2.1. Direct numerical simulations methodology

The base flow for the equilibrated vortex rings (see § 2.2) are obtained from DNS using Incompact3d with Gaussian vorticity core initial conditions. For a Gaussian distribution, the base flow azimuthal vorticity (see figure 1c) is

(2.1)\begin{equation} \bar \omega_{\theta}(s) = \frac{1}{{\rm \pi} \epsilon^2} \exp{(-s^2/\epsilon^2)}, \end{equation}

where $\epsilon = a/R$ is the slenderness ratio, a key parameter for the vortex rings considered here, with $a$ being the radius of the vortex core. The DNS for the vortex ring is carried out on the region $L_x \times L_y \times L_z = 15 \times 10 \times 10$ via ensuring errors near the outflow region do not affect the flow field in the neighbourhood of the ring. During the simulation, following Archer et al. (Reference Archer, Thomas and Coleman2008), a time varying inflow condition is enforced to maintain the ring stationary with respect to its centroid of enstrophy. Here, two distinct simulations are performed, one for obtaining the equilibrated base flow (discussed next in § 2.2), while the other for extracting the DNS growth rates. For the latter, the initial vortex ring is perturbed along the core centreline via a sum of the first $24$ Fourier components, each with unit amplitude and a random phase. For relatively thicker rings, for example, vorticity is shed from the initial ring, following nearly stationary flow before perturbations acquire significant amplitude. Growth rates are then extracted using the procedure detailed in § 3.2 during this nearly stationary flow. Note that our streamwise direction is not periodic so that there is no contamination from adjacent rings of a periodic solution.

2.2. Equilibrated vortex rings

The primary base flow for our stability calculations is also obtained via the technique described in § 2.1, where the initial velocity distribution corresponding to the Gaussian vorticity (2.1) is now evolved for a finite time in the incompressible Navier–Stokes solver without imposing any centreline perturbations. The ring eventually attains a quasi-steady state, after an initial vorticity shedding, and the velocity field at this state is extracted from the region $x \ge 5$, to ensure consistent geometric properties across all cases considered here, which is then transformed and interpolated using cubic splines to a cylindrical domain $L_x \times L_r = 10 \times 5$ (centred on the ring) and averaged azimuthally. This technique yields an equilibrated velocity field (a quasi-steady solution of Euler equations) on the cylindrical $(r,\theta ,x)$ grid (see figure 1b) for our stability calculations. This grid is clustered about the vortex ring core using a mapping strategy by Bayliss & Turkel (Reference Bayliss and Turkel1992) and Bayliss, Class & Matkowsky (Reference Bayliss, Class and Matkowsky1995). In this work cases utilizing this particular type of base flow are indicated by the notation $E_I()$ or $E_V()$, depending on whether the inviscid or viscous stability equations, respectively, are used. For the equilibrated ring, the ring and core radii are obtained from the integrals (Saffman Reference Saffman1970)

(2.2)\begin{gather} R_{\theta} = \frac{1}{\varGamma} \int{r \bar\omega_{\theta} \, {\rm d}r \, {\rm d}x}, \end{gather}

(2.3)\begin{gather}a_{\theta}^2 = 2(R_2^2 - R_{\theta}^2), \end{gather}

respectively, where

(2.4a–c)\begin{equation} R_2^2 = \frac{1}{\varGamma} \int{r^2 \bar\omega_{\theta} \, {\rm d}r \, {\rm d}x},\quad \varGamma = \int{\bar\omega_{\theta} \, {\rm d}r \, {\rm d}x}\quad\textrm{and}\quad r^2 = y^2 + z^2. \end{equation}

For these rings, the slenderness parameter is $\epsilon ^\ast = a_{\theta }/R_{\theta }$, where such equilibrated quantities for the equilibrated base flow calculations are distinguished with an $()^\ast$. In this work, results of stability analysis of the equilibrated ring are compared with the asymptotic results of Widnall & Tsai (Reference Widnall and Tsai1977) and Blanco-Rodríguez & Le Dizés (Reference Blanco-Rodríguez and Le Dizés2016), which would show the growth rates for equilibrated rings to be quite close to that from the asymptotic theories.

2.3. Gaussian vortex rings

A second type of base flow is obtained by adding a uniform axial velocity (Saffman Reference Saffman1970)

(2.5)\begin{equation} {\rm \Delta} \bar u_x = \frac{1}{4 {\rm \pi}}\left[\ln \frac{8}{\epsilon} - 0.558\right] \end{equation}

to the velocity distribution corresponding to (2.1), which we also use in this work and denote such cases as $G_I()$ or $G_V()$. Streamlines of a typical Gaussian base flow are shown in figure 1(a). Axial and radial velocity profiles on lines through the centre of the ring's core are shown in figure 1(d). It proved convenient to use this base flow to understand grid requirements and for mode convergence studies that are discussed later in appendix B.2. Gaussian vortex ring results are also used here to compare with previous DNS results of Shariff et al. (Reference Shariff, Verzicco and Orlandi1994) and Archer et al. (Reference Archer, Thomas and Coleman2008) that use a Gaussian vorticity core as the initial condition. A more detailed study of the instability of Gaussian vortex rings are found in Naveen (Reference Naveen2021).

2.4. Comparison of base flows

Recall that the asymptotic theory of Widnall & Tsai (Reference Widnall and Tsai1977) yielded an inviscid steady base flow for the vortex ring for a uniform circular core of azimuthal vorticity along with corrections to $O(\epsilon )$ and $O(\epsilon ^2)$, where the latter correction accounts for the strain rate field that supports the elliptical instability. Later, Blanco-Rodríguez et al. (Reference Blanco-Rodríguez, Le Dizés, Selçuk, Delbende and Rossi2015) obtained the corresponding base flow for a Gaussian core to the leading order. When the velocity field due to a vortex ring with a Gaussian core is obtained by solving the Poisson problem, as described in § 2.3, an $O(\epsilon ^2)$ contribution due to non-local effects gets included. A further contribution to $O(\epsilon ^2)$ is required due to the $O(\epsilon ^2)$ part of the vorticity distribution, which, however, is absent for this base flow. This local $O(\epsilon ^2)$ part of the vorticity comes from the deformation of the vortex ring core, which is due to the strain field of induced velocity. Although it naturally appears in the perturbation solutions of Widnall & Tsai (Reference Widnall and Tsai1977) and Blanco-Rodríguez et al. (Reference Blanco-Rodríguez, Le Dizés, Selçuk, Delbende and Rossi2015), being exact solutions of the Euler equation at all orders, the fact that the vorticity distribution (2.1) has only the leading order part precludes its inclusion in the calculated Poisson solutions. A fully consistent base flow is still obtained when the Gaussian vortex is allowed to equilibrate, as described in § 2.2.

For further proof, we now compare the velocity profile from the asymptotic analysis of Blanco-Rodríguez et al. (Reference Blanco-Rodríguez, Le Dizés, Selçuk, Delbende and Rossi2015) with those obtained via solving the Poisson equations. Figures 2 and 3 show this comparison for the equilibrated and Gaussian rings, respectively, up to $O(\epsilon ^2)$. This is obtained by polar transforming the individual velocity components to $(s,\phi )$ (see figure 1a) with its origin at the core centre (i.e. stagnation point), followed by azimuthal Fourier transforms on concentric circles with their centres coinciding with the core centre (more details are in Blanco-Rodríguez et al. (Reference Blanco-Rodríguez, Le Dizés, Selçuk, Delbende and Rossi2015)). In this coordinate system if the radial and azimuthal velocity components are

(2.6)\begin{equation} \bar u_s(s,\phi) = \bar u_s^{(0)}(s) + \epsilon \bar u_s^{(1)}(s,\phi) + \epsilon^2 \bar u_s^{(2)}(s,\phi) + \cdots \end{equation}

and

(2.7)\begin{equation} \bar u_\phi(s,\phi) = \bar u_\phi^{(0)}(s) + \epsilon \bar u_\phi^{(1)}(s,\phi) + \epsilon^2 \bar u_\phi^{(2)}(s,\phi) + \cdots, \end{equation}

respectively, the assumption of dipole and quadrupole fields at $O(\epsilon )$ and $O(\epsilon ^2)$ yield

(2.8a,b)\begin{equation} \bar u_s^{(1)}(s,\phi) = \check{u}_s^{(1)}(s) \sin{\phi},\quad \bar u_\phi^{(1)}(s,\phi) = \check{u}_\phi^{(1)}(s) \cos{\phi} \end{equation}

and

(2.9a,b)\begin{equation} \bar u_s^{(2)}(s,\phi) = \check{u}_s^{(2)}(s) \sin{2\phi},\quad \bar u_\phi^{(2)}(s,\phi) = \check{u}_\phi^{(2)}(s) \cos{2\phi}, \end{equation}

respectively, where lengths and velocities are scaled with $a$ and $\varGamma /(2 {\rm \pi}a)$, respectively.

Figure 2.

Comparison of velocity fields obtained from the asymptotic analysis of Blanco-Rodríguez et al. (Reference Blanco-Rodríguez, Le Dizés, Selçuk, Delbende and Rossi2015) with the equilibrated ring at (a) leading order, (b) $O(\epsilon )$ and (c) $O(\epsilon ^2)$ for $\epsilon ^\ast = 0.1866$. The radial and azimuthal velocity components are shown in thin and thick lines, respectively, with , case ${E_I5}$ from table 3 and $\mbox {- - - -}$, asymptotic analysis (Blanco-Rodríguez et al. Reference Blanco-Rodríguez, Le Dizés, Selçuk, Delbende and Rossi2015). The grey shaded region indicates the vortex core.

Figure 3.

Same as in figure 2 but for the Gaussian ring, showing , the $\epsilon = 0.1545$ case from table 6.

From figure 2, it is clear that the equilibrated ring base flow matches quite well with the asymptotic base flow of Blanco-Rodríguez et al. (Reference Blanco-Rodríguez, Le Dizés, Selçuk, Delbende and Rossi2015), at all $\epsilon$ orders, especially inside the vortex core. On the other hand, as expected, figure 3 shows that the Gaussian ring base flow profile shows large differences, more so inside the core, except at the leading order, where there is an exact match.

As the elliptic instability occurs due to the velocity field at $O(\epsilon ^2)$, the difference between the asymptotic and Gaussian base flows at this order is plotted in figure 4 for different $\epsilon$ using the relation

(2.10)\begin{equation} ()|_d = \sqrt{\{()|_t - ()|_g\}^2}, \end{equation}

where $()|_t$ and $()|_g$ denote base velocity fields from asymptotic analysis and Gaussian rings, respectively. Clearly, the difference at $O(\epsilon ^2)$ reduces as $\epsilon \rightarrow 0$ (see figure 4), although the form of the velocity profile for the Gaussian rings remains unchanged owing to the absence of core deformation for all $\epsilon$.

Figure 4.

Difference between the velocity fields obtained from the asymptotic analysis and Gaussian rings at $O(\epsilon ^2)$ for (a) radial and (b) azimuthal velocity components for different values of $\epsilon$: (thin), $0.4033$; $\mbox {- - - -}$ (thin), $0.2368$; (thick), $0.1545$ and $\mbox {- - - -}$ (thick), $0.0504$. All these cases are tabulated in table 6. The grey region shows the vortex core, while the arrow indicates the direction of increase of $\epsilon$.

2.5. Comparison with vortex dipoles

In this context, it is worthwhile to mention the work of Sipp, Jacquin & Cosssu (Reference Sipp, Jacquin and Cosssu2000) which has shown the existence of a family of quasi-steady Euler solutions for the growth of vortex dipoles. They obtained a family of vortex dipoles characterized solely via $a/b$ (core radius $a$, separation $b$), irrespective of their initial vorticity distributions (including Gaussian). This follows from the fact that with time, there is very little change in either $b$ or $\varGamma$, while $a$ follows the simple viscous diffusion law $a^2 = a_0^2 + 4\nu t$ ($a_0$ is the initial core radius). Similar evolution of peak vorticity is also observed, which follows via viscous effects. Unlike in vortex dipoles, the initial slenderness ratio $\epsilon$ of the vortex rings is a deciding factor on the nature of their subsequent evolution, which is via two distinct stages.

1. (a) Vortex shedding. Unlike thinner rings, for thicker vortex rings, the initial period of evolution is dominated by vigorous vorticity shedding leading to a skewed Gaussian distribution (from an initial Gaussian vorticity distribution), which is also sensitive to $Re$. This stage is completely absent in vortex dipoles.

2. (b) Viscous diffusion. Even the effect of viscous diffusion is different for thin and thick rings. For thicker rings, $\epsilon$ is almost unchanged, as both $a$ and $R$ change at similar rates. For the thinner ring evolution, $a$ directly follows viscous diffusion law, while $R$ remains almost unchanged, thus, very similar to vortex dipole evolution.

It is thus clear that the slenderness ratio of the vortex ring $\epsilon$ has an important role to play in its evolution (also reported by Archer et al. Reference Archer, Thomas and Coleman2008) and a family of Euler solutions (like vortex dipoles) parametrized via $\epsilon$ is not possible here. This may be further confirmed from table 3 in § 3.3 where the thicker ring cases, ${E_{I}2}$–${E_{I}4}$, after undergoing equilibration at different initial conditions lead to different solutions at the same $\epsilon ^*$, which is even true for the thinner cases (${E_{I}5}$–${E_{I}7}$), although the differences are smaller in the latter (see also figure 10 in § 3.4). Of course, if $\epsilon \ll 1$ it is possible to obtain a family of quasi-steady Euler solutions, but our present work is not restricted to this limit.

3.1. Numerical methods and solutions

Both our base flow Poisson equation solutions and DNS use Incompact3d, which implements high-resolution, sixth-order compact differences for derivatives and third-order Runge–Kutta for time stepping to solve the incompressible Navier–Stokes equations. The solver is known to scale well for up to ${O}(10^5)$ processors (Laizet & Li Reference Laizet and Li2011). The primary purpose of the DNS is to compare some of the growth rates extracted from these studies with the modal growth rates from our stability calculations, as means of validating the latter (see § 3.2). We also compare our results with the earlier DNS of Shariff et al. (Reference Shariff, Verzicco and Orlandi1994) and Archer et al. (Reference Archer, Thomas and Coleman2008), both of whom had used second-order finite differences in their simulations.

The matrices of the stability equations (A 2) and (A 4) (see (A 5), (A 6), (A8), (A9) and (A11) in appendix A) are discretised using Chebyshev polynomials at their respective Gauss–Lobatto points. In our parallel computations, sub-matrices of (A 5), (A 6) and (A11) are first obtained via the PETSc libraries (see Balay et al. Reference Balay, Gropp, McInnes, Smith, Arge, Bruaset and Langtangen1997; Balay Reference Balay2015a,Reference Balayb) which are then solved using the SLEPc libraries (see Hernandez, Roman & Vidal Reference Hernandez, Roman and Vidal2005; Roman et al. Reference Roman, Campos, Romero and Tomas2017) and MATLAB, which uses the Krylov–Schur iterative algorithm with shift-invert spectral transformation to overcome the singular nature of the matrix (A11).

Boundary conditions on $\hat {\boldsymbol {q}}$ are specified at the $r = 0$ plane as compatibility conditions (Batchelor & Gill Reference Batchelor and Gill1962; Khorrami, Malik & Ash Reference Khorrami, Malik and Ash1989) at the $r= L_r$ and $x = 0$ surfaces to be $\hat {\boldsymbol {q}} = 0$, while values at $x = L_x$ are obtained via linearly extrapolating from the two neighbouring interior points.

The stability solver is checked first via reproducing the stability spectrum for the classical Hagen–Poiseuille flow (see appendix B.1). Next, convergence of modes and growth rates are examined with respect to the Gaussian core rings, using cases from table 4 in § 3.3. In each case of our vortex ring stability calculations, as discussed in this section, converged eigenvalues are identified for a specified tolerance with spurious ones discarded first by varying the number of points along the radial and axial directions and then using different shift values during the iterative procedure (full details appear in appendix B.2). The results presented in this work are from discretisations with at least 100 points along both $x$ and $r$ directions. Our inviscid calculations are converged up to a numerical tolerance of at least $10^{-2}$, with the thinner rings showing better convergence (${\sim }10^{-4}$), while the viscous calculations are converged up to at least $10^{-4}$.

A typical eigenspectrum from these calculations are shown in figure 5 for a Gaussian core vortex ring of $\epsilon = 0.4131$ (see case ${G_{V}1}$ of table 4 in § 3.3) with $n = 6$, $N_x \times N_r = 100 \times 100$ and $s_h = 0.1$. The three unstable discrete eigenmodes are separately marked in the figure, which are now classified based on their respective frequencies $\sigma$. The one with $\sigma = 0$ is labelled as the stationary mode ${ S}$. The other two modes, appearing in pairs, with $\sigma \neq 0$ are referred to as rotating modes ${ R}$. Depending upon their sense of rotation, these may be further classified into clockwise ($\sigma /n < 0$) and counter-clockwise (${\sigma /n > 0}$) rotating modes. In this work the zero frequency stationary modes appear to be the dominant mode type; therefore, this is the mode type we will be referring to throughout unless explicitly mentioned otherwise.

Figure 5.

The converged eigenspectrum of a Gaussian core vortex ring with $\epsilon = 0.4131$ (case ${G_{V}1}$ of table 4 in § 3.3) for $n=6$, $N_x \times N_r = 100 \times 100$ and $s_h = 0.1$, identifying the following modes: $\bullet$, stationary mode ${ S}$; $\blacksquare$, rotating modes ${ R}$ and $\circ$, stable modes with , $\alpha = 0$ and $\mbox{- - - -}$, $\sigma = 0$.

3.2. Comparing global modes with DNS

We first compare our global stability results with DNS that should provide further confidence in our stability calculations. The comparison is done between two pairs of cases, whose details are in table 1, the first for a thick ring at $\epsilon ^\ast = 0.315$ and then for a thin ring at $\epsilon ^\ast = 0.213$.

Table 1.

Growth rates of the fastest growing modes at $Re = 5500$ extracted from DNS during an interval of linear growth (cases ${ D1}$ and ${ D2}$) compared with the corresponding growth rates from modal analyses over the same interval (cases ${ {E_V1}}$ and ${ {E_V2}}$) (see text).

Growth rates are computed from the DNS solutions of a perturbed ring, as described in § 2.1, by first interpolating the velocity fields onto a cylindrical domain $L'_x \times L_r = 5 \times 5$, centred on the ring. The energy of perturbations in this cylindrical domain is

(3.1)\begin{equation} E = \frac{1}{2}\int_{-L'_x/2}^{L'_x/2}\int_{0}^{L_r}\int_{0}^{2 {\rm \pi}} [{u'_r}^2 + {u'_{\theta}}^2 + {u'_x}^2] r\,{\rm d}\theta \,{\rm d}r \,{\rm d}x, \end{equation}

where the primed quantities denote the velocity perturbations obtained via subtracting the instantaneous velocity from the initial unperturbed velocity field. Now, on taking the azimuthal Fourier transform of the velocity fields, the energy of mode $n$ is simply

(3.2)\begin{equation} E_n = {\rm \pi}\int_{-L'_x/2}^{L'_x/2}\int_{0}^{L_r} [\mathcal{F}\{u'_r\}^2_n + \mathcal{F}\{u'_{\theta}\}^2_n + \mathcal{F}\{u'_x\}^2_n]r \,{\rm d}r \, {\rm d}x, \end{equation}

where $\mathcal {F}\{*\}_n$ denotes the Fourier coefficient of mode $n$. The growth rate of mode $n$ is then $\alpha _n = (1/2)(1/E_n)\,{\rm d}E_n/{\rm d}t$. The growth rates so-obtained are averaged over the interval $45 \le t \le 60$, where linear growth is observed in the simulations, which is thus expected to correspond to the modal growth rates obtained next. The DNS cases are listed as ${D()}$ in table 1.

For the stability cases of table 1, in order to use a comparable base flow, the velocity field of the unperturbed equilibrated vortex ring is extracted at the midpoint of the aforementioned DNS averaging interval ($t=52.5$). Here, after an initial rapid decrease of core radius from $\epsilon = 0.4131$ due to vigorous vorticity shedding, followed by a steady increase due to viscous diffusion of vorticity, at $t^\ast = 52.5$, the slenderness ratio attains $\epsilon ^\ast = 0.315$, based on local values of ring radius and core size, as computed using (2.2) and (2.3). A second pair of cases ($D2$ and ${E_V2}$) is done starting from a thinner ring ($\epsilon = 0.2$ at $t=0$), which grows slightly to $\epsilon ^\ast = 0.213$ at $t^\ast =52.5$.

Table 1 clearly shows that in both cases the maximum growths occur at identical azimