## 1. Introduction

Most atmospheric scientists are familiar with the type of inertia-gravity wave that is trapped along coastlines or the equator, known as a Kelvin wave (e.g., Thomson 1879; Gill 1982, chapter 10; Wang 2003). These waves play a prominent role in tropical dynamics including El Niño–Southern Oscillation (e.g., Battisti 1988). Perhaps less familiar within the atmospheric science community is the work by Lord Kelvin on vortex waves (Thomson 1880), in which he laid out the general strategy for obtaining a wide variety of solutions of waves traveling within infinitely extended, concentrated cylindrical vortices resembling the flow in tornadoes or dust devils some distance away from the lower boundary. These waves are usually referred to as “Kelvin waves” in the fluid physics literature, and they are also known as vortex waves, centrifugal waves, or inertial waves (Lugt 1989). Such waves play an important role in the vortex breakdown phenomenon as well as in the development of multiple-vortex tornadoes (e.g., Lewellen 1993; Rotunno 2013). Examples of these waves within tornadoes are shown in Fig. 1. Here so-called bending waves (to be introduced in section 3) are visible, which are often well pronounced during the rope stage of the tornado (Fig. 1b). The effects of vortex waves are also seen in Fig. 2, where Fig. 2a shows a multivortex structure that results from unstable growth of spiral modes (section 3), and Fig. 2b displays what appears to be a vortex breakdown. This phenomenon often occurs in vortices with nonzero axial velocity and it will be reviewed in section 8.

The need for this review arises from the observation that, despite the ubiquity of Kelvin vortex waves in tornado-like vortices, a detailed introduction to the structure and behavior of these waves appears to be absent from the atmospheric science literature. Although this topic is introduced in some fluid physics texts (e.g., Thomson 1880; Lamb 1932, section 158; Chandrasekhar 1961, chapter VII, especially p. 284; Drazin and Reid 1981, p. 75; Saffman 1992, chapters 11 and 12; Lim and Redekopp 1998; Fritts et al. 1998; Rossi 2000; Batchelor 2002, p. 559; Alekseenko et al. 2007, chapter 4), in a given presentation either only limited, or rather advanced, aspects of vortex waves are discussed. As a consequence, there is a large gap between the knowledge base offered in mainstream texts about atmospheric dynamics (e.g., Gill 1982; Holton and Hakim 2013; Markowski and Richardson 2010) and the advanced peer-reviewed literature on the remarkably rich Kelvin vortex wave dynamics. Thus, the present paper seeks to

narrow the gap between the atmospheric-dynamics literature and the advanced fluid-physics literature by providing a detailed introduction to these waves. Except where the mathematical steps are readily available in mainstream texts, the complete mathematical development is included in this paper;

retrace Kelvin’s original approach, and point to possible applications of his contributions to tornado-like flows, including multiple-vortex development and vortex breakdown.

To achieve this, an almost trivial generalization of Kelvin’s equations is included by allowing for a piecewise constant axial flow, which admits a wide class of unstable solutions not considered in Kelvin’s original work. These solutions have been arrived at previously, but along with a review of these topics, it is demonstrated that the results follow from Kelvin’s approach. In the context of stability analyses, Kelvin’s (slightly modified) approach presented in this review has been superseded by much more advanced analysis techniques, but these are still based on analyses such as those presented here, and will be touched upon in section 8.

The remainder of the paper is structured as follows. In the next section a brief history of vortex waves will be offered and the relevance of these waves will be described, and section 3 offers an overview of the classification of vortex waves. In section 4 the governing equations describing infinitely extended columnar vortices will be presented following Kelvin’s approach. Subsequently, consecutively more refined scenarios are introduced that follow directly from Kelvin’s equations. Starting with the scenario of vanishing base-state axial flow, a cylindrical domain bounded by rigid walls will be considered (section 5) to gain an intuition for the structure and dynamics of these waves. Thereafter the Rankine vortex will be discussed in section 6, and two scenarios that allow for unstable wave growth will be considered in section 7, i.e., a Rankine vortex with upward motion in its core as well as a two-celled vortex with descending motion in its irrotational core, and rising motion outside of it. The unstable waves in the latter scenario provide a rudimentary model for multiple-vortex formation in tornado-like vortices. These ideas will then be applied to vortex breakdown in section 8. Limitations of Kelvin’s approach and its applicability to tornadoes will be addressed in section 9. Finally, section 10 offers concluding remarks and possible directions for future investigations.

## 2. A brief history of Kelvin vortex waves

Around the mid-1800s, Sir William Thomson, who was granted the title Lord Kelvin in 1892, pursued the idea of describing the previously discovered atoms in terms of microscopic knotted vortex rings, called “vortex atoms” (Thomson 1867; Fabre et al. 2006). The medium in which these vortex rings were thought to exist was the hypothesized all-pervading, homogeneous perfect fluid known as the aether. This idea was fueled by Hermann von Helmholtz’s discovery of the laws of vortex motion (von Helmholtz 1858), specifically his second law, which implies that these vortex rings could persist forever in perfect homogeneous (and thus barotropic) fluids. Falconer (2019) offers a detailed summary of Kelvin’s vortex atom theory. Kelvin was particularly interested in the vibrational modes of these vortex rings, postulating that different vibrational modes could account for the different atomic spectra that had previously been discovered. As a first step, Kelvin formulated the equations describing wave motions within an infinitely long, cylindrical vortex. This effort led to the paper entitled “Vibrations of a columnar vortex” (Thomson 1880), which is the basis of this review. Although the idea of vortex atoms did not survive past the 1890s, it does bear an intriguing resemblance to string theory, and Kelvin vortex waves still do play an important role in fundamental physics, e.g., in the dynamics of quantum vortices in superfluids (e.g., Fonda et al. 2014).

Kelvin’s work entered the field of aeronautical engineering following the discovery that lift-generating devices produce a pair of trailing vortices, which pose a hazard to aircraft that encounter these vortices, which consequently reduce airspace capacity (Hallock and Holzäpfel 2018). The stability of the wake vortices is directly related to the unstable growth of bending Kelvin modes (the different Kelvin modes will be introduced in the next section). This instability is known as “cooperative instability” and is sometimes visibly manifest as a contortion of aircraft contrails and the formation of contrail lobes (Lewellen and Lewellen 2001; Wu et al. 2006, p. 499; Schultz and Hancock 2016).

In the late 1950s, another phenomenon was discovered by aerodynamicists, i.e., the vortex breakdown. Peckham and Atkinson (1957) are generally credited for first documenting this phenomenon during the investigation of lift-generating vortices produced at the leading edge of ogival delta wings. At large angles of attack, the observation included a disintegration of the vortex structure, which appeared to “bell out before disappearing—as though the core was becoming more diffuse” (Peckham and Atkinson 1957, p. 5), indicating turbulence and an undesirable drop of lift. (The reader may skip ahead to Fig. 31 for an example of such a breakdown in a tornado.)^{1} On the other hand, vortex breakdown was found to have beneficial effects in some applications such as stabilizing the flame in combustion chambers of gas turbines (e.g., Escudier 1988; Spencer and McGuirk 2008). Vortex breakdown has thus garnered considerable interest in the engineering and fluid physics community, and soon after its official discovery, Squire (1960) and Benjamin (1962) explained the vortex breakdown phenomenon as an axisymmetric analog of the hydraulic jump, which is related to the inability of wave perturbations, in this case axisymmetric Kelvin vortex waves, to propagate upstream (Squire 1960; Benjamin 1962). Among other vortex profiles, these studies also considered the Rankine vortex, and the vortex breakdown criterion discovered in these studies directly follows from Kelvin’s analysis. Vortex breakdown and its relation to Kelvin vortex waves will be reviewed in section 8.

Aside from the application to engineering problems, the roles of Kelvin vortex waves have been studied in the general context of transition to turbulence in rotating flows (e.g., Hopfinger et al. 1982). Importantly, these experiments revealed the existence of solitary-like vortex waves (Maxworthy et al. 1985). In the atmospheric science literature, Kelvin’s work does not appear to be mentioned frequently. One exception is Fultz (1959), who studied oscillations in convective motion. He observed vortex waves in his rotating tank experiments and pointed to a textbook by Bjerknes et al. (1933), where vortex waves are introduced. Apparently, however, Bjerknes et al. (1933) had been unaware of Kelvin’s treatment, prompting Fultz to note that “its significance in meteorological dynamics had to be rediscovered.” Kelvin vortex waves feature prominently in the studies by Andreassen et al. (1998), Fritts et al. (1998) and Fritts and Alexander (2003), which are concerned with turbulence generation as a result of shear instability as well as in breaking internal gravity waves (Fritts and Alexander 2003). These authors refer to vortex waves as “twist waves” (Arendt et al. 1997). Kieu (2016) extended Kelvin’s solutions with the goal of describing waves in the inner core region of tropical cyclones.

In the field of tornado research, centrifugal waves attracted some attention in the 1970s and 1980s when research using tornado vortex chambers flourished (Ward 1972; Church et al. 1977; Rotunno 1979; Snow 1982; Church and Snow 1993). However, the focus of these analyses was the determination of how the flow parameters, most notably the swirl ratio (Davies-Jones 1973), led to different tornado structures including vortex breakdown and multiple vortices. The importance of centrifugal waves is mentioned in these studies in the context of vortex breakdown (Church et al. 1977; Rotunno 1979; Snow 1982; Fiedler and Rotunno 1986), but the structure and dynamics of these waves are not analyzed further. Nolan and Farrell (1999) analyzed these waves in detail, and they observed downward propagating axisymmetric waves in their numerically simulated tornado-like vortex. Their analysis suggests that these waves are less likely to be observed in flows with low swirl ratios, consistent with such flows being supercritical (section 8), but their linear analysis did not fully explain the behavior of these waves. More recently, Nolan (2012) revisited this phenomenon by studying the linear instability of fully nonlinear flows.

Because the vortex chamber experiments reproduced the observed transition of a single tornado vortex into multiple vortices, investigations into the stability of the tornado followed (e.g., Rotunno 1978; Walko and Gall 1984; Nolan 2012). This instability is a manifestation of unstable centrifugal waves, and will be discussed in section 7. With the exception of Nolan’s analyses (Nolan and Farrell 1999; Nolan 2012), these centrifugal waves apparently have not played important roles in interpreting tornadic flows since the 1980s, and Snow (1982) at the end of his review on tornado dynamics, added that “features that should be investigated…include the nature of the wavelike features often seen on the walls of condensation funnels” (Snow 1982, p. 963). It is fair to say that his recommendation has not materialized, and consideration of centrifugal waves in the context of tornadoes appears to be scarce. Aside from Nolan and Farrell (1999), some examples include Lugt (1989) and Trapp (2000), who discuss vortex breakdown in tornadoes and mesocyclones, respectively, and in this context briefly mention centrifugal waves; Bluestein et al. (2003) observed oscillations of a tornado’s intensity and as a possible explanation, centrifugal waves were invoked; in their textbook, Markowski and Richardson (2010) introduce the stability criterion for centrifugal oscillations, which correspond to an axisymmetric Kelvin mode.

This brief overview reveals that, while the relevance of vortex waves in tornadoes is generally acknowledged, the association with Kelvin’s work is not widely appreciated, and the structure and behavior of these waves has attracted somewhat limited attention in the atmospheric science community. To start the discussion of vortex wave dynamics, the basic types and naming conventions of these waves are introduced in the next section. The main distinction between the different wave types is the azimuthal wavenumber *m*, which determines whether a wave is manifest as, e.g., axisymmetric or spiral perturbation. The azimuthal wavenumber is dimensionless and must be an integer to guarantee continuity of the wave perturbation. The vortex column serves as waveguide, and for infinitely extended vortex columns, each wave type (determined by *m*) may generally attain an arbitrary axial wavenumber *k*.

## 3. Classification of Kelvin vortex waves

### a. Axisymmetric modes, m = 0

If *m* = 0, the perturbation is axisymmetric, resulting in a widening and narrowing of the vortex (e.g., in a Rankine vortex the radius of maximum winds extends and contracts). The axial wavenumber must be nonzero for axisymmetric waves to exist. A commonly used name of this mode goes back to Lord Rayleigh (Strutt 1902, p. 444). When discussing photographs of a liquid jet undergoing axisymmetric oscillations before breaking up into drops (as happens to a water stream emanating from a faucet), he noted: “…I have often been embarrassed for want of an appropriate word to describe the condition in question. But a few days ago, during a biological discussion, I found that there is a recognised, if not very pleasant, word. The cylindrical jet may be said to become *varicose*, and varicosity goes on increasing with time, until eventually it leads to absolute disruption.” Chandrasekhar (1961, p. 515), after citing this passage from Rayleigh, adds: “In recent times, ‘sausage instability’ has been used to describe the same condition; but this is also not a very ‘pleasant’ description, and varicose instability would seem preferable.” The designations “sausage” or “sausaging” as well as “varicose” modes have been adopted widely to describe axisymmetric vortex waves, though in this paper these waves will be referred to mainly as axisymmetric waves. This wave can only propagate in the axial direction and it plays an important role in the onset of vortex breakdown (section 8).

### b. Spiral modes, |m| > 0

Waves with azimuthal wavenumbers |*m*| ≥ 1 are referred to as helical or spiral modes. If |*m*| = 1, the waves displace the vortex axis and are called “bending” modes. For all other choices of *m* the vortex axis remains centered. An example of how an azimuthal wave perturbation is related to the vortex structure is shown in Fig. 3, demonstrating that a wavenumber of two corresponds to an elliptic deformation of the vortex core. The sign of the azimuthal wavenumber determines the handedness of the spiral. As shown in Fig. 4, *m* > 0 implies left-handedness and *m* < 0 implies right-handedness.^{2}

If the slope of the phase lines in the (*θ*, *z*) plane wrapping around the cylindrical vortex is taken to be *dh*/*dθ*, where *h* is the local height of the phase line and *θ* is the azimuthal angle, this slope is identical to the pitch of the wave, given by −*m*/*k* (Fabre et al. 2006). Defined this way, pitch is proportional to the distance by which the helix advances during one revolution (i.e., to the axial wavelength of the helix). Saffman (1992) and Alekseenko et al. (2007) use slightly different definitions of pitch, which only apply to |*m*| = 1 but are proportional to each other as well as to the definition used here. It follows that, e.g., left-handed spirals have a negative pitch and axisymmetric waves have zero pitch. For a given axial (vertical) wavenumber, the pitch becomes larger in magnitude as the number of spirals winding around the vortex (i.e., *m*) is increased.

An azimuthal wavenumber of *m* = 2 implies a double helix or double spiral structure. Waves with *k* = 0 and *m* ≠ 0 are sometimes called “fluted” modes (Maxworthy 1988), which are 2D waves that only propagate in azimuthal direction. Hopfinger et al. (1982), Hopfinger and Browand (1982), and Maxworthy et al. (1985) coined the term “kink wave” for helical solitary waves (*m* = ±1) they observed in their experiments. These waves resembled the soliton solution by Hasimoto (1972).^{3}

The azimuthal angular phase speed of spiral modes is given by *ω*/*m*, where *ω* is the wave frequency. Assuming that the angular velocity of the vortex Ω is positive, implying cyclonic rotation, then if the azimuthal phase speed is larger than the angular velocity of the vortex, the waves are said to be “cograde,” meaning these waves propagate downstream relative to the motion within the vortex. If the angular speed of the wave is less than Ω, but the waves still move in the same sense as the azimuthal vortex flow, the wave is said to be “retrogade.” These waves propagate upstream relative to the vortex flow, but are advected downstream. Finally, the waves are “countergrade” if they move in the opposite sense than the vortex, now also for a stationary observer (Fabre et al. 2006). From these basic definitions and conventions the rich structure of vortex waves is already becoming apparent. The next section introduces the governing equations describing the dynamics of these waves.

## 4. Governing equations

One goal of this paper is to reintroduce Kelvin’s original mathematical treatment, which is summarized in Fig. 5. Starting with the linearized, inviscid, incompressible equations of motion in cylindrical coordinates, the existence of a base state flow is assumed, and the equations are linearized about this base state. Subsequently, a normal-mode solution is inserted, and the resulting simplified equation set is solved for the radial and azimuthal flow components. One then specifies the base state flow for a given vortex configuration, uses the mass continuity equation, and arrives at a linear ordinary differential equation (ODE) whose solution governs the radial wave structure. Application of the boundary conditions finally leads to the complete flow field as well as the dispersion relation, from which the wave frequencies and speeds, as well as the existence and growth rates of unstable modes, may be inferred.

### a. Linearization of the momentum and mass-continuity equations

*u*,

*υ*,

*w*). The azimuthal velocity component is also called swirl velocity. The equation for the radial velocity

*u*is given by (e.g., Drazin and Reid 1981, p. 71):

*r*is the radial distance from the center,

*θ*is the azimuth, and

*p*is the pressure perturbation relative to a hydrostatic reference state. The density

*ρ*is assumed to be constant in this treatment. The last term on the lhs is the centripetal acceleration, which appears when writing the equations of motion in cylindrical coordinates. For the azimuthal component

*υ*we have

*u*/

*r*term arises from the axisymmetry of the coordinate system (i.e., whenever the radial velocity

*u*is nonzero there is nonzero radial divergence).

*V*(

*r*) to be a solution of the above equations, a state of cyclostrophic balance is implied. The azimuthal base-state velocity

*V*(

*r*) may have an arbitrary dependence on

*r*. The axial base-state flow

*W*is constant, and there is no radial base-state flow. Then,

*P*(

*r*) is the base-state pressure and the primed variables denote perturbations from the base state. Throughout the analysis,

*V*is assumed to be positive, so the base-state vortex spins cyclonically. To linearize the governing equations, this decomposition is inserted into Eqs. (1)–(4). For the radial momentum equation, one finds

*r*, and

*ω*=

*ω*

_{r}+

*iω*

_{i}is the complex phase speed.

^{4}The axial wavenumber

*k*is assumed to be real in this analysis. It is implied that only the real part of the solution has physical relevance (e.g., Markowski and Richardson 2010, chapter 6; Holton and Hakim 2013, chapter 5), which is sometimes written explicitly by adding the complex conjugate to the above normal modes (e.g., Fabre et al. 2006). Upon inserting these normal modes into the linearized governing equations, one finds that

### b. Equations for the radial and azimuthal velocity

*V*(

*r*) and a constant base-state axial flow

*W*. While the axial flow is uniform within each of the two regions, it may be different in each region. This simplifies Eqs. (24) and (26), which are subsequently inserted into the continuity equation, Eq. (22), to obtain an ordinary differential equation for

### c. Conditions at the free boundary

*r*<

*R*, variables will have the suffix 1. The variables pertaining to the outer region,

*r*>

*R*, will be marked with the suffix 2. The free boundary

*R*

_{b}between these two regions is perturbed by the waves, and its location given by

#### 1) Kinematic boundary condition

*r*′ [Eq. (27)] in Eq. (29) results in

#### 2) Dynamic boundary condition

*R*approximately gives

*P*

_{1}=

*P*

_{2}, and the condition reduces to (see also Gallaire and Chomaz 2003)

*r*′ may be eliminated using the kinematic boundary condition, Eq. (31), giving

## 5. Solid-body rotation in a bounded domain

### a. General solution and boundary condition

*r*=

*R*. With this,

*dV*/

*dr*=

*V*/

*r*= Ω. These conditions are inserted in Eqs. (24) and (26). First, it is noted that

*d*[Eq. (25)] becomes

*r*=

*x*/

*β*, so that

*d*/

*dr*=

*β*(

*d*/

*dx*), which leads to

*β*

^{2}as the eigenvalue, and with

*μ*and the coefficients

*a*

_{n}. The power series then takes the following form (e.g., Arfken et al. 2013, p. 352; Deal 2018):

*m*the general solution of Eq. (54) is

*A*and

*B*are arbitrary constants, and

*J*

_{m}(

*βr*) and

*Y*

_{m}(

*βr*) are, respectively, the Bessel functions of the first and second kind, of order

*m*. These are shown in Fig. 6. The Bessel function of the first kind of order

*m*= 0,

*J*

_{0}(

*x*), (red curve in Fig. 6a) qualitatively behaves like a damped cosine wave, and

*J*

_{1}(

*x*) (dashed black curve in Fig. 6a) behaves similarly to a damped sine wave. Since

*Y*

_{m}(

*βr*) is negatively infinite at

*r*= 0, this solution is rejected, so that

*β*

^{2}represents the squared radial wavenumber. Because the boundary is rigid, the kinematic boundary condition is simply that the normal velocity at the boundary should vanish, so

^{5}Since the flow is inviscid, the flow parallel to the boundary, (

*υ*,

*w*), is free-slip. Using the equation for the radial velocity, Eq. (46), and the solution, Eq. (57), the boundary condition becomes

*βr*). Equation (58) then becomes

*β*values for which this equation is fulfilled. One then solves Eq. (52) for the wave frequency

*ω*(which is defined via

*g*) and obtains the dispersion relation. The ratio of the Bessel functions behaves qualitatively like the (negative of the) tangent function, implying that this equation has a countably infinite set of solutions.

### b. Axisymmetric modes in a bounded domain

*m*= 0, one can find the solution of Eq. (61) right away. In this case the rhs is zero and use can be made of the fact that

*d*/

*dx*(cos

*x*) = −sin

*x*. The fraction on the lhs of Eq. (61) becomes zero only where the numerator

*J*

_{1}(

*βr*) is zero [the zeros of the denominator, which make the equation singular, do not coincide with

*J*

_{1}(

*βr*) = 0]. It follows that the requirement for a solution is

*β*

_{j}

*R*≡

*α*

_{j}= 3.8317, 7.0156, 10.1735,

*j*= 1, 2, 3. The different

*β*

_{j}values correspond to the different discrete

*radial modes*of the waves, which will be introduced in the next paragraph. Since

*β*

_{j}may be interpreted as radial wavenumber, the root

*α*

_{0}= 0 is not considered here. Now Eq. (52) can be used to determine

*g*

_{j}(

*k*):

*g*

_{j}(

*k*) =

*ω*

_{j}(

*k*), yields

*fundamental mode*, moves faster than the higher-order modes. Moreover, the phase and group speeds coincide for small

*k*, and the longest axisymmetric waves of the fundamental radial mode propagate the fastest.

*r*,

*z*)] plane, one uses the equations for

*w*

_{1}≡

*A*. Since the problem at hand is governed by a linear, homogeneous ODE, the amplitude

*w*

_{1}may be selected arbitrarily. Then, starting with the radial velocity equation, Eq. (46), the solution

*β*

^{2}

*g*

^{2}/

*k*, and one may write

*m*= 0 this equation becomes

*u*and

*w*waves, the vertical motion has a different magnitude than the radial motion; for instance, for small

*k*,

*u*

_{1}≪

*w*

_{1}, and the perturbation flow is dominated by the vertical perturbation velocity (i.e., the perturbations are manifest mainly as axial jets; the reader may skip ahead to Figs. 17e,f for an example), and for large

*k*, the perturbation flow is mainly in the radial direction.

*R*, the boundary condition,

*β*

_{j}=

*α*

_{j}/

*R*= 3.8317/

*R*, 7.0156/

*R*, 10.1735/

*R*, …. Since

*β*may be interpreted as a radial wavenumber, these solutions dictate the radial structure of the wave. For the smallest eigenvalue one obtains a single cell in the radial direction, which is the classic “sausage” mode, shown in Figs. 8a and 8b. For the third eigenvalue, the result is three circulation cells in the radial direction, because the radial velocity distribution now has three roots. This configuration is shown in Figs. 8c and 8d. The higher the order of the solution, the more cells appear in the radial direction. The magnitude of these cells slowly decays with increasing distance from the axis. These different solutions are the radial modes (Alekseenko et al. 2007). Fultz (1959) presents laboratory observations of the fundamental and the second radial modes of these waves. Theoretically, for each azimuthal wavenumber there are infinitely many radial solutions [in reality, the high-order modes tend to be dampened by viscous effects (Arendt et al. 1997; Fabre et al. 2006)]. Shapiro (2001a) arrived at solutions for nonlinear axisymmetric centrifugal waves with a similar radial structure as Kelvin vortex waves, although he assumed no radial boundary and the wave perturbations increased in magnitude with

*r*.

#### Restoring force and propagation mechanism of the axisymmetric mode

*Rayleigh discriminant*(e.g., Chandrasekhar 1961). The restoring forces are the radial pressure gradient force and the centrifugal force. The oscillation frequency

*ω*is related to the rate at which the ambient pressure gradient changes in the radial direction, which is linked to

*ω*

^{2}> 0, the radially displaced ring will find itself in a stronger pressure gradient than required for cyclostrophic balance. The result is an inward radial acceleration (i.e., stability; e.g., Markowski and Richardson 2010, p. 49). If the Rayleigh discriminant is negative, the displacement is unstable, which results in radial accelerations leading to toroidal circulation cells (e.g., Kundu and Cohen 2008, p. 486). For common vortex profiles, such as the Rankine and Burgers–Rott profiles (e.g., Wood and White 2011), or the Lamb–Oseen vortex (e.g., Fabre et al. 2006),

To understand why the phase speed depends on the axial wavenumber, consider the variation of the width of the vortex (e.g., as measured by the radial displacement of a concentric material surface relative to the base state), which serves as a measure of vertical vorticity. The narrower the vortex, the larger the vertical vorticity due to angular-momentum conservation. As shown in Fig. 9, which depicts an upward propagating wave, the regions of horizontal convergence near the vortex axis are shifted upward by a quarter wavelength relative to the axial vorticity maxima, implying that the vorticity extrema are propagating upward via vortex stretching and compression (see also Shapiro 2001a; Fabre et al. 2006). The regions of horizontal divergence and convergence are associated with meridional vorticity, which arises from tilting of the axial vorticity into and out of the meridional plane by the axial gradients of azimuthal velocity. The periodic vertical motion induced by the meridional vorticity stretches/compresses the axial vorticity. This is associated with a twisting and untwisting of the vortex lines (Melander and Hussain 1994; Arendt et al. 1997). The longer the wave, the larger are the pressure perturbations that accompany the vertical vorticity extrema [see, e.g., Markowski and Richardson (2010, p. 27) for the relationship between vorticity and pressure] or alternatively, the larger the induced velocity magnitudes (e.g., Dahl 2020) associated with the meridional vorticity extrema. As a consequence, as the vertical wavelength increases there are stronger vertical pressure-gradient accelerations and *w*′ gradients, resulting in more vigorous stretching/compression of vertical vorticity and hence faster wave motion.

### c. Spiral modes in a bounded domain

*m*≠ 0, the wave frequency, using Eq. (52) with

*g*=

*ω*−

*m*Ω, is given by

*m*→ −

*m*with

*k*> 0, the sign of

*ω*

_{j}changes, but the dispersion behavior remains the same. Changing the sign of

*k*does not change the wave frequency. Because of this symmetry, we may restrict the analysis without loss of generality to the case of

*k*> 0 and

*m*> 0, which describes the left-handed spiral modes (e.g., Saffman 1992, p. 232; Alekseenko et al. 2007). The dispersion relation, with

*α*

_{j}obtained via Eq. (61), of the first three radial modes of the bending wave is shown in Fig. 10. Also displayed (Fig. 10a) is the graphical solution of Eq. (61). The cograde modes are represented by solid lines and have a positive oscillation frequency and axial phase speed (i.e., they move upward), and vice versa for the retrograde modes (Figs. 10b,c). The axial phase speeds behave like

*k*

^{−1}for small

*k*, and thus attain infinite values in the longwave limit (which results from the fact that the phase lines are almost parallel to the

*z*axis). The group speeds, Fig. 10d, remain finite and are always slower than the phase speeds.

*w*(

*r*) =

*w*

_{1}

*J*

_{1}(

*βr*). This gives, using again Eq. (52):

*V*(

*r*) = Ω

*r*must be added to the last equation. These flow fields are plotted in Figs. 11 and 12 . For

*m*= 1, the perturbation flow (Fig. 11a) features a dipole, and the total flow (Fig. 11b) exhibits a displacement of the vortex center. For higher-order radial modes, the perturbation flow becomes more structured in the radial direction, but the total flow is still characterized mainly by a shift of the vortex center (Figs. 11c,d). To demonstrate the richness of structure of the higher-order modes, Fig. 12 shows the

*m*= 4 spiral mode for the fundamental (Figs. 12a,b) and third (Figs. 12c,d) radial modes.

#### Propagation mechanism of the spiral mode

To gain an understanding of the propagation of these modes, the velocity and the pressure fields of the fundamental spiral mode (*m* = 1) are shown in Fig. 13. The fields are displayed on an unrolled cylindrical surface, i.e., in the (*θ*, *z*) plane and for a small axial wavenumber. Like in the axisymmetric case, positive axial vorticity perturbations are associated with a negative pressure perturbation, and negative axial vorticity perturbations are associated with a positive pressure perturbation, so *p*′ ~ −*ζ*′, where *ζ*′ is the axial vorticity perturbation (Fabre et al. 2006). It is apparent that there is axial stretching a quarter wavelength above the perturbation vorticity maxima for the cograde (upward propagating) mode (Fig. 13a). For the retrograde mode, (Fig. 13b), the maximum stretching occurs below the perturbation vorticity maxima. Similarly to the axisymmetric mode, these waves thus propagate due to stretching and compression of axial perturbation vorticity.

To summarize this section, each of the azimuthal modes (i.e., axisymmetric and spiral modes) have an infinite number of radial modes, which are characterized by an increasing number of circulation cells in the radial direction with increasing order of the radial mode. The perturbation flow of the fundamental mode has only one circulation cell in radial direction. Moreover, it was highlighted that the longwave limit of the axisymmetric fundamental mode moves the fastest. A better approximation of a tornado-like flow is a vortex not bounded by cylindrical walls, which will be considered next.

## 6. Rankine vortex

### a. General solution and boundary conditions

For the Rankine vortex, one combines the solution for solid-body rotation of the previous section with an irrotational outer region. As mentioned in section 4c, the inner solution is designated by the suffix “1” and the outer solution is designated by the suffix “2.” For now, it is assumed that *W*_{1} = *W*_{2} = *W*, such that *g*_{1} = *g*_{2}.

*R*may be interpreted as core radius. Inserting this into the equation for the radial velocity, Eq. (24) with

*d*reducing to −

*g*

^{2}per Eq. (25), leads to

*modified*Bessel differential equation upon substituting

*r*=

*x*/

*k*. The modified Bessel ODE equals the regular Bessel ODE if the argument is imaginary (Arfken et al. 2013, 680–683). If

*m*is an integer, the solution is a linear combination of the Bessel functions encountered in the previous section but with imaginary argument (Arfken et al. 2013, 680–683). The solution of Eq. (81) becomes

*A*and

*B*are arbitrary constants, and

*I*

_{m}(

*kr*) and

*K*

_{m}(

*kr*) are the modified Bessel functions of the first and second kind, respectively. These functions loosely resemble exponential functions, becoming infinite for large arguments in case of

*I*

_{m}and for small arguments in case of

*K*

_{m}, as shown in Fig. 14. In the present case, the

*I*

_{m}solution is unphysical because the perturbations are required to be finite at radial infinity, so only the modified Bessel function of the second kind,

*K*

_{m}, is retained. Then,

*g*

_{1}=

*g*

_{2}simply enforces continuity of the normal velocities on each side of the boundary. Using Eq. (67) for the inner side, one finds that

*r*=

*R*,

*r*=

*R*,

*J*

_{m}(

*βR*)

*K*

_{m}(

*kR*) and some minor rearrangement gives the desired expression:

*β*, which yields

*ω*via Eq. (52). A few words about the stability of the waves described by this dispersion relation are in order. For unstable growth, the equation must be fulfilled for an imaginary wave frequency

*ω*

_{i}, making

*g*imaginary. However, the rhs of Eq. (96) is always real, so the imaginary part of the lhs must also be zero. Kelvin vortex waves in a Rankine vortex with uniform or vanishing axial base-state flow are thus stable. The solutions of this transcendental equation may be carried out graphically or numerically. Before proceeding with the numerical solutions, an important and in Kelvin’s words “curiously interesting” limit is considered.

### b. Axisymmetric modes in the Rankine vortex: Longwave limit (k ≪ 1)

*m*= 0), finding the solutions of Eq. (96) is straightforward. Figure 15 shows the rhs as blue horizontal lines for several choices of

*k*. The smaller

*k*, the larger the

*y*intercept of the line, so for

*k*→ 0, the rhs → ∞. The lhs is also plotted, revealing singularities where

*J*

_{0}has its roots. As suggested by the graphic, for small

*k*(long axial wavelength), the intersection of the lhs and rhs approaches the roots of

*J*

_{0}. Solutions thus exist for

*α*

_{j}=

*Rβ*

_{j}= 2.4048, 5.5201, … To obtain the wave frequency, Eq. (52) is used again and solved for

*g*=

*ω*(if

*W*= 0; else

*g*=

*ω*−

*Wk*),

*k*one may use a Taylor-series expansion around

*k*= 0, which gives

*ω*

_{j}(0) = 0, this yields

*k*coincides with the group speed, is thus

*V*

_{max}=

*V*(

*R*) is the maximum azimuthal base-state wind. For the smallest root,

*α*

_{1}= 2.4048, corresponding to the fundamental radial mode,

*V*at the radius of maximum winds (RMW), so the wave speed increases with the swirl velocity. Moreover, like in the bounded case discussed in section 5, the fundamental mode propagates the fastest (smallest

*α*

_{j}). More generally (i.e., outside the limit of

*k*≪ 1), the phase speed must be determined numerically, as will be done next.

### c. Axisymmetric modes in the Rankine vortex: Beyond the longwave limit

To drop the restriction of small wavenumbers, one needs to find the intersection of the lhs and rhs of Eq. (96) numerically, an example of which is shown in Fig. 16a. The dispersion relations for axisymmetric waves in the Rankine vortex are shown in Figs. 16b–16d, revealing that the waves qualitatively behave like those in the bounded vortex case (section 5), the main difference being that in the longwave limit the waves in the Rankine vortex move faster by about 60% (cf. Figs. 7c and 16c for *k* = 0). Incidentally, these solutions is reminiscent of Love wave^{6} dispersion involving the tangent function, which behaves similarly to the ratio of Bessel functions.

*β*

_{j}

*R*value, with the outer solution at the RMW using either the kinematic [Eq. (33)] or the dynamic [Eq. (41)] boundary condition. This way the constant

*w*

_{2}≡

*B*in Eq. (83) may be expressed in terms of

*w*

_{1}=

*A*. For the pure Rankine-vortex case the dynamic boundary condition,

*r*=

*R*, gives

*w*

_{1}may be chosen arbitrarily, this completely determines the inner and outer solutions. The results for the fundamental mode are shown for decreasing axial wavenumbers in Fig. 17. It is evident that the circulation cells widen with decreasing wavenumber, transitioning from a Couette-type solution that mainly affects the core, to an “open cell” solution where the outer part of the cell extends to radial infinity in the longwave limit. In addition, the flow becomes dominated by the axial perturbation velocity (Figs. 17e,f) as already inferred for the Couette solution. Regarding higher-order radial modes, the flow pattern resembles that of the Couette flow (Figs. 8c,d), except that the outermost cell intersects the RMW, “opening up” like the fundamental mode as the axial wavenumber decreases (not shown).

### d. Spiral modes in the Rankine vortex

The spiral modes in the Rankine vortex behave similarly to those in a vortex bounded by a rigid cylindrical wall (section 5). The dispersion relation for *ω* > 0 (cograde modes) is shown in Fig. 18, and for *ω* < 0 (retrograde/countergrade modes) in Fig. 19. The only fundamental difference to the bounded vortex is that now a countergrade mode appears (dashed black curves in Fig. 19). Both the phase and group speeds of this mode approach zero for small wavelengths (large *k*, Fig. 19).

The structure of the fundamental cograde modes for *m* = 1 and *m* = 2 is shown in Fig. 20, resembling the corresponding modes in the Couette flow, although the perturbation flow generally extends beyond the RMW, outside of which the waves rapidly decay. For higher-order radial modes, additional cells appear in the perturbation flow, just like in the case of the bounded vortex (not shown). The velocity perturbations are discontinuous across at *r* = *R* as indicated by the kinks in the streamlines. This is a result of the discontinuous radial derivative of the base-state azimuthal velocity (i.e., the “cusp” in the velocity profile) of the Rankine vortex and is not seen in more realistic, smooth profiles (see, e.g., Walko and Gall 1984).^{7}