1: \documentclass[12pt,preprint]{aastex}

2: 

3: %\usepackage{amsmath,amssymb}

4: 

5: \shorttitle{Predicting the Solar Cycle}

6: \shortauthors{Bushby \& Tobias}

7: 

8: % ....................................................

9: \newcommand{\ovr}[2]{\ensuremath{\frac{{#1}}{r^{#2}}}}

10: \newcommand{\dprime}{\ensuremath{{\prime\prime}}}

11: \newcommand{\etal}{{\it et al }}

12: % ....................................................

13: 

14: 

15: %=============================================================================

16: \begin{document}

17: 

18: %% \bibliographystyle{apj}

19: 

20: 

21: \title{On Predicting the Solar Cycle using Mean-Field Models}

22: 

23: 

24: \author{Paul J. Bushby\altaffilmark{1}}

25: \email{P.J.Bushby@damtp.cam.ac.uk}

26: 

27: 

28: \author{Steven M. Tobias\altaffilmark{2}}

29: \email{smt@maths.leeds.ac.uk}

30: 

31: 

32: \altaffiltext{1}{DAMTP, Centre for Mathematical Sciences, University of

33:   Cambridge, Wilberforce Road, Cambridge CB3 0WA, U.K.}

34: 

35: \altaffiltext{2}{Department of Applied Mathematics, University of Leeds,

36:   Leeds LS2 9JT, U.K.}

37: 

38: \begin{abstract}

39: We discuss the difficulties of predicting the solar cycle using

40: mean-field models. Here we argue that these difficulties arise owing

41: to the significant modulation of the solar activity cycle, and that

42: this modulation arises owing to either stochastic or deterministic

43: processes. We analyse the implications for predictability in both of

44: these situations by considering two separate solar dynamo models. The

45: first model represents a stochastically-perturbed flux transport

46: dynamo. Here even very weak stochastic perturbations can give rise to

47: significant modulation in the activity cycle. This modulation leads to

48: a loss of predictability. In the second model, we neglect stochastic

49: effects and assume that generation of magnetic field in the Sun can be

50: described by a fully deterministic nonlinear mean-field model --- this

51: is a best case scenario for prediction. We designate the output from

52: this deterministic model (with parameters chosen to produce

53: chaotically modulated cycles) as a target timeseries that subsequent

54: deterministic mean-field models are required to predict. Long-term

55: prediction is impossible  even if a model that is correct in all

56: details is utilised in the prediction. Furthermore, we show that even

57: short-term prediction is impossible if there is a small discrepancy in

58: the input parameters from the fiducial model. This is the case even if

59: the predicting model has been tuned to reproduce the output of

60: previous cycles. Given the inherent uncertainties in determining  the

61: transport coefficients and nonlinear responses for mean-field models,

62: we argue that this makes predicting the solar cycle using the output

63: from such models impossible.

64: \end{abstract}

65: 

66: \keywords{(magnetohydrodynamics:) MHD -- Sun: activity -- Sun: magnetic fields}

67: 

68: %=============================================================================

69: \section{Introduction}

70: 

71: Magnetic activity in the Sun is known to play a central role in

72: driving both long-term and short-term dynamics (Tobias 2002; Weiss

73: 2002). The magnetic field is responsible for spectacular

74: events such as sunspots, solar flares, and coronal mass ejections, and for

75: heating the solar corona to high temperatures. Large-scale magnetic

76: activity is known to be dominated by the eleven year activity

77: cycle. This cycle has been systematically observed since the early

78: seventeenth century and its properties are well documented (see

79: e.g. Ossendrijver 2003). Of particular current interest is the impact

80: of magnetic activity on solar irradiance that might have significant

81: implications for the terrestrial climate (see Solanki \etal 2004).

82: 

83: Given the importance of solar activity, it is not surprising that

84: there has been a continued interest in understanding the mechanisms

85: responsible for generating the solar magnetic field. The Sun's

86: magnetic field is believed to be generated by a hydromagnetic dynamo

87: in which motion of the solar plasma (advection) is able to sustain a

88: magnetic field against the continued action of ohmic dissipation (see

89: e.g Moffatt 1978; Charbonneau 2005). Progress in understanding this

90: fundamental problem of solar magnetohydrodynamics is slow owing to the

91: difficulties of the dynamo problem. The extreme parameters of the

92: solar interior and the inherent three-dimensionality of the dynamo

93: problem make it impossible to solve the equations accurately on a

94: computer. Much effort has therefore focused on {\it mean-field} dynamo

95: models (Steenbeck, Krause \& R\"adler 1966; Krause \& R\"adler 1980),

96: which describe the evolution of the mean magnetic field,

97: parameterising the effects of the small-scale fields and flows in

98: terms of tensor transport coefficients. These transport coefficients

99: include $\alpha_{ij}$ (which leads to a regenerative term in the

100: mean-field equations --- the so-called $\alpha$-effect) and the

101: turbulent diffusivity ($\beta_{ijk}$). We stress here that there is no

102: mechanism within the theory for determining the form of these

103: coefficients, except for flows at low magnetic Reynolds number or with

104: short correlation time, and in solar models these are usually chosen

105: in a plausible but ad-hoc manner (often, for simplicity, adopting

106: isotropic representations in which $\alpha_{ij}=\alpha \delta_{ij}$

107: and $\beta_{ijk}=\beta\epsilon_{ijk}$). Much attention has been

108: focused upon determining these transport coefficients in both the

109: linear and nonlinear dynamo regimes from numerical simulations

110: (Cattaneo \& Hughes 1996; Brandenburg \& Subramanian 2005) but there

111: is still no consensus over the nature of these, even to within an order of

112: magnitude (see Courvoisier, Hughes \& Tobias 2006). Mean-field

113: models have, however, proved successful in providing illustrations of

114: the type of behaviour that might be expected to occur in the Sun (and

115: other stars). It is often argued that, {\it although these models have

116: no predictive power}, understanding the underlying mathematical form

117: of the equations can lead to the identification of robust patterns of

118: behaviour.

119: 

120: Many different models have been proposed for the solar dynamo. In the

121: distributed dynamo model, the $\alpha$-effect operates throughout the

122: convection zone and interacts with the latitudinal shear (or the

123: sub-surface shear layer, see Brandenburg 2005) to generate magnetic

124: field. Alternatively, the dynamo could be operating near the

125: tachocline, where an $\alpha$-effect might be driven either by a

126: tachocline-based instability or by turbulent convection. This, in

127: conjunction with the strong shear, could drive an ``interface''

128: dynamo (Parker 1993). Finally, there are flux transport models, in

129: which the (so-called) Babcock-Leighton mechanism produces an

130: $\alpha$-effect (or source term) at the surface. This surface

131: $\alpha$-effect is coupled to the radial shear in the tachocline

132: (where another $\alpha$-effect may be operating) via a meridional flow

133: (Choudhuri, Sch\"ussler \& Dikpati 1995; Dikpati \& Charbonneau

134: 1999). The relative merits of these models are discussed elsewhere in

135: the literature (see, e.g. Charbonneau 2005) --- the only comment we

136: make here is that this plethora of models arises because of the lack

137: of available constraints on the form of the transport coefficients in

138: the mean-field formalism. We note further that it is not clear that

139: any of the above scenarios capture the essential dynamo processes

140: correctly or that these processes can ever be captured by a mean-field

141: model.

