Machine learning techniques have been widely used in attempts to forecast several solar datasets. Most of these approaches employ supervised machine learning algorithms which are, in general, very data hungry. This hampers the attempts to forecast some of these data series, particularly the ones that depend on (relatively) recent space observations. Here we focus on an attempt to forecast the solar surface longitudinally averaged radial magnetic field distribution using a form of spatial-temporal neural networks. Given that the recording of these spatial-temporal datasets only started in 1975 and are therefore quite short, the forecasts are predictably quite modest. However, given that there is a potential physical relationship between sunspots and the magnetic field, we employ another machine learning technique called transfer learning which has recently received considerable attention in the literature. Here, this approach consists in first training the source spatial-temporal neural network on the much longer time/latitude sunspot area dataset, which starts in 1874, then transferring the trained set of layers to a target network, and continue training the latter on the magnetic field dataset. The employment of transfer learning in the field of computer vision is known to obtain a generalized set of feature filters that can be reused for other datasets and tasks. Here we obtain a similar result, whereby we first train the network on the spatial-temporal sunspot area data, then the first few layers of the neural network are able to identify the two main features of the solar cycle, i.e. the amplitude variation and the migration to the equator, and therefore can be used to train on the magnetic field dataset and forecast better than a prediction based only on the historical magnetic field data.

Using neural networks as a prediction method, we attempt to demonstrate that forecasting of the Sun's sunspot time series can be extended to the spatial-temporal case. We employ this machine learning methodology to forecast not only in time but also in space (in this case the latitude) on a spatial-temporal dataset representing the solar sunspot diagram extending to a total of 142 years. The analysis shows that this approach seems to be able to reconstruct the overall qualitative aspects of the spatial-temporal series, namely the overall shape and amplitude of the latitude and time pattern of sunspots. This is, as far as we are aware, the first time neural networks have been used to forecast the Sun's sunspot butterfly diagram, and although the results are limited in the quantitative prediction aspects, it points the way to use the full spatial-temporal series as opposed to just the time series for machine learning approaches to forecasting. Further to that, we use the method to predict that the upcoming Cycle 25 maximum sunspot number will be around $R_{25}=57 \pm 17$. This implies a very weak cycle and that it will be the weakest cycle on record.

We attempt to forecast the Sun's sunspot butterfly diagram in both space (i.e. in latitude) and time, instead of the usual one-dimensional time series forecasts prevalent in the scientific literature. We use a prediction method based on the non-linear embedding of data series in high dimensions. We use this method to forecast both in latitude (space) and in time, using a full spatial-temporal series of the sunspot diagram from 1874 to 2015. The analysis of the results shows that it is indeed possible to reconstruct the overall shape and amplitude of the spatial-temporal pattern of sunspots, but that the method in its current form does not have real predictive power. We also apply a metric called structural similarity to compare the forecasted and the observed butterfly cycles, showing that this metric can be a useful addition to the usual root mean square error metric when analysing the efficiency of different prediction methods.

Increasing evidence is becoming available about not only the surface differential rotation of rapidly rotating cool stars but, in a small number of cases, also about temporal variations, which possibly are analogous to the solar torsional oscillations. Given the present difficulties in resolving the precise nature of such variations, due to both the short length and poor resolution of the available data, theoretical input is vital to help assess the modes of behaviour that might be expected, and will facilitate interpretation of the observations. Here we take a first step in this direction by studying the variations in the convection zones of such stars, using a two dimensional axisymmetric mean field dynamo model operating in a spherical shell in which the only nonlinearity is the action of the azimuthal component of the Lorentz force of the dynamo generated magnetic field on the stellar angular velocity. We consider three families of models with different depths of dynamo-active regions. For moderately supercritical dynamo numbers we find torsional oscillations that penetrate all the way down to the bottom of the convection zones, similar to the case of the Sun. For larger dynamo numbers we find fragmentation in some cases and sometimes there are other dynamical modes of behaviour, including quasi-periodicity and chaos. We find that the largest deviations in the angular velocity distribution caused by the Lorentz force are of the order of few percent, implying that the original assumed `background' rotation field is not strongly distorted.

Analyses of recent helioseismic data have produced ample evidence for substantial dynamical variation of the differential rotation within the solar convection zone. Given the inevitable difficulties in resolving the precise nature of variations at deeper layers, much effort has recently gone into determining theoretically the expected modes of behaviour, using nonlinear dynamo models. Two important limitations of these models are that they have so far included only one form of nonlinearity, and as yet they have not taken into account the density stratification in the solar convection zone. Here we address both of these issues by studying the effects of including density stratification, as well as including an alpha--quenching nonlinearity in addition to the previously studied effects of the Lorentz force on the differential rotation. We find that observationally important features found in the earlier uniform density models remain qualitatively unchanged, although there are quantitative differences. This is important as it provides more realistic theoretical predictions to be compared with and guide observations, especially in the deeper regions where the uncertainties in the inversions are larger. However the presence of an effective alpha-quenching nonlinearity significantly reduces the amplitudes of the oscillations.

Recent analyses of the helioseismic data have produced evidence for a variety of interesting dynamical behaviour associated with torsional oscillations. What is not so far clear is whether these oscillations extend all the way to the bottom of the convection zone and, if so, whether the oscillatory behaviour at the top and the bottom of the convection zone is different. Attempts have been made to understand such modes of behaviour within the framework of nonlinear dynamo models which include the nonlinear action of the Lorentz force of the dynamo generated magnetic field on the solar angular velocity. One aspect of these models that remains uncertain is the nature of the boundary conditions on the magnetic field. Here by employing a range of physically plausible boundary conditions, we show that for near-critical and moderately supercritical dynamo regimes, the oscillations extend all the way down to the bottom of the convection zone. Thus, such penetration is an extremely robust feature of the models considered. We also find parameter ranges for which the supercritical models show spatiotemporal fragmentation for a range of choices of boundary conditions. Given their observational importance, we also make a comparative study of the amplitude of torsional oscillations as a function of the boundary conditions.

We report our first results on comparing the variations of the solar internal rotation with solar activity, as predicted by non-linear solar dynamo modelling, with helioseismic measurements using the SOHO MDI data.

In a recent paper [M. A. J. H. Ossendrijver, A&A {\bf 359}, 364 (2000)] numerical simulations of a 2D mean--field model where shown to produce grand minima, typical of the long-term behavior of solar magnetic activity. The model consisted of dynamo that features an $\alpha$ effect based on the buoyancy instability of magnetic fluxtubes, which gives rise to the switching back and forth from grand minima to ``regular'' solar behavior. In this Letter, we report evidence from a time-series analysis of the model for the presence of crisis--induced intermittency due to attractor--widening. We support this finding by showing that the average duration of the minima, $<\tau>$, follows the theoretically predicted scaling $<\tau> \sim (C_{\delta\alpha}-C_{\delta\alpha}^*)^{-\gamma}$, where $C_{\delta\alpha}$ is the bifurcation parameter of interest, together with other statistical evidence. As far as we are aware, this is the first time concrete and detailed evidence has been produced for the occurrence of this type of crisis--induced intermittency -- due to attractor widening -- for such dynamo models.

Analyses of recent helioseismic data indicate that the dynamical regimes at the base of the convection zone can be different from those observed at the top, having either significantly shorter periods or non--periodic behaviour. Recently spatiotemporal fragmentation/bifurcation has been proposed as a dynamical mechanism to account for the multi-mode behaviour that is possibly observed in the solar convection zone, without requiring separate physical mechanisms with different time scales at different depths. Here we study the robustness of this mechanism with respect to changes to the zero order rotation profile, motivated by the uncertainties of and differences between the various reductions of the helioseimic data. We find that spatiotemporal fragmentation is a common feature of the reductions investigated

Observations of the Sun and solar-type stars continue to reveal phenomena whose understanding is very likely to require a nonlinear framework. Here we shall concentrate on two such phenomena, namely the grand minima type behaviour observed in the Sun and solar-type stars and the recent dynamical variations of the differential rotation in the solar convection zone, deduced from the helioseismic observations, and discuss how their explanations have recently motivated the development/employment of novel ideas from nonlinear dynamics.

We find concrete evidence for a recently discovered form of intermittency, referred to as in--out intermittency, in both PDE and ODE models of mean field dynamos. This type of intermittency (introduced in Ashwin et al 1999) occurs in systems with invariant submanifolds and, as opposed to on--off intermittency which can also occur in skew product systems, it requires an absence of skew product structure. By this we mean that the dynamics on the attractor intermittent to the invariant manifold cannot be expressed simply as the dynamics on the invariant subspace forcing the transverse dynamics; the transverse dynamics will alter that tangential to the invariant subspace when one is far enough away from the invariant manifold. Since general systems with invariant submanifolds are not likely to have skew product structure, this type of behaviour may be of physical relevance in a variety of dynamical settings. The models employed here to demonstrate in--out intermittency are axisymmetric mean--field dynamo models which are often used to study the observed large scale magnetic variability in the Sun and solar-type stars. The occurrence of this type of intermittency in such models may be of interest in understanding some aspects of such variabilities.

Recent analyses of helioseismological observations seem to suggest the presence of two new phenomena connected with the dynamics of the solar convective zone. Firstly, there are present torsional oscillations with periods of about 11 years, which penetrate significantly into the solar convection zone and secondly, oscillatory regimes exist near the base of the convection which are markedly different from those observed near the top, having either significantly reduced periods or being non-periodic. Recently spatiotemporal fragmentation/bifurcation has been proposed as a possible dynamical mechanism to account for such observed multi-mode behaviours in different parts of the solar convection zone. Evidence for this scenario was produced in the context of an axisymmetric mean field dynamo model operating in a spherical shell, with a semi-open outer boundary condition and a zero order angular velocity obtained by the inversion of the MDI data, in which the only nonlinearity was the action of the Lorentz force of the dynamo generated magnetic field on the solar angular velocity. Here we make a detailed study of the robustness of this model with respect to plausible changes to its main ingredients, including changes to the alpha and eta profiles as well as the inclusion of a nonlinear alpha quenching. We find that spatiotemporal fragmentation is present in this model for different choices of the rotation data and as the details of the model are varied. Taken together, these results give strong support to the idea that spatiotemporal fragmentation is likely to occur in general dynamo settings.

Recent analyses of the helioseismic observations indicate that the previously observed surface torsional oscillations with periods of about 11 years extend significantly downwards into the solar convective zone. Furthermore, there are indications that the dynamical regimes at the base of the convection zone are different from those observed at the top, having either significantly shorter periods or non-periodic behaviour. We propose that this behaviour can be explained by the occurrence of {\it spatiotemporal fragmentation}, a crucial feature of which is that such behaviour can be explained solely through nonlinear spatiotemporal dynamics, without requiring separate mechanisms with different time scales at different depths. We find evidence for this mechanism in the context of a two dimensional axisymmetric mean field dynamo model operating in a spherical shell, with a semi-open outer boundary condition, in which the only nonlinearity is the action of the azimuthal component of the Lorentz force of the dynamo generated magnetic field on the solar angular velocity.

Recent analysis of the helioseismic observations indicate that the previously observed surface torsional oscillations extend significantly downwards into the solar convection zone. In an attempt to understand these oscillations, we study the nonlinear coupling between the magnetic field and the solar differential rotation in the context of a mean field dynamo model, in which the nonlinearity is due to the action of the azimuthal component of the Lorentz force of the dynamo generated magnetic field on the solar angular velocity. The underlying zero order angular velocity is chosen to be consistent with the most recent helioseismic data. The model produces butterfly diagrams which are in qualitative agreement with the observations. It displays torsional oscillations that penetrate into the convection zone, and which with time migrate towards the equator. The period of these oscillations is found to be half that of the period of the global magnetic fields. This is compatible with the observed period of the surface torsional oscillations. Inside the convection zone, this is a testable prediction that is not ruled out by the observations so far available.

We find concrete evidence for the presence of crisis-induced and Pomeau-Manneville Type-I intermittencies in an axisymmetric PDE mean-field dynamo model. These findings are of potential importance for two different reasons. Firstly, as far as we are aware, this is the first time detailed evidence has been produced for the occurrence of these types of intermittency for such deterministic PDE models. And secondly, despite the rather idealised nature of these models, the concrete evidence for the occurrence of more than one type of intermittency in such models makes it in principle possible that different types of intermittency may occur in different solar-type stars or even in the same star over different epochs. In this way a multiple intermittency framework may turn out to be of importance in understanding the mechanisms responsible for grand-minima type behaviour in the Sun and solar-type stars and in particular in the interpretation of the corresponding observational and proxy evidence.

We briefly discuss the status of the intermittency hypothesis, according to which the grand minima type variability in solar-type stars may be understood in terms of dynamical intermittency. We review concrete examples which establish this hypothesis in the mean-field setting. We discuss some difficulties and open problems regarding the establishment of this hypothesis in more realistic settings as well as its operationally decidability.

We study the changes in the dynamical behaviour of axisymmetric spherical mean-field dynamo models produced by changes in their geometry and topology, by considering a two parameter family of models, ranging from a full sphere to spherical shell, torus and disc-like configurations, within a unified framework. We find that the two parameter space of the family of models considered here separates into at least three different regions with distinct characteristics for the onset of dynamo action. In two of these regions, the most easily excited fields are oscillatory, in one case with dipolar symmetry, and in the other with quadrupolar, whereas in the third region the most easily excited field is steady and quadrupolar. In the nonlinear regime, we find that topological changes can alter significantly the dynamical behaviour, whilst modest changes in geometry can produce qualitative changes, particularly for thin disc-like configurations. This is of potential importance, since the exact shapes of astrophysical bodies, especially accretion discs and galaxies, are usually not precisely known.

Calculations for mean field dynamo models (in both full spheres and spherical shells), with both algebraic and dynamic $\alpha$--quenchings, show qualitative as well as quantitative differences and similarities in the dynamical behaviour of these models. We summarise and enhance recent results with extra examples. Overall, the effect of using a dynamic $\alpha$ appears to be complicated and is affected by the region of parameter space examined.

An important question regarding the study of mean field dynamo models is how to make precise the nature of their underlying dynamics. This is difficult both because relatively little is known about the dynamical behaviour of infinite dimensional systems and also due to the numerical cost of studying the related partial differential equations. As a first step towards their understanding, it is useful to consider the corresponding truncated models. Here we summarise some recent results of the study of a class of truncated axisymmetric mean field dynamo models. We find conclusive evidence in these models for various types of intermittency as well as multiple attractors and final state sensitivity. We also find that the understanding of the underlying dynamics of such dynamo models requires the study of a new class of dynamical systems, referred to as the {\it non-normal} systems. Current work demonstrates that these types of systems are capable of a novel type of intermittency and also of relevance for the understanding of the full axisymmetric PDE dynamo models.

We study axisymmetric mean field spherical and spherical shell dynamo models, with both dynamic and algebraic $\alpha$--quenchings. Our results show that there are qualitative as well as quantitative differences and similarities between these models. Regarding similarities, both groups of models exhibit symmetric, antisymmetric and mixed modes of behaviour. As regards differences, the important feature in the full sphere models is the occurrence of chaotic behaviour in the algebraic $\alpha$--quenching models. For the spherical shell models with dynamic $\alpha$ the main features include the possibility of multi-attractor regimes with final state sensitivity with respect to small changes in the magnitude of $\alpha$ and the initial parity. We find the effect of introducing a dynamic $\alpha$ is likely to be complicated and depend on the region of the parameter space considered, rather than a uniform change towards simplicity or complexity.

We investigate the detailed dynamics of a truncated $\alpha\omega$ dynamo model with a dynamic $\alpha$ effect. We find the presence of multiple attractors, including two chaotic attractors with a fractal basin boundary which merge to form a single attractor as the control parameter is increased. By considering phase portraits and the scaling of averaged times of transitions between the two attractors, we demonstrate that this merging is accompanied by a crisis-induced intermittency. We also find a range of parameter values over which the system has a fractal parameter dependence for fixed initial conditions. This is the first time this type of intermittency has been observed in a dynamo model and it could be of potential importance in accounting for some forms of intermittency in the solar and stellar output.

In a recent work (Covas et al. 1997), the behaviour and the robustness of truncated $\alpha\Omega$ dynamos with a dynamic $\alpha$ were studied with respect to a number of changes in the driving term of the dynamic $\alpha$ equation, which was considered previously by Schmalz and Stix (1991) to be of the form $\sim A_{\phi}B_{\phi}$. Here we review and extend our previous work and consider the effect of adding a quadratic quenching term of the form $\alpha|B|^2$. We find that, as before, such a change can have significant effects on the dynamics of the related truncated systems. We also find intervals of (negative) dynamo numbers, in the system considered by Schmalz and Stix (1991), for which there is sensitivity with respect to small changes in the dynamo number and the initial conditions, similar to what was found in our previous work. This latter behaviour may be of importance in producing the intermittent type of behaviour observed in the sun.

We investigate the behaviour of $\alpha\Omega$ dynamos with a dynamic $\alpha$, whose evolution is governed by the imbalance between a driving and a damping term. We focus on truncated versions of such dynamo models which are often studied in connection with solar and stellar variability. Given the approximate nature of such models, it is important to study how robust they are with respect to reasonable changes in the formulation of the driving and damping terms. For each case, we also study the effects of changes of the dynamo number and its sign, the truncation order and initial conditions. Our results show that changes in the formulation of the driving term have important consequences for the dynamical behaviour of such systems, with the detailed nature of these effects depending crucially on the form of the driving term assumed, the value and the sign of the dynamo number and the initial conditions. On the other hand, the change in the damping term considered here seems to produce little qualitative effect.