Models for the long-term variations of solar activity

Abstract

One obvious feature of the solar cycle is its variation from one cycle to another. In this article, we review the dynamo models for the long-term variations of the solar cycle. By long-term variations, we mean the cycle modulations beyond the 11-year periodicity and these include, the Gnevyshev–Ohl/Even–Odd rule, grand minima, grand maxima, Gleissberg cycle, and Suess cycles. After a brief review of the observed data, we present the dynamo models for the solar cycle. By carefully analyzing the dynamo models and the observed data, we identify the following broad causes for the modulation: (1) magnetic feedback on the flow, (2) stochastic forcing, and (3) time delays in various processes of the dynamo. To demonstrate each of these causes, we present the results from some illustrative models for the cycle modulations and discuss their strengths and weakness. We also discuss a few critical issues and their current trends. The article ends with a discussion of our current state of ignorance about comparing detailed features of the magnetic cycle and the large-scale velocity from the dynamo models with robust observations.

1 Introduction

The most prominent and fundamental feature of the solar magnetic field is its 11-year cyclic oscillation. Systematic observations of the large-scale solar magnetic field, available since the 1950 s, revealed the reversals of the field. However, the times of the reversals and the strength of the field are not the same for all the cycles. Time series of sunspot number and the sunspot area, for which we have direct observations for longer durations (group sunspot number since 1610 and the area since 1874), also show cycle-to-cycle variations (Hathaway 2015); Fig. 1. Thus, there is no doubt that the 11-year solar cycle is not regular and that makes the prediction of the future cycle a formidable task (Petrovay 2020). The prediction, however, is essential as the Sun’s magnetic field drives the space weather which sometimes poses serious problems to us—e.g., by damaging satellite’s electronics, modern-day technologies such as telecommunications, GPS networks, and electric power grids at high latitudes, making polar routes dangerous for aviation, increasing the radiation dose to astronauts in space (Temmer 2021). Evidence suggests that the variable solar activity may also drive changes in the Earth’s global temperature (Solanki et al. 2013).

Simply saying that the solar cycle is irregular is not enough to describe its true nature; there are many distinct features—such as grand minima and grand maxima—which can be considered as extreme examples of irregularity; Fig. 2. Additionally, there are some long-term patterns beyond the usual 11-year variation, such as Gnevyshev–Ohl rule and Gleissberg cycle. Below we briefly discuss these long-term variations. However, the readers can check the excellent reviews (Hathaway 2015; Usoskin 2023; Biswas et al. 2023) for extensive discussion.

2 Long-term variations of the solar cycle

2.1 Grand minima and maxima

Grand minima are the extended episodes of considerably lower magnetic activity than the normal one. The best example of these is the Maunder minimum in the 17th century when solar activity was considerably weaker than the normal one for about 70 years (Eddy 1976); Fig. 1. We emphasize that this is not an artifact due to few observations but a real and well-observed event (Hoyt and Schatten 1996). Observations have shown that the Maunder minimum was not complete lack of activity—the Sun was still producing spots and even cycles but at a lower rate (Beer et al. 1998; Zolotova and Ponyavin 2015; Usoskin et al. 2015; Vaquero et al. 2015; Zolotova and Ponyavin 2016. This makes the Maunder minimum (and possibly all grand minima) a special state of solar activity. Further distinct aspects of Maunder minimum are the followings. (1) There was a strong hemispheric asymmetry during the latter half of Maunder minimum; sunspots were observed mostly in the southern hemisphere (Ribes and Nesme-Ribes 1993). (2) Recovery of Maunder minimum is gradual, however, the onset is somewhat uncertain (Fig. 1). Some previous results suggested that the onset is abrupt (Usoskin et al. 2000), while later studies showed that it is likely to be gradual (Vaquero et al. 2011). (3) A proxy of solar activity inferred from a cosmogenic isotope $^{14}$C showed that the cycle length before the onset of the Maunder minimum was significantly longer than its usual value (Miyahara et al. 2021; Usoskin et al. 2021).

Analyses of $^{14}$C for the last 11,400 years revealed the following important results (Usoskin et al. 2007; Usoskin 2023; Usoskin et al. 2021). (1) The Sun spent about 17% of its time in the grand minimum state. (2) Grand minima are of two types: short minima of Maunder type with duration 30–90 years and long ($>100$ years) minima of Spörer type. (3) The grand minima recur aperiodically. (4) The waiting time distribution of the occurrence of grand minima displays a deviation from an exponential distribution. However, Moss et al. (2008) showed that feature (5) can be an artifact of poor statistics (it is based on only 27 grand minima identified in the observed data).

Fig. 1

Yearly variation of the monthly mean sunspot number smoothed using a Gaussian filter of FWHM $=7$ months (red curve), available since 1749 and the yearly mean group sunspot number (black curve) available during 1610–2015. Note that group number is scaled by a factor of 18 to bring it to the scale of sunspot number. The blue curve with 98-year period guides the Gleissberg cycle. Cycle numbers for which the Even–Odd effect is obeyed are shown by tagging the number on the odd cycles. Data source: WDC-SILSO, Royal Observatory of Belgium, Brussels

Grand maxima, on the other hand, are extended periods with appreciably higher magnetic activity than the normal one. The modern maximum that occurred around 1960 is an example of the same. In about last 11,000 years, 23 grand maxima were detected and the Sun spent about $12\%$ of its time in this phase (Usoskin et al. 2007; Usoskin 2023). Grand maxima are more short-lived than the grand minima (Solanki et al. 2004; Usoskin et al. 2021). The distribution of the duration of grand maxima shows a smooth variation with an exponential fall at longer durations. The distribution of the waiting time between the consecutive grand maxima is not conclusive but there is an indication of deviation from the exponential law.

Fig. 2

Reconstructed (decadal) sunspot number along with its $68\%$ confidence interval (gray shading) over nine millennia derived from $^{14}$C data using a multi-proxy Bayesian method by Wu et al. (2018). The red curve shows the decadally resampled international sunspot number for the last 300 years (version 2, scaled by 0.6). The dashed line marks the zero spot number. Asterisks and circles mark the times of occurrences of grand maxima and minima, respectively

2.2 Cycles and modulations beyond 11 years

Beyond the regular 11-year solar cycle, the following longer cycles or modulations are detected in the solar activity data.

• Gnevyshev–Ohl rule/Even–Odd effect: This says that if the cycles are arranged in pairs with the even cycle and the following odd cycle, then the sum of the sunspot number in the odd cycle is higher than the even cycle (Gnevyshev and Ohl 1948). We note that this is not a strict rule, it is violated in cycle pairs: 4–5, and 22–23 (Hathaway 2015).

• Gleissberg cycle: a modulation in the solar activity with a mean period of about 90 years (Gleissberg 1939; Hathaway 2015). Recent data shows that a sinusoidal fit to the detrended amplitude gives an approximate period of 98 years as shown in Fig. 1; also see Forgács-Dajka et al. (2004) who found somewhat closer value (104 years) in the area-weighted sunspot group data.

• Suess/de Vries cycle: Cycle with a period of 205–210 years detected in the cosmogenic isotopes (Suess 1980).

Some other cycles like the millennial Eddy cycle and 2400-year Hallstatt cycle are also noticed in the cosmogenic isotope, however, their signals are poor and require longer data to confirm (Usoskin 2023).

We would like to mention that not only the solar cycle is irregular, but the magnetic cycles that are observed in other sun-like stars are also irregular (Boro Saikia et al. 2018; Garg et al. 2019). Analyzing the data of 111 stars of spectral type F2–M2, Baliunas et al. (1995) showed that the slowly rotating (old) stars show a smoother variability in the magnetic cycle and possibly occasional grand minima, whereas the rapidly rotating young stars show irregular activity and no grand minima. Recently, Shah et al. (2018) and Baum et al. (2022) claim that HD 4915 and HD 166620 are possibly entering into the grand minimum phase. Hence, stellar cycles are also interesting in terms of their modulations.

Now we shall come to the models for these long-term variations of solar activity. By models in this article, we mean the dynamo models that are used to explain the long-term variations of solar activity. Below we first present an introductory discussion of the solar dynamo (Sect. 3) and models (Sect. 4). Then we identify the causes of the long-term modulations (Sect. 5), followed by some illustrative models for the same (Sects. 6 and 7). Finally, we discuss a few open questions with current trends (Sect. 8) and end the article with concluding remarks (Sect. 9).

3 Solar dynamo: an overview

Dynamo is a process in which a sufficiently strong and complex plasma flow maintains a magnetic field by overcoming its Ohmic dissipation. The magnetic fields that give rise to sunspots, global dipole magnetic field, and 11-year cycle are essentially of the large-scale (global) type which usually requires a non-zero net helicity (mirror asymmetry) in the flow. Due to the global rotation, the convective motion of plasma in the Sun is helical and the rotation is differential (nonuniform). Through this helical flow, the poloidal field in the Sun is primarily generated from the toroidal field, while the poloidal field acts as a source for the toroidal field through the differential rotation (Parker 1955a). Thus, the solar dynamo is essentially a cyclic oscillation between two fields and the turbulent transport plays an essential role in this oscillation (Sect. 4). For a detailed discussion of the solar dynamo, we refer the readers to the comprehensive reviews by Ossendrijver (2003) and Charbonneau (2020).

To study the solar dynamo, we need to begin with at least two basic equations of the magnetohydrodynamics, namely,

$$\begin{aligned} \frac{\partial \varvec{B}}{\partial t}= & {} \varvec{\nabla } \times (\varvec{v} \times \varvec{B} - \eta \varvec{\nabla } \times \varvec{B}), \end{aligned}$$

(1)

$$\begin{aligned} \rho \left[ \frac{\partial \varvec{v}}{\partial t} + (\varvec{v} \cdot \varvec{\nabla }) \varvec{v} \right]= & {} -\varvec{\nabla }P + \varvec{J}\times \varvec{B}+ \varvec{\nabla }\cdot (2 \nu \rho \varvec{S}) + \varvec{F}, \end{aligned}$$

(2)

where $\varvec{B}$ and $\varvec{v}$ are the magnetic and velocity fields, respectively, $\eta$ is the magnetic diffusivity, $\rho$ is the density, P is the pressure, $\varvec{J} = \varvec{\nabla }\times \varvec{B}/\mu _0$, the current density, $\nu$ is the kinematic viscosity, $S_{ij} = \frac{1}{2} ( \nabla _i v_j + \nabla _j v_i) - \frac{1}{3} \delta _{ij} \varvec{\nabla }\cdot \varvec{v}$ is the rate-of-strain tensor and the term $\varvec{F}$ includes gravitational, Coriolis and any other body forces acting on the fluid.

The above equations must be solved along with the mass continuity, internal energy and equation of state in the solar CZ with appropriate boundary conditions. This approach to studying the solar dynamo—so-called global MHD simulations—was pioneered by Gilman (1983) and Glatzmaier (1984) in the 1980 s. While these simulations gave a few positive results (e.g., large-scale flows and field and a bit of polarity reversal), being computationally expensive, these simulations were hardly applied to explore the long-term variations of the solar cycle. Further, the applicability of these simulations in the Sun is questionable due to their operation in a completely different parameter regime. In recent years, however, we have got some encouraging results in the global MHD simulations, a few of them were run for several magnetic cycles to explore the cycle modulations. In Sect. 7, we shall discuss the modulations of cycles found in these simulations. Probably the biggest problem in these simulations is to explore and understand the cause of solar cycle variabilities. In fact, often an equivalent mean-field model is set up to identify dynamics of the magnetic field in these global MHD simulations. On the other hand, the mean-field models have been extensively employed in the past to explore the cause of solar cycle variability. Therefore, below we shall consult the mean-field version of the above equations to identify various mechanisms that can possibly lead to cycle modulations.

Writing the magnetic and velocity fields in terms of mean/large-scale and fluctuating/small-scale parts and applying suitable approximations in Eqs. (1) and (2), one can obtain the following equations (Krause and Rädler 1980).

$$\begin{aligned} \frac{\partial \varvec{\overline{B}}}{\partial t}= & {} \varvec{\nabla } \times ( \varvec{\overline{v}} \times \varvec{ \overline{B}} + \varvec{ \overline{{\mathcal {E}}}}), \end{aligned}$$

(3)

$$\begin{aligned} \rho \left[ \frac{\partial \varvec{\overline{v}}}{\partial t} + (\varvec{\overline{v}} \cdot \varvec{\nabla }) \varvec{\overline{v}} \right]= & {} -\varvec{\nabla }P + \varvec{\overline{J}}\times \varvec{\overline{B}} + \overline{ \varvec{ J^\prime } \times \varvec{B^\prime } } - \varvec{ \nabla } \cdot \rho \varvec{\overline{Q}} + \varvec{\overline{F}}, \end{aligned}$$

(4)

where the quantities with overline and prime respectively denote the mean and fluctuating components. The mean electromotive force $\varvec{ \overline{{\mathcal {E}}}}$ is given by

$$\begin{aligned} \varvec{ \overline{{\mathcal {E}}} } = \overline{ \varvec{v^\prime } \times \varvec{B^\prime } }. \end{aligned}$$

(5)

In the mean-field theory, this $\varvec{ \overline{{\mathcal {E}}}}$ is written in terms of the mean magnetic field in some limiting case (which holds at small Strouhal and magnetic Reynolds numbers) as follows.

$$\begin{aligned} \overline{{\mathcal {E}}}_i = \hat{ \alpha }_{ij} \overline{B}_j + \hat{ \eta }_{ijk} \frac{\partial \overline{B}_j}{\partial x_k}, \end{aligned}$$

(6)

Some components of $\hat{\alpha }$ tensor give the dynamo action and some of the components of $\hat{ \eta }$ tensor are responsible for the diffusion of fields. For homogeneous and isotropic turbulence, one can show that

$$\begin{aligned} \varvec{ \overline{{\mathcal {E}}} } = \alpha \varvec{\overline{B}} - \eta _t ( \varvec{\nabla } \times \varvec{\overline{B}}), \end{aligned}$$

(7)

where $\alpha = - \frac{1}{3} \tau _{\rm{corr}} ( \overline{ \varvec{ v^\prime } \cdot \varvec{\nabla }\times \varvec{v^\prime } } - (\rho \mu _0)^{-1} \overline{ \varvec{ B^\prime } \cdot \varvec{\nabla }\times \varvec{B^\prime } } )$ and $\eta _t \approx \frac{1}{3} \tau _{\rm{corr}} \overline{ \varvec{ {v^\prime } } \cdot \varvec{{v^\prime }} }$. In the classical $\alpha {\Omega }$ dynamo model, this $\alpha$ coefficient is the one which is responsible for the generation of the poloidal magnetic field from the toroidal one (Parker 1955a; Steenbeck et al. 1966). The turbulent diffusivity $\eta _t$ is several orders of magnitude larger than the molecular $\eta$ and that is the reason we have dropped out the $\eta$ term in Eq. (3).

If the turbulence is inhomogeneous, then there will be an additional term $\varvec{\gamma } \times \varvec{\overline{B}}$ in the above $\varvec{ \overline{{\mathcal {E}}}}$. This $\varvec{\gamma }$ is the magnetic pum** which is usually ignored in most of the kinematic mean-field dynamo models, but found to be important in the solar dynamo (e.g., Guerrero and de Gouveia Dal Pino 2016; Karak and Miesch 2017) and has also been detected in global convection simulations (Racine et al. 2011; Augustson et al. 2015; Simard et al. 2016; Warnecke et al. 2018).

The value of Q in Eq. (4) can be written in terms of Reynolds and Maxwell stresses as

$$\begin{aligned} \overline{Q}_{ij} = \overline{ v_i^\prime v_j^\prime } - (\rho \mu _0)^{-1} \overline{ B_i^\prime B_j ^\prime } = \overline{Q}_{ij}^\lambda - \mathcal{N}_{ijkl} \frac{\partial \overline{v}_k }{\partial x_l}, \end{aligned}$$

(8)

where the tensor $\varvec{{{\mathcal {N}}}}$ gives the turbulent viscosity $\nu _t$. While $\varvec{{{\mathcal {N}}}}$ is rotation dependent and has a complex form, it has two simple terms in the case of isotropic turbulence. Again like $\eta _t$, $\nu _t$ is usually much larger than the molecular viscosity ($\nu$) and thus the latter is neglected in Eq. (4).

The term $\overline{Q}_{ij}^\lambda$ is called the $\Lambda$ effect which drives angular momentum in the rotating CZ (Kippenhahn 1963; Rüdiger 1989; Kichatinov and Rüdiger 1993) to give rise to the differential rotation. While the second term in Eq. (8) tends to smooth out the nonuniformity in rotation, $\overline{Q}_{ij}^\lambda$ makes the rotation nonuniform.

4 Some historical developments of the dynamo models

4.1 Axisymmetric kinematic dynamo equations

Despite tremendous contribution to the field, the pioneering simulations of Gilman (1983) and Glatzmaier (1984, 1985) were discouraging for the dynamo modellers as they failed to reproduce most of the basic features of the solar magnetic field. Due to this failure, the dynamo modellers paid more attention to the mean-field models to study the solar cycle. Motivated by the observed large-scale magnetic and velocity fields, the coronal structure, and to make calculations simple, historically the mean-field solar dynamo was mostly studied under the axisymmetric approximation. With this approximation, the large-scale magnetic field can be written as

$$\begin{aligned} \varvec{\overline{B}}(r,\theta ,t) = \varvec{B_{\rm{p}}} + \varvec{B_{\rm{t}}} = \varvec{\nabla }\times \left[ A(r,\theta ,t)\hat{\varvec{\phi }} \right] + B (r,\theta ,t) \hat{\varvec{\phi }}, \end{aligned}$$

(9)

where $\varvec{B_{\rm{p}}} = B_r (r,\theta ,t) \hat{{\varvec{r}}} + B_\theta (r,\theta ,t) \hat{\varvec{\theta }}$ is the poloidal component of the magnetic field and $\varvec{B_{\rm{t}}} = B (r,\theta ,t) \hat{\varvec{\phi }}$ is the toroidal component. Similarly, the velocity can be written as

$$\begin{aligned} \varvec{\overline{v}}(r,\theta ) = \varvec{v_{\rm{m}}} (r,\theta ) + v_\phi (r,\theta ) \hat{\varvec{\phi }} = v_r (r,\theta ) \hat{{\varvec{r}}} + v_\theta (r,\theta ) \hat{ \varvec{\theta }} + r \sin \theta {{\Omega }} (r,\theta ) \hat{ \varvec{\phi }}, \end{aligned}$$

(10)

where $\varvec{v_{\rm{m}}} = v_r (r,\theta ) \hat{ {\varvec{r}}} + v_\theta (r,\theta ) \hat{ \varvec{\theta }}$ is the meridional circulation and ${\Omega } (r,\theta )$ is the angular frequency.

We note that here we have taken the velocity as the time-independent (steady), which is the case when we make the kinematic approximation. The kinematic approach has been adopted extensively in the literature to study the solar dynamo because in this case, we do not have to consider the equation for the flow (Eq. 4) and thus it makes the dynamo problem linear (see Eqs. 11, 12) below) and thus simpler. Nevertheless, observations provide us with the azimuthal flow in the whole CZ and the meridional flow in the near-surface layer. Given the fact that the differential rotation shows a little variation (in the form of torsional oscillation), one would expect that the kinematic approach is not a bad assumption for the sun.

After substituting the above forms of the fields in Eq. (3) and using the value of $\overline{{\mathcal {E}}}$ from Eq. (7) one can derive the following equations.

$$\begin{aligned}{} & {} \frac{\partial A}{\partial t} + \frac{1}{s}(\varvec{v_m} \cdot \varvec{\nabla })(s A) = \eta _t\left( \nabla ^2 - \frac{1}{s^2}\right) A + \alpha B, \end{aligned}$$

(11)

$$\begin{aligned}{} & {} \frac{\partial B}{\partial t} + \frac{1}{r}\left[ \frac{\partial (r v_r B)}{\partial r}+ \frac{\partial (v_\theta B)}{\partial \theta } \right] = \eta _t\left( \nabla ^2 - \frac{1}{s^2}\right) B + s(\varvec{B_p} \cdot \varvec{\nabla }){\Omega } + \frac{1}{r}\frac{d\eta _t}{dr}\frac{\partial (rB)}{\partial r}, \end{aligned}$$

(12)

where $s= r\sin {\theta }$ and $\eta _t$ is assumed to depend only on r.

In the above equations, the second term involving $\varvec{v_m}$ (or equivalently $v_r, v_\theta$) corresponds to advection of poloidal field by meridional flow and the first terms on the RHS of both equations represent the diffusion. In Eq. (11), the term $\alpha B$ is the source for the poloidal magnetic field. We discuss more about it in the next section. The term $s(\varvec{ B_p} \cdot \varvec{\nabla }){\Omega }$ in Eq. (12) is the source for the toroidal field, in which the nonuniformity of the rotation along the direction of poloidal field induces a toroidal field (the ${\Omega }$ effect). We note that while deriving above equations we have neglected a term ${\hat{\varvec{{\phi }}}} \cdot [\varvec{\nabla }\times (\alpha \varvec{B_p})]$. This is also a source for the toroidal field through the $\alpha$-effect. However, in Sun we believe that this term is negligible compared to the source due to differential rotation. The dynamo model constructed based on this assumption is traditionally called the $\alpha {\Omega }$ type dynamo in which the poloidal and toroidal fields maintain each other through a feedback loop. Finally, the last term in Eq. (12) gives rise to an advection due to nonuniform turbulent diffusion.

Fig. 3

Demonstration of Babcock–Leighton process. Decay and dispersal of two BMRs deposited symmetrically at $25^{\circ }$ are shown for three years. Note that due to finite tilts ($\approx 14^{\circ }$ as assigned by Joy’s law) of the BMRs, a net poloidal field near the pole is produced (see the weak field near the pole in the last snapshot). Snapshots are taken from a 3D model (Karak and Miesch \(\alpha \frac{\partial {\Omega } }{\partial r} < 0 \) in the northern hemisphere (Parker 1955a; Yoshimura 1975). The observed profile of ${\Omega }$ shows that $\frac{\partial {\Omega } }{\partial r}$ is positive in the low latitudes where sunspots emerge. Furthermore, the observations of BMR tilts also suggest that the $\alpha$ corresponding to the Babcock–Leighton process is also positive in the northern hemisphere. Hence, Parker–Yoshimura sign rule suggests a poleward migration in contrast to observations. This problem was resolved by introducing a meridional flow such that it is equatorward in the deeper CZ. A sufficiently strong flow can overpower the poleward dynamo wave and can explain the equatorward migration of the sunspot belt (Wang et al. 1991; Choudhuri et al. 1995; Durney 1995; Hazra et al. 2014a). While the poleward component of the meridional flow on the surface has been well-known for many years, recent helioseismic observations find some indication of the equatorward (return) flow near the base of CZ (Rajaguru and Antia 2015; Gizon et al. 2020). Also, there is controversy about the depth of the return flow and the number of cells that exist in the CZ.

The dynamo models in which the equatorward migration of the toroidal field at the BCZ is driven by the meridional flow (or some other flow), rather than dynamo waves, are popularly known as the flux transport dynamo models (Wang et al. 1991; Choudhuri et al. 1995; Durney 1995); see Karak et al. (2014a) for a review on this topic. Usually, the Babcock–Leighton dynamo models (in which the poloidal source is due to the Babcock–Leighton process) include a meridional flow to produce the equatorward migration of the toroidal field and thus they are of flux transport type, however, an $\alpha {\Omega }$ type dynamo can also be of flux transport type if a sufficiently strong meridional flow is present. Usually, these flux transport dynamo models, consider a single cell (in each hemisphere) meridional circulation profile with a return flow of about a few meters per second at a depth of about $0.7\,R_\odot$. Most of the existing Babcock–Leighton type flux transport dynamo models are kinematic, however, see Rempel (2006), Inceoglu et al. (2017) and Bekki and Cameron (2022) for exceptions.

5 Mechanisms of long-term variations

With the above introduction to the solar dynamo theory, we shall explore how the irregular variations in the solar cycle can occur. Broadly, we can think of the following three major causes for these.

• Magnetic feedback on the flow

• Stochastic forcing

• Time delays in various processes of the dynamo

5.1 Magnetic feedback on the flow

We have seen in Sect. 3 that the flows are essential for the dynamo mechanism. The large-scale component of velocity, as appearing in Eq. (3), is observed in the form of differential rotation and meridional circulation in the Sun. The differential rotation induces a strong toroidal field from the poloidal one in the CZ. The meridional circulation transports the magnetic field near the surface from low to high latitudes where it pushes the field to the deeper CZ, and it possibly transports the field towards the equator near the BCZ. Hence, it is natural that any dynamical change in these large-scale flows can cause variation in the solar cycle.

In Eq. (4), we find several dynamical terms through which modulation in the flow can arise. First is the mean Lorentz force $\varvec{\overline{J}} \times \varvec{ \overline{B}}$, which arises through the interaction between the mean magnetic and the mean current. This is also called the Malkus-Proctor effect (Malkus and Proctor 1975). The second is the small-scale feedback, which consists of two parts. One is the direct small-scale Lorentz forcing $\overline{\varvec{J^\prime } \times \varvec{B^\prime }}$ appearing in Eq. (4) (through the fluctuating current and magnetic field). Another is the dynamical modulation in the $\Lambda$ effect, which comes from the anisotropic turbulence (appearing through $\overline{Q}$ in Eq. 8). This modulation arises because the mean magnetic field also gives rise to the Lorentz force on the small-scale turbulence (Kitchatinov et al. 1994b). This, so-called micro (small-scale) feedback, has been captured through a simple quenching in the $\Lambda$ effect in many mean-field dynamo models (Küker et al. 1999). The dynamo-induced small-scale magnetic field also affects the large-scale flows and the turbulent transport; see Eqs. (4, 7, 8) and Käpylä (2019).

Next, the magnetic field gives feedback on the dynamo coefficients. The magnetic field dependence of the $\alpha$ coefficient is popularly studied in the literature (Pouquet et al. 1976; Field and Blackman 2002; Subramanian and Brandenburg 2004). While we do not have an analytical theory for the magnetic field dependence of the dynamo coefficients from the first principle, Rüdiger and Kichatinov (1993) and Kitchatinov et al. (1994a) respectively gave the dependences of the $\alpha$ and $\eta$ on the magnetic field using the quasi-linear and quasi-isotropic turbulence. Based on this theory, when the magnetic field is much larger than the equipartition field strength, the $\alpha$ falls as $1/B^3$, while many kinematic dynamo models traditionally use a quenching factor $1/\left( {1 + (B/B_{{{\text{eq}}}} )^{2} } \right)$ with $B_{\rm{eq}}$ being the equipartition field. MHD turbulent (Käpylä and Brandenburg 2009; Karak et al. 2014b) and global convection simulations (Racine et al. 2011; Simard et al. 2016; Warnecke et al. 2018) also do show some magnetic quenching in $\alpha$. Whatever be the exact magnetic field-dependent form of $\alpha$, in all these results one thing is clear: as the magnetic field tries to grow, it suppresses $\alpha$ and this in turn reduces the generation of the magnetic field. Hence, the nonlinearity in $\alpha$ tries to make the cycle regular, rather than producing irregularity in the cycle, provided the strength of $\alpha$ is not much higher than the critical $\alpha$.

The feedback on the small-scale turbulence can also be seen in the modulation of turbulent viscosity and magnetic diffusivity, which can in turn give some variation in the magnetic field. However, due to difficulties in computing these turbulent coefficients, we have limited knowledge on how much cycle modulation can arise due to magnetic feedback on the turbulent coefficients; anyhow, see the results of quasi-linear approximation (Kitchatinov et al. 1994a) and the convection simulations cited above. Unlike $\alpha$ quenching, the diffusivity quenching, however, tends to make the model unstable by increasing the magnetic field when the field strength is large (Kitchatinov and Olemskoy 2010) and thus unless some other mechanism, say $\alpha$ quenching is included, the dynamo usually does not produce stable cycle (Kitchatinov and Olemskoy 2010; Vashishth et al. 2021). The last two references have also shown that the nonlinearities in $\alpha$ and $\eta$ produce dynamo hysteresis—strong oscillatory magnetic field in the subcritical regime if the dynamo is started with a strong field and decaying solution otherwise if started with a weak field. This was also confirmed in numerical simulations of turbulent dynamos (Karak et al. 2015b; Oliveira et al. 2021).

We would like to emphasize that the recent observations of stellar rotation (Metcalfe et al. 2016) show that the rate of solar rotation is close to the minimum rate for the onset of the large-scale dynamo (also see Rengarajan 1984). Furthermore, Cameron and Schüssler (2017) and Cameron and Schüssler (2019) showed that the variability seen in the cosmogenic isotope for the last 10,000 years is consistent with the results from the generic normal form model for a noisy and weakly nonlinear limit cycle. Grand minima are only produced when the dynamo is not highly supercritical and the Sun and solar-like slowly rotating stars do produce grand minima (Kitchatinov and Olemskoy 2010; Kumar et al. 2021a; Vashishth et al. 2021, 2023). All these suggest that the solar dynamo is only slightly supercritical and weakly nonlinear. Thus, possibly the nonlinear effects are not very important in producing long-term modulations in the solar cycle.

5.2 Stochastic forcing

Solar CZ is highly turbulent and the turbulent quantities (appearing in Eqs. (7) and (8)) are subjected to fluctuations around their means in a time scale equal to the correlation time of the turbulent convection. As there is a finite number of convection cells over the longitudes at a given latitude in the Sun, the fluctuations in the turbulent coefficients are significant compared to their mean values. Hoyng (1988) argued that the fluctuations in the $\alpha$-effect can be larger than its mean and thus they can produce variation in the solar cycle (Choudhuri 1992; Hoyng 1993; Ossendrijver et al. 1996). In fact, due to small-scale dynamo, there are always fluctuations around $\overline{{\mathcal {E}}}$ (Brandenburg et al. 2008; Brandenburg and Spiegel 2008). The fluctuations in the angular momentum transport (as parameterized by the $\Lambda$ effect; Eq. 8) can also alter the differential rotation and meridional circulation and thus can produce modulations in the solar cycle (Rempel 2005; Inceoglu et al. 2017).

Fig. 4

Tilt angles of BMRs computed by “tracking” the MDI line-of-sight magnetograms covering the Cycle 23 (1996 September–2008 December). Here each BMRs are tracked over their lifetimes and the tilt and latitude of a BMR are taken by averaging their values over its time evolution when the flux is more than 60% of its maximum. In a solid line guides Joy’s law: $\gamma = 17.8 \sin \lambda$. b Shows the tilt distribution (with $5^{\circ }$ bin size) with fitted Gaussian (solid line) of $\mu = 10.1^{\circ }$ and $\sigma = 18.7^{\circ }$. The figure is produced using the data presented in Sreedevi et al. (2023)

Babcock–Leighton process, in which decay and dispersal of tilted BMRs generate a poloidal field in the Sun, involves some intrinsic fluctuations. The tilts of BMRs have a considerable amount of scatter around Joy’s law (Howard 1991; Stenflo and Kosovichev 2012; McClintock et al. 2014; Senthamizh Pavai et al. 2015; Jha et al. 2020). As seen in Fig. 4, the scatter is indeed much larger than the mean. Also, a large number of BMRs are having opposite tilts (negative in the northern hemisphere) due to non-Joy and anti-Hale configurations which generate opposite polarity field. Not only the tilt, but the rate of emergence and flux content of BMR also have considerable variations around their means. The cumulative effect of the fluctuations of all these parameters of BMRs can have a large impact on the polar field or the dipole moment at the end of a cycle which can lead to a considerable variation in the solar cycle (Nagy et al. 2017). On average, in the Sun, only a few (new) BMRs per day are produced and thus the short-term variation in the poloidal field is considerably large. This we can also identify by carefully observing the magnetic field on the solar surface (Cameron et al. 2013; Jiang et al. 2014a; Mordvinov et al. 2016; Kitchatinov et al. 2018; Karak et al. 2018a; Mordvinov et al. 2022). The variation in the inflows around BMR (Jiang et al. 2010; Martin-Belda and Cameron 2017; Nagy et al. 2020) and the meridional flow (Baumann et al. 2004; Karak 2010; Upton and Hathaway 2014a) also can change the amount of poloidal field generated. Like fluctuations in the classical-$\alpha$, there is a long list of work which reported the fluctuations in the Babcock–Leighton process and have utilized these in the dynamo models to reproduce various aspects of the long-term modulation of the solar cycle (e.g., Charbonneau and Dikpati 2000; Charbonneau et al. 2004; Charbonneau 2005; Charbonneau et al. 2007; Choudhuri and Karak 2009; Karak and Choudhuri 2011; Choudhuri and Karak 2012; Olemskoy and Kitchatinov 2013; Passos et al. 2014; Lemerle and Charbonneau 2017; Karak and Miesch 2017; Nagy et al. 2017)

5.3 Time delay in various processes of the dynamo

Time delays are involved in various processes of the dynamo action. Yoshimura (1978) argued that the adjustment of the velocity field due to the back reaction of the dynamo-generated magnetic field is not instantaneous and involves a bit of delay—at low Prandtl number, the fluctuations in the large-scale flow lag behind the Lorentz force. Furthermore, the modification of the thermodynamics due to magnetic feedback alters the velocity field and this involves again a time delay. Yoshimura (1978) showed that a time delay in the dynamo model produces a long-term modulation with occasional ceased activity. In the Babcock–Leighton dynamo framework, some time delays are unavoidable because the sources for the poloidal and toroidal fields are spatially segregated. The poloidal field from the surface layer has to be transported down to the deeper CZ to be sheared by the differential rotation, thus there involves a long time lag between the poloidal to the toroidal field. This lag is comparable to the solar cycle. Similarly the toroidal to the poloidal field conversion process also involves a time lag as the toroidal field needs to rise to the surface to form BMRs and then BMRs decay to give rise to the poloidal field. This delay however is short compared to the solar cycle length as the BMR eruption takes a few days to month and the decay takes another few months. While these two lags are captured by default in the numerical dynamo models of Babcock–Leighton and interface (in cases where the regions of shear and $\alpha$-effect are spatially segregated, MacGregor and Charbonneau 1997) types, a time delay is included by hand in the iterative map and the time-delay dynamo models (Sect. 5.4 of Charbonneau 2010). This time delay in the nonlinear dynamo model produces a variety of cycle modulations, including Gnevyshev–Ohl rule and intermittent cycles like the grand minima in certain parameter regimes (Durney 2000; Charbonneau 2001; Wilmot-Smith et al. 2006; Charbonneau et al. 2007). While in most of the Babcock–Leighton models, the toroidal to poloidal process is assumed to be instantaneous, Jouve et al. (2010) and Fournier et al. (2018) included a short delay in their model and made it magnetic field dependent regarding the fact that the flux tube with a strong magnetic field rises fast due to high magnetic buoyancy. This magnetic field-dependent time delay during the flux emergence in their flux transport dynamo with nonlinear $\alpha$-effect can produce some modulation in the cycle amplitude. We here note that time delays in all these models produce cycle variability only when the nonlinearity becomes important; it is the nonlinearity which is essential to produce cycle modulation. Thus, the time delay alone cannot produce a variability in the solar cycle.

With these basic discussions of the causes of the modulation of the solar cycle, we are now ready to discuss some illustrative models for the long-term variation in the solar cycle.

6 Mean-field models for long-term cycle variabilities

6.1 Models with nonlinear feedback on the large-scale flows

As discussed in Sect. 5.1, the large-scale flows are subject to change dynamically due to direct Lorentz feedback on the flow (Malkus and Proctor 1975) or through the feedback on the angular momentum transport (like $\Lambda$ effect; Kitchatinov et al. 1994b). Extensive research has been done on this topic to capture the Lorentz feedback of the dynamo-generated magnetic field (Spiegel 1977; Tavakol 1978; Ruzmaikin 1981). Particularly, using a simplified dynamo model (Küker et al. 1999) showed that a modification of the differential rotation by the large-scale Lorentz force produces strong modulation in the magnetic cycle including grand minima (also see, Moss and Brooke 2000). However, when the $\alpha$ quenching is added, the modulation is drastically suppressed (which is expected as the $\alpha$ quenching tends to limit the growth of the magnetic field). Some amount of cycle modulation and grand minima are again recovered if a strong $\Lambda$ quenching is included in this model; also see Kitchatinov et al. (1999) for a similar study. In a somewhat improved dynamo model but by including only the feedback of the large-scale magnetic field on the differential rotation, Bushby (2006) find some modulation in the magnetic cycle including grand minima like phases. Chaotic solutions are also produced in the highly truncated dynamo model (Weiss et al. 1984) which produces modulation in the cycle.

Basically, in all these models, the modulations happen in two ways (Knobloch et al. 1998). In one, the large-scale magnetic field of one parity (dipolar or quadrupole) drives velocity perturbations and energy is exchanged between the magnetic field and the flow. In this case, a large variation in the flow velocity is observed with no change in the parity. In the second case, there exists a nonlinear interaction between the dipole and quadrupole modes, mediated via the velocity perturbation which is driven by the Lorentz force. This modulation is associated with changes in the parity with almost no change in the velocity (Thelen 2000). These two mechanisms of cycle modulation in the literature are referred as Type II and I, respectively (e.g., Tobias 1997; Knobloch et al. 1998). Based on a nonlinear extension of the Parker (1993) model, Beer et al. (1998) showed that the latter type of modulation is the cause of Maunder-like grand minima. In Fig. 5, we see that during the grand minimum the parity of the magnetic field is changed. Based on a highly idealized simple model of the nonlinear dynamo equations, Weiss and Tobias (2016) and Beer et al. (2018) showed that the long time-scale ‘supermodulation’ apparent in the cosmogenic isotope data can be ascribed to switching of the dynamo between two different modulational patterns i.e., from dipole or quadrupole symmetry to mixed-mode solutions.

Fig. 5

Image reproduced with permission from Beer et al. (1998), copyright by Springer

Butterfly diagrams showing the toroidal field at a fixed radius as a function of time and latitude. In a the parity is interrupted by the occurrence of a grand minimum. The dynamo recovers from the grand minimum with a strong hemispheric asymmetry. b The grand minimum triggers a flip from a dipolar to quadrupolar parity.

Nevertheless, these models are still preliminary and fail to produce many detailed features of the observed magnetic field and the large-scale flows, particularly the correct amount of variation in the differential rotation. Furthermore, in some studies, the amount of feedback on the differential rotation and meridional circulation are tuned.

6.1.1 Variation in differential rotation

Observations find almost no variation in the differential rotation (Gilman and Howard 1984; Jha et al. 2021) except a tiny one ($< 0.5\%$) around the mean, which is known as the torsional oscillation (Howe 2009). Early dynamo models tried to explain torsional oscillation using the variation of Reynolds stresses due to dynamo-generated magnetic field (Kueker et al. 1996). Some other models tried to explain it using the mean Lorentz force of the dynamo-generated magnetic field on the momentum equation (Schüssler 1981; Chakraborty et al. 2009); see Pipin and Kosovichev (2019) who included both magnetic feedbacks on the turbulent angular momentum transport and the large-scale Lorentz force. In spite of that, none of these models could successfully explain both the equatorward and poleward branches of the torsional oscillation. A comprehensive model of Rempel (2006) showed that an enhanced surface cooling of the active region belt as proposed by Spruit (2003), in addition to the Lorentz forcing, is needed to explain the equatorward branch of torsional oscillation. This model expectedly finds almost no long-term modulation in the magnetic cycle due to this tiny variation in the differential rotation. Thus, we may expect that the observed tiny change in the differential rotation may not be a potential cause of the long-term modulation in the solar cycle. However, a series of mean-field dynamo calculations have demonstrated that a variety of cycle modulations including grand minima can be produced due to the nonlinear back reaction of the magnetic field on the large-scale flow through the so-called Type I modulation, which leaves a little imprint in the differential rotation (e.g., Beer et al. 1998; Knobloch et al. 1998; Bushby 2006; Weiss and Tobias 2016). Therefore, it is subtle to answer how much is the role of the tiny variation in differential rotation in producing cycle irregularity.

6.1.2 Variation in meridional circulation

The meridional circulation is also subject to vary due to the Lorentz forcing of the dynamo-generated magnetic field acting directly on it or through the alteration of the differential rotation. Models including the magnetic feedback often find a cyclic variation in the meridional flow (Rempel 2006; Passos et al. 2012; Hazra and Choudhuri 2017), in agreement with some observations (Hathaway and Rightmire 2010). Inflows around active regions also cause a cyclic perturbation in the meridional flow (Gizon and Rempel 2008; González Hernández et al. 2008, 2010). This cyclic change in the meridional circulation with no overall modulation cannot produce much variation in the magnetic cycle (Karak and Choudhuri 2012). However, when the amount of perturbation in meridional flow varies with the solar cycle strength, it can cause a significant modulation in the solar cycle (Jiang et al. 2010). Observations find some temporal variations in the meridional circulation (González Hernández et al. 2006), although there is no consensus on its long-term trend due to limited data. If the steady-state meridional circulation is maintained by a slight imbalance between two large terms—the non-conservative part of the centrifugal force and the baroclinic forcing (which arises due to a latitudinal temperature difference), a slight change in the balance can produce a large variation in the meridional circulation (Kitchatinov and Rüdiger 1999). In fact, the global convection simulations do find a tiny variation in the differential rotation but a considerable variation in the meridional circulation (Karak et al. 2015a; Passos et al. 2017), somewhat consistent with the available observations. By assimilating the synthetic magnetic proxies in the variational data assimilation method based on flux transport dynamo model, Hung et al. (2015) and Hung et al. (2017) also find a time-varying meridional circulation.

In the models, particularly in the flux transport dynamo models, the variation in the meridional circulation has been found to produce a profound effect on the solar cycle. In these models, the meridional flow regulates the cycle duration, weaker flow makes the cycle longer and vice versa (Dikpati and Charbonneau 1999). The meridional circulation has also an effect on the cycle strength, however, the effect depends on the diffusivity used in the model. If the dynamo operates in the diffusion-dominated regime (relative importance of the diffusion is more with respect to the advection due to flow), then a weaker flow allows the poloidal magnetic field to diffuse for a longer time and thus makes the magnetic field weak (Yeates et al. 2008). The opposite scenario happens when the flow is strong. Using this idea, Karak (2010) discovered that if we want to match the cycle duration by adjusting the speed of the meridional flow, then the cycle amplitudes are also matched up to some extent. Thus, a significant part of the variation of the solar cycle can easily be modelled simply by varying the speed of the meridional flow. Karak (2010) also showed that a sudden weakening of the meridional flow can trigger a Maunder-like grand minimum as shown in Fig. 6.

Fig. 6

Image reproduced with permission from Karak (2010), copyright by AAS

Figure showing that a sufficient drop in the meridional flow can trigger a Maunder-like grand minimum. a Shows the required meridional circulation speed in m s$^{-1}$ (solid/dashed for north/south). b Shows the location of the sunspots from the dynamo model.

Although the results from some of the flux transport dynamo models with variation in the meridional flow are very promising, in terms of modelling long-term variations in the solar cycle, it remains to be answered whether there was any large variation in the meridional flow in the past, particularly during the Maunder minimum.

6.1.3 Joint models with multiple nonlinearities

A few mean-field dynamo models were developed by considering full MHD equations with multiple possible nonlinearities (Brandenburg et al. 1989, 1991; Barker and Moss 1994; Thelen 2000; Jennings 1993; Muhli et al. 1995; Rempel 2006; Pipin and Kosovichev 2019; Sraibman and Minotti 2019). Most of these models included $\alpha$ quenching and Lorentz force feedback (in some form) in the momentum equation. The aim of these models was mostly to study the nonlinear stability and the operation of the dynamo. These models do not produce a considerable long-term modulation and grand minima unless some stochastic fluctuations in the dynamo parameter are included (Inceoglu et al. 2017). This will be discussed in the later sections.

6.2 Models with fluctuations

As discussed in Sect. 5.2, stochastic fluctuations in the solar dynamo is unavoidable and thus using these fluctuations numerous dynamo models have been constructed to explain the variable solar cycle.

6.2.1 Fluctuations in $\alpha$-effect

There is a long history studying the modulation of the solar cycle utilizing the stochastic fluctuations in the dynamo model. Choudhuri (1992), Hoyng (1993), Ossendrijver and Hoyng (1996), Ossendrijver et al. (1996), Gómez and Mininni (2006), Brandenburg and Spiegel (2008) and Moss et al. (2008) are some examples from a long list of publications in which stochastic fluctuations in the $\alpha$ parameter in their dynamo model were included and found long-term modulations including quiescent period like grand minima in some parameter regimes. We would like to mention that most of these models also include some nonlinearities, usually the $\alpha$ quenching to stabilize the dynamo. Therefore, it was found that when this $\alpha$ quenching was included, the variability was decreased. In Fig. 7, we present cycles from a simplified mean-field $\alpha {\Omega }$ dynamo model of Ossendrijver and Hoyng (1996) with stochastic fluctuations in the $\alpha$ term. They showed that with a certain amount of fluctuations in $\alpha$, the variability in the modelled cycle closely resembles the variability seen in the observed sunspot data.

Fig. 7

A representative case of the solar cycle (as measured by the toroidal field) from a simplified $\alpha {\Omega }$ dynamo model with stochastic noise in the $\alpha$-effect. Image reproduced with permission from Ossendrijver and Hoyng (1996), copyright by ESO

6.2.2 Fluctuations in $\alpha$-effect coupled with dynamic $\alpha$-effect

When the classical $\alpha$-effect is combined with another $\alpha$-effect having a magnetic field-dependent lower threshold, a large modulation is expected. The best example for this is the dynamo model coupled with the dynamic $\alpha$-effect which is produced due to the instability in the flux tube at the BCZ (Schmitt 1985; Chatterjee et al. 2011). In a mean-field dynamo model, Schmitt et al. (1996) included this dynamic $\alpha$-effect in the overshoot layer below the CZ in addition to the classical $\alpha$-effect. As the dynamic $\alpha$-effect works only when the magnetic field is greater than a threshold field strength, it stops operating when the field falls below this threshold. Schmitt et al. (1996) and Ossendrijver (2000) showed that when the magnetic field is strong in a normal cycle, both $\alpha$ operate concurrently. However, due to stochastic fluctuations, the magnetic field can occasionally fall below the threshold and the dynamical $\alpha$ stops operating. This caused the magnetic field to fall drastically—that is the beginning of a grand minimum; see Fig. 8. Note that in this case, the magnetic field can suddenly drop to a considerably lower value. During this quiescent period, the classical $\alpha$ alone slowly grows the field and recovers the model from grand minimum.

Fig. 8

Cycle modulations and grand minima (as measured by the magnetic energy) in the dynamo model with stochastic fluctuations in the $\alpha$-effect combined with a (threshold field dependent) dynamic $\alpha$ produced due to the instability in the flux tube at the BCZ. Image reproduced with permission from Schmitt et al. (1996), copyright by ESO

Fig. 9

The durations vs the times of their occurrence of the grand minima in the 2D Babcock–Leighton dynamo model of Choudhuri and Karak (2012). Image reproduced with permission from Choudhuri and Karak (2012), copyright by APS

6.2.3 Fluctuations in Babcock–Leighton process

For about the last two decades, Babcock–Leighton type flux transport dynamo models have been extensively used to explain the variabilities in the solar cycle. The first landmark paper in this series came from Charbonneau and Dikpati (2000) who included stochastic fluctuations in their 2D (axisymmetric) flux transport dynamo model. For this, they added a stochastic term with a coherence time of a month in the $\alpha$ parameter, the Babcock–Leighton source term in their axisymmetric model. They essentially replaced $\alpha$ by $\left[ 1+ s~ \sigma (t)\right] \alpha$ in Eq. (11), where $\sigma$ is random deviate within $[-1,1]$ whose value is updated every month and s determines the level of fluctuations. They found some modulation in the solar cycle, including the observed weak anti-correlation between the cycle duration and the amplitude with 200% fluctuations ($s=1$) in $\alpha$. Later Charbonneau et al. (2007) showed that the long time delay inbuilt in this type of Babcock–Leighton dynamo model naturally reproduces a Gnevyshev–Ohl like pattern in the modelled solar cycle (more in Sects. 6.4 and 8.3). Fluctuations in the Babcock–Leighton process can also lead to a large variation in the poloidal field with occasional dips in the polar field (as observed in the solar magnetic field), which can lead to double peaks (Gnevyshev gaps) and spikes in the following cycle (Karak et al. 2018a). Large fluctuations in the Babcock–Leighton process also lead to Maunder-like grand minimum as shown initially by Charbonneau et al. (2004) and later by many other authors (Choudhuri and Karak 2009; Passos et al. 2012; Hazra et al. 2014b; Passos et al. 2014) in different Babcock–Leighton type dynamo models. In particular, Choudhuri and Karak (2012) estimated the amount of variation in the polar field at the end of a cycle (a cumulative effect of the fluctuations in the Babcock–Leighton process) and the variation in the meridional circulation based on indirect observations and included those into their high diffusivity dynamo model. They found the correct frequency of the grand minima in the last 11,000 years (Fig. 9). Another work was by Olemskoy and Kitchatinov (2013) who also made an estimate of the level of fluctuations in the Babcock–Leighton process by computing the contribution to the polar field from the sunspot group data of Royal Greenwich, Kodaikanal and Mount Wilson Observatories. They found that the statistic of grand minima are consistent with the Poisson random process, which indicates that the initiation of grand minima is independent of the history of the past minima (also see Karak and Choudhuri 2013). They also showed that there is a correlation between the occurrence of grand minima and the deviation from the dipolar parity and thus the hemispheric asymmetry; also see Nagy et al. (2017) and Hazra and Nandy (2019) for the same conclusion from different Babcock–Leighton models. Figure 10a shows the smoothed (in the same way as done in, Usoskin et al. 2007) sunspot number from an 11,000-year long simulation done by Olemskoy and Kitchatinov (2013) in which the red and blue shaded areas represent the grand maxima and minima, respectively.

The hemispheric asymmetry which is a robust feature during grand minima is also reflected in a typical grand minimum as presented in Fig. 10b. The fluctuations in the Babcock–Leighton process of north and south hemispheres are uncorrelated and thus hemispheric asymmetry is unavoidable during grand minima in this model and also other dynamo models with fluctuations in the Babcock–Leighton process (Passos et al. 2014; Karak and Miesch 2018). The hemispheric asymmetry, however, cannot remain for multiple cycles as the diffusive coupling at the equator tends to smooth out the asymmetry acquired due to fluctuations (Chatterjee and Choudhuri 2006; Karak and Miesch 2017). Another way that north–south asymmetry in the magnetic field can come about in these models is due to the random excitation of the quadrupolar mode by the stochastic fluctuations in the Babcock–Leighton process (Schüssler and Cameron 2018).

Fig. 10

a A proxy of the smoothed sunspot number from the dynamo model of Olemskoy and Kitchatinov (2013). Blue and red shaded regions correspond to the grand minima and maxima (using the same definition as used in, Usoskin et al. 2007). b Toroidal field as a function of latitude and time, highlighting a grand minimum. Images reproduced with permission from Olemskoy and Kitchatinov (2013), copyright by AAS

Further support for the stochastic origin of the long-term modulation of the solar cycle came from Cameron and Schüssler (2017) who studied the following ‘stochastic’ normal form model.

$$\begin{aligned} {d X } = \left( \beta + i \omega _0 - (\gamma _r + i \gamma _i) |X|^2\right) X {d t} + \sigma X dW_c = 0, \end{aligned}$$

(13)

where X is a complex quantity whose real and imaginary components give the toroidal and poloidal fields, $\beta$ determines the growth rate of the dynamo and thus the supercriticality, $\omega _0$ sets the magnetic cycle frequency, $\gamma _r$ and $\gamma _i$ regulate the nonlinearity of the model and determine the cycle amplitude, $W_c$ represents a complex Wiener process and $\sigma$ is a measure of added noise. The values of all these parameters are fixed by the observations. Cameron and Schüssler (15). The poloidal field generated from these few BMRs is alone sufficient to recover the model to the normal phase. Their model reproduces most of the features of the grand minima (including frequency of grand minima, longer cycles and strong hemispheric asymmetry during grand minima).

6.2.5 Variability vs dynamo supercriticality

Whenever there is any change in the dynamo number D ($= \alpha _0 { \Delta {\Omega }} \,R_\odot ^3 /\eta _0^2$, where $\alpha _0$ is the strength of $\alpha$-effect, ${\Delta } {\Omega }$ is the amount of shear in the CZ, and $\eta _0$ is the diffusivity), there will be a change in the amplitude of the magnetic field. Thus, the cycle modulation due to fluctuations in the dynamo parameter is obvious. However, for a given level of fluctuations and the form of nonlinearity, the amount of variability depends on the value of D or the regime operation of the dynamo. This is apparently seen in Fig. 16 that the same amount of variation in D causes a large variation in magnetic field when the dynamo operates near the critical transition ($\delta |B_c|$) and a small variation when the dynamo operates in supercritical regime ($\delta |B_s|$).³ The reason for this is not difficult to understand. When the dynamo operates near the critical transition, a small D makes the growth rate of the magnetic field small and the dynamo weakly nonlinear. Now consider a scenario when the magnetic field has become weak due to a reduction of D (or $\alpha _0$) and after some time due to fluctuations, D has increased. Then the dynamo will (almost linearly) amplify the field for a long time before the nonlinearity becomes important and thus the net growth of the field will be large. On the other hand, if the dynamo operates in a highly supercritical regime, then the nonlinearity will quickly suppress the dynamo growth and the net amplification of the field will be small.

Fig. 16

A Hopf bifurcation diagram, showing the transition from a fixed point to dynamo instability. This is a typical variation of the magnetic field strength (|B|) vs dynamo number (D) in dynamo model with any nonlinear quenching mechanism as long as D is not much larger than $D_c$. Here, $D_{\rm{c}}$ is the critical D. $\delta |B_c|$ and $\delta |B_s|$ are the amplitude variations of the magnetic field for a given change in D in two different regimes of the dynamo

The above discussion also means that in the near-critical (or weakly supercritical) regime, we expect long-term modulation in the cycle and extended grand minima. In this regime when the field becomes weak due to fluctuations, the dynamo will take a long time (several cycles) to grow the field and this will tend to produce a smooth long-term variation. In contrast, in the super-critical regime, we do not expect much long-term modulation in the cycle amplitude and no extended grand minima because when the magnetic field falls to a low value, the dynamo will quickly increase the field in a cycle. This is clearly seen in Fig. 17; also see Vashishth et al. (\({\Omega }\) effect. There is also a short time delay between the toroidal and poloidal field conversion as the toroidal flux tubes take finite time to rise to the surface to form BMR and a finite time is spend to decay and disperse the BMR. All these two delays in the nonlinear dynamo model can produce a variety of modulations.

6.4.1 Iterative map

Durney (2000) assumed that there is a delay of one cycle between the poloidal field of cycle n and the toroidal field of cycle $n+1$ and brilliantly reduced the Babcock–Leighton dynamo equations into a iterative map. He writes,

$$\begin{aligned} T_{n+1} = { \Delta } {\Omega } { \Delta } t P_n = a P_n, \qquad n = 0, 1, 2, \dots \end{aligned}$$

(14)

(Here ${ \Delta } {\Omega }$ is the shear in the CZ and ${ \Delta } t$ is the time interval during which the poloidal field acts on the shear.) Neglecting the time delay in the toroidal to poloidal fields conversion, we can write the following nonlinear relation:

$$\begin{aligned} P_{n+1} = f(T_{n+1}) T_{n+1} \end{aligned}$$

(15)

Here $f(T_{n+1})$ is a measure of the efficiency of the poloidal field generation from the toroidal field (Babcock–Leighton process) which depends on the toroidal field. Substituting Eqs. (14) into (15) and normalizing the fields appropriately, we find

$$\begin{aligned} p_{n+1} = a\,f(p_{n}) p_{n} \end{aligned}$$

(16)

(where normalizing factors are absorbed in a). For different nonlinear functions (f) of the Babcock–Leighton process, different maps can be constructed. Durney (2000) chose it $1 + \beta ( 1 - p_n)$ and thus the map became

$$\begin{aligned} p_{n+1} = p_{n} \left( 1 + \beta ( 1 - p_n) \right) \qquad \beta > 0. \end{aligned}$$

(17)

Charbonneau (2001) chose it $\gamma (1 - p_n) p_n$ which produced a map

$$\begin{aligned} p_{n+1} = \gamma ~ p_{n} ^2 ( 1 - p_n) \qquad \gamma > 0. \end{aligned}$$

(18)

By capturing a lower cut off in the Babcock–Leighton process, Charbonneau et al. (2005) produced another map

$$\begin{aligned} p_{n+1} = \alpha\,f(p_n) p_n \qquad \alpha > 0, \end{aligned}$$

(19)

where

$$ f(p_{n} ) = \frac{1}{4}\left[ {1 + erf\left( {\frac{{p_{n} - p_{1} }}{{w_{1} }}} \right)} \right]\left[ {1 - erf\left( {\frac{{p_{n} - p_{2} }}{{w_{2} }}} \right)} \right]{\text{ }} $$

(20)

(Here $p_1 = 0.6$, $w_1 = 0.2$ $p_2 = 1.0$, and $w_1 = 0.8$.)

Charbonneau (2001) and Charbonneau et al. (2005) showed that as the map parameter increases, the transition from the fixed amplitude oscillation to the chaotic solution occurs through a sequence of period doubling; see left panel of Fig. 20 for the map given by Eq. (19). Durney (2000) showed that in the parameter regime of the doubly periodic oscillations, the Gnevyshev–Ohl rule can be explained. Later Charbonneau (2001) showed that this is indeed not necessary, stochastic perturbation outside this region can also produce Gnevyshev–Ohl rule as a consequence of the oscillatory nature of the convergence to the fixed point; even the map without showing limit cycle for $1 / (1 + B^2)$ type nonlinearity also show Gnevyshev–Ohl rule. Intermittent and chaotic solutions are produced in all these maps, as long as the map parameter is above a certain value; also see Fig. 5 of Charbonneau (2001) for an illustration.

Fig. 20

Bifurcation diagrams. a Cycle amplitude (iterate $P_n$) versus the map parameter $\alpha$ from Eq. (19). b Same as left one but obtained from 2D numerical dynamo model of Charbonneau et al. (2005) and shows the magnetic energy as function of the dynamo number $C_s$. Images reproduced with permission from Charbonneau et al. (2005), copyright by AAS

6.4.2 1D time-delay dynamo

Finite delays are also included in the 1D dynamo models in which the equations for the toroidal and poloidal fields are truncated by removing the spatial dependences in the following way.

$$\begin{aligned} \frac{\rm{d} B}{\rm{ d} t} = \frac{\omega }{L} A (t - T_0) - \frac{B}{\tau _d} \end{aligned}$$

(21)

$$\begin{aligned} \frac{\rm{d} A}{\rm{ d} t} = \alpha _0 f(B(t - T_1)) B(t-T_1) - \frac{A}{\tau _d}. \end{aligned}$$

(22)

Here, $T_0$ and $T_1$ represent the time delays required for the generation of toroidal and poloidal fields, respectively. $\omega$ and L are the contrast in differential rotation and the length scale in the tachocline, $\tau _d$ is the diffusion time scale, $\alpha _0$ is the amplitude of the Babcock–Leighton source. $f(B(t - T_1))$ is the nonlinear function which represents the suppression of the Babcock–Leighton mechanism. By considering $f(B(t - T_1))$ of the from of Eq. (20) (with lower quenching), Wilmot-Smith et al. (2006) found irregular cycles in a certain parameter regime (when the time delay is larger than the diffusion time). Later, including fluctuations in the $\alpha _0$ term, Hazra et al. (2014b), Kumar et al. (2021a) and Tripathi et al. (2021) obtained long-term modulations and grand minima like intermittent solutions in a range of parameters.

6.4.3 2D time-delay dynamo

In models like the Babcock–Leighton type flux transport and the interface dynamos, in which the source regions for the fields are spatially segregated, the time delays are by default inbuilt into the equations. Thus all the Babcock–Leighton dynamo models discussed in this review are also time-delay models. Charbonneau et al. (2005) showed that the 2D Babcock–Leighton dynamo models also show the same type of behaviour as seen in the reduced map. Figure 20b, shows the bifurcation diagram of a 2D Babcock–Leighton dynamo model with a nonlinear quenching function of the form given by Eq. (20). Again, with this type of nonlinearity, we observe that the solution goes to a chaotic regime through a sequence of period doubling with the increase of dynamo number. The model is capable to produce Gnevyshev–Ohl rule in a wide range of parameter regimes with stochastically forced $\alpha$ (Charbonneau et al. 2007); more in Sect. 8.3.

Fig. 21

Cycle modulations (as measured by the toroidal field at $r = 0.7\,R_{\odot }$ and $\theta = 20^{\circ }$) in the flux transport dynamo model of Jouve et al. (2010) with magnetic field-dependent delay in the Babcock–Leighton source for the poloidal field generation process (Eq. 23). The left and right panels are for short (14 days on 1 kG fields) and long delay (14 days on 50 kG fields so that 10 kG fields will be delayed by almost a year in this case compared to a few hours in the previous case). Images reproduced with permission from Jouve et al. (2010), copyright by ESO

Jouve et al. (2010) went one step ahead of this and included the short time delay associated with the flux emergence from the deep-seated toroidal flux that is usually ignored in the dynamo models. They realized the fact that the buoyancy time delay depends on the magnetic field strength—strong flux tubes experience strong buoyancy and thus rise quickly compared to the weaker ones (Fan et al. 1994). Jouve et al. (2010) captured this delay in their flux transport dynamo model by replacing the poloidal source term: $\alpha B$ in Eq. (11) by

$$\begin{aligned} \frac{\alpha {B}(0.7\,R_\odot , \theta , t - \tau _{\rm{B}})}{1 + \left( { {B}(0.7\,R_\odot ,\theta , t - \tau _{\rm{B}}) }/{B_0}\right) ^2}, \end{aligned}$$

(23)

where $\tau _{\rm{B}}$ is the delay time which they took to be equal to $\tau _0 / B(0.7\,R_\odot , \theta , t)^2$ and $\tau _0$ is used to regulate the amount of delay. Figure 21 shows the results from two simulations with different amounts of delay as presented by Jouve et al. (2010). We evidently see that just the addition of this delay in the poloidal source produces a considerable amount of modulation in the cycle and the amount of modulation increases with the increase of delay (also see Fournier et al. 2018).

However, we note that in these models, the flux loss due to magnetic buoyancy is ignored (Sect. 6.3). The flux loss tries to keep the magnetic field around the equipartition value. In that case, the toroidal flux tube will not have much different field strength from one another and the delay times will not be very different. Hence, if the flux loss due to magnetic buoyancy is incorporated in these models then we do not expect much modulation in the cycle. Biswas et al. (2022) included flux loss due to magnetic buoyancy in the dynamo model with local $\alpha$ prescription and they did not find noticeable modulation in the cycle due to field-dependent delay with respect to the case without delay.

7 MHD simulations for long-term cycle variabilities

In the MHD simulations one needs to solve the following continuity equation for the mass and the energy equation in addition to Eqs. (1) and (2).

$$\begin{aligned}{} & {} \frac{\partial \rho }{\partial t} + \varvec{\nabla }\cdot ( \rho \varvec{v}) = 0, \end{aligned}$$

(24)

$$\begin{aligned}{} & {} T \left[ \frac{\partial s}{\partial t} + (\varvec{ v} \cdot \varvec{\nabla }) s \right] = \frac{1}{\rho } \left[ \eta \mu _0 \varvec{J}^2 - \varvec{\nabla }\cdot (F_{\rm{rad}} + F_{\rm{SGS}}) - F_{\rm{cool}} \right] + 2 \nu \varvec{S}^2, \end{aligned}$$

(25)

where $\rho$ is the density, $\nu$ is the kinematic viscosity, $F_{\rm{rad}}$ is the radiative diffusive flux and is given by $- K \varvec{\nabla }T$, K being the heat conductivity, $F_{\rm{SGS}}$ represents the additional subgrid scale (SGS) diffusion which is used to keep the simulation numerically stable (usually taken as $-\chi _{\rm{SGS}} \rho T \varvec{\nabla }s^\prime$ with $\chi _{\rm{SGS}}$ being the SGS diffusion coefficient and $s^\prime$ is the fluctuations of entropy), and $F_{\rm{cool}}$ is the radiative cooling near the surface. These equations are numerically solved using appropriate initial and boundary conditions to study the dynamo problem; see for example Käpylä et al. (2020) for the detailed profiles of all the model parameters and the boundary conditions.

Fig. 22

Cycles from the global MHD convection simulations using EULAG-MHD code. Image reproduced with permission from Passos and Charbonneau (2014), copyright by ESO

In the last one decade, global MHD convection simulations have reached to a somewhat realistic level (not in terms of the Reynolds numbers but in terms of the level of turbulence and the realistic value of Rossby number). They produced some basic features of the large-scale flows and the magnetic field. We refer the readers to Sect. 6 of Charbonneau (2020) for a review on this subject. An advantage of these simulations is that all the nonlinear and stochastic effects are included by default in these simulations, in contrast to the mean-field models where these effects need to be included by hand. However, due to limited computation facilities, global convection simulations were rarely run for a longer time so that a long-term cycle modulations can be studied. Furthermore, being extremely complicated in nature, identifying the mechanisms of the long-term modulations are not trivial. Fan and Fang (2014), Karak et al. (2015a), Viviani et al. (2018) and Viviani et al. (2019) presented some simulations which were run for a somewhat longer duration. Here we discuss three important results for the cycle modulations gleaned from three different numerical codes. The first one is from Passos and Charbonneau (2014) who presented cycles from a simulation run of 1650 years in which 40 excellent cycles with average period of 40 years were seen (Fig. 22). This simulation ‘broadly’ reproduces some observed features of the solar magnetic cycle including the regular polarity reversal and dipole-dominated large-scale field. Interestingly a good amount of modulation in the cycle amplitude is naturally produced in this simulations. There is also a pattern for the Gnevyshev–Ohl rule (watch the peaks after $t=800$ yr) and a hint for the Gleissberg modulation in these cycles. However, in this 1650 years of simulations no grand minimum or maximum is seen. With respect to the solar observations, there are some discrepancies as well which include an in-phase variation of the poloidal and toroidal components, magnetic activity confined to high latitudes, and the low degree of hemispheric coupling.

Fig. 23

Left top: Time-latitude distributions of the longitude average $B_r$ and $B_\phi$ at $0.92\,R_\odot$ from the ASH simulation code of Augustson et al. (2015). The grand minimum identified in this simulation is marked by vertical lines. Right panel shows the parity of the magnetic field computed at $0.75\,R_\odot$ (orange curve) and $0.95\,R_\odot$ (black). Images reproduced with permission from Augustson et al. (2015), copyright by AAS

The second important result came from Augustson et al. (2015) utilizing the 3D MHD ASH code which is shown in Fig. 23. This is from a model of one solar mass rotating at three times the Sun. Again this simulation produces many features that are consistent with observation and the important one is the equatorward migration of the toroidal field belt at low latitudes which is caused by the nonlinear modulation of the differential rotation. The interesting feature in their simulation is that a clear grand minimum was identified with a considerably reduced magnetic field for about five cycles. However during this minimum, the large-scale field failed to reverse, although its amplitude oscillated cyclically. The grand minimum in this simulation is possibly caused by the interplay between the symmetric and anti-symmetric dynamo families. During the regular cycle, the anti-symmetric dynamo family is greater than the symmetric family but during grand minimum phase, the symmetric family dominates over the anti-symmetric one; see the right panel of Fig. 23 for the increased parity of the radial field at two different depths. The mechanism for the generation of the grand minimum in this simulation is similar to the one proposed by Tobias (1997) and Moss and Brooke (2000) based on the nonlinear mean-field dynamo model.

Fig. 24

Temporal variation of the mean $B_r$ average over the longitudes on the surface from global convection simulation using Pencil Code (Käpylä et al. 2016). The color scale is saturated at the half of the extrema i.e., at [$-11.7$, 10.4] kG. Note the disturbed magnetic activity during $t =$ 20–45 years. Image reproduced with permission from Käpylä et al. (2016), copyright by ESO

The final one is from Käpylä et al. (2016) utilizing the Pencil Code. They performed this simulation for a model of one solar mass rotating at five times the Sun. This simulation also produced some solar-like features, including regular polarity reversals and an equatorward migration of the toroidal field at low latitudes (due to a nonlinear dynamo wave). In addition to the dominant global magnetic cycle of an average period of 4.9 years, there are two other prominent cycles, one having a higher frequency mode near the surface and at low latitudes with poleward migration, and the other one having low frequency residing at the BCZ. Their simulation also finds an episode of reduced magnetic field for about three cycles, happening asynchronously in hemispheres ($t =$ 20–45 years). Interestingly, the magnetic field in the deeper CZ is stronger during this period and thus the global magnetic field during this period is larger than during the normal phase. The dynamics of the magnetic fields in this simulation are extremely complex and the grand episode of the reduced activity is caused through the interplay of various dynamo modes.

Although the global convection simulations produce some long-term modulations in the cycles and the grand minimum-like reduced activity which are consistent with solar observations, there are few caveats that we should keep in mind. The cycle modulations observed in the above convection simulations are many ways far from the actual Sun. The large-scale flows produced in these simulations are also quite far from the real Sun. Importantly, the power in the sub-surface convective flow at large scales are much stronger than that is obtained from the measurements, so-called the convective conundrum (Hanasoge et al. 2012; Lord et al. 2014; also see e.g., Hotta et al. 2015; Karak et al. 2018b for the studies that attempt to resolve this.) Furthermore, we do not have any information about whether these long-term modulations found in the convection simulations are robust in the model parameter regimes. Finally, the simulations do not produce BMRs which are important component of the solar cycle and are responsible for the generation of the poloidal field in the Sun, at least the field that is observed on the surface. Therefore, future work is needed to make the global convection simulations more realistic so that they can be utilized to study the long-term cycle modulation.

Ideal MHD simulations in the local box are also performed to study the large-scale dynamos. With helical forcing and imposed shear, Karak et al. (2015b) performed HMD simulations in the local Cartesian geometry and found modulations in the large-scale magnetic cycle. The interesting fact about their study was that they found grand minima like intermittent activity only when the dynamo operates in the subcritical and critical regime but not in the supercritical regime. This independent study thus supports the idea that the variability and the grand minima are less probable in the supercritical dynamo; see Sect. 6.2.5 for details. Local simulations are also useful to study the cycle modulations in presence of the small-scale dynamo. The small-scale magnetic field generated from the small-scale dynamo affects the flows and thus the global dynamo (Karak and Brandenburg 2016). Global dynamos are usually performed at low Reynolds numbers and thus small-scale dynamo is not excited, except a few (e.g., Nelson et al. 2013; Käpylä et al. 2017; Hotta and Kusano 2021), however, they are not ran for many cycles.

8 Some open questions and current trends

8.1 Do grand minima represent different states of the solar dynamo?

By analyzing the solar activity data for the past, Usoskin et al. (2014) showed that the distribution of the solar activity is bi-modal. Of this, the dominant mode corresponds to the regular activity phase and the reduced-activity mode which corresponds to the grand minima is distinct from the dominant regular one. As seen from Fig. 25, the distribution is clearly bimodal.

Fig. 25

Probability distribution function of the reconstructed sunspot number (filled grey curve) and the observed group sunspot number (red curve). Blue is the bi-Gaussian curve. Figure is modified after Usoskin et al. (2014); also see Wu et al. (2018) for the distribution from longer data

In terms of the dynamo theory, of course, the physics during grand minima is not quite the same as that during the regular cycle. For example, during the grand minima phases, the magnetic field falls to a low value and then the Sun takes some time to grow its magnetic field to the normal level. During these grand minima phases, the generation of the poloidal field is low because the Babcock–Leighton process which is the dominant source for the poloidal field in the Sun becomes less efficient due to fewer BMRs (Sect. 6.2.4). The strength and morphology of the flow can also be different during this phase. Therefore, the Babcock–Leighton dynamo models coupled with weak $\alpha$-effect and/or the dynamo models coupled with the dynamic $\alpha$-effect produced by the instability of flux tube (discussed in Sect. 6.2.2) can naturally explain the bimodal distribution. The dominant source of the poloidal field (Babcock–Leighton process or dynamic $\alpha$-effect) maintains the normal mode of the solar activity while the grand minima phase is maintained by the weak $\alpha$-effect. Time-delay models or the iterative map with low-amplitude additive noise and fluctuating map parameter/dynamo number with specific nonlinearity can also naturally produce the bimodal distribution of solar activity; see Fig. 1 of Charbonneau (2001) and Tripathi et al. (2021). Also, see Petrovay (2007) for a possible explanation of a bimodal solar dynamo based on an interface dynamo model coupled with a fast tachocline model. Karak et al. (2015b) using 3D simulations of turbulent dynamo suggested that the bimodal distribution of solar activity can be produced in the subcritical dynamo.

8.2 Do grand maxima require different mechanisms for their origin?

In most of the previous studies, the origin of the grand maxima is ignored. However, the mechanism for its generation is subtle because the dynamo is more nonlinear during the grand maxima phase. When the Sun tries to produce a strong magnetic field, the nonlinearity tries to quench its generation process (efficiencies of both the poloidal and toroidal field generations are reduced with magnetic field). Incidentally, the Sun spent less time in the grand maxima phase than in the grand minima phase (Usoskin et al. 2007; Solanki et al. 2004). Thus, producing extended grand maxima using the stochastic fluctuations in the dynamo parameters is less obvious. However, stochastic fluctuations still can produce a very strong magnetic cycles and grand maxima if they occur in a certain phase of the cycle. Kitchatinov and Olemskoy (2016) showed that at the beginning of a cycle (when the field at the poles is strongest) if the generation of the poloidal field is reversed (say due to the emergence of some wrongly tilted BMRs), then it will produce the same polarity field as it was there in the pole. Consequently, instead of reversing the old polarity polar field, it will amplify. This strong polar field will make the current cycle very strong. By introducing stochastic fluctuations in a 2D flux transport dynamo model, Kitchatinov and Olemskoy (2016) showed that this mechanism can occasionally produce a much stronger cycle which corresponds to the grand maxima phase. However, in this mechanism, not more than one strong cycle at a time is produced, while in the solar grand maximum, at least two consecutive cycles are strong (Usoskin et al. 2007). Also, this study is based on a kinematic model, in which the nonlinear feedback of the magnetic field on the flow is ignored.

Another way of generating the grand maxima is through the combined effect of multiple poloidal field generation processes. Ölçek et al. (2019) find that when the deep-seated $\alpha$-effect is coupled with the surface Babcock–Leighton process in a dynamo model, these two processes more or less contribute equally to the generation of the poloidal field through a sort of constructive interference. This could be the mechanism of grand maxima in their dynamo model. However again this model is kinematic and the magnetic feedback is not taken care of.

8.3 What is the origin of Gnevyshev–Ohl/Even–Odd rule?

One plausible explanation for the Gnevyshev–Ohl rule is the fossil field hypothesis. A steady large-scale magnetic field of fossil origin (Boruta 1996) can interfere with the oscillating magnetic field from the CZ. In one cycle the oscillating magnetic field appears in the same polarity as that of the fossil field and it makes the cycle strong. In the next cycle, the oscillating magnetic field becomes of opposite polarity and thus the cycle becomes weak. To explain the Even–Odd rule using this idea, the fossil field has to be comparable to the dynamo-generated oscillating magnetic field at the BCZ which is of the order of 10 kG. However, the present observations do not confirm this strong fossil field. Furthermore, if the Gnevyshev–Ohl rule is caused by the fossil field, then there should be infinite memory in it, in a sense that once this rule is established, odd (even) cycles will always be stronger than the previous even (odd) cycles, even if there are some violations due to other effects. However, studies show that there was a possible reversal in the even–odd pattern during 1745–1850 (Mursula et al. 2001; Tlatov 2013; Zolotova and Ponyavin 2015).

There is another possible explanation for the Gnevyshev–Ohl rule which was proposed by Durney (2000) using the nonlinear period-doubling effect. He suggested that the solar dynamo is operating in the region of period doubling beyond the bifurcation point and in this region the alternating amplitude modulation is unavoidable. Later Charbonneau (2001) showed that this is indeed not essential; stochastic forcing in the dynamo can lead to Even–Odd effect even outside this parameter range of period doubling. The time delays (Sect. 6.4) involved in the solar dynamo including simple amplitude limiting nonlinearity can produce the same period-doubling and Gnevyshev–Ohl rule as seen in the complex nonlinear system (Charbonneau et al. 2007). The dynamo model and map with various types of nonlinearity show Gnevyshev–Ohl rule under fluctuations in the poloidal source ($\alpha$). However, this Gnevyshev–Ohl rule becomes evident only when the nonlinearity becomes important (large map parameter or dynamo number) and it seems to be little restricted to a narrow range of the diffusivity in the BCZ (Sec 4.5 of Charbonneau et al. 2005). Thus, there is still room to explore the robustness of this feature.

8.4 What are the causes of Gleissberg and Suess/de Vries cycles?

Gleissberg and Suess/de Vries cycles are not strictly cycles, rather they are modulation over the dominant 11-year period and they are detected in the cosmogenic data having ranges of periods from 90 to 100 years and from 205 to 210 years, respectively. If these modulations are the true nature of the solar cycle, then they are probably coming from the nonlinear interaction between the magnetic field and the flows in the lower part of CZ. In a axisymmetric $\alpha \Omega$ dynamo model coupled with the angular momentum equation and ignoring meridional circulation, Pipin (1999) showed that the Gleissberg cycle is a results from the magnetic feedback on the angular momentum fluxes which maintains the differential rotation in the CZ. In this model, the period of the Gleissberg cycle is determined by the time associated with the re-establishment of the differential rotation after the magnetic perturbations of the angular momentum transport. If this is the mechanism of the Gleissberg cycle in the sun, then the observed differential rotation should show a variation in the Gleissberg timescale. However, the available observations does not provide a conclusive evidence on it (Howard 1978). Passos and Charbonneau (2014) found a hint of Gleissberg modulation in their computed proxy sunspot number but not in the radial field data and they require data for a longer duration to confirm its existence. Cameron and Schüssler (2019) showed that Gleissberg and Suess/de Vries cycles are consistent with realization noise and the noisy normal form model can reproduce these modulations.

9 Summary and discussion

Besides the 11-year (a)periodic variation of the amplitude, the most prominent variation of the solar cycle is the long-term modulation which has been unambiguously identified in the direct and indirect (cosmogenic data) observations of the solar activity. Examples of the long-term modulation include the Gnevyshev–Ohl/Even–Odd rule, grand minima, grand maxima, Gleissberg cycle and Suess cycles. In this review, we have presented comprehensive discussions on the origins and models of long-term variations. To do so, we have broadly identified the following three major causes for the cycle modulations: (1) magnetic feedback on the flow, (2) stochastic forcing, and (3) time delays in various dynamo processes. Problems in the nonlinear mean-field models are that not all possible nonlinearities are always included in the model and the resulting mean flows and magnetic fields are not compared carefully with observations. Global MHD convection simulations are very useful in this respect because all nonlinearities are captured by default. They have begun to produce some correct results of the large-scale field and the flow, however, there are big discrepancies as well. Until global convection simulations reach somewhat realistic parameter regimes and resolve the major issues (convective conundrum, observed large-scale flows, nonappearance of BMRs), we need to rely on the mean-field models only.

Stochastic fluctuations in the mean-field models are enough to explain many features of the long-term variabilities. The Babcock–Leighton dynamo models are promising models for the solar cycle, in terms of their success in reproducing the long-term modulations in the solar cycle. Stochastic fluctuations in these models are due to randomness in the BMR properties (primarily due to scatter around Joy’s law, BMR emergence rates and emergence latitudes). Babcock–Leighton models are also nonlinear because at least the toroidal to poloidal field generation step includes several essential nonlinearities (tilt quenching, latitude quenching, toroidal flux loss due to magnetic buoyancy). While these nonlinearities have the tendency of stabilizing the magnetic field, they can lead to fluctuating cycles including Gnevyshev–Ohl rule and grand minima at highly supercritical regimes (large dynamo numbers) due to the inherent time delay in the dynamo models with spatially segregated source regions. Observations indicate that the solar dynamo is possibly operating in a weakly nonlinear regime (slightly above the dynamo transition) and thus cycle modulations are caused by stochastic effects.

One way to pick up the correct model out of all the possible models for long-term modulation is to carefully compare the model results with the observations. Observational results include the followings (but not limited to). (1) There is a strong hemispheric asymmetry during the second half of the Maunder Minimum (Ribes and Nesme-Ribes 1993). (2) However, the asymmetry during the normal cycle is less, appears randomly, and is smoothed out in a few cycles (the memory of the asymmetry does not remain for multiple cycles; Goel and Choudhuri 2009; McIntosh et al. 2013; Das et al. 2022). (3) The differential rotation in the whole CZ of the Sun is well-measured for about last four decades and it has only a tiny variation (surface differential is measured even for about 200 years and also shows little variation; Gilman and Howard 1984; Jha et al. 2021.)

Grand minima are possibly triggered by the stochastic fluctuations in the dynamo parameters or/and the nonlinear interaction of the Lorentz force. The recovery to the normal phase from grand minima is trivial in any model which includes the $\alpha$-effect because it can operate in the weak-field regime. Recovery through the Babcock–Leighton process is also possible because recent analyses revealed spots during Maunder minimum and the Babcock–Leighton process can produce a poloidal field with a few BMRs or smaller BMRs having nonzero tilts (including ephemeral regions). Grand maxima are probably more special events and less frequent than grand minima. Again nonlinear modulation of the flow via Lorentz force and stochastic fluctuations can trigger these events. A dual source of poloidal field generation occurring constructively or reversed generation of the poloidal field due to stochastic fluctuations can produce prominent grand maxima. Occurrences of grand minima and maxima in the dynamo models can be described by stochastic processes and these are consistent with the observations. The waiting time distributions of the grand minima and maxima in the dynamo models are also described by the memoryless stochastic processes, which however disagree with the available observations.