142: 

143: It is also possible to construct predictions of solar activity without

144: using dynamo theory, and there is a long literature describing these

145: predictive methods (see e.g. Zhang 1996; Hathaway, Wilson \& Reichmann

146: 1999; Sello 2003; Zhao \etal 2004; Saba, Strong \& Slater 2005). One class

147: of prediction techniques uses statistical and timeseries analysis

148: methods (see e.g. Tong 1995 for details). These methods, which are

149: also applicable in many other areas of physics, vary in complexity

150: from simple linear methods to methods that use dynamical systems

151: theory to reconstruct nonlinear attractors in phase space. However,

152: these methods have the drawback that they do not utilise any of the

153: ``physics'' of the problem. Predictions can also be made by using

154: precursor methods (see e.g. Schatten 2002), which do utilise some of

155: the physical features of the system in addition to the timeseries data.

156: 

157: In recent papers (Dikpati, de Toma \& Gilman 2006; Dikpati \& Gilman

158: 2006), an attempt has been made to unify these two approaches by

159: utilising a mean-field model in order to make predictions about the

160: future activity of the Sun. These papers describe an axisymmetric,

161: mean-field model of a flux transport dynamo. Here the authors make use

162: of the observations of magnetic flux at the solar surface to feed into

163: a model of solar activity.  The flux that is observed at the solar

164: surface is advected by a parameterised meridional flow (which can be

165: observed down to a certain depth) and interacts with a differential

166: rotation profile that has been inferred from helioseismology. The

167: magnetic flux also interacts with turbulence, the effects of which are

168: parameterised by certain turbulent transport coefficients

169: (representing the turbulent diffusivity and the $\alpha$-effect). As

170: with all current mean field models these turbulent transport effects

171: have been parameterised in a plausible but ad-hoc manner, and are

172: unconstrained by observations and indeed theory. The simplest predictive scheme

173: proposed by Dikpati {\it et al} (2006) therefore takes the form of a

174: parameterised linear system forced by boundary observations. The

175: implicit underlying philosophy here is that by reducing the correct

176: physics for the generation of the solar activity cycle (i.e. a

177: nonlinear self-excited dynamo) to such a scheme, predictions about

178: future solar activity can be made. 

179: 

180: In this paper we shall investigate the predictability of various

181: dynamo models. We demonstrate that even when all the nonlinear

182: physics of the solar dynamo is removed, problems remain for prediction

183: owing to the increased importance of stochastic effects --- even very

184: weak stochastic perturbations can produce significant modulation in

185: these linear-type models. We also discuss the best-case

186: scenario for prediction where stochastic effects can be ignored, and

187: demonstrate that in these cases prediction is still difficult owing to

188: uncertainties in the input parameters of these parameterised

189: mean-field models.   

190: 

191: The paper is organised as follows. In the next section we describe (in

192: a general way) the importance of modulation and the role of

193: stochasticity and nonlinearity in solar dynamo models. In section 3 we

194: investigate a flux transport model and demonstrate how the presence of

195: even extremely weak noise can render predictions useless. In section 4

196: we consider the ``best-case" scenario for prediction where noise does

197: not play a role in the modulation --- we demonstrate that more

198: accurate prediction schemes may arise by using basic timeseries

199: analysis techniques rather than from constructing mean-field models of

200: the solar cycle. Finally, in section 5 we discuss the implications of

201: our work for predictions of the solar cycle. 

202: 

203: 

204: \section{Problems for prediction and mechanisms for modulation}

205: 

206: In this section, we discuss the problems that must be overcome by

207: schemes designed to yield a prediction of future solar magnetic

208: activity. Some of these problems arise owing to the nature of solar

209: magnetic activity whilst others arise from the lack of a detailed

210: theory that is capable of describing solar magnetic activity in such

211: extreme conditions as those that exist in the solar interior.

212: 

213: It is clear that if the solar cycle were strictly periodic, with a

214: constant amplitude, then it would be straightforward to predict future

215: behaviour. However, all measurements of solar magnetic activity (both

216: direct observations and evidence from proxy data) indicate that the

217: variations in the magnetic activity do not follow a periodic

218: pattern. Departures from periodicity may be driven either by

219: perturbations or by modulation. For the case of a weakly perturbed

220: periodic system, the dynamics is essentially captured by the periodic

221: signal, with the small perturbations playing a secondary role. We

222: distinguish this behaviour from a modulated signal in which there are

223: significant departures from periodicity (often occurring on longer

224: timescales), with large variations in the observed amplitude of the

225: signal. All the evidence from direct observations indicates that the

226: solar cycle is strongly modulated. The amplitude of the solar cycle

227: varies enormously over long timescales, an extreme example of this

228: modulation was a period of severely reduced activity in the

229: seventeenth century known as the Maunder Minimum. Proxy data from

230: records of terrestrial isotopes, such as $^{10}$Be and $^{14}$C (see

231: e.g. Beer 2000, Weiss \& Tobias 2000, Wagner {\it et al} 2001),

232: demonstrate that this modulation has been a characteristic feature of

233: the solar magnetic activity over (at least) the last 20,000 years.

234: 

235: Mathematically there are only two possible sources for this strong

236: modulation of the basic solar cycle (Tobias 2002). The modulation may

237: arise either as a result of stochastic effects (see e.g.  Ossendrijver

238: \& Hoyng 1996) or by deterministic processes (see e.g. Tobias, Weiss

239: \& Kirk 1995). In this context we define deterministic processes to be

240: those that {\it are} captured by the differential equations of dynamo

241: theory, with no random elements. Stochastic processes are those

242: that occur on an unresolved length or timescale, and so can not be

243: described by the differential equations without including a random

244: element into the model.

245: 

246: It is well known that stochastic modulation can arise even if the

247: deterministic physics that leads to the production of the basic cycle

248: is essentially linear. This parameter regime is generally considered to be a

249: good one for prediction, since any nonlinear effects are only playing

250: a secondary role. However, in this stochastically-perturbed case, the

251: small random fluctuations that lead to the modulation will have large

252: short-term effects and render prediction extremely difficult, if not

253: impossible. Conversely, if the modulation arises purely as a result of

254: deterministic processes, then the underlying physics is nonlinear (or

255: potentially non-autonomous) and this leads to difficulty in prediction

256: owing to the possible presence of deterministic chaos and (more

257: importantly) the difficulty of constructing accurate nonlinear models

258: with large numbers of degrees of freedom. 

259: 

260: In the next two sections we demonstrate the problems for prediction

261: for dynamo models in both of the classes described above. In the next

262: section we describe a flux transport model of the same type as the one

263: used in the prediction scheme of Dikpati {\it et al} (2006) and we

264: demonstrate that even very small random fluctuations can produce

265: significant modulation, leading to extreme difficulties for

266: prediction. We then, in section 4, go on to describe a model where the

267: modulation arises owing to the presence of deterministic chaos and show that in

268: this case, prediction using model fitting is a poor way to proceed, but some

269: prediction is possible if it is possible to reconstruct the attractor

270: for activity. 

271: 

272: \section{Prediction using a stochastically-perturbed flux transport

273:   dynamo model} 

274: 

275: \subsection{The dynamo model}

276: 

277: We assume initially that the modulated solar magnetic activity can be

278: described by a stochastically-perturbed mean-field

279: dynamo model. In this model, nonlinear effects are playing a secondary

280: role, and all the modulation is being driven by the stochastic

281: effects. The aim of this section is to assess whether or not models

282: of this type can be used to make meaningful predictions of the solar

283: magnetic activity. In these models, the evolution of the large-scale

284: magnetic field is described by the standard mean-field equation (see,

285: for example, Moffatt 1978),   

286: 

287: \begin{equation}

288: \frac{\partial \mathbf{B}}{\partial t} = \nabla \times \left(

289: \alpha\mathbf{B}+\mathbf{U} \times \mathbf{B} - \beta \nabla \times \mathbf{B} \right).\label{eqn:1}

290: \end{equation}

291: 

292: \noindent Here, $\mathbf{B}$ represents the large-scale magnetic

293: field and $\mathbf{U}$ corresponds to the mean velocity field, $\beta$ is

294: the (turbulent) magnetic diffusivity, and the $\alpha\mathbf{B}$ term

295: corresponds to the mean-field $\alpha$-effect. Using the well-known

296: $\alpha\omega$ approximation, we solve this equation numerically in an

297: axisymmetric spherical shell ($0.6R_{\odot} \le r \le

298: R_{\odot}$ and $0 \le \theta \le \pi$). In solving

299: Equation~(\ref{eqn:1}) we need to ensure that $\mathbf{B}$ remains

300: solenoidal (i.e. $\nabla \cdot \mathbf{B}=0$). To achieve

301: this, we decompose the magnetic field into its poloidal and toroidal

302: components,  

303: 

304: \begin{equation}

305: \mathbf{B}= B(r,\theta,t)\mathbf{e_{\phi}} + \nabla \times \left(

306: A(r,\theta,t)\mathbf{e_{\phi}}\right),\label{eqn:2}

307: \end{equation}

308: 

309: \noindent where $B(r,\theta,t)$ denotes the toroidal (azimuthal) field

310: component and the scalar potential $A(r,\theta,t)$ relates to the

311: poloidal component of the magnetic field. So, rather than solving

312: Equation~(\ref{eqn:1}) directly, the problem has been reduced to solving two

313: coupled partial differential equations for the scalar quantities

314: $A(r,\theta,t)$ and $B(r,\theta,t)$. We adopt idealised boundary

315: conditions, in which $A=B=0$ at $\theta=0$ and $\theta=\pi$ and

316: $r=0.6R_{\odot}$ and $A$ and $B$ are smoothly matched to a potential

317: field at $r=R_{\odot}$.  

318: 

319: \par This particular dynamo model is closely related to the flux

320: transport model described by Dikpati \& Charbonneau (1999). The

321: large-scale velocity field, $\mathbf{U}$, is given by    

322: 

323: \begin{equation}

324: \mathbf{U}= u_r(r,\theta)\mathbf{e_{r}} +

325: u_{\theta}(r,\theta)\mathbf{e_{\theta}} +

326: \Omega(r,\theta)r\sin\theta\mathbf{e_{\phi}},\label{eqn:3}

327: \end{equation}

328: 

329: \noindent where $\Omega(r,\theta)$ is a prescribed analytic fit to the

330: helioseismologically-determined solar rotation profile (see, for

331: example, Bushby 2006) and $u_r$ and $u_{\theta}$ correspond to a

332: prescribed meridional circulation. We assume that the meridional

333: circulation pattern in each hemisphere consists of a single cell, with

334: a polewards flow at the surface and an (unobservable) equatorwards

335: flow at the base of the convection zone --- the flow is confined to the

336: region $R_b \le r \le R_{\odot}$. The functional form that we adopt

337: for this flow is similar in form to the one described by Dikpati \&

338: Charbonneau (1999),  

339: 

340: \begin{eqnarray}

341: u_r(r,\theta)&=&U_o \left( \frac{R_{\odot}}{r} \right)^2 \left[

342:   -\frac{2}{3} + \frac{1}{2}c_1\xi^{0.5}-

343:   \frac{4}{9}c_2\xi^{0.75}\right] \xi \sin \theta \left( 3\cos^2

344: \theta - \sin^2 \theta \right),\label{eqn:4}\\

345: u_{\theta}(r,\theta)&=&U_o \left( \frac{R_{\odot}}{r} \right)^3 \left[

346:   -1 + c_1\xi^{0.5}-c_2\xi^{0.75}\right] \cos \theta \sin^2

347: \theta,\label{eqn:5}  

348: \end{eqnarray}

349:  

350: \noindent where $\xi(r)=[(R_{\odot}/r)-1]$, $c_1=4[\xi(R_b)]^{-0.5}$,

351: $c_1=3[\xi(R_b)]^{-0.75}$, and $U_o$ is some characteristic flow

352: speed. This flow pattern can be stochastically perturbed by setting

353: $R_b=0.7R_{\odot}+\epsilon(t)$, where $\epsilon(t)$ is a time-dependent,

354: randomly fluctuating variable in the range $-0.005R_{\odot} \le

355: \epsilon \le 0.005R_{\odot}$. The aim here is to assess whether or not

356: such weak stochastic variations in the flow pattern could give rise to

357: significant modulation in the activity cycle, and if so what are the

358: consequences for prediction.

359: 

360: \par In order to complete the specification of the model, we need to

361: choose plausible functional forms for the $\alpha$-effect and the

362: turbulent magnetic diffusivity. It should be emphasised again that

363: these mean-field coefficients are poorly constrained by theory and

364: observations, although plausible assumptions can be made. Defining

365: $\beta_o$ to be a characteristic value of the turbulent magnetic

366: diffusivity within the solar convection zone, we adopt a similar

367: spherically-symmetric profile to that adopted by Dikpati \&

368: Charbonneau (1999),  

369: 

370: \begin{equation}

371: \beta(r) = \frac{1}{2} (\beta_o-\beta_c)\left[ 1 + \mbox{erf}\left(\frac{r-0.7R_{\odot}}{0.025R_{\odot}} \right)\right] +

372:   \beta_c, \label{eqn:6}

373: \end{equation}

374: 

375: \noindent where erf corresponds to the error function and $\beta_c$

376: (here taken to be $1\%$ of $\beta_o$) represents the magnetic

377: diffusivity below the turbulent convection zone. Following Dikpati

378: \& Charbonneau (1999), rather than prescribing a simple functional

379: form for $\alpha$ we neglect the $\alpha$-effect term in the toroidal

380: ($B$) field equation and replace the corresponding $\alpha B$ term in

381: the poloidal ($A$) equation by a non-local, nonlinear source of

382: poloidal flux,  

383: 

384: \begin{eqnarray}

385: S(r,\theta,t)&=&\frac{S_o}{2}\left[ 1 +

386:   \mbox{erf}\left(\frac{r-0.95R_{\odot}}{0.01R_{\odot}} \right)\right]

387: \left[ 1 - \mbox{erf}\left(\frac{r-R_{\odot}}{0.01R_{\odot}}

388:   \right)\right] \\ \nonumber &&\left[

389:   1+\left(\frac{B(0.7R_{\odot},\theta,t)}{B_o}\right)^2\right]^{-1}\sin

390:   \theta \cos \theta B(0.7R_{\odot},\theta,t).\label{eqn:7}

391: \end{eqnarray}

392: 

393: \noindent Here, $S_o$ is a characteristic value of this poloidal

394: source and $B_o$ represents the (somewhat arbitrarily chosen) field

395: strength at which this non-local source becomes suppressed by the

396: magnetic field. This source term parameterises the contribution to the

397: poloidal magnetic flux due to the decay of active regions --- the

398: non-locality reflects the fact that active regions are believed to

399: form as the result of buoyant magnetic flux rising from the base of

400: the convection zone to the solar photosphere. See Dikpati \&

401: Charbonneau (1999) for a more detailed discussion of this source term, though

402: again it must be stressed that the functional form and the nonlinear dependence

403: are chosen in a plausible yet ad-hoc manner. 

404: 

405: \subsection{Numerical results}

406: 

407: In order to carry out numerical simulations, we first

408: non-dimensionalise this flux transport model. By using scalings similar

409: to those described by Dikpati \& Charbonneau (1999), it can be shown

410: that the model solutions are fully determined by two non-dimensional

411: parameters (once other parameters such as $B_o$ have been selected).

412: Denoting the equatorial angular velocity at the solar

413: surface by $\Omega_{eq}$, these non-dimensional parameters are the

414: Dynamo number, $D=S_o\Omega_{eq}R_{\odot}^3/\beta_o^2$, and the

415: magnetic Reynolds number corresponding to the meridional flow,

416: $Re=U_oR_{\odot}/\beta_o$. Here, we set $D=7 \times 10^6$ and

417: $Re=5600$. In the absence of stochastic noise, this set

418: of parameters produces a strong circulation-dominated dynamo in which the

419: magnetic energy is a periodic function of time. Although the dynamo

420: number is not weakly supercritical, nonlinear effects are not strong

421: enough here to produce a modulated activity cycle --- the primary role

422: of the nonlinearity is to prevent the unstable dynamo mode from

423: growing exponentially. We term such a model a ``linear-type'' model. 

424: 

425: \par When weak stochastic effects are included in the model, the

426: resulting activity cycle is indeed weakly modulated. This is

427: illustrated in Figure~1, which shows the time-dependence of this

428: solution. The time-series clearly illustrates that, although the

429: amplitude of the ``cycle minimum'' only appears to be weakly

430: time-dependent, there are significant variations in the peak amplitude

431: of the magnetic energy time-series. These variations are qualitatively

432: similar to those observed by Charbonneau \& Dikpati (2000), who

433: considered large amplitude random fluctuations in the flow pattern

434: within the solar convection zone --- the peak amplitude of these

435: fluctuations was comparable with the peak amplitude of the flow. In this

436: particular model, we have shown that even very weak stochastic

437: variations in the centre of mass of the flow pattern can still produce

438: significantly modulated behaviour. These stochastic effects are

439: expected to become increasingly significant for dynamo numbers

440: approaching critical. So, these models are obviously highly sensitive

441: to the addition of stochastic noise.   

442: 

443: \par In the absence of stochastic noise, the attractor (in phase

444: space) for this solution is two-dimensional, and the future behaviour

445: of the solution at any instant in time is entirely determined by the

446: current position of the system on the attractor. The same is not true

447: when this system is perturbed by stochastic effects, and it clearly

448: becomes much more difficult to predict the future behaviour of the

449: system. Since the attractor of this stochastically perturbed solution

450: cannot be unambiguously defined, another possible way of assessing the

451: ``predictability'' of this solution is to look for a correlation

452: between successive cycle maxima. Defining $T_n$ to be the magnitude of

453: the $n^{th}$ cycle maximum, Figure~2 shows $T_{n+1}$ as a function of

454: $T_n$. It is clear from this scatter plot that there is no obvious

455: correlation between the amplitudes of successive cycle maxima in this

456: case. Since the modulation is being driven entirely by random

457: stochastic forcing, this result is not surprising. This lack of

458: correlation suggests that the behaviour of previous cycles cannot be

459: used to infer the magnitude of the following one. This implies that

460: even weak stochastic effects may seriously reduce the possibilities

461: for solar cycle prediction in this linear-type regime.

462: 

463: \section{Predictions using a deterministic dynamo model}

464: 

465: \subsection{The dynamo model}

466: 

467: In the previous section, we demonstrated that even very weak

468: stochastic perturbations to the meridional flow pattern can lead to a

469: loss of predictability in a linear-type flux transport dynamo

470: model. In that model, the modulation of the activity cycle was driven

471: entirely by stochastic effects. As discussed in Section~2, the only other

472: possible scenario is that the observed modulation is driven by

473: nonlinear effects. This scenario, where the observed modulation is

474: deterministic in origin, is the ``best-case'' scenario for prediction,

475: as in this case the entirely unpredictable stochastic elements may be

476: ignored. We stress again that, given that solar magnetic activity is

477: significantly modulated, either deterministic or stochastic modulation must be

478: considered in any realistic model (predictive or otherwise) of the

479: solar cycle. So, in this section, we completely neglect stochastic effects and

480: assume that the observed (chaotic) modulation in the solar magnetic

481: activity can be described by a fully deterministic model in which any activity

482: modulation (e.g. solar-like ``Grand minima'') is driven entirely by

483: nonlinear effects. The model that we use was described in detail in

484: two recent papers (Bushby 2005, 2006), so we only present a brief

485: description here. The exact details of the model are unimportant for

486: our main conclusions.   

487: 

488: \par Like the flux transport dynamo model from the previous section,

489: this model describes an axisymmetric, mean-field,

490: $\alpha\omega$-dynamo in a spherical shell. Unlike the previous model,

491: this model represents an ``interface-like'' dynamo that is operating

492: primarily in the region around the base of the solar convection

493: zone. It is worth mentioning again that (as discussed in

494: the introduction) there is still no general consensus regarding which of

495: these dynamo scenarios is more likely to be an accurate representation

496: of the solar dynamo. For this interface-like dynamo model, we neglect

497: meridional motions, since they are poorly determined near the base of the solar

498: convection zone. Like several earlier models (e.g. Tobias 1997;

499: Moss \& Brooke 2000; Covas \etal 2000), this dynamo model includes the

500: feedback (via the azimuthal component of the Lorentz force) of the

501: mean magnetic field upon the differential rotation (Malkus \& Proctor

502: 1975). This nonlinear feedback is a crucial element of the model and, in the

503: absence of stochastic effects, is the sole driver of modulation in the

504: magnetic activity cycle. Denoting this magnetically-driven velocity

505: perturbation by $V(r,\theta,t)$, the large-scale velocity field is

506: given by

507: 

508: \begin{equation}

509: \mathbf{U} = \left[\Omega(r,\theta)r\sin\theta +

510:   V(r,\theta,t)\right]\mathbf{e_{\phi}},

511: \label{eqn:8}

512: \end{equation}

513: 

514: \noindent where (as in the previous model) $\Omega(r,\theta)$

515: represents an analytic fit to the solar differential

516: rotation. Whilst the evolution of the large-scale magnetic field is

517: again governed by Equation~(1), an additional evolution equation is

518: required for the velocity perturbation, $V$. This equation is given by 

519: 

520: \begin{eqnarray}

521: \frac{\partial V}{\partial t}&=& \frac{1}{\mu_o

522: \rho} \left[(\nabla \times \mathbf{B}) \times \mathbf{B} \right]\cdot

523: \mathbf{e_{\phi}} +

524:  \frac{1}{r^3}\frac{\partial}{\partial r}\left[\nu r^4

525: \frac{\partial}{\partial r} \left(\frac{V}{r}\right)\right] \\

526: \nonumber &&+

527: \frac{1}{r^2 \sin^2 \theta}\frac{\partial}{\partial \theta}\left[\nu

528:   \sin^3 \theta \frac{\partial}{\partial \theta} \left(\frac{V}{\sin

529: \theta}\right)\right],\label{eqn:9}

530: \end{eqnarray}

531:  

532: \noindent where $\rho$ represents the fluid density (here taken to be

533: constant), $\mu_o$ is the permeability of free space and $\nu$

534: represents the (turbulent) fluid viscosity.

535: 

536: 

537: \par In order to complete the model, the spatial dependence of the

538: transport coefficients ($\alpha$, $\beta$ and $\nu$) must also be

539: specified. Again, we emphasise that there are no direct

540: observational constraints relating to these coefficients --- as noted

541: in the introduction, there is no consensus as to their form and there

542: is still a debate as to their order of magnitude (and even their

543: sign). Having said that, it is possible to make some plausible

544: assumptions for an ``interface-like'' dynamo model (see Bushby 2006

545: for more details). The precise choices of these parameters are unimportant for

546: our main conclusions.  

547: 

548: \par Having set up this model, it is possible to choose a set of

549: parameters so that the solutions {\it do} reproduce some salient

550: features of the solar dynamo (Bushby 2005, 2006). We stress here that,

551: although the parameters have been chosen in a plausible manner, this

552: dynamo model should not be regarded as an accurate representation of

553: the solar interior and is subject to many uncertainties. Furthermore

554: we stress again that this is the case with {\it all} mean-field solar

555: dynamo models. However we use this model as a useful tool to analyse

556: the possibility of producing predictive models of the solar cycle. We

557: proceed by choosing fiducial parameters and profiles for the turbulent

558: transport coefficients that lead to ``solar-type" magnetic activity,

559: with chaotically modulated cycles and recurrent ``Grand Minima''. We

560: then integrate this model forward in time to produce a timeseries and

561: designate this timeseries as the ``target" run, which any subsequent

562: model should be able to predict. This target run is shown in Figure~3,

563: which shows a timeseries of the activity together with a

564: reconstruction of the dynamo attractor in phase space. Although this

565: solution is chaotically modulated, it is certainly no more chaotic than

566: the equivalent attractor for the $^{10}$Be data, which is a well-known

567: proxy for solar magnetic activity (e.g. Beer 2000). Whilst the

568: nonlinear effects are significant enough to drive the modulation, they

569: are actually very difficult to detect. In this model, the cyclic

570: component of the fluctuations in the differential rotation (which are

571: driven by the nonlinear Lorentz force) are small compared with the

572: mean differential rotation. This is consistent with observations of

573: the (so called) torsional oscillations in the solar convection zone. Finally,

574: note once more that, since the modulation is driven entirely by

575: nonlinear effects, this model is specified exactly. 

576: 

577: \subsection{Numerical results}

578: 

579: The question is then posed as to whether {\it any} mean-field model

580: can be constructed that leads to meaningful predictions of the future

581: behaviour of the target run. Clearly the best chance for a mean-field

582: model being capable of predicting the future behaviour of the target

583: run is to use {\it the exact model} that led to the target run

584: data. Hence we test this model first, as all subsequent models will be

585: inferior to this. We proceed by setting the model parameters to be

586: those that generated the long test run, and consider the behaviour of

587: solutions that are started from very similar points on the

588: attractor. Some of the solutions are shown in Figure~4. This figure

589: shows clearly that although the predictor solutions are able to track

590: the target solution for a couple of activity cycles, the nature of the

591: solutions means that the predictors and target solution diverge

592: quickly after this time. This is not surprising behaviour. It is

593: well-known that chaotic solutions have a sensitive dependence on

594: initial conditions and that long-term prediction of such solutions is

595: fraught with problems (see e.g. Tong 1995). What is clear is that

596: simply using a model that is based upon mean-field theory will not

597: work in the long term {\it even if the model is correct in every

598: detail}. One might be able to predict one or two cycles ahead {\it if

599: one has solved the problem of constructing an exact representation of

600: the solar dynamo} but as noted above this is not an easy task.

601: 

602: We now turn to the related problem of short-term prediction. As

603: discussed in the introduction there are large uncertainties in the

604: form and amplitude of the input parameters for all mean-field models.

605: What we investigate here is whether these uncertainties lead to

606: significant difficulties in prediction even in the short term. Again

607: we examine the best case scenario and consider a mean-field model for

608: the predictor runs that has correctly parameterised the {\it form} of

609: all the input variables (differential rotation, $\alpha$-effect,

610: turbulent diffusivity and nonlinear response). In addition these

611: predictors have been given the correct input {\it values} for

612: all-but-one of the parameters. Hence the predictor models are exactly

613: the same as the target model with the exception of one input parameter

614: that has been altered by $5\%$. This would be a staggeringly good

615: representation should it be possible to achieve this for solar

616: activity. Furthermore we increase the chances of the predictor being

617: able to predict the future behaviour of the target solution by

618: matching the two timeseries over a number of cycles. This is analogous

619: to the procedure employed by Dikpati \etal (2006) who cite support for

620: their forecasting model by assuring that their model agrees with the

621: solar cycle data for eight solar cycles --- in reality this is not

622: difficult to achieve with enough model parameters at one's

623: disposal. Figure~5 shows the results of integrating the predictor

624: models for two different choices of incorrect parameter. Note that

625: even though the predictor has been designed to reproduce the target

626: over a number of cycles and that the predictor is very closely related

627: to the target, there is still a good chance that it can get the next

628: cycle incorrect, with significant errors in (particularly) the cycle

629: amplitude. There are also clear variations in the cycle period, which

630: obviously implies that the exact time between successive cycle maxima

631: is also an unpredictable feature of the system.   

632: 

633: We stress again that any mean-field model of solar activity includes

634: transport coefficients that are still uncertain possibly to an order of

635: magnitude (and certainly not to $5\%$ accuracy). Although the

636: incredible success of global and local helioseismology is placing

637: restrictions on the form of the differential rotation and the

638: meridional flows, it is unlikely in the foreseeable future that

639: significant constraints will be put on the transport coefficients or their

640: nonlinear response to the mean magnetic field.

641: 

642: \section{Predictions using a reconstruction of the attractor}

643: 

644: Having established that there are difficulties in obtaining reliable

645: predictions by fitting mean-field models (even if the modulation is

646: deterministic in origin), it is of interest to determine

647: whether or not more reliable predictions could be obtained by

648: utilising more general timeseries analysis techniques. In order to

649: reconstruct an attractor from a given timeseries, it is necessary to

650: define a corresponding phase space. There are various ways of doing

651: this, but given (any) discrete timeseries, $x(t)$, in which the data

652: is sampled at intervals of $\Delta t$, the vector 

653: 

654: \begin{equation} 

655: \mathbf{X}(t) = \left[ x(t), x(t-\Delta t)..., x(t-(d-1)\Delta t)\right]

656: \label{eqn:10}

657: \end{equation}

658: 

659: \noindent defines a

660: point in a $d$-dimensional ``embedded'' phase space (see, e.g., Farmer

661: \& Sidorowich 1987; Casdagli 1989). Given a time $T$, the idea of a

662: prediction algorithm is to find a mapping $f$ such that $f

663: \left(\mathbf{X}(T)\right)$ gives a good approximation to $x(T+\Delta

664: t)$. The predictive mapping technique that is used here uses a local

665: approximation method (see, e.g., Casdagli 1989), which considers the

666: behaviour of the nearest neighbours, in phase space, to

667: $\mathbf{X}(T)$. By using a least squares fit, the subsequent

668: evolution of each of these neighbouring points in phase space is used

669: to construct a piecewise-linear approximation to the predictive map,

670: $f$. This approximate mapping can then be applied to $\mathbf{X}(T)$

671: to obtain an estimate for $x(T+\Delta t)$. This algorithm can then be

672: repeated to find estimates for $x(T+2 \Delta t)$ and subsequent

673: points. The optimal value for $d$ can be determined by minimising the

674: error of this predictive algorithm over the known segment of the

675: timeseries.

676: 

677: The results of applying this predictor algorithm to the target

678: solution are also shown in Figure~5, where the timeseries

679: predictions are shown as crosses. The prediction is started from the

680: cycle maximum before the mean-field predictor diverges from the

681: target. Longer training timeseries lead to a more densely-populated

682: reconstructed attractor, which increases the probability of making

683: more accurate predictions. However, rather than using the

684: entire target run, these predictions are based upon (approximately)

685: $50$ cycles --- this will give a fairer comparison between these

686: results and timeseries predictions that are based upon the real sunspot

687: data. The application of the algorithm to earlier segments of

688: the timeseries suggests that a value of $d \ge 5$ is required in

689: order to minimise predictive errors. As can be seen from Figure~5,

690: this algorithm appears to predict the magnitude of the maximum of the

691: following cycle to a reasonable degree of accuracy, although the

692: predictions subsequently diverge from the target. Whilst neither of

693: these techniques are capable of producing reliable long-term predictions,

694: these results do suggest that for the short-term prediction of solar

695: magnetic activity, timeseries analysis techniques may provide a

696: viable alternative to predictions based simply upon mean-field dynamo

697: models (provided stochastic effects can be neglected). 

698: 

699: \section{Conclusions}

700: 

701: Solar magnetic activity arises as a result of a hydromagnetic dynamo

702: --- that much we believe  to be true. As yet, there is no consensus on

703: the location of the dynamo, the dominant nonlinear or stochastic

704: effects, or even the fundamental processes that are responsible for

705: the operation of such a dynamo. Although plausible mechanisms have

706: been proposed, as yet none of these are entirely satisfactory. Against

707: this background, there is a drive to be able to predict solar activity

708: with greater accuracy, due to the importance of this activity in

709: driving solar events.  

710: 

711: What we have demonstrated here is that no meaningful predictions can

712: be made from illustrative mean-field models, no matter how they are

713: constructed. If the mean-field model is constructed to be a driven linear

714: oscillator then the small stochastic effects that lead to the modulation

715: will have an extremely large

716: effect on the basic cycle and make even short-term prediction

717: extremely difficult. The second scenario, where the modulation arises

718: as a result of nonlinear processes rather than stochastic

719: fluctuations, is clearly a better one for prediction --- though here

720: too, prediction is fraught with difficulties. Owing to the inherent

721: nonlinearity of the dynamo system, long-term predictions are

722: impossible (even if the form of the model is completely correctly

723: determined). Furthermore, even short-term prediction from mean-field

724: models is meaningless because of fundamental uncertainties in the form

725: and amplitude of the transport coefficients and nonlinear

726: response. Any deterministic nonlinear model that produces chaotically

727: modulated activity cycles will be faced with the same difficulties. 

728: 

729: The equations that describe dynamo action in the solar interior are

730: known to be nonlinear partial differential equations --- the momentum

731: equation is nonlinear in both the velocity and the magnetic field. One

732: indication of the role played by nonlinear effects in the solar dynamo

733: is the presence of cyclic variations in the solar differential

734: rotation (the ``torsional oscillations''). Furthermore estimates of

735: the field strength at the base of the convection zone consistent with

736: the observed formation of active regions yield fields of sufficient

737: strength ($10^4-10^5$G) for the nonlinear Lorentz force to be

738: extremely significant, whilst the flows are vigorously nonlinear and

739: turbulent. It therefore seems extremely unlikely that the dynamics of

740: the solar interior can be described by a forced linear system without

741: throwing away much (if not all) of the important physics. In this case

742: it must be argued {\it not only} that this discarded physics is

743: irrelevant to the dynamo process but also that the parameterisation of

744: the unresolved physics should not include a stochastic component, as

745: this would have an extremely large effect on such a relinearised system.

746: 

747: It is certainly tempting to try to use the observed magnetic flux at

748: the solar surface as an input to a model for prediction (whether

749: nonlinear or stochastic, mean-field or full MHD). Certainly any fully

750: consistent solar activity model constructed in the future should be

751: capable of reproducing the observed pattern of magnetic activity at

752: the solar surface, although this will require a complete understanding

753: not only of the generation process via dynamo action, but also the

754: processes which lead to the formation and subsequent rise of

755: concentrated magnetic structures from the solar interior to the

756: surface.  However it is not clear what role the flux at the solar

757: surface plays in the basic dynamo process. Is it inherent to the

758: process (as modelled by flux transport dynamos) or simply a by-product

759: of the dynamo process that is occurring deep within the sun? Estimates

760: suggest that between $5$ and $10 \%$ of the solar flux generated in

761: the deep interior makes it to the solar surface (e.g. Galloway \&

762: Weiss 1981). For the flux at the solar surface to be the key for

763: dynamo action, it must be explained why the majority of the magnetic

764: flux that resides in the solar interior plays such a little part in

765: the dynamics (to such an extent that it does not even appear as a

766: small stochastic perturbation to the large-scale flux transport

767: dynamo). 

768: 

769: Finally it is important to stress that {\it even if} a model has been

770: tuned so as to reproduce results over a number of solar activity

771: cycles, then there is a good chance of error in the prediction for the

772: next cycle. Any advection-diffusion system in which one is free to

773: specify  not only the sources and the sinks but also the transport

774: processes can be tuned to reproduce any required features of activity.

775: Moreover, the formulation of a prediction in terms of a parameterised 

776: mean-field model does not inherently put the prediction on a sounder

777: scientific basis than a prediction based on methods of timeseries

778: analysis alone (some of which use very sophisticated mathematical

779: techniques). This, of course, is not to say that any given prediction

780: from such a model will be incorrect, just that the basis for making

781: the prediction has no strong scientific support.

782: 

783: \acknowledgments

784: 

785: We would like to thank Nigel Weiss for useful discussions and for

786: providing helpful comments and suggestions. PJB would like to

787: acknowledge the support of PPARC.   

788: 

789: \clearpage

790: %=============================================================================

791: %% \bibliography{apj-jour,ms}

792: 

793: \begin{thebibliography}{34}

794: \expandafter\ifx\csname natexlab\endcsname\relax\def\natexlab#1{#1}\fi

795: 

796: \bibitem[{{Beer}(2000)}]{Beer00}

797: Beer, J. 2000, Space Science Rev., 94, 53

798: 

799: \bibitem[{{Brandenburg}(2005)}]{Brand05}

800: Brandenburg, A. 2005, ApJ, 625, 539

801: 

802: \bibitem[{{Brandenburg \& Subramanian}(2005)}]{BrandS05}

803: Brandenburg, A., Subramanian, K. 2005, Physics Reports, 417, 1

804: 

805: \bibitem[{{Bushby}(2005)}]{Bush05}

806: Bushby, P. J. 2005, Astron. Nachr., 326, 218

807: 

808: \bibitem[{{Bushby}(2006)}]{Bush06}

809: Bushby, P. J. 2006, MNRAS, 371, 772

810: 

811: \bibitem[{{Casdagli}(1989)}]{Casd89}

812: Casdagli, M. 1989, Physica D, 35, 335

813: 

814: \bibitem[{{Cattaneo \& Hughes}(1996)}]{Catt96}

815: Cattaneo, F., Hughes, D. W. 1996, Phys. Rev. E, 54, 4532

816: 

817: \bibitem[{{Charbonneau \& Dikpati}(2000)}]{Charb00}

818: Charbonneau, P., Dikpati, M. 2000, ApJ, 543, 1027

819: 

820: \bibitem[{{Charbonneau}(2005)}]{Charb05}

821: Charbonneau, P. 2005, Living Reviews in Solar Physics, Vol. 2, No. 2

822: 

823: \bibitem[{{Choudhuri, Sch\"ussler \& Dikpati}(1995)}]{Choud95}

824: Choudhuri, A. R., Sch\"ussler, M., Dikpati, M. 1995, A\&A, 303, L29 

825: 

826: \bibitem[{{Courvoisier, Hughes \& Tobias}(2006)}]{Courv06}

827: Courvoisier, A., Hughes, D. W., Tobias, S. M. 2006, Phys. Rev. Lett.,

828: 96, 034503

829: 

830: \bibitem[{{Covas \etal}(2000)}]{Covas2000}

831: Covas, E., Tavakol, R., Moss, D., Tworkowski, A. 2000, A\&A, 360, L21

832:  

833: \bibitem[{{Dikpati \& Charbonneau}(1999)}]{Dikp99}

834: Dikpati, M., Charbonneau, P. 1999, ApJ, 518, 508

835: 

836: \bibitem[{{Dikpati \& Gilman}(2006)}]{Dikgil06}

837: Dikpati, M., Gilman, P. A. 2006, ApJ, 649, 498

838: 

839: \bibitem[{{Dikpati, de Toma \& Gilman}(2006)}]{Dikp06}

840: Dikpati, M., de Toma, G., Gilman, P. A. 2006, Geophys. Research Lett., 33,

841: L05102 

842: 

843: \bibitem[{{Farmer \& Sidorowich}(1987)}]{Farm87}

844: Farmer, J. D., Sidorowich, J. J. 1987, Phys Rev. Lett., 59, 845

845: 

846: \bibitem[{{Galloway \& Weiss}(1981)}]{GallWeiss81}

847: Galloway, D.~J., Weiss, N.~O. 1981, ApJ, 243, 945

848: 

849: \bibitem[{{Hathaway, Wilson \& Reichmann}(1999)}]{Hath99}

850: Hathaway, D. H., Wilson, R. M., Reichmann, E. J. 1999, Journal of

851: Geophys. Research, 104, 22,375

852: 

853: \bibitem[{{Krause \& R\"adler}(1980)}]{Krause80}

854: Krause, F., R\"adler, K. H. 1980, Mean-field Magnetohydrodynamics and

855: Dynamo Theory (Oxford: Pergamon Press)

856: 

857: \bibitem[{{Malkus \& Proctor}(1975)}]{Malk75}

858: Malkus, W. V. R., Proctor, M. R. E. 1975, J. Fluid Mech., 67, 417

859: 

860: \bibitem[{{Moffatt}(1978)}]{Moff78}

861: Moffatt, H. K. 1978, Magnetic Field Generation in Electrically

862: Conducting Fluids (Cambridge: Cambridge University Press) 

863: 

864: \bibitem[{{Moss \& Brooke}(2000)}]{Moss00}

865: Moss, D., Brooke, J. 2000, MNRAS, 315, 521

866: 

867: \bibitem[{{Ossendrijver}(2003)}]{Oss03}

868: Ossendrijver, M. 2003, Astron. and Astrophys. Rev., 11, 287

869: 

870: \bibitem[{{Ossendrijver \& Hoyng}(1996)}]{OsHo96}

871: Ossendrijver, A. J. H., Hoyng, P. 1996, A\&A, 313, 959

872: 

873: \bibitem[{{Parker}(1993)}]{Park93}

874: Parker, E. N. 1993, ApJ, 408, 707

875: 

876: \bibitem[{{Saba, Strong \& Slater}(2005)}]{Saba05}

877: Saba, J. L. R., Strong, K. T., Slater, G. L. 2005, Memorie della

878: Societa Astronomica Italiana, 76, 1034

879: 

880: \bibitem[{{Schatten}(2002)}]{Schat02}

881: Schatten, K. 2002, J. Geophys. Res., 107(A11), 1377

882: 

883: \bibitem[{{Sello}(2003)}]{Sello03}

884: Sello, S. 2003, A\&A, 410, 691

885: 

886: \bibitem[{{Solanki et al.}(2004)}]{Sol04}

887: Solanki, S. K., Usoskin, I. G., Kromer, B., Sch\"ussler, M., Beer,

888: J. 2004, Nature, 431 (7012), 1084

889: 

890: \bibitem[{{Steenbeck, Krause \& R\"adler}(1966)}]{Steen66}

891: Steenbeck, M., Krause, F., R\"adler, K. H. 1966, Z. Naturforsch., 21a,

892: 369

893: 

894: \bibitem[{{Tobias, Weiss \& Kirk}(1995)}]{TWK95}

895: Tobias, S.~M., Weiss, N.~O., Kirk, V. 1995, MNRAS, 273, 1150 

896: 

897: \bibitem[{{Tobias}(1997)}]{Tobias97}

898: {Tobias}, S.~M. 1997, A\&A, 322, 1007

899: 

900: \bibitem[{{Tobias}(2002)}]{Tobias02}

901: {Tobias}, S.~M. 2002, Phil. Trans. Roy. Soc. Lond. A, 360, 2741

902: 

903: \bibitem[{{Tong}(1995)}]{Tong95}

904: Tong, H. 1995, Chaos and Forecasting (Singapore: World Scientific

905: Publications)

906: 

907: \bibitem[{{Wagner \etal}{2001}}]{Wagner}

908: Wagner, G., Beer, J., Masarik, J., Muscheler, R., Kubik, P.~W., Mende,

909: W., Laj, C., Raisbeck, G.~M., Yiou, F. 2001, Geophys. Res. Lett., 28,

910: 303  

911: 

912: \bibitem[{{Weiss}(2002)}]{Weiss02}

913: {Weiss}, N.~O. 2002, A\&G, 43, 3.09

914: 

915: \bibitem[{{Weiss \& Tobias}{2000}}]{WeissTobias}

916: {Weiss}, N.~O., Tobias, S.~M. 2000, Space Science Rev., 94, 99

917: 

918: \bibitem[{{Zhang(1996)}}]{Zhang96}

919: Zhang, Q. 1996, A\&A, 310, 646

920: 

921: \bibitem[{{Zhao et al.}(2004)}]{Zhaoetal04}

922: Zhao, H., Liang, H., Zhan, L., Zhong, S. 2004, ChA\&A, 28, 67 

923: 

924: \end{thebibliography}

925: 

926: \clearpage

927: 

928: \begin{figure}

929: \epsscale{0.68}

930: \plotone{f1a.eps}

931: \epsscale{0.68}

932: \plotone{f1b.eps}

933: \caption{The time evolution of the stochastically-perturbed flux

934:   transport dynamo. Top: Timeseries

935:   for the mean of the squared toroidal field ($B^2$) at the base of

936:   the convection zone. Bottom: In this figure, $B^2$ at the base of

937:   the convection zone is plotted

938:   against the mean of the squared values of the poloidal magnetic

939:   potential ($A^2$) at the surface of the domain.\label{fig1}}    

940: \end{figure}

941: 

942: \begin{figure}

943: \plotone{f2.eps}

944: \caption{The lack of correlation between successive maxima in the

945:   stochastically perturbed timeseries (as shown in Figure~1). Defining

946:   $T_n$ to be magnitude of the $n^{th}$ maximum, this plot shows the

947:   sequential behaviour of these maxima, plotting $T_{n+1}$ as a

948:   function of $T_n$.\label{fig2}}  

949: \end{figure}

950: 

951: \clearpage

952: 

953: \begin{figure}

954: \plotone{f3a.eps}

955: \plotone{f3b.eps}

956: \caption{The time evolution of the target solution. Top: Timeseries

957:   for the mean of the squared toroidal field ($B^2$) in the dynamo

958:   region. Bottom: An attractor for the target solution,

959:   in which $B^2$ is plotted against the mean of the squared values of the

960:   poloidal magnetic potential ($A^2$) and the velocity

961:   perturbation ($V^2$).\label{fig3}}   

962: \end{figure}

963: 

964: \clearpage

965: 

966: \begin{figure}

967: \plotone{f4.eps}

968: \caption{Three timeseries showing the evolution of the mean squared

969:   toroidal field in the dynamo region. The solid line shows a

970:   segment of the target solution timeseries; the dashed and dotted

971:   lines show the time-evolution of solutions that are started from nearby

972:   points on the same attractor. Although all solutions have the same

973:   model parameters, the timeseries rapidly diverge after a couple of

974:   cycles.\label{fig4}} 

975: \end{figure}

976: 

977: \clearpage

978: 

979: \begin{figure}

980: \begin{center}

981: \epsscale{0.99}

982: \plotone{f5a.eps}

983: \vspace{0.1cm}

984: \plotone{f5b.eps}

985: \caption{Attempts to predict two different segments of the target

986: solution timeseries: In each plot, the solid line shows a segment of

987: the target solution timeseries, the dashed line shows the behaviour of

988: the chosen mean-field predictor (different predictors are used for

989: each plot), whilst the crosses show the predictions that are obtained

990: by reconstructing the nonlinear attractor for the target solution. In each

991: case, the mean-field predictors have been optimised by ensuring that

992: the chosen predictor closely matches the target timeseries segment for

993: a large number of cycles (6 cycles in the upper plot, 9 in the

994: lower).\label{fig5}}   

995: \end{center}

996: \end{figure}

997: 

998: 

999: \end{document}

1000: