda|$, corresponding 
677: to larger scales. At late times the system is dominated by magnetic 
678: energy, and most of it is concentrated in the largest available scale 
679: in the domain.
680: 
681: \begin{figure*}
682: \includegraphics[width=14cm]{fig7.eps}
683: \caption{Above: kinetic energy density and velocity field lines in run 
684:     II, at $t=1.5$ (left) and at $t=15$ (right). Below: magnetic energy 
685:     density and magnetic field lines in the same run, at $t=1.5$ (left) 
686:     and at $t=15$ (right). \ad{For convenience, intensity and field lines 
687:     are always shown in pairs, with intensity on the left and field lines 
688:     on the right.}}
689: \label{fig:decay3D}
690: \end{figure*}
691: 
692: Based on the definition of the energy spectrum, we can also 
693: introduce an energy-containing lengthscale as
694: \begin{equation}
695: \ell = \frac{R \, \min\{|\lambda|\}}{E} \int{E(|\lambda|) |\lambda|^{-1}
696:     \, {\textrm d}|\lambda|} ,
697: \end{equation}
698: and a Taylor lengthscale
699: \begin{equation}
700: \lambda_T = R \, \min\{|\lambda|\} \left[ E \bigg/ \int{E(|\lambda|) 
701:     |\lambda|^2 \, {\textrm d}|\lambda|} \right]^{1/2} ,
702: \end{equation}
703: where $E$ is the energy, and $E(|\lambda|)$ is the energy spectrum as 
704: a function of $|\lambda|$ (the sums are represented symbolically as 
705: integrals). Using the kinetic and magnetic energies, these characteristic 
706: lengths can be computed for the velocity and magnetic fields. In both 
707: runs, at $t=0$ $\ell \approx \lambda_T \approx 0.33$ for both fields. As 
708: the system evolves these quantities grow monotonically, but while at 
709: $t=20$ for the velocity field $\ell \approx 0.5$, for the magnetic field 
710: $\ell \approx 1$. As a criterion to decide if the simulations were well 
711: resolved in spectral space, the scales where the kinetic and magnetic 
712: enstrophy spectrum peaked were observed as a function of time, and it 
713: was asked that their corresponding wavenumbers $|\lambda|$ were smaller 
714: than $\max\{|\lambda|\}$ at all times.
715: 
716: Figure \ref{fig:decay_xi} shows the amplitude of the individual 
717: modes $\xi^B$ in run II at $t=10$ as a function of $q$ and $l$ (all 
718: values of $m$ for each value of $l$ are summed). Most of the magnetic 
719: energy is concentrated in the shell with $l=1$, and the modes with 
720: $q = \pm 1$ in this shell have the largest amplitude. Note the 
721: imbalance between the mode with $q=1$ and $q=-1$, indicating some 
722: helicity is present in the magnetic field.
723: 
724: The dominance of a helical large scale magnetic field at late 
725: times can also be identified in an inspection of the fields in 
726: configuration space. Figure \ref{fig:decay3D} shows field intensity 
727: and field lines for the velocity and magnetic field at $t=1.5$ 
728: (left) and $t=15$ (right) in run II. While at early times both fields 
729: show small scale features, at late times the velocity field looks 
730: reminiscent of a quadrupole and the magnetic field looks like a 
731: dipole oriented roughly in the $z$ direction. However, the magnetic 
732: field at $t=15$ is helical and the magnetic field lines are not 
733: purely poloidal. There is a small toroidal component to the magnetic 
734: field, and the magnetic field lines proceed slowly in the 
735: $\phi$-direction in a helical fashion.
736: 
737: In Fig. \ref{fig:decay3D} and in the following visualizations 
738: the labels are as follows. The $x$, $y$, and $z$ directions are 
739: indicated by the arrows (in the online version, these are respectively 
740: red, green, and blue). \ad{The color and opacity are proportional to the 
741: field intensity, colorbars are given as a reference.} Field lines are 
742: computed taking a snapshot of the field at a fixed instant in time, 
743: \ad{and integrating a trajectory from twenty random points} in the 
744: surroundings of the center of the sphere. The field is not evolved 
745: in time as the lines are integrated. To indicate the direction of the 
746: fields, in the online version the lines change color according to the 
747: distance integrated from the initial point; from \ad{red to yellow, and 
748: finally blue.}
749: 
750: \section{\label{sec:application2}DYNAMO EFFECT}
751: 
752: \begin{figure}
753: \includegraphics[width=7cm]{fig8.eps}
754: \caption{Intensity of the forcing function ${\bf f}$ used in the dynamo 
755:     simulations (left), and associated field lines (right). Above: 
756:     function used in run III. Below: function used in runs IV and V.}
757: \label{fig:force}
758: \end{figure}
759: 
760: \begin{figure}
761: \includegraphics[width=9cm]{fig9.eps}
762: \caption{Kinetic (dashed line) and magnetic energy (solid line) as a 
763:     function of time in run III.}
764: \label{fig:dynamo_ener1}
765: \end{figure}
766: 
767: \begin{figure}
768: \includegraphics[width=9cm]{fig10.eps}
769: \caption{\ad{Trace of the magnetic dipole orientation on the unit sphere 
770:     (above) and magnitude of the dipole moment as a function of time 
771:     (below) in run III.}}
772: \label{fig:dynamo_dip1}
773: \end{figure}
774: 
775: \add{In an MHD dynamo, an initially small ``seed'' magnetic field is 
776: amplified and sustained against Ohmic dissipation solely by the motions 
777: of a conducting fluid. Magnetic fields observed in planets and stars 
778: are believed to be the result of a dynamo process. The mechanical 
779: mechanisms proposed vary widely, including thermal convection, Ekman 
780: pumping due to rotation, precession, irregularities on the inner 
781: surface of the planetary mantles, and so on. In this Section, we 
782: study three simple examples of forced dynamo action in the sphere 
783: with relatively simple realizations of the forcing function ${\bf f}$ 
784: of Eq. (5).} \ad{Figure \ref{fig:force} shows visualizations of the two 
785: expressions used for ${\bf f}$. The geometry of the forcing function is 
786: not intended to mimic any particular process in the planetary or stellar 
787: dynamos, but is rather inspired in a common practice in dynamo simulations 
788: with periodic boundary conditions: a few spectral modes are forced in the 
789: mechanical energy, and for Reynolds numbers large enough generic 
790: properties of the simulations are studied.}
791: 
792: \add{In all cases to be described here, a purely hydrodynamic 
793: (${\bf B}=0$) computation was carried out first until a steady state 
794: (laminar or turbulent) was reached for the velocity field. The 
795: amplitude of ${\bf f}$ was chosen to have rms velocity of order 
796: one in the steady state. Then a random magnetic field with energy 
797: $E_M \sim 10^{-6}$ was loaded into the modes with $|q|=l=4$.} The 
798: simulations were then continued in order to observe the amplification 
799: and subsequent development of the magnetic field.
800: 
801: Three dynamo simulations were done in which the Reynolds numbers and 
802: the number of modes excited were progressively increased. \add{In the 
803: first case, the system reaches a steady state with a stable dipole 
804: moment which varies little in time. In the second case, the dipole 
805: moment forms, then spontaneously changes direction. In the third, the 
806: Reynolds number is high enough that the velocity field might be called 
807: turbulent, and many modes are excited; for this case, the dipole moment 
808: changes erratically in time.}
809: 
810: A resolution of $\max \{|q|\} = \max \{l\} = 9$ was used in all the 
811: runs. The same criteria as in the previous section was used to decide 
812: if a simulation was well resolved. For the last run, a simulation with 
813: higher resolution was also carried out to see if the results would be 
814: modified by the change in resolution.
815: 
816: \begin{figure}
817: \includegraphics[width=7cm]{fig11.eps}
818: \caption{Above: kinetic energy density and velocity field lines in run 
819:     III, at $t=915$. Below: magnetic energy density and magnetic field 
820:     at the same time. \ad{The same conventions than in Fig. \ref{fig:decay3D} 
821:     are used.}}
822: \label{fig:3Dlaminar}
823: \end{figure}
824: 
825: \subsection{\label{sec:laminar}Laminar runs}
826: 
827: In this section we present results from two runs with 
828: $\nu=\eta=3\times10^{-3}$. In run III, the external forcing ${\bf f}$ 
829: in Eq. (\ref{eq:momentum}) is given by the coefficient
830: \begin{equation}
831: \xi^f_{2,2,1} = f_0(1+i),
832: \end{equation}
833: with $f_0 = 1.4$, which corresponds to one C-K mode and as a result 
834: injects maximum kinetic helicity in the system. In run IV, the external 
835: forcing is
836: \begin{eqnarray}
837: & \xi^f_{2,2,0} = 5 \xi^f_{-2,2,0} = f_0 , 
838:     \label{eq:force1} \\
839: & \xi^f_{2,2,m} = 5 \xi^f_{-2,2,m} = f_0(1+i) ,
840:     \label{eq:force2}
841: \end{eqnarray}
842: where $m$ runs from 1 to 2 and negative values of $m$ are again given 
843: by $\xi_{q,l,-m} = (-1)^m \xi^*_{qlm}$. The amplitude of the forcing 
844: is $f_0 = 0.9$. This forcing injects non-maximal kinetic helicity (plus, 
845: of course, kinetic energy) into the system. \ad{Figure \ref{fig:force} shows 
846: visualizations of the two forcing functions in configuration space. In 
847: both cases the forcing is stronger in the center of the sphere, and a 
848: modulation due to $m=2$ modes in the forcing can be easily identified.}
849: The phase and amplitude of the external force ${\bf f}$ were kept 
850: constant during the entire simulations.
851: 
852: Figure \ref{fig:dynamo_ener1} shows the time history of the 
853: magnetic and kinetic energy in run III. The Reynolds numbers based on 
854: the length $R=1$ and the rms velocity are $R_V = R_M \approx 290$, and 
855: the energy containing scale of the flow is $\ell \approx 0.5$. Before 
856: the magnetic field is introduced, only the forced mode is excited.
857: 
858: After the magnetic seed is introduced, the magnetic energy is 
859: amplified exponentially in a kinematic regime. Then the magnetic field 
860: saturates around $t\approx 150$ and the Lorentz force quenches the 
861: velocity field. In the final steady state, more mechanical modes 
862: besides the forced mode are excited. This is the result of an 
863: instability of the flow, triggered by the Lorentz force as the 
864: magnetic field grows exponentially.
865: 
866: Figure \ref{fig:dynamo_dip1} shows \ad{the trace of the orientation of 
867: the dipole moment in the surface of the unit sphere, and the magnitude 
868: of the dipole moment as a function of time for run III.} The dipole 
869: moment grows during the kinematic regime, \ad{but its orientation changes 
870: erratically in time. At $t \approx 200$ $|\vmu|$ reaches its maximum 
871: amplitude and converges slowly to a steady state. Also it direction 
872: changes slowly, and after $t\approx 400$ almost no change is observed. 
873: At late times} the dipole shows no inclination toward further systematic 
874: dynamical development.
875: 
876: \begin{figure}
877: \includegraphics[width=9cm]{fig12.eps}
878: \caption{Kinetic (dashed line) and magnetic energy (solid line) as a 
879:     function of time in run IV.}
880: \label{fig:dynamo_ener2}
881: \end{figure}
882: 
883: \begin{figure}
884: \includegraphics[width=9cm]{fig13.eps}
885: \caption{\ad{Trace of the magnetic dipole orientation on the unit sphere 
886:     (above) and magnitude of the dipole moment as a function of time 
887:     (below) in run IV.}}
888: \label{fig:dynamo_dip2}
889: \end{figure}
890: 
891: A visualization of the velocity and magnetic fields in configuration 
892: space in the steady state of run III is shown in Fig. \ref{fig:3Dlaminar}. 
893: The kinetic energy is concentrated in two counter-rotating regions, located 
894: in the center of each hemisphere. \ad{Note the $m=2$ modulation in the 
895: forcing is still visible in the kinetic energy density.} The velocity 
896: field in these regions is mostly toroidal, as indicated by the velocity 
897: field lines. The magnetic energy is larger in the center of the sphere, 
898: and along the axis defined by the two counter-rotating eddies. In the 
899: interior, but away from the axis, magnetic field lines are mostly 
900: toroidal, as the result of the stretching by the two counter-rotating 
901: eddies. Along the axis and close to the boundaries, the flow is mostly 
902: poloidal.
903: 
904: \begin{figure}
905: \includegraphics[width=9cm]{fig14.eps}
906: \caption{Kinetic energy spectrum at $t=52$ [thick (blue) solid line] and 
907:     at $t=500$ [thick (blue) dash-dotted line] in run IV; the thin lines 
908:     correspond to the magnetic energy spectrum at $t=52$ (solid), $t=82$ 
909:     (dotted), $t=250$ (dashed), and $t=500$ (dash-dotted).}
910: \label{fig:dynamo_spc2}
911: \end{figure}
912: 
913: \begin{figure*}
914: \includegraphics[width=14cm]{fig15.eps}
915: \caption{Above: kinetic energy density and velocity field lines in run 
916:     IV, at $t=315$ (left) and at $t=1065$ (right). Below: magnetic energy 
917:     density and magnetic field lines in the same run, at $t=315$ (left) 
918:     and at $t=1065$ (right). \ad{The same conventions than in Fig. 
919:     \ref{fig:decay3D} are used.}}
920: \label{fig:reversal}
921: \end{figure*}
922: 
923: The time history of the kinetic and magnetic energy in run IV 
924: is shown in Fig. \ref{fig:dynamo_ener2}. The Reynolds numbers for this 
925: run (based on the length $R=1$) are $R_V = R_M \approx 280$, and the 
926: energy containing scale of the flow is $\ell \approx 0.48$. Although 
927: the kinematic viscosity and magnetic diffusivity are the same as in 
928: run III, the external forcing injects energy in a larger number of 
929: modes and even before the magnetic field is introduced non-forced modes 
930: have some mechanical energy. After the magnetic seed is introduced, the 
931: magnetic energy grows exponentially until at $t \approx 150$ saturates. 
932: The system seems to reach a steady state but suddenly at $t\approx 500$ 
933: the magnetic energy decreases by a factor of $\approx 1.8$, the kinetic 
934: energy increases by $\approx 1.1$, and the system reaches a second 
935: steady state.
936: 
937: \begin{figure}
938: \includegraphics[width=9cm]{fig16.eps}
939: \caption{Kinetic (dashed line) and magnetic energy (solid line) as a 
940:     function of time in run V.}
941: \label{fig:dynamo_ener3}
942: \end{figure}
943: 
944: \begin{figure}
945: \includegraphics[width=9cm]{fig17.eps}
946: \caption{Time evolution of the three components of the dipole moment 
947:     in run V. Labels are as in Fig. \ref{fig:dynamo_dip1}.}
948: \label{fig:dynamo_dip3}
949: \end{figure}
950: 
951: The abrupt change in the kinetic and magnetic energy in run IV at 
952: $t \approx 500$ is associated with a reorientation of the magnetic dipole 
953: moment. Figure \ref{fig:dynamo_dip2} shows the time evolution \ad{of the 
954: direction and amplitude of $\vmu$. As in run III, in the kinematic stage 
955: the dipole moment grows rapidly and its direction fluctuates erratically, 
956: until reaching a first quasi-steady state at $t \approx 200$. The dipole 
957: moment then stays approximately constant until at $t \approx 400$} the 
958: magnetic field evolves rapidly, and the dipole moment changes direction 
959: to a second attractor \ad{(reached at $t \approx 600$)}. The angle the 
960: dipole flips by is close to $\pi/2$. \ad{After $t \approx 600$ only a slow 
961: change in $\vmu$ is observed.} The amplitude of the dipole moment 
962: also changes rapidly as the dipole shifts at $t \approx 400$; while at 
963: $t \approx 300$ $|\vmu| \approx 0.36$, at 
964: $t \approx 1000$ $|\vmu| \approx 0.54$.
965: 
966: \begin{figure}
967: \includegraphics[width=9cm]{fig18.eps}
968: \caption{Kinetic energy spectrum at $t=75$ [thick (blue) solid line] and 
969:     at $t=600$ [thick (blue) dash-triple dotted line] in run V; the thin 
970:     lines correspond to the magnetic energy spectrum at $t=75$ (solid), 
971:     $t=168$ (dotted), $t=240$ (dashed), $t=544$ (dash-dotted), and $t=500$ 
972:     (dash-triple dotted).}
973: \label{fig:dynamo_spc3}
974: \end{figure}
975: 
976: Figure \ref{fig:dynamo_spc2} shows the kinetic and magnetic energy 
977: spectra in run IV at different times. The kinetic energy spectrum peaks 
978: at $|\lambda| \approx 9$, corresponding to the forced modes with $q=2$ 
979: and $l=2$. The energy in the remaining modes is at least two orders
980: of magnitude smaller than in the forced modes. At late times, some 
981: kinetic energy is excited in the largest available scale, as well as a 
982: small angular momentum ($|{\bf L}|^2/E_V \approx 1.5 \times 10^{-4}$ 
983: after $t \approx 600$). The angle between the dipole moment and this 
984: small angular momentum remains constant after $t=100$ and is $\pi/2$. 
985: Visualizations of the velocity and magnetic fields in configuration 
986: space are shown in Fig. \ref{fig:reversal}. The geometry of the 
987: velocity field is more complex than in run III\ad{, although the $m=2$ 
988: modulation in the kinetic energy can still be recognized.} Note at 
989: $t=315$ the magnetic field lines are mostly poloidal in the center 
990: of the sphere, and toroidal close to the boundary. \ad{The change in 
991: the orientation of $\vmu$ at $t \approx 400$ takes place without a 
992: strong change in the velocity field, and a relatively small change in 
993: the magnetic field configuration.}
994: 
995: \begin{figure*}
996: \includegraphics[width=14cm]{fig19.eps}
997: \caption{Above: kinetic energy density and velocity field lines in run 
998:     V, at $t=75$ (left) and at $t=1035$ (right). Below: magnetic energy 
999:     density and magnetic field lines in the same run, at $t=75$ (left) 
1000:     and at $t=1035$ (right). \add{We were unable to find an ordered 
1001:     pattern in either field.}}
1002: \label{fig:dynamo3D}
1003: \end{figure*}
1004: 
1005: \subsection{\label{sec:chaos}A \add{rapidly-varying dipole}}
1006: 
1007: For run V, the external forcing f is given by Eqs. (\ref{eq:force1}) and 
1008: (\ref{eq:force2}), but the kinematic viscosity and magnetic diffusivity 
1009: are dropped to $\nu = \eta = 3 \times 10^{-4}$. The resulting kinetic 
1010: and magnetic Reynolds numbers based on the length $R=1$ are 
1011: $R_V=R_M \approx 2300$, and the energy-containing and Taylor scales are 
1012: respectively $\ell \approx 0.4$ and $\lambda_T = 0.35$.
1013: 
1014: The evolution of the kinetic and magnetic energy in run V is shown in 
1015: Fig. \ref{fig:dynamo_ener3}. Again, after an initial kinematic 
1016: regime where the magnetic energy is amplified exponentially, the system 
1017: reaches a statistical steady state. Note that in this simulation both 
1018: the kinetic and magnetic energy fluctuate strongly with time, indicating 
1019: the nonlinear coupling between modes is stronger than in runs III 
1020: and IV, as a result of the higher Reynolds numbers.
1021: 
1022: \ad{Figure \ref{fig:dynamo_dip3} shows the trace of the dipole moment in 
1023: the surface of the unit sphere, and its intensity as a function of time, 
1024: in run V. The direction of $\vmu$ appear to fluctuate randomly, changing 
1025: hemisphere with a characteristic time of the order of 100 eddy turnover 
1026: times. In the meantime, the whole surface of the unit sphere seems to be 
1027: explored by the fluctuations in $\vmu$.} The angular momentum is small 
1028: and fluctuates around $|{\bf L}|^2/E_V \approx 5 \times 10^{-3}$. However, 
1029: unlike in Run III, in this simulation the angle between the dipole moment 
1030: and the angular momentum is not even approximately constant and fluctuates 
1031: seemingly randomly between $0$ and $\pi$. The question of under what 
1032: circumstance the dipole favors one or another alignment appears to be 
1033: quite unsettled, and deserves further study in higher-resolution 
1034: simulations with larger angular momentum \ad{and rotation}.
1035: 
1036: The kinetic and magnetic energy spectra in run V at several 
1037: times are shown in Fig. \ref{fig:dynamo_spc3}. More modes are excited 
1038: in the velocity field, in accordance with the strong fluctuations in 
1039: time observed in the kinetic energy. While in runs III and IV the 
1040: magnetic energy spectrum peaks at large scales even during the kinematic 
1041: regime, in run V at early times the magnetic energy peaks at scales 
1042: smaller than the forcing scale. Also, after the nonlinear saturation of 
1043: the dynamo, a fluctuation in the amplitude of the large scale magnetic 
1044: field is observed. The minima are correlated with times of minima 
1045: of $\mu^2$, when the three components of the dipole moment fluctuate 
1046: around zero. Most of the activity in this run is in small scales 
1047: and fluctuations in the flow are larger than in runs III and IV. Even 
1048: at late times when a large scale magnetic field has developed, 
1049: intermediate scales give a large contribution to the magnetic energy. 
1050: This also explains the strong fluctuations observed in the dipole 
1051: moment. Since $\vmu$ is proportional to the current density, the 
1052: small scales give a large contribution to the dipole moment.
1053: 
1054: Figure \ref{fig:dynamo3D} shows the magnetic and velocity field 
1055: in real space at two different times in run V. The fields have more 
1056: small scale structure than in run IV. \ad{Any trace of the $m=2$ 
1057: modulation of the forcing has been completely lost.} Also, in the 
1058: kinematic regime (see e.g. the magnetic field at $t=75$) the magnetic 
1059: energy is mostly in the small scales, as also indicated by the 
1060: magnetic energy spectrum.
1061: 
1062: \section{\label{sec:conclusions}DISCUSSION}
1063: 
1064: \add{We have introduced some computational machinery that is intended for 
1065: a quantitative discussion of nonlinear and incompressible 
1066: magnetohydrodynamics inside a sphere for a wide range of initial 
1067: conditions and varieties of mechanical forcing. It is concerned 
1068: essentially with the analytical and computational aspects of 
1069: dynamo action in this geometry in a somewhat abstract setting, and 
1070: is not the same as periodic dynamo computations with rectangular 
1071: symmetry, or with planetary-dynamo or solar-dynamo computations 
1072: whose central focus is reproduction of observations of magnetic 
1073: behavior of a real system. There are other features yet to be 
1074: included and it is our intent to introduce them one at a time: rigid 
1075: rotation and Ekman pumping, an insulating but non-conducting boundary 
1076: to permit the magnetic field's penetration into the surrounding vacuum 
1077: region, and different forms of mechanical excitations.}
1078: 
1079: \add{The operation of the wholly spectral code, which has had a precedent 
1080: in axially periodic circular-cylinder geometry has been illustrated by a 
1081: few simple examples, not in any sense a comprehensive study. First, 
1082: decaying turbulence has been studied involving the relaxation of helical 
1083: initial conditions to a magnetically dominated state whose spectrum is 
1084: dominated by the longest allowed spatial scales and whose kinetic energy 
1085: is essentially gone. Also, dynamo computations have been done for three 
1086: successively higher sets of Reynolds numbers, revealing a different magnetic 
1087: behavior in each case. In the first case, a magnetic dipole formed in what 
1088: was essentially a laminar velocity field and appears willing to persist for 
1089: as long as the code is run. In the second case, another dipole formed, but 
1090: after the passage of a few hundred large-scale eddy-turnover times, it 
1091: changed its magnitude to a somewhat larger value, and flipped its 
1092: orientation, for reasons we do not know but that are worth exploring. 
1093: These fluctuations have required neither a preferred direction enforced 
1094: by rigid rotation nor thermally convective rolls. Finally, in the third 
1095: case, the velocity field had Reynolds numbers (based on the radius of 
1096: the sphere) of about $2300$ and could be said to be turbulent. In this 
1097: turbulent case, there was a dipole moment, but it exhibited no systematic 
1098: or regular behavior as far as we could tell, and changed its orientation} 
1099: \ad{from one hemisphere to the other} \add{every $100$ or so eddy turnover 
1100: times,} \ad{exploring in the meantime all possible orientations.} \add{It 
1101: did appear to be a resolved computation, despite the turbulence, and we 
1102: believe represents a bona fide solution of the MHD equations. All these 
1103: runs, it should be stressed, were carried out with a magnetic Prandtl 
1104: number of unity, and may well change for values far from that. One 
1105: outcome to be noted is that neither turbulence nor rotation have been 
1106: a necessary ingredient for the development and maintenance of a 
1107: magnetic dipole, but the presence of mechanically helicity has 
1108: helped a lot. Spontaneous changes in magnetic dipole orientation 
1109: have been easy to observe, both in laminar and turbulent cases.}
1110: 
1111: \add{We should also stress that the code as presently constituted is 
1112: limited to a relatively small number of degrees of freedom, far 
1113: fewer than well-resolved high Reynolds number mechanical turbulence 
1114: would demand. It will also be attempted to design wall-friction 
1115: terms to permit a more efficient exchange of angular momentum from 
1116: the fluid to the wall \cite{Shan94}, once rigid rotation is introduced.} 
1117: \ad{Also rigid rotation and different boundary conditions for the magnetic 
1118: fields can be implemented. Given the properties and limitations of 
1119: the method described (purely spectral, non-dispersive, and 
1120: conservative), we believe this method can be used as a testbed to 
1121: explore the effect of different physical effects, boundary conditions, 
1122: subgrid models (several models, such as the Lagrangian averaged model 
1123: \cite{Holm02a,Holm02b,Mininni05d} are easier to implement in spectral 
1124: space), etc., before trying these ideas in more complex and realistic 
1125: codes to reach high Reynolds numbers.}
1126: 
1127: \begin{acknowledgments}
1128: The authors would like to express their gratitude to A. Pouquet for 
1129: valuable discussions and his careful reading of the manuscript. Computer 
1130: time was provided by NCAR. The NSF grants CMG-0327888 at NCAR and 
1131: ATM-0327533 at Dartmouth College supported this work in part and are 
1132: gratefully acknowledged. Three-dimensional visualizations of the flow 
1133: were done using VAPoR \cite{vapor}, a software for interactive 
1134: visualization and analysis of large datasets.
1135: \end{acknowledgments}
1136: 
1137: % BIBLIOGRAPHY %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1138: 
1139: %\bibliography{ms}
1140: \begin{thebibliography}{10}
1141: 
1142: \bibitem{Moffatt}
1143: H.~K. Moffatt, {\em Magnetic field generation in electrically conducting
1144:   fluids}.
1145: \newblock Cambridge, Cambridge Univ. Press (1978).
1146: 
1147: \bibitem{Glatzmaier95}
1148: G.~A. Glatzmaier and P.~H. Roberts, ``A three-dimensional self-consistent
1149:   computer simulation of a geomagnetic field reversal,'' Nature {\bf 377},
1150:   203--209 (1995).
1151: 
1152: \bibitem{Dikpati99}
1153: M.~Dikpati and P.~Charbonneau, ``A babcock-leighton flux transport dynamo with
1154:   solar-like differential rotation,'' Astrophys.\ J. {\bf 518}, 508--520
1155:   (1999).
1156: 
1157: \bibitem{Kono02}
1158: M.~Kono and P.~H. Roberts, ``Recent geodynamo simulations and observations of
1159:   the geomagnetic field,'' Rev.\ Geophys. {\bf 40}, 1--53 (2002).
1160: 
1161: \bibitem{Nandy02}
1162: D.~Nandy and A.~R. Choudhuri, ``Explaining the latitudinal distribution of
1163:   sunspots with deep meridional flow,'' Science {\bf 296}, 1671--1673 (2002).
1164: 
1165: \bibitem{Mininni04}
1166: P.~D. Mininni and D.~O. G\'omez, ``A new technique for comparing solar dynamo
1167:   models and observations,'' Astron.\ Astrophys. {\bf 426}, 1065--1073
1168:   (2004).
1169: 
1170: \bibitem{Gailitis01}
1171: A.~Gailitis, O.~Lielausis, E.~Platacis, S.~Dement'ev, A.~Cifersons, G.~Gerbeth,
1172:   T.~Gundrum, F.~Stefani, M.~Christen, and G.~Will, ``Magnetic field saturation
1173:   in the riga dynamo experiment,'' Phys.\ Rev.\ Lett. {\bf 86}, 003024
1174:   (2001).
1175: 
1176: \bibitem{Steiglitz01}
1177: R.~Steiglitz and U.~M\"uller, ``Experimental demonstration of a homogeneous
1178:   two-scale dynamo,'' Phys.\ Fluids {\bf 13}, 561--564 (2001).
1179: 
1180: \bibitem{Noguchi02}
1181: K.~Noguchi, V.~I. Pariev, S.~A. Colgate, H.~F. Beckley, and J.~Nordhaus,
1182:   ``Magnetorotational instability in liquid metal couette flow,'' Astrophys.\
1183:   J. {\bf 575}, 1151--1162 (2002).
1184: 
1185: \bibitem{Petrelis03}
1186: F.~P\'etr\'elis, M.~Bourgoin, L.~Mari\'e, J.~Burguete, A.~Chiffaudel,
1187:   F.~Daviaud, S.~Fauve, P.~Odier, and J.-F. Pinton, ``Nonlinear magnetic
1188:   induction by helical motion in a liquid sodium turbulent flow,'' Phys.\ Rev.\
1189:   Lett. {\bf 90}, 174501 (2003).
1190: 
1191: \bibitem{Sisan03}
1192: D.~R. Sisan, W.~L. Shew, and D.~P. Lathrop, ``Lorentz force effects in
1193:   magneto-turbulence,'' Phys.\ Earth Plan.\ Int. {\bf 135}, 137--159 (2003).
1194: 
1195: \bibitem{Spence05}
1196: E.~J. Spence, M.~D. Nornberg, C.~M. Jacobson, R.~D. Kendrick, and C.~B. Forest,
1197:   ``Observation of a turbulence-induced large scale magnetic field,'' Phys.\
1198:   Rev.\ Lett. {\bf 96}, 055002 (2006).
1199: 
1200: \bibitem{Roberts01}
1201: P.~H. Roberts and G.~A. Glatzmaier, ``The geodynamo, past, present and
1202:   future,'' Geophys.\ Astrophys.\ Fluid Dyn. {\bf 94}, 47--84 (2001).
1203: 
1204: \bibitem{Brandenburg05}
1205: A.~Brandenburg and K.~Subramanian, ``Astrophysical magnetic fields and
1206:   nonlinear dynamo theory,'' Phys.\ Rep. {\bf 417}, 1--209 (2005).
1207: 
1208: \bibitem{Ponty05}
1209: Y.~Ponty, P.~D. Mininni, D.~C. Montgomery, J.-F. Pinton, H.~Politano, and
1210:   A.~Pouquet, ``Numerical study of dynamo action at low magnetic prandtl
1211:   numbers,'' Phys.\ Rev.\ Lett. {\bf 94}, 164502 (2005).
1212: 
1213: \bibitem{Mininni05a}
1214: P.~D. Mininni, Y.~Ponty, D.~C. Montgomery, J.-F.Pinton, H.~Politano, and
1215:   A.~Pouquet, ``Dynamo regimes with a nonhelical forcing,'' Astrophys.\ J.
1216:   {\bf 626}, 853--863 (2005).
1217: 
1218: \bibitem{Mininni05b}
1219: P.~D. Mininni and D.~C. Montgomery, ``Low magnetic prandtl number dynamos with
1220:   helical forcing,'' Phys.\ Rev.\ E {\bf 72}, 056320 (2005).
1221: 
1222: \bibitem{Alexakis05}
1223: A.~Alexakis, P.~D. Mininni, and A.~Pouquet, ``Shell-to-shell energy transfer in
1224:   magnetohydrodynamics. i. steady state turbulence,'' Phys.\ Rev.\ E {\bf 72},
1225:   046301 (2005).
1226: 
1227: \bibitem{Mininni05c}
1228: P.~D. Mininni, A.~Alexakis, and A.~Pouquet, ``Shell-to-shell energy transfer in
1229:   magnetohydrodynamics. ii. kinematic dynamo,'' Phys.\ Rev.\ E {\bf 72},
1230:   046302 (2005).
1231: 
1232: \bibitem{Chandrasekhar57}
1233: S.~Chandrasekhar and P.~C. Kendall, ``On force-free magnetic fields,''
1234:   Astrophys.\ J. {\bf 126}, 457--460 (1957).
1235: 
1236: \bibitem{Turner83}
1237: L.~Turner, ``Statistical mechanics of a bounded, ideal magnetofluid,'' Ann.\
1238:   Phys.\ (NY) {\bf 149}, 58--161 (1983).
1239: 
1240: \bibitem{Yoshida91}
1241: Z.~Yoshida, ``Discrete eigenstates of plasmas described by the
1242:   chandrasekhar-kendall functions,'' Prog.\ Theor.\ Phys. {\bf 86}, 45--55
1243:   (1991).
1244: 
1245: \bibitem{Yoshida92}
1246: Z.~Yoshida, ``Eigenfunction expansions associated with the curl derivatives in
1247:   cylindrical geometries: Completeness of chandrasekhar-kendall,'' J.\ Math.\
1248:   Phys. {\bf 33}, 1252--1256 (1992).
1249: 
1250: \bibitem{Cantarella00}
1251: J.~Cantarella, D.~DeTurck, H.~Gluck, and M.~Teytel, ``The spectrum of the curl
1252:   operator on spherically symmetric domains,'' Phys.\ Plasmas {\bf 7},
1253:   2766--2775 (2000).
1254: 
1255: \bibitem{Mueller}
1256: C.~Mueller, {\em Foundations of the mathematical theory of electromagnetic
1257:   waves}.
1258: \newblock New York, Springer Verlag (1969).
1259: 
1260: \bibitem{Montgomery78}
1261: D.~C. Montgomery, L.~Turner, and G.~Vahala, ``Three-dimensional
1262:   magnetohydrodynamic turbulence in cylindrical geometry,'' Phys. Fluids {\bf
1263:   21}, 757--764 (1978).
1264: 
1265: \bibitem{Shan91}
1266: X.~W. Shan, D.~Montgomery, and H.~D. Chen, ``Nonlinear magnetohydrodynamics by
1267:   galerkin-method computation,'' Phys.\ Rev.\ A {\bf 44}, 6800--6818 (1991).
1268: 
1269: \bibitem{Shan93a}
1270: X.~W. Shan and D.~Montgomery, ``On the role of the hartmann number in
1271:   magnetohydrodynamic activity,'' Plasma Phys.\ Cont.\ Fus. {\bf 35},
1272:   619--631 (1993).
1273: 
1274: \bibitem{Shan93b}
1275: X.~W. Shan and D.~Montgomery, ``Global searches of hartmann-number-dependent
1276:   stability boundaries,'' Plasma Phys.\ Cont.\ Fus. {\bf 35}, 1019--1032
1277:   (1993).
1278: 
1279: \bibitem{Shan94}
1280: X.~W. Shan and D.~Montgomery, ``Magnetohydrodynamic stabilization through
1281:   rotation,'' Phys.\ Rev.\ Lett. {\bf 73}, 1624--1627 (1994).
1282: 
1283: \bibitem{Li96a}
1284: S.~J. Li and D.~Montgomery, ``Decaying two-dimensional turbulence with rigid
1285:   walls,'' Phys.\ Lett.\ A {\bf 218}, 281--291 (1996).
1286: 
1287: \bibitem{Li97}
1288: S.~J. Li, D.~Montgomery, and W.~B. Jones, ``Two-dimensional turbulence with
1289:   rigid circular walls,'' Theor.\ Comp.\ Fluid Dyn. {\bf 9}, 167--181 (1997).
1290: 
1291: \bibitem{Kress00}
1292: B.~T. Kress and D.~C. Montgomery, ``Pressure determinations for incompressible
1293:   fluids and magnetofluids,'' J.\ Plasma Phys. {\bf 64}, 371--377 (2000).
1294: 
1295: \bibitem{Gallavotti}
1296: G.~Gallavotti, {\em Foundations of fluid dynamics}.
1297: \newblock Berlin, Springer Verlag (2002), 
1298: \newblock pp. 83--87 ff.
1299: 
1300: \bibitem{vapor}
1301: J.~Clyne and M.~Rast, ``A prototype discovery environment for analyzing and
1302:   visualizing terascale turbulent fluid flow simulations,'' in {\em
1303:   Visualization and data analysis 2005} (R.~F. Erbacher, J.~C. Roberts, M.~T.
1304:   Grohn, and K.~Borner, eds.), pp. 284--294, SPIE, Bellingham, Wash. (2005). 
1305: 
1306: \bibitem{Ivers03}
1307: D.~J. Ivers and C.~G. Phillips, ``A vector spherical harmonic spectral code for
1308:   linearised magnetohydrodynamics,'' ANZIAM J. {\bf 44}, C423--C442 (2003).
1309: 
1310: \bibitem{Press}
1311: W.~H. Press, B.~P. Flannery, S.~A. Teukolsky, and W.~T. Vetterling, {\em
1312:   Numerical Recipes in FORTRAN 77}.
1313: \newblock Cambridge, Cambridge Univ. Press (1992).
1314: 
1315: \bibitem{Canuto}
1316: C.~Canuto, Y.~Hussaini, A.~Quarteroni, and T.~Zang, {\em Spectral methods in
1317:   fluid dynamics}.
1318: \newblock New York, Springer Verlag (1988).
1319: 
1320: \bibitem{Matthaeus80}
1321: W.~H. Matthaeus and D.~Montgomery, ``Selective decay hypothesis at high
1322:   mechanical and magnetic reynolds numbers,'' Ann.\ N.\ Y.\ Acad.\ Sci. {\bf
1323:   357}, 203 (1980).
1324: 
1325: \bibitem{Ting86}
1326: A.~C. Ting, W.~H. Matthaeus, and D.~Montgomery, ``Turbulent relaxation
1327:   processes in magnetohydrodynamics,'' Phys.\ Fluids {\bf 29}, 3261--3274
1328:   (1986).
1329: 
1330: \bibitem{Kinney95}
1331: R.~Kinney, J.~C. McWilliams, and T.~Tajima, ``Coherent structures and turbulent
1332:   cascades in two-dimensional incompressible magnetohydrodynamic turbulence,''
1333:   Phys.\ Plasmas {\bf 2}, 3623--3639 (1995).
1334: 
1335: \bibitem{Mininni05d}
1336: P.~D. Mininni, D.~C. Montgomery, and A.~Pouquet, ``Numerical solutions of the
1337:   three-dimensional magnetohydrodynamic $\alpha$ model,'' Phys.\ Rev.\ E {\bf
1338:   71}, 046304 (2005).
1339: 
1340: \bibitem{Holm02a}
1341: D.~D. Holm, ``Averaged lagrangians and the mean effects of fluctuations in
1342:   ideal fluid dynamics,'' Physica D {\bf 170}, 253--286 (2002).
1343: 
1344: \bibitem{Holm02b}
1345: D.~D. Holm, ``Lagrangian averages, averaged lagrangians, and the mean effects
1346:   of fluctuations in fluid dynamics,'' Chaos {\bf 12}, 518--530 (2002).
1347: 
1348: \end{thebibliography}
1349: 
1350: \end{document}
1351: