1. Introduction
The interaction of horizontal shear flows and magnetic fields in stably stratified layers is central to many problems in astrophysical fluid dynamics – involving, for example, planetary interiors, stellar radiative zones and accretion discs. An important example of such a flow, which has received considerable attention recently, is that of the solar tachocline (see Hughes, Rosner & Weiss Reference Hughes, Rosner and Weiss2007). The tachocline, discovered via helioseismic observations, is a thin layer in the Sun, extending downwards from the (neutrally stable) base of the convective zone to the (stably stratified) top of the radiative interior, characterised by radial velocity shear and also planetary-scale horizontal shears associated with the equator to pole differential rotation of the Sun. Most models of the solar dynamo invoke the tachocline as the site for the storage and generation of the Sun’s strong, predominantly toroidal, magnetic field.
Here we are interested in the stability of stably stratified flows with horizontal shear and an aligned field in a quite general context. It is possible to examine the stability of such flows in a continuously stratified three-dimensional setting (e.g. Miura & Pritchett Reference Miura and Pritchett1982; Cally Reference Cally2003). However, we adopt the alternative approach of considering the dynamics of a thin fluid layer under the shallow-water approximation, which is valid when the horizontal length scale of the motion is long compared with the depth of the fluid layer, as is typically the case in large-scale astrophysical flows. This leads to a set of two-dimensional partial differential equations, with no explicit dependence on the vertical coordinate, which offers a considerable mathematical simplification. Such shallow-water equations capture the fundamental dynamics of density stratification, including gravity waves, and allow the interaction of stratification with horizontal shear flows and magnetic fields to be analysed in the simplest possible setting.
Hydrodynamic shallow-water models, which date back to Laplace, are derived by considering a thin fluid layer of constant density bounded below by a rigid medium and above by a fluid of negligible inertia (e.g. Vallis Reference Vallis2006, § 3.1). The corresponding reduction for electrically conducting fluids – the shallow-water magnetohydrodynamic (SWMHD) equations of Gilman (Reference Gilman2000) – additionally requires the fluid layer to be a perfect conductor, and to be bounded above and below by perfect conductors. There are few direct astrophysical analogues for such a configuration. However, we can borrow an important idea from planetary atmospheric dynamics, where the hydrodynamic shallow-water equations are widely used to understand waves and instabilities in a continuously stratified atmospheric layer. This is justified because there is a formal mathematical analogy between the linearised equations in the two systems, provided the layer depth in the shallow-water model is taken to be a so-called equivalent depth (e.g. Gill Reference Gill1982, § 6.11), so that the shallow-water gravity wave speed (in the horizontal) matches that of (the fastest) gravity waves in a continuously stratified layer. We have this analogy in mind throughout this study.
The SWMHD equations have been widely studied in recent years. They have been shown to possess a hyperbolic as well as a Hamiltonian structure (De Sterck Reference De Sterck2001; Dellar Reference Dellar2002), and to support wave motions such as inertia–gravity waves and Alfvén waves (Schecter, Boyd & Gilman Reference Schecter, Boyd and Gilman2001; Zaqarashvili et al. Reference Zaqarashvili, Oliver, Ballester and Shergelashvili2007; Heng & Spitkovsky Reference Heng and Spitkovsky2009). As reviewed by Gilman & Cally (Reference Gilman, Cally, Hughes, Rosner and Weiss2007), they have also been used to study the linear instability of shear flows in the tachocline; these studies considered basic states that were functions only of latitude, investigating the dependence of the instabilities on the strength and spatial structure of the magnetic field and on a (reduced) gravity parameter (Gilman & Dikpati Reference Gilman and Dikpati2002; Dikpati, Gilman & Rempel Reference Dikpati, Gilman and Rempel2003).
Here we investigate the simplest general model of shear flow instabilities in SWMHD. Specifically, we consider the linear instability of a steady parallel flow and aligned magnetic field, both sheared in the horizontal cross-stream direction, in the inviscid and perfectly electrically conducting limit. We use a planar geometry, which is immediately applicable to local shear zones, and which could also apply to planetary-scale shears with appropriate care. In this first study, we consider the case with no background rotation, which will be the best model for certain astrophysical objects. Even if background rotation is dynamically important, it is nonetheless crucial to understand the non-rotating problem before rotation is added. We are thus left to investigate how the prototypical flows are modified by the combined action of magnetic fields and stratification, which, in isolation, are generally thought to be stabilising. The limits with no field (i.e. hydrodynamic shallow water) or no stratification (i.e. incompressible two-dimensional MHD) are both well studied, as is the classical case with neither magnetic field nor stratification. We are thus able to draw upon ideas and methods from a substantial literature (e.g. Drazin & Howard Reference Drazin and Howard1966; Drazin & Reid Reference Drazin and Reid1981; Vallis Reference Vallis2006).
We start, in § 2, by formulating the linear instability problem for plane-parallel basic states with the flow and field dependent on the cross-stream direction. In § 3 we derive extensions of classical growth-rate bounds, semi-circle theorems, stability criteria and parity results for modal solutions for quite general basic states. In § 4 we develop the fundamental idea that background flows belong to one of two distinct classes – shear flows and jets – possessing contrasting instabilities. This classification is first investigated analytically for the simplest shear flow (the vortex sheet, § 4.1) and the simplest jet (the rectangular jet, § 4.2). The justification for this is that analogous behaviour can be found for smooth profiles with long-wavelength disturbances; this is demonstrated in § 4.3 by extending the analysis of Drazin & Howard (Reference Drazin and Howard1962) to SWMHD. Together, these results introduce certain key ideas: the existence of wavenumber cutoffs, a tendency for stabilisation by field or stratification, and the possibility of destabilisation by field and stratification acting together. We then show that these characteristics also hold for less idealised cases, i.e. smooth flows with finite wavenumber disturbances. This is demonstrated numerically for a shear flow (the hyperbolic
$\tanh$
profile, § 5.1) and a jet (the Bickley jet, § 5.2), with a discussion of the instability mechanisms where possible. Possible applications to the solar tachocline and future extensions of the work are discussed in § 6.
2. Mathematical formulation
2.1. Governing equations
We consider a thin layer of perfectly electrically conducting fluid moving under the influence of gravity. We use a Cartesian geometry, with horizontal coordinates
$x$
and
$y$
, and an upwards pointing coordinate
$z$
. At time
$t$
, the fluid, which is taken to be inviscid and of constant density
${\it\rho}$
, has a free surface at
$z=h(x,y,t)$
and is bounded below by a rigid impermeable boundary at
$z=-H(x,y)$
.
We consider motions with a characteristic horizontal length scale
$L_{0}$
that is long compared with a characteristic layer depth
$H_{0}$
. One can then make a shallow-water reduction in which the vertical momentum balance is taken to be magnetohydrostatic, and for which the horizontal velocity
$\boldsymbol{u}$
and horizontal magnetic field
$\boldsymbol{B}$
are independent of
$z$
. When the bottom boundary is perfectly electrically conducting (and is thus a magnetic field line) and the free surface remains a field line, the magnetic shallow-water equations of Gilman (Reference Gilman2000) are obtained. These are an extension of the classical shallow-water equations of geophysical fluid dynamics.
We use these equations in non-dimensional form. We denote the characteristic horizontal velocity of the basic state by
$U_{0}$
, and the characteristic magnetic field strength by
$B_{0}$
. Non-dimensionalising
$x$
and
$y$
by
$L_{0}$
,
$t$
by the advective time scale
$L_{0}/U_{0}$
,
$H$
by
$H_{0}$
,
$h$
by
$U_{0}^{2}/g$
(where
$g$
is the acceleration due to gravity), velocity by
$U_{0}$
and magnetic field by
$B_{0}$
, the SWMHD equations are
However, (2.2) need not be considered explicitly, since, if it is satisfied at some initial time, then (2.1a – c ) guarantee that it remains satisfied for all time.
The system has two non-dimensional parameters. The Froude number
$F$
is the ratio of the characteristic horizontal velocity of the basic state to the gravity wave speed
$\sqrt{gH_{0}}$
(and is related to the reduced gravity parameter
$G$
of Gilman & Dikpati (Reference Gilman and Dikpati2002) via
$G=F^{-2}$
). The parameter
$M$
is the ratio of the Alfvén wave speed
$B_{0}/\sqrt{{\it\mu}{\it\rho}}$
to the characteristic horizontal velocity of the basic state. When
$H$
is constant and
$F\rightarrow 0$
, (2.1c
) and (2.2) become
$\boldsymbol{{\rm\nabla}}\boldsymbol{\cdot }\boldsymbol{u}=0$
and
$\boldsymbol{{\rm\nabla}}\boldsymbol{\cdot }\boldsymbol{B}=0$
, respectively, and we recover the equations for two-dimensional incompressible magnetohydrodynamics, with
$h$
playing the role of pressure. When
$M\rightarrow 0$
, (2.1b
) decouples from (2.1a
) and (2.1c
), and we recover the hydrodynamic shallow-water equations; these have a well-known correspondence with two-dimensional compressible hydrodynamics (e.g. Vallis Reference Vallis2006, § 3.1), which we exploit from time to time.
As an example of astrophysical parameter values, we estimate
$M$
and
$F$
in the solar tachocline, using data from Gough (Reference Gough, Hughes, Rosner and Weiss2007). We set
$U_{0}$
to be the equator-to-pole difference in the zonal velocity, implying
$U_{0}\approx 200~\text{m}~\text{s}^{-1}$
. This is appropriate for planetary-scale disturbances; for local shear zones, one might take
$U_{0}\approx 20~\text{m}~\text{s}^{-1}$
. There is considerable uncertainty in the strength of the magnetic field in the tachocline (Hughes et al.
Reference Hughes, Rosner and Weiss2007), although a likely range is
$10^{3}~\text{G}\lesssim B_{0}\lesssim 10^{5}~\text{G}$
. Then, taking
${\it\rho}=210~\text{kg}~\text{m}^{-3}$
, we find
$0.03\lesssim M\lesssim 3$
. To estimate
$F$
, we must choose a gravity wave speed
$\sqrt{gH_{0}}$
for the layer. One means of doing this is to take
$H_{0}$
to be the depth of the tachocline and to interpret
$g$
as a reduced gravity, accounting for the fractional density difference of the overlying fluid, as in Dikpati & Gilman (Reference Dikpati and Gilman2001). The reduced gravity may also be constructed by considering a thin active layer between two deeper passive layers (cf. Vallis Reference Vallis2006, § 3.2). However, here we pursue the analogy between shallow-water flows and those of a continuously stratified layer with buoyancy frequency
$N$
and depth
$H_{1}$
, and choose
$\sqrt{gH_{0}}$
to be the speed of the fastest gravity wave in such a layer, which is
$NH_{1}/{\rm\pi}$
(Gill Reference Gill1982, § 6.11). Taking
$H_{1}\approx 2\times 10^{7}~\text{m}$
(i.e.
$0.03R_{\odot }$
, where
$R_{\odot }$
is the solar radius) and
$N\approx 8\times 10^{-4}~\text{s}^{-1}$
, which are bulk values that might describe a mode spanning the entire tachocline, gives a gravity wave speed
$NH_{1}/{\rm\pi}=\sqrt{gH_{0}}\approx 5000~\text{m}~\text{s}^{-1}$
, corresponding to an equivalent depth
$H_{0}\approx 50~\text{km}$
(taking
$g\approx 540~\text{m}~\text{s}^{-2}$
). Again taking
$U_{0}\approx 200~\text{m}~\text{s}^{-1}$
, we thus estimate
$F\approx 0.04$
, although it is clear that
$F$
would be somewhat smaller or larger if one considered motions towards the top of the radiative zone (with stronger stratification) or towards the base of the convection zone (with weaker stratification).
2.2. The linear instability problem
Above a topography of the form
$H=H(y)$
, we consider a basic state
$h=0$
,
$\boldsymbol{u}=U(y)\boldsymbol{e}_{x}$
and
$\boldsymbol{B}=B(y)\boldsymbol{e}_{x}$
, so that the magnetic field is initially aligned with the flow. We then consider the linear evolution of perturbations in
$h$
,
$\boldsymbol{u}=(u,v)$
and
$\boldsymbol{B}=(b_{x},b_{y})$
, governed by
where
${\it\alpha}$
is the (real) wavenumber and
$c$
is the (complex) phase speed. Manipulation of (2.3), and dropping hats, leads to the single equation
where
$Q=-U^{\prime }/H$
is the background potential vorticity, and
Following Howard (Reference Howard1961), under the transformation
$Hv=(U-c)G$
, (2.5) becomes
We shall use this more compact form for the remainder of this study. In the non-magnetic shallow-water limit (
$M=0$
), (2.5) reduces to equation (3.4) of Balmforth (Reference Balmforth1999). In the two-dimensional incompressible magnetohydrodynamic limit (
$F=0$
and
$H=1$
), (2.7) reduces to equation (3.5) of Hughes & Tobias (Reference Hughes and Tobias2001).
We shall consider (2.7) in either an unbounded domain, for which
$|G|\rightarrow 0$
as
$|y|\rightarrow \infty$
, or in a bounded domain with rigid side walls at
$y=\pm L$
, where
$G=0$
and hence
$b_{y}=0$
via (2.3d
). Either way, for given real
${\it\alpha}$
, (2.7) is then an eigenvalue problem for the unknown phase speed
$c=c_{r}+\text{i}c_{i}$
. We will focus on instabilities, i.e.
$c_{i}\neq 0$
, in which case (2.7) has no singularities for real values of
$y$
. Since the transformation
${\it\alpha}\rightarrow -{\it\alpha}$
leaves (2.7) unchanged, we may take
${\it\alpha}\geqslant 0$
without loss of generality. Instability then occurs if
$c_{i}>0$
, with growth rate
${\it\alpha}c_{i}$
.
3. General theorems
In this section we derive three results that hold for general shear flows
$U(y)$
: two provide bounds on the growth rate of any instability, whereas the third concerns implications of the parity of the basic state flow.
3.1. Growth-rate bound
A bound on the instability growth rate may be obtained by calculating the rate of change of the total disturbance energy using the combination
where
$^{\ast }$
denotes complex conjugate. On adopting the form (2.4) for the perturbations, the real part of this expression gives (on dropping hats)
$$\begin{eqnarray}\displaystyle & & \displaystyle {\it\alpha}c_{i}(H(|u|^{2}+|v|^{2}+M^{2}|b_{x}|^{2}+M^{2}|b_{y}|^{2})+F^{2}|h|^{2})\nonumber\\ \displaystyle & & \displaystyle \quad =-\text{Re}(HU^{\prime }(vu^{\ast }-M^{2}b_{x}^{\ast }b_{y})+M^{2}HB^{\prime }(vb_{x}^{\ast }-u^{\ast }b_{y}))-\text{Re}\,\left(\frac{\text{d}}{\text{d}y}(Hvh^{\ast })\right).\end{eqnarray}$$
On integrating over the
$y$
domain, employing the boundary condition on
$v$
and manipulating the remaining terms on the right-hand side using
$\pm 2\text{Re}\,(pq^{\ast })\leqslant |p|^{2}+|q|^{2}$
, we obtain the following bound on the growth rate:
In the absence of a magnetic field, this reduces to the well-known bound in hydrodynamics (Høiland Reference Høiland1953; Howard Reference Howard1961).
3.2. Semi-circle theorems
In a classic paper, Howard (Reference Howard1961) proved that for incompressible hydrodynamic parallel shear flows, the wave speed
$c$
of any unstable mode must lie within a semi-circle in the complex plane determined by properties of the basic state flow. Subsequently, semi-circle theorems have been derived for several other hydrodynamical and hydromagnetic systems (e.g. Collings & Grimshaw Reference Collings and Grimshaw1980; Hayashi & Young Reference Hayashi and Young1987; Shivamoggi & Debnath Reference Shivamoggi and Debnath1987; Hughes & Tobias Reference Hughes and Tobias2001). In a similar manner, a semi-circle theorem may be derived for the SWMHD system.
Multiplying (2.7) by
$G^{\ast }$
, integrating over
$y$
and using the boundary condition on
$v$
(and hence
$G$
) gives the relation
where
$S$
and
$K$
are functions of
$c$
given in (2.6). The imaginary part of (3.4) gives
Equation (3.5) immediately yields Rayleigh’s result that for unstable modes (i.e.
$c_{i}>0$
),
$c_{r}$
lies in the range of
$U$
(i.e.
$U_{min}\leqslant c_{r}\leqslant U_{max}$
, where the subscripts ‘min’ and ‘max’ refer to the minimum and maximum values across the domain).
On using (3.5), the real part of (3.4) gives
which implies that
This gives the first semi-circle bound: the complex wave speed
$c$
of an unstable eigenfunction must lie within the region defined by
The second semi-circle bound is obtained, in the standard manner, from the inequality
$0\geqslant \int (U-U_{max})(U-U_{min}){\it\chi}\,\text{d}y$
. Substituting from (3.5) and deriving an inequality from (3.6) leads to the expression
which gives the second semi-circle bound: the speed
$c$
of an unstable eigenfunction must lie within the region defined by
Thus, taking these results together, the eigenvalue
$c$
of an unstable mode must lie within the intersection of the two semi-circles defined by (3.8) and (3.10). In the absence of a magnetic field, semi-circle (3.10) lies wholly within semi-circle (3.8), and we recover the well-known result of Howard (Reference Howard1961). However, as observed by Hughes & Tobias (Reference Hughes and Tobias2001), who considered the stability of aligned fields and flows in incompressible MHD, for non-zero magnetic field there is the possibility of the two semi-circles overlapping, being disjoint or indeed ceasing to exist; thus, in addition to giving eigenvalue bounds for unstable modes, these results also provide sufficient conditions for stability. From (3.8) and (3.10) it therefore follows that the basic state is linearly stable if any one of the following three conditions is satisfied:
A drawback of the growth-rate bound (3.3) and the semi-circle bounds (3.11)–(3.13) is that they do not contain the Froude number
$F$
. Although it is possible to introduce
$F$
into the semi-circle bounds using similar manipulations to those employed by Pedlosky (Reference Pedlosky1964), this does not sharpen the bound (Mak Reference Mak2013), and we thus omit it.
3.3. Consequences of basic state parity
For the hydrodynamic case, it can be shown that symmetries of the basic state lead to symmetries in the stability problem (Howard Reference Howard1963). These results may be generalised to SWMHD if we make the further assumptions that
$B^{2}(y)$
and
$H(y)$
are even functions about
$y=0$
.
We first consider the case when
$U(y)$
is odd about
$y=0$
. Equation (2.7) is unchanged under
$c\rightarrow -c$
and
$G(y)\rightarrow G(-y)$
. Since the equation is also unchanged under
$c\rightarrow c^{\ast }$
and
$G\rightarrow G^{\ast }$
, it follows that an eigenfunction with eigenvalue
$c=c_{r}+c_{i}$
must be accompanied by eigenfunctions with
$c=\pm c_{r}\pm \text{i}c_{i}$
. Thus unstable solutions either have
$c_{r}=0$
or are a pair of counter-propagating waves with the same phase speed. As argued by Howard (Reference Howard1963), the symmetry in the basic state implies that there is no preferred direction for wave propagation, consistent with the form of the eigenvalues.
Now consider the case when
$U(y)$
is even about
$y=0$
. Then
are also eigenfunctions of (2.7). Following Drazin & Howard (Reference Drazin and Howard1966), if we now take
$G_{o}$
multiplied by (2.7) with
$G=G_{e}$
and subtract this from
$G_{e}$
multiplied by (2.7) with
$G=G_{o}$
, integrating over
$-L\leqslant y\leqslant L$
gives
owing to the imposed boundary conditions on the eigenfunction. The vanishing of the Wronskian
$W$
implies that the functions
$G_{e}$
and
$G_{o}$
are linearly dependent throughout the domain, which is possible only if one of them is identically zero. Thus an unstable eigenfunction corresponding to a particular eigenvalue is either an even or odd function about
$y=0$
.
4. Shear flows and jets: instability to long-wavelength disturbances
In classical hydrodynamic stability (i.e.
$F=M=0$
), there are universal scalings for the instability of an arbitrary shear flow in the long-wavelength limit
${\it\alpha}\ll 1$
(Drazin & Howard Reference Drazin and Howard1962, Reference Drazin and Howard1966). In this limit, any flow can be classified as either a shear layer (with a complex phase speed of order unity) or a jet (with a complex phase speed of order
${\it\alpha}^{1/2}$
). Long-wavelength theory therefore naturally identifies two classes of flows, each with fundamentally distinct modes of instability. Here we extend this analysis to SWMHD, obtaining explicit analytical expressions for the phase speed and growth rate for both shear layers and jets, which may be used to identify various stabilisation and destabilisation effects in terms of
$F$
and
$M$
. It turns out that equivalent expressions can be obtained by considering the simpler problems of the instability of the vortex sheet and the top-hat jet. We therefore first analyse these two piecewise constant flows in §§ 4.1 and 4.2, before extending the analysis of Drazin & Howard (Reference Drazin and Howard1962) to arbitrary smooth flows in § 4.3. Throughout this section, for simplicity, we take
$H=1$
(no topography) and
$B=1$
(a uniform magnetic field); we also impose
4.1. Vortex sheet
The vortex sheet velocity profile is given by
Then, for
$y\neq 0$
, (2.7) becomes
$G^{\prime \prime }-{\it\alpha}^{2}K^{2}G=0$
. Since
$U(y)$
is discontinuous at
$y=0$
, the eigenfunction
$G$
must satisfy two jump conditions at
$y=0$
. In the usual way, the (linearised) kinematic boundary condition implies
The pressure (or free-surface displacement) is also continuous at
$y=0$
. The corresponding condition on
$G$
is most easily derived by integrating (2.7) across
$y=0$
, yielding
Using (4.1) and (4.3a ), we thus find
where
The second jump condition (4.3b
) then implies the eigenvalue relation for
$c$
:
Note that
$c$
is independent of the wavenumber
${\it\alpha}$
, so any unstable mode with
$c_{i}>0$
has an unbounded growth rate as
${\it\alpha}\rightarrow \infty$
. This is an artefact of considering ideal fluids; viscosity will preferentially suppress small scales and remove this unphysical behaviour.
There are several special cases. When
$F=M=0$
, we recover the classical Kelvin–Helmholtz instability with
$c=\pm \text{i}$
. When
$F=0$
but
$M\neq 0$
, (4.6) reduces to the incompressible MHD case of Michael (Reference Michael1955), with
$c^{2}=-(1-M^{2})$
; thus, the Kelvin–Helmholtz instability is stabilised when
$M\geqslant 1$
, since the disturbance has to do work to bend the field lines. When
$M=0$
but
$F\neq 0$
, (4.6) gives the classical hydrodynamic shallow-water dispersion relation, which is analogous to that of two-dimensional compressible hydrodynamics. The Kelvin–Helmholtz instability is stabilised when
$F\geqslant \sqrt{2}$
(Miles Reference Miles1958; Bazdenkov & Pogutse Reference Bazdenkov and Pogutse1983), since the disturbance has to do work to move the free surface against gravity. Thus, increasing
$F$
or
$M$
in the absence of the other is stabilising.
In the general case where
$F$
and
$M$
are both non-zero, (4.6) can be rearranged and squared to yield a quartic equation for
$c$
:
Here we have ignored the degenerate case with
$c=0$
, which is a solution of (4.6) when
$M=1$
. By comparing solutions of (4.7) with those of (4.6) found using a Newton iteration method, we find that only two roots of (4.7) also satisfy (4.6): these are
$c=\pm c_{v}$
, where
A contour plot of
$\text{Im}\,(c_{v})$
is shown in figure 1. From (4.8), there is instability only if
Although increasing
$F$
is always stabilising at fixed
$M$
, the critical value of
$F$
above which the flow is stable increases as
$M$
increases towards 1. Thus, although magnetic field and free-surface effects are stabilising in isolation, together they can lead to instabilities at arbitrarily large values of
$F$
, provided
Using an asymptotic analysis, it is possible to investigate these instabilities further at large
$F$
and with
$M$
just smaller than unity. Rewriting (4.8) in terms of
$1-M^{2}$
and expanding for
$|1-M^{2}|\ll 1$
, we obtain
where terms of
$O((1-M^{2})^{3})$
have been neglected. When
$F=O(1)$
, the first term on the right-hand side of (4.11) dominates. However, in the regime of interest (4.10), with
$F^{2}\sim (1-M^{2})^{-1}\gg 1$
, the two terms on the right-hand side of (4.11) have the same order of magnitude, and instead we obtain
This simple formula is consistent with both stability boundaries in (4.9), and, as shown in figure 2, closely predicts
$c_{i}$
in this weak-instability regime, even when
$F$
is of order unity. Using (4.12), it is straightforward to show that
$\text{Im}\,(c_{v})$
is maximised when
$M^{2}=1-1/F^{2}$
, with
$c_{v}\sim \text{i}/(2F^{2})$
, so that the growth rate of the most unstable mode decays like
$F^{-2}$
in this regime.
4.2. Rectangular jet
We now consider the top-hat velocity profile
$$\begin{eqnarray}G=\left\{\begin{array}{@{}cl@{}}A_{+}\exp (-{\it\alpha}K_{0}(y-1)), & y>+1,\\ A_{e}\cosh ({\it\alpha}K_{1}y)+A_{o}\sinh ({\it\alpha}K_{1}y), & |y|<1,\\ A_{-}\exp (+{\it\alpha}K_{0}(y+1)), & y<-1,\end{array}\right.\end{eqnarray}$$
for some
$A_{+}$
,
$A_{-}$
,
$A_{e}$
and
$A_{o}$
, where
For the odd mode, we set
$A_{e}=0$
,
$A_{+}=-A_{-}$
and write
$c=c_{o}$
. Then the jump conditions and (4.14) give
In contrast to the vortex sheet dispersion relation (4.6), here
$c$
depends upon
${\it\alpha}$
.
The case
$F=0$
may be solved exactly. Then
$K_{0,1}=1$
, so (4.16) and (4.17) give
where
$T=\tanh {\it\alpha}$
. Gedzelman (Reference Gedzelman1973) gave formulae for these two cases (his (3.4) and (3.5)), but both are missing factors of 2; the case
$M=0$
was given in Rayleigh (Reference Rayleigh1878). Both modes are stable for all
${\it\alpha}$
when
$M\geqslant 1/2$
; otherwise, both modes are unstable (with the same growth rate) provided that
Thus, the magnetic field introduces a long-wave cutoff.
For non-zero
$F$
and
$M$
, solutions to (4.16) and (4.17) exhibit a wide variety of behaviour, as described in Mak (Reference Mak2013). Here we focus on the long-wave regime (
${\it\alpha}\ll 1$
), in which solutions exhibit fundamentally jet-like characteristics.
We consider first the even mode. Assuming that
$K_{1}$
remains bounded (which may be confirmed a posteriori),
$\tanh ({\it\alpha}K_{1})\approx {\it\alpha}K_{1}$
when
${\it\alpha}\ll 1$
, so that (4.16) becomes
Suppose that
$F^{2}=O(1)$
. If
$M=O(1)$
, then
$c_{e}=\pm M$
at leading order, and the next correction in
${\it\alpha}$
is also real. However, if
$M^{2}\sim {\it\alpha}\ll 1$
, then at leading order, (4.20) gives
which is consistent with the small
$M$
and
${\it\alpha}$
limits of (4.18). Thus a weak magnetic field reduces the growth rate of the hydrodynamic instability and eventually suppresses it, with instability when
A dependence on
$F$
appears only at the next order in
${\it\alpha}$
.
To capture the influence of
$F$
, the analysis may be extended to larger values of
$F$
with
$F^{2}\sim {\it\alpha}^{-1}\gg 1$
. In this case, the square root in (4.20) also enters the balance at leading order, and we obtain
which formally reduces to (4.21) when
$F$
is of order unity. Again there is a long-wave cutoff due to the magnetic field, with instability only when
Comparing (4.22) with (4.24) shows the destabilising influence of increasing
$F$
. Indeed, in contrast to the result (4.9) for the vortex sheet, (4.23) does not predict stabilisation for large
$F$
when
${\it\alpha}\ll 1$
.
Now consider the odd mode with
${\it\alpha}\ll 1$
. The dispersion relation (4.17) reduces to
Suppose that
$F^{2}=O(1)$
. If
$M=O(1)$
, then
$c_{o}$
is real at the first two orders in
${\it\alpha}$
. However, if
$M^{2}\sim {\it\alpha}\ll 1$
, then at leading order, (4.25) gives
When
$F<1$
, again there is a cutoff due to the magnetic field, with instability only when
These cutoffs show a strong destabilisation as
$F\rightarrow 1$
. When
$F$
is close to
$1$
, the asymptotic ordering leading to (4.26) breaks down, and one must instead seek solutions with
$c=1+O({\it\alpha}^{2/5})$
. When
$F>1$
, expression (4.26) shows that
$c_{o}$
always has a positive imaginary part, so that there is no cutoff at small
$M$
(when
${\it\alpha}\ll 1$
). So again, in contrast to the vortex sheet, there is no stabilisation at large
$F$
.
4.3. Smooth profiles
We now consider smooth profiles of the basic state velocity, again with uniform field and no topography. For long-wave disturbances, the leading-order instability behaviour is determined by
$U(y)$
as
$|y|\rightarrow \infty$
, with higher-order corrections determined by the flow at finite
$y$
. Assuming that
$U_{\pm }=U(\pm \infty )$
are well defined, the analysis reveals two distinct balances: a shear layer (if
$U_{+}\neq U_{-}$
) or a jet (if
$U_{+}=U_{-}$
), which are analogous to the piecewise constant profiles studied in §§ 4.1 and 4.2. On choosing an appropriate frame of reference and suitable normalisation for the basic flow, any velocity profile may thus be designated as either a shear layer with
$U_{\pm }=\pm 1$
, or as a jet with
$U_{\pm }=0$
.
We consider the governing equation (2.7), written as
with
$S^{2}$
and
$K^{2}$
depending on
$y$
and
$c$
as defined by (2.6). We assume that
$U^{\prime }$
(and so
$(Z^{2})^{\prime }$
) decays sufficiently rapidly as
$|y|\rightarrow \infty$
. Adopting the same notation as Drazin & Howard (Reference Drazin and Howard1962), we consider solutions of the form
where
$K_{\pm }^{2}=1-F^{2}S_{\pm }^{2}=1-F^{2}((U_{\pm }-c)^{2}-M^{2})$
. The perturbations must decay as
$|y|\rightarrow \infty$
; hence
$\text{Re}\,(K_{\pm })>0$
. We consider expansions of the form
with
${\it\chi}_{0},{\it\theta}_{0}\rightarrow \text{const.}\;(\neq 0)$
and
${\it\chi}_{n},{\it\theta}_{n}\rightarrow 0$
as
$|y|\rightarrow \infty$
$(n\geqslant 1)$
. It turns out to be most convenient to fix
${\it\chi}_{0}(\infty )={\it\theta}_{0}(-\infty )=1$
, and then to accommodate the necessary degree of freedom in the matching conditions for
$G$
at
$y=0$
, namely
$G_{+}(0)={\it\Gamma}G_{-}(0)$
and
$G_{+}^{\prime }(0)={\it\Gamma}G_{-}^{\prime }(0)$
for some constant
${\it\Gamma}$
. Consistency thus implies
Without loss of generality, we shall focus on the equations for
${\it\chi}$
; those for
${\it\theta}$
follow in a similar fashion. On substituting (4.29) (with (4.30)) into (4.28), equating the coefficients at each order of
${\it\alpha}$
gives
$$\begin{eqnarray}\left.\begin{array}{@{}c@{}}{\it\chi}_{1}=\displaystyle \int _{\infty }^{y}\left(1-\frac{Z_{+}^{2}}{Z^{2}(y_{1})}\right)\text{d}y_{1},\\ {\it\chi}_{2}=\displaystyle \int _{\infty }^{y}\left(\frac{1}{Z^{2}(y_{1})}\int _{\infty }^{y_{1}}(S^{2}(y_{2})-S_{+}^{2})\,\text{d}y_{2}+K_{+}^{2}\int _{\infty }^{y_{1}}\left(1-\frac{Z_{+}^{2}}{Z^{2}(y_{2})}\right)\text{d}y_{2}\right)\text{d}y_{1}.\end{array}\right\}\end{eqnarray}$$
The matching condition (4.31) then leads to the result
$$\begin{eqnarray}\displaystyle 0 & = & \displaystyle \left(\frac{S_{+}^{2}}{K_{+}}+\frac{S_{-}^{2}}{K_{-}}\right)+{\it\alpha}\left(\int _{0}^{\infty }(S^{2}-S_{+}^{2})\,\text{d}y+\int _{-\infty }^{0}(S^{2}-S_{-}^{2})\,\text{d}y\right.\nonumber\\ \displaystyle & & \displaystyle \left.-\,\frac{S_{+}^{2}}{K_{+}K_{-}}\int _{-\infty }^{0}\left(1-\frac{S_{-}^{2}}{S^{2}}\right)\text{d}y-\frac{S_{-}^{2}}{K_{+}K_{-}}\int _{0}^{\infty }\left(1-\frac{S_{+}^{2}}{S^{2}}\right)\text{d}y\right)+O({\it\alpha}^{2}).\end{eqnarray}$$
On expressing the eigenvalue
$c$
as
$c=c^{(0)}+{\it\alpha}c^{(1)}+{\it\alpha}^{2}c^{(2)}+\cdots \,$
, (4.34) then determines the successive
$c^{(j)}$
.
Although we have focused on the case of a uniform magnetic field, it is possible to include a non-uniform field, subject to imposing conditions analogous to those for
$U(y)$
. With underlying topography, other assumptions on
$H(y)$
are required, as described by Collings & Grimshaw (Reference Collings and Grimshaw1980).
4.3.1. Shear layers
For a shear layer,
$U_{\pm }=\pm 1$
, and the leading-order term of expression (4.34) gives
This is exactly the eigenvalue equation of the vortex sheet (4.6); hence, for any shear layer,
$c\rightarrow c_{v}$
, as defined in (4.8), as
${\it\alpha}\rightarrow 0$
. This is not surprising; sufficiently long waves see the shear layer as a discontinuity in the flow.
When
$F>F_{c}$
, as defined in (4.9),
$c_{v}$
is real. However there could be an instability at higher order. Following Blumen, Drazin & Billings (Reference Blumen, Drazin and Billings1975), we may calculate
$c^{(1)}$
, although this requires
$U(y)$
to be specified. Such a calculation is performed in § 5.1.3 for
$U(y)=\tanh y$
, revealing a secondary instability for large values of
$F$
.
4.3.2. Jets
For a jet,
$U_{\pm }=0$
; at leading order the flow is thus seen to be uniform and hence there is no instability. It is therefore essential to retain the
$O({\it\alpha})$
terms in (4.34), which becomes
where
$S_{0}^{2}=(0-c)^{2}-M^{2}$
,
$K_{0}^{2}=1-F^{2}S_{0}^{2}$
. Here, for a fixed value of
$F$
, we need to consider different regimes for
$M$
.
For
$F^{2}=O(1)$
, if
$M^{2}=O(1)$
then
$c^{(0)}$
and
$c^{(1)}$
are real. To find an instability we need to consider the regime
$M^{2}\sim {\it\alpha}$
, which implies
$c^{(0)}=0$
. At the next order, we choose to balance the first two terms on the right-hand side of (4.36), assuming that the second integral is negligible; this is confirmed by the analysis of appendix A. Defining
$E=\int _{-\infty }^{+\infty }U^{2}/2\,\text{d}y$
, where
$U$
is assumed to decay sufficiently rapidly that
$E$
is finite, we obtain
The corresponding result for compressible hydrodynamics was derived by Gill & Drazin (Reference Gill and Drazin1965), and for incompressible MHD by Gedzelman (Reference Gedzelman1973).
For large
$F$
, the regime of interest is
$M^{2}\sim {\it\alpha}$
,
$F^{2}\sim {\it\alpha}^{-1}$
. Considering the same balance as above gives
This result reduces to (4.37) in the limit of small
$F$
. The presence of a magnetic field introduces a long-wave cutoff.
Note that expressions (4.37) and (4.38) take the same form as (4.21) and (4.23), found for the even mode of the rectangular jet. Indeed, if we take
$U(y)$
defined by (4.13), then
$E=1$
and the correspondence is exact. The discussion of the rectangular jet concerning long-wave cutoffs and destabilisation at large
$F$
thus applies to arbitrary smooth jets for
${\it\alpha}\ll 1$
.
Interestingly, the dispersion relation (4.26) for the odd mode of the rectangular jet is not captured by the long-wave analysis. The reason is that, for these modes,
$c_{r}$
remains
$O(1)$
as
${\it\alpha}\rightarrow 0$
(as in (4.26)). This leads to a problem in (4.28) since, when
$M^{2}\sim {\it\alpha}$
,
$Z^{2}$
becomes small when
$U\approx c_{r}$
, and the standard asymptotic procedure leading to (4.32) breaks down. In the hydrodynamic case it is found that
$c_{i}\sim {\it\alpha}^{2/3}$
(Drazin & Howard Reference Drazin and Howard1962); our numerical results suggest the same scaling when
$F$
and
$M$
are non-zero.
5. Instability of smooth profiles at finite wavenumber
Having identified the existence of two classes of flows for instability at small wavenumbers, we now consider how these instabilities extend to finite wavenumber. We consider one smooth shear layer (the hyperbolic tangent profile) and one smooth jet (the Bickley jet). Linear instability calculations involving these two profiles are well documented in a wide variety of contexts (e.g. Lipps Reference Lipps1962; Howard Reference Howard1963; Michalke Reference Michalke1964; Sutherland & Peltier Reference Sutherland and Peltier1992; Hughes & Tobias Reference Hughes and Tobias2001) and these provide a comparison and check on our results. We again restrict attention (for simplicity) to the case of a uniform background magnetic field,
$B(y)\equiv 1$
, and with no underlying topography,
$H(y)\equiv 1$
. Even so, the eigenvalue problem (2.7) must be solved numerically; we employ a shooting method with a continuation approach, as detailed in appendix B.
5.1. Hyperbolic-tangent shear layer
In this subsection, we consider the basic state velocity defined by
5.1.1. General instability characteristics
From inequality (3.3) we know that the growth rates
${\it\alpha}c_{i}$
are bounded above by
$|U^{\prime }|_{max}/2=1/2$
; furthermore, from the stability criteria (3.11) or (3.12), this profile is stable when
$M\geqslant 1$
. When
$F=M=0$
, instability exists only when
$0<{\it\alpha}<1$
, with a neutral mode at
${\it\alpha}=1$
(Drazin & Reid Reference Drazin and Reid1981, § 31.10). In this case, there is a single primary mode of instability, which may be classified as an inflection-point instability and attributed to interacting waves supported by the background shear (see, for example, the review by Carpenter et al.
Reference Carpenter, Tedford, Heifetz and Lawrence2012). When
$M=0$
, but
$F\neq 0$
, there is a secondary mode of instability, first found by Blumen et al. (Reference Blumen, Drazin and Billings1975) for two-dimensional compressible hydrodynamics. This has a smaller growth rate and a less pronounced spatial decay than the primary mode. Further, whilst the primary mode has
$c_{r}=0$
, there are two branches for the secondary mode with equal and opposite (non-zero) phase speeds, consistent with the parity results in § 3.3, which ensure that
$c=\pm c_{r}+\text{i}c_{i}$
for unstable modes. The secondary mode can be attributed to interacting gravity waves (e.g. Satomura Reference Satomura1981; Hayashi & Young Reference Hayashi and Young1987; Takehiro & Hayashi Reference Takehiro and Hayashi1992; Balmforth Reference Balmforth1999) and, indeed, can occur for linear shear flows, explicitly filtering out the possibility of waves due to a background vorticity gradient. This is consistent with the theorem of Ripa (Reference Ripa1983), which states that for instability (with
$M=0$
), either the associated potential vorticity profile possesses an inflection point, or
$F>1$
, both of which encourage instability through interacting waves.
Both modes have been found in our SWMHD system. Figure 3 shows contours of
$c_{i}$
over
$(M,F)$
space at selected values of
${\it\alpha}$
for the profile (5.1), distinguishing between the primary mode (with
$c_{r}=0$
, shown as solid contours) and the secondary mode (with
$c_{r}\neq 0$
, shown as dashed contours). In figure 3(a), at
${\it\alpha}=0.7$
, and figure 3(b), at
${\it\alpha}=0.44$
(which is the most unstable mode when
$F=M=0$
(Michalke Reference Michalke1964)), only the primary mode exists. In figure 3(c), at
${\it\alpha}=0.2$
, and figure 3(d), at
${\it\alpha}=0.01$
, both modes exist, although in different parts of
$(M,F)$
space. We observe that figure 3(d) and figure 1 are remarkably similar, confirming the results from the long-wave asymptotic analysis in § 4.
Contours of
$c_{i}$
over
$(M,F)$
space for
$U(y)=\tanh y$
; (a)
${\it\alpha}=0.70$
, (b)
${\it\alpha}=0.44$
, (c)
${\it\alpha}=0.20$
, (d)
${\it\alpha}=0.01$
. The primary modes (here classified by
$|c_{r}|<10^{-3}$
) are contoured as solid lines. The secondary modes are contoured as dashed lines.
The growth rate
${\it\alpha}c_{i}$
is shown in figure 4 as a function of
${\it\alpha}$
; secondary modes generally have weaker growth rates than the primary modes, consistent with the results of Blumen et al. (Reference Blumen, Drazin and Billings1975). As we shall see in § 5.1.3, where we extend the long-wave asymptotic analysis beyond the leading order, the relation between the two types of unstable modes can be explored in some detail in the long-wavelength limit.
Growth rate versus
${\it\alpha}$
at selected parameter values for
$U(y)=\tanh y$
. The upper four curves are primary modes, the lower two are secondary modes.
5.1.2. Instability mechanism of the primary mode
As mentioned above, inflection-point instabilities can be attributed to interacting waves supported by the vorticity gradient of the background shear. These waves are normally called Rossby waves, although they should not be confused with Rossby waves that could arise from a background vorticity gradient due to planetary rotation. The constructive interference of a pair of counter-propagating Rossby waves has been proposed as the mechanism leading to instability of shear flows in a variety of settings (e.g. Bretherton Reference Bretherton1966; Hoskins, McIntyre & Robertson Reference Hoskins, McIntyre and Robertson1985; Caulfield Reference Caulfield1994; Harnik & Heifetz Reference Harnik and Heifetz2007; Heifetz et al. Reference Heifetz, Mak, Nycander and Umurhan2015). For the SWMHD system, it is therefore natural to enquire how this underlying mechanism is modified by magnetic and shallow-water effects.
Let us first consider the case of
$F=M=0$
. The background vorticity profile, shown in figure 5, supports two Rossby waves, propagating in the negative (positive)
$x$
-direction on the positive (negative) vorticity gradient in
$y>0$
(
$y<0$
). Viewed individually, the Rossby waves are neutral and propagate against the mean flow. If, however, they become phase locked, they can interfere constructively, leading to mutual amplification and hence instability. This is shown schematically in figure 5, where the two Rossby waves are represented as perturbed vorticity contours (or equivalently perturbed material contours) for
$y>0$
and
$y<0$
. The resulting positive and negative vorticity anomalies are also shown. In this configuration, the transverse flow induced by each wave acts to amplify the existing transverse material displacement of the other. There is thus a mutual amplification and instability. This is also consistent with numerical solutions for instabilities of the flow (5.1) when
$F=M=0$
; figure 6(a) shows the vorticity perturbation of the most unstable mode (with
${\it\alpha}=0.44$
), in agreement with figure 5.
Modified counter-propagating Rossby wave mechanism in schematic form. Shown are two displaced material contours. The associated vorticity anomalies when
$F=M=0$
are shown by the closed solid curves; the effect of these on the other contours, leading to instability, is shown by the longer arrows. The closed dashed curves represent the additional vorticity anomalies when
$F$
and
$M$
are non-zero. The stabilising effect of these, which opposes the driving of the instability, is shown by the shorter arrows.
We now quantify how free-surface and magnetic effects modify this mechanism for SWMHD. We use the SWMHD vorticity equation, which is given by
where
${\it\omega}$
and
$j$
are the
$z$
-components of the vorticity and electric current. Linearising about the basic state
$\boldsymbol{U}_{0}=U(y)\boldsymbol{e}_{x}$
,
$\boldsymbol{B}_{0}=\boldsymbol{e}_{x}$
, taking modal solutions of the form (2.4) and noting that
$v=(\partial /\partial t+U\partial /\partial x){\it\eta}$
, where
${\it\eta}$
is the cross-stream displacement, we obtain the vorticity budget
where
${\it\Omega}=-U^{\prime }$
is the basic state vorticity. The three contributions to
${\it\omega}$
arise from the advection of the background vorticity, vortex stretching in shallow water, and magnetic tension.
Inspection of figure 3 shows that the instability is most vigorous when
$F=M=0$
; we therefore expect that the vorticity anomalies from the magnetic and shallow-water effects will be stabilising. The vorticity
${\it\omega}$
and the decomposition (5.3) are shown in figure 7 for a mode at
$F=0.5$
,
$M=0.25$
. Even though this eigenfunction results from a calculation with non-zero
$F$
and
$M$
, the
$-{\it\eta}{\it\Omega}^{\prime }$
contribution has the same structure as that of figure 6(a). The extra contributions from non-zero
$F$
and
$M$
are shown in figure 7(c,d); both of these terms are maximised at
$y=0$
, where they are approximately in phase. Thus, at the simplest level, they lead to the vorticity anomalies shown by the dashed circles in figure 5. The transverse flow induced by these vorticity anomalies counteracts the mutually amplifying transverse flow of the Rossby waves, and is thus stabilising; a similar observation has been made for hydrodynamic stratified flows by Rabinovich et al. (Reference Rabinovich, Umurhan, Harnik, Lott and Heifetz2011) and for incompressible MHD by Heifetz et al. (Reference Heifetz, Mak, Nycander and Umurhan2015).
Eigenfunctions of (a) vorticity and (b) height for the most unstable modes for
$U(y)=\tanh y$
at
$F=0$
,
$M=0$
.
A generic vorticity budget breakdown for the case where neither
$F$
nor
$M$
is zero; shown here is
$F=0.5$
and
$M=0.25$
, with (a)
${\it\omega}$
, (b)
$-{\it\eta}{\it\Omega}^{\prime }$
, (c)
$F^{2}h{\it\Omega}$
, (d)
$M^{2}j/(U-c)$
. Notice that the vorticity contribution from the magnetic term is substantially larger than that from the shallow-water term.
As a further verification of these ideas, we have also adopted a perturbative approach to the analysis of expression (5.3), approximating the shallow-water and magnetic contributions using the eigenfunction for
$F=M=0$
. It can be seen that calculating
$F^{2}h{\it\Omega}$
using
$h$
in figure 6(b) is consistent with the vorticity anomaly from the full linear equations (figure 7
c). To obtain an estimate of the magnetic contribution, it is necessary to calculate
$j$
using the governing equations (2.3) with the velocity obtained when
$F=M=0$
. This is slightly more involved than for
$h$
, but can be shown to provide a vorticity anomaly consistent with figure 7(d) (see Mak Reference Mak2013).
5.1.3. Long-wave asymptotics for the secondary mode
We can gain some understanding of the secondary mode, and its connection to the primary mode, by considering the limit
${\it\alpha}\ll 1$
. The mathematical framework was developed in § 4.3.1, where an expression
$c_{v}$
was obtained for the leading-order phase speed
$c^{(0)}$
, which, as noted, is real for
$F>F_{c}$
. In this case, following Blumen et al. (Reference Blumen, Drazin and Billings1975), we use expression (4.34) to calculate
$c^{(1)}$
, the next term in the expansion of the eigenvalue. For the particular flow (5.1), and after considerable algebra (Mak Reference Mak2013), we obtain
$$\begin{eqnarray}\displaystyle c^{(1)} & = & \displaystyle \text{i}\frac{\sqrt{F^{2}((1-c_{v})^{2}-M^{2})-1}}{4c_{v}^{2}}\frac{(1+c_{v})^{2}-M^{2}}{\sqrt{1+4F^{2}+4F^{4}M^{2}}}\nonumber\\ \displaystyle & & \displaystyle \times \left(1+\frac{c_{v}}{2}\left(\log \left(\frac{(1+c_{v})^{2}-M^{2}}{(1-c_{v})^{2}-M^{2}}\right)-{\it\delta}_{1}-{\it\delta}_{2}\right)\right.\nonumber\\ \displaystyle & & \displaystyle \left.+\,\frac{1-M^{2}-c_{v}^{2}}{4M}\left[\log \left(\frac{(1+M)^{2}-c_{v}^{2}}{(1-M)^{2}-c_{v}^{2}}\right)+{\it\delta}_{1}-{\it\delta}_{2}\right]\right),\end{eqnarray}$$
where
${\it\delta}_{1}={\it\delta}_{2}=0$
when
$F<F_{c}$
, and where
when
$F>F_{c}$
for the root with
$c^{(0)}=c_{v}>0$
(there are corresponding expressions for the other root with
$c^{(0)}=-c_{v}<0$
). It may be shown that for
$M\rightarrow 0$
, the expression inside the square brackets is equal to
$4M(1-c_{v})^{-2}+O(M^{2})$
, and that (5.4) is equivalent to equation (21) in Blumen et al. (Reference Blumen, Drazin and Billings1975). For sufficiently large
$F$
,
$\text{Im}\,(c^{(1)})>0$
; this is the weak secondary instability. By expanding
$c_{v}$
up to powers of
$F^{-4}$
, it may be shown from (5.4) that
$\text{Im}\,(c^{(1)})\downarrow 0$
as
$F\rightarrow \infty$
, so there is no cutoff at finite
$F$
.
The analysis leading to expression (5.4) is valid only when
$c_{v}$
is not small, i.e. for
$F^{2}$
not close to
$F_{c}^{2}=2(1-M^{2})^{-1}$
or
$M$
not close to
$1$
. When
$F^{2}-F_{c}^{2}$
is small,
$O({\it\alpha}^{2/3})$
to be precise, we have
$c_{v}\sim {\it\alpha}c^{(1)}\sim {\it\alpha}^{1/3}$
. Rescaling and choosing the appropriate branch so that
$\text{Re}\,(\sqrt{\cdots })>0$
gives
$c^{(1)}$
as the solution of the cubic equation
for
${\it\alpha}^{2/3}\sim 2(1-M^{2})^{-1}-F^{2}\rightarrow 0$
. When
$M=0$
, (5.6) reduces to equation (23) in Blumen et al. (Reference Blumen, Drazin and Billings1975). There are two admissible roots with positive imaginary parts. There is a transition to non-zero real parts when
This expression reduces to
$F_{cusp}^{2}=2-3(6{\it\alpha}^{2})^{1/3}$
when
$M\rightarrow 0$
, as given by Blumen et al. (Reference Blumen, Drazin and Billings1975).
The asymptotic results, together with the numerical computations, are presented in figure 8. The inner expansion, including the location of the cusp given by (5.7), is in good agreement when
$F\approx F_{c}$
, and the outer expansion is in good agreement otherwise. Note that the inner expansion establishes the existence of two unstable modes when
$F<F_{cusp}$
, whereas the outer expansion only picks up the mode with larger
$c_{i}$
, i.e. the primary mode. It is the coalescence of the two modes revealed by the inner expansion that leads to the pair of secondary unstable modes. Thus the primary and secondary modes are continuously connected, but in a non-trivial manner.
Transition from the primary to secondary mode for
$U(y)=\tanh y$
. Shown are
$c_{i}$
and
$c_{r}$
versus
$F$
at
${\it\alpha}=0.01$
for (a)
$M=0$
, (b)
$M=0.25$
, (c)
$M=0.5$
. The crosses are the numerical results, the grey lines show the outer expansion
$c_{v}+{\it\alpha}c^{(1)}$
, with
$c^{(1)}$
given by (5.4), and the black lines the inner expansion given by relevant solutions to the cubic equation (5.6). For
$F>F_{cusp}$
, solutions with both
$c_{r}>0$
and
$c_{r}<0$
are shown.
5.2. Bickley jet
In this subsection we consider the basic state velocity defined by
From inequality (3.3), the growth rate
${\it\alpha}c_{i}$
is bounded above by
$|U^{\prime }|_{max}/2=2/(3\sqrt{3})$
; furthermore, from stability criterion (3.12), this flow is stable when
$M\geqslant 1/2$
. The Bickley jet is even about
$y=0$
and hence the parity result of § 3.3 holds – i.e. the eigenfunctions are either even or odd. In this case, the numerical method in appendix B may be adapted to integrate over only half the domain, with the imposition of either
$G^{\prime }(0)=0$
(even mode) or
$G(0)=0$
(odd mode).
When
$F=M=0$
there is a single even mode and a single odd mode, which are unstable only in the respective bandwidths
$0<{\it\alpha}<2$
and
$0<{\it\alpha}<1$
(Drazin & Reid Reference Drazin and Reid1981, § 31.9). Figure 9 shows contours of
$c_{i}$
for the continuation of these modes over
$(M,F)$
space for selected values of
${\it\alpha}$
. The values of
${\it\alpha}$
correspond to: (i) the most unstable mode when
$F=M=0$
(a,b); (ii) the mode with highest
$c_{i}$
when
$F=M=0$
(c,d); (iii) a long-wave disturbance (e,f). Increasing
$F$
and
$M$
is generally stabilising; however, there are small regions where this does not hold, as can be seen from figure 9(d,f). Figure 10 shows the growth rate as a function of
${\it\alpha}$
for selected values of
$F$
and
$M$
. The short-wave cutoffs, which for
$F=M=0$
are at
${\it\alpha}=2$
(even mode) and
${\it\alpha}=1$
(odd mode), are reduced as
$M$
increases, with a non-trivial dependence on
$F$
.
Contours of
$c_{i}$
over
$(M,F)$
space at selected
${\it\alpha}$
for the even (a,c,e) and odd (b,d,f) modes of the Bickley jet, with (a)
${\it\alpha}=0.9$
, (b)
${\it\alpha}=0.52$
, (c)
${\it\alpha}=0.3$
, (d)
${\it\alpha}=0.23$
, (e,f)
${\it\alpha}=0.05$
. The cutoff from the asymptotic result (5.10) is also plotted in red in (e).
Instability of the Bickley jet: growth rate of the mode tracked from
$F=M=0$
.
The behaviour at small
${\it\alpha}$
can be clarified using the long-wave analysis of § 4.3.2, although, as discussed there, this holds only for the even mode. Noting that for the Bickley jet,
$2E=\int _{-\infty }^{+\infty }(\text{sech}^{2}y)^{2}\,\text{d}y=4/3$
, the long-wave asymptotic results (4.37) and (4.38) become
and
There is a long-wave cutoff due to the magnetic field (cf. (4.24)). The accuracy of this is verified in figure 11, which shows the comparison between the numerically determined growth rates and the asymptotic results (5.9) and (5.10). As discussed in § 4.3.2, the long-wave analysis does not capture the odd mode since
$c_{r}$
remains
$O(1)$
as
${\it\alpha}\rightarrow 0$
. This has been verified numerically, where it is found that
$c_{i}\sim {\it\alpha}^{2/3}$
, a result derived for the hydrodynamic case by Drazin & Howard (Reference Drazin and Howard1962).
Comparison of the computed growth rates for the even modes (crosses; circles denote solutions with zero growth rates) and the predicted growth rates from the asymptotic results (5.9) and (5.10) (dot-dashed line and dashed line respectively), for
$U(y)=\text{sech}^{2}y$
. Here,
$M=0.1$
, and (a)
$F=0$
, (b)
$F=1$
, (c)
$F=5$
, (d)
$F=10$
.
As discussed above, when
$M=0$
the instability problem is equivalent to that of two-dimensional compressible hydrodynamics. There it is known that the rectangular jet supports multiple modes of instability at sufficiently large Mach number (here corresponding to sufficiently large
$F$
), which are due to over reflexion (Gill Reference Gill1965). It is therefore natural to ask if there are corresponding instabilities for the smooth Bickley jet profile and, if so, how they may change when
$M$
is non-zero. We have performed a scan over
$(M,F)$
space at various values of
${\it\alpha}$
, with initial guesses for
$c$
within the semi-circle (3.10), solving the governing eigenvalue equation with no parity imposed. We do indeed find multiple modes of instability, with a quantised character suggestive of over reflexion, for large
$F$
when
${\it\alpha}$
is sufficiently large. Sample plots of
$c_{i}$
and cross-sections of the eigenfunctions for the even modes are shown in figure 12; the results for the odd mode are similar. As
$F$
increases, the mode tracked from
$F=0$
changes character, turning into the most unstable quantised mode; furthermore, additional modes appear, although the growth rates are small.
Even mode of the Bickley jet at
${\it\alpha}=1$
. (a)
$c_{i}$
as a function of
$F$
at
$M=0$
,
${\it\alpha}=1$
. (b)
$c_{i}$
as a function of
$M$
at
$F=20$
. Crosses, circles and triangles show the three most unstable modes. (c–e) Cross-sections of the vorticity perturbation for
$F=20$
,
$M=0.05$
, for these three modes. The locations where
$U=c_{r}$
are labelled on the axes. The spikes reach an amplitude of approximately 0.5; the vertical axis has been chosen in order that the oscillations within the core are clearly visible.
There are two distinctive features of the spatial structure of the eigenfunctions, as shown in figure 12(c–e). First, there is a core region consisting of approximately quantised oscillations: the mode tracked from
$F=0$
has one oscillation, the
$n$
th additional mode has
$n+1$
oscillations. Second, the boundary of the core region is approximately located where
$U(y)=c_{r}$
. Here the eigenfunctions are highly oscillatory and of larger amplitude, although they are well resolved in the numerics underlying figure 12(c–e).
6. Conclusions and discussion
The SWMHD equations, introduced by Gilman (Reference Gilman2000), are a useful model for studying MHD in thin stratified fluid layers. We have investigated the linear instability of parallel shear flows with an aligned magnetic field in planar geometry with no background rotation. The instability of hydrodynamic shear flows is a classical problem, with well-known results such as Rayleigh’s inflection point criterion, Howard’s semi-circle theorem and Høiland’s growth-rate bound, supplemented by extensive asymptotic and numerical results for various idealised flows. Motivated by geophysical and astrophysical considerations, previous authors have extended separately the analysis to include the influence of shallow-water dynamics and a magnetic field. Here, for the first time, we have applied this classical approach to the SWMHD system, complementing previous work on the instability of specific flow configurations in spherical geometry (Gilman & Dikpati Reference Gilman and Dikpati2002; Dikpati et al.
Reference Dikpati, Gilman and Rempel2003). In our analysis, the stratification is measured by the Froude number
$F$
, while the imposed field strength is measured by the parameter
$M$
.
In § 3, we consider the stability of arbitrary flow and field profiles, leading to a growth-rate bound, semi-circle theorems and explicit stability criteria. In these, increasing
$M$
is generally stabilising. However, none of the criteria involve
$F$
, and so it is not possible to explore the combined effects of varying
$F$
and
$M$
.
One way to address these combined effects for quite general flows is via a long-wave analysis (
${\it\alpha}\ll 1$
). In § 4 we generalise to SWMHD the asymptotic approach of Drazin & Howard (Reference Drazin and Howard1962), in which all flows may be classified as either shear layers or jets. These possess contrasting instabilities. For the shear-layer profile, a key finding is that, although increasing
$F$
or
$M$
in the absence of the other is seen to be stabilising, their combined effects can offset each other, resulting in a tongue of instability for arbitrarily large
$F$
and
$M$
approaching unity. However, the strongest instabilities, with phase speeds of order unity, are found at smaller
$F$
and
$M$
. In contrast, for jets the phase speed is
$O({\it\alpha}^{1/2})$
, so the instabilities are weaker. Also, whereas for shear layers the leading-order phase speed is independent of
${\it\alpha}$
, for jets it is
${\it\alpha}$
-dependent. There is a long-wave cutoff when
$M$
is non-zero, although this disappears with increasing
$F$
.
The interesting combined effects of
$F$
and
$M$
revealed by the long-wave analysis lead naturally to a numerical investigation of two canonical smooth flows at finite wavenumbers (§ 5). For the hyperbolic-tangent shear layer there are two modes of instability, a stronger primary mode and a weaker secondary mode, which are, respectively, extensions of inflection point and supersonic instabilities. We interpret the primary instability in terms of the counter-propagating Rossby waves (CRWs) mechanism, which has frequently been used to understand hydrodynamic shear instabilities (e.g. Carpenter et al.
Reference Carpenter, Tedford, Heifetz and Lawrence2012). Here we show that the vorticity anomalies associated with shallow-water effects and magnetic tension oppose those for the underlying CRW mechanism; these ideas, first discussed in Mak (Reference Mak2013), are consistent with the later results of Heifetz et al. (Reference Heifetz, Mak, Nycander and Umurhan2015) for incompressible MHD. For the Bickley jet, the symmetry of the basic state flow implies the existence of competing even and odd modes of instability. Both even and odd modes have long- and short-wave cutoffs; in between, the even mode is generally more unstable. Note that the odd mode is not picked up by the long-wave analysis, as discussed in § 4.3.2. At large
$F$
, there are multiple modes of instability, due to over reflexion.
Given the motivation for this study, it is important to consider the implications of our results for the solar tachocline. For local shear zones, with characteristic length scale
$L_{0}=0.1R_{\odot }$
and characteristic velocity
$U_{0}=20~\text{m}~\text{s}^{-1}$
, taking
$10^{3}~\text{G}\lesssim B_{0}\lesssim 10^{5}~\text{G}$
gives
$0.3\lesssim M\lesssim 30$
. Thus only for the weakest fields in this range will there be instability. For planetary-scale shear flows, with
$L_{0}=R_{\odot }$
and
$U_{0}=200~\text{m}~\text{s}^{-1}$
, we have
$0.03\lesssim M\lesssim 3$
and
$F\approx 0.04$
. Here instability is suppressed only by fields at the strong end of the possible spectrum. The
$e$
-folding time for an instability is given by
$\hat{t}=({\it\alpha}c_{i})^{-1}(L_{0}/U_{0})$
, which, for both local and planetary shears, gives
$\hat{t}=O(10^{6}({\it\alpha}c_{i})^{-1})~\text{s}$
. With
${\it\alpha}c_{i}\approx 0.1$
, for both shear layers and jets (see figures 4 and 10),
$\hat{t}=O(10^{7})~\text{s}$
. Since
${\it\alpha}=O(1)$
for the unstable modes, the length scale of any instability is comparable to that of the shear.
Having investigated the quite complicated behaviour that results from our idealised model, there are several natural extensions of our study. Perhaps the most important additional effect would be the inclusion of background rotation. For solar dynamics, for which we have estimated an instability time scale of
$O(10^{7})~\text{s}$
, i.e. several solar days, we might expect rotation to modify the instabilities. In the extreme case of strong rotation, some of these effects may be similar to those of Gilman (Reference Gilman1967), who considered quasi-geostrophic baroclinic instabilities in a thin MHD layer. Although the formulation of the problem considered here, together with the general stability criteria of § 3, does allow for non-uniform magnetic fields, all our detailed calculations have assumed the magnetic field to be uniform. Thus we have not explored the possibility of non-uniform fields destabilising hydrodynamically-stable velocity profiles, as has been studied in two-dimensional incompressible MHD (Stern Reference Stern1963; Chen & Morrison Reference Chen and Morrison1991) and spherical SWMHD (Gilman & Dikpati Reference Gilman and Dikpati2002; Dikpati et al.
Reference Dikpati, Gilman and Rempel2003). Combining the effects of rotation and non-uniform fields would allow a direct comparison with previous studies of the solar tachocline, where the main emphasis was on the destabilisation of hydrodynamically-stable velocity profiles by non-uniform toroidal fields (Gilman & Cally Reference Gilman, Cally, Hughes, Rosner and Weiss2007). Finally, the nonlinear evolution of these types of shear flow instability is clearly of interest. It typically leads to turbulent flows; these may be important for dynamo action, through some mean-field
${\it\alpha}$
-effect, and also for the transport of mass and momentum, which can feed back on the large-scale flow.
Acknowledgements
This work was supported by the STFC doctoral training grant ST/F006934/1. We thank the referees for their useful comments. J.M. thanks E. Heifetz for helpful discussions.
Appendix A. Consistency checks for the long-wavelength jet analysis
The aim of this appendix is to show that
is
$O({\it\alpha}\log {\it\alpha})$
, and hence that the asymptotic analysis of § 4.3.2 is consistent.
Following Drazin & Howard (Reference Drazin and Howard1962), we assume that
$|U|\leqslant A\text{e}^{-a|y|}$
, which is satisfied for the Bickley jet. They make the additional assumption that
$c$
is ‘almost pure imaginary’; here we adopt the modified assumption that
which is supported by both our numerical and asymptotic results.
Consider first the case of
$F=0$
for which (A 1) is given by
For
$y>0$
, we split the range of integration into
$(0,{\it\lambda})$
and
$({\it\lambda},\infty )$
, where
since
$c^{2}=O({\it\alpha})$
. Then
$$\begin{eqnarray}\displaystyle (c^{2}-M^{2})\int _{0}^{{\it\lambda}}(\cdots \,)\,\text{d}y & {\leqslant} & \displaystyle |c^{2}-M^{2}|\int _{0}^{{\it\lambda}}\left(1+\frac{|c|^{2}+M^{2}}{c_{i}^{2}}\right)\text{d}y\nonumber\\ \displaystyle & {\leqslant} & \displaystyle |c^{2}-M^{2}|{\it\lambda}(1+N)=O({\it\alpha}\log {\it\alpha}),\end{eqnarray}$$
where we have used the usual integral inequalities, the inequality
$|(U-c)^{2}-M^{2}|\geqslant c_{i}^{2}$
, the assumption that
$M^{2}=O({\it\alpha})$
and the derived result (4.37) that
$c=O({\it\alpha}^{1/2})$
.
Similarly,
$$\begin{eqnarray}\displaystyle (c^{2}-M^{2})\int _{{\it\lambda}}^{\infty }(\cdots \,)\,\text{d}y & = & \displaystyle (c^{2}-M^{2})\int _{{\it\lambda}}^{\infty }\left(\frac{U^{2}-2Uc}{(U-c)^{2}-M^{2}}\right)\text{d}y\nonumber\\ \displaystyle & {\leqslant} & \displaystyle \frac{|c^{2}-M^{2}|}{c_{i}^{2}}\int _{{\it\lambda}}^{\infty }(U^{2}+2|U||c|)\,\text{d}y\leqslant N\left(\frac{A^{2}\text{e}^{-2a{\it\lambda}}}{2a}+\frac{2A|c|\text{e}^{-a{\it\lambda}}}{a}\right)\nonumber\\ \displaystyle & = & \displaystyle O({\it\alpha}).\end{eqnarray}$$
The dominant contribution is from (A 5), and hence
$I=O({\it\alpha}\log {\it\alpha})$
.
When
$F$
is non-zero, the only difference is in the pre-factor to the integral (A 1). However, with
$c$
from either (4.37) or (4.38) the pre-factor is
$O(1)$
and we may conclude that
$I=O({\it\alpha}\log {\it\alpha})$
.
Appendix B. Numerical method
We seek a numerical solution of the eigenvalue equation (2.7). Although the two velocity profiles considered are defined over the entire real line, we solve (2.7) on
$y\in [-L,L]$
. We employ a shooting method, which is numerically more efficient than solving the resulting linear algebra problem through a discretisation of the operators, particularly when the solution has small-scale features (see figure 12
c–e for example). We shoot from the boundaries, with matching imposed at
$y=0$
, employing a generalised Newton method as the root-finding algorithm. Since the solutions decay exponentially as
$|y|$
becomes large, namely
where
we adopt expression (B 1) as the boundary condition to be implemented at
$y=\pm L$
. Since our interest is in instabilities, singularities in the governing equation are avoided. To ensure negligible influence of the finite domain,
$L$
is doubled until the computed eigenvalue changes by less than 0.5 %. The routines are written in MATLAB, using ode113 as the integrator (an Adams–Bashforth method with adaptive grid). Although the boundary conditions are functions of
$c$
, changing at every iteration, we generally have no problems with convergence provided that the initial guess is close to the true value. Solutions are initialised from
$(M,F)=(0,0)$
at some fixed
${\it\alpha}$
using a known numerical result documented in, for example, Drazin & Reid (Reference Drazin and Reid1981). Runs at new parameter values are then initialised using a linear extrapolation for
$c$
from previously calculated values at nearby parameters.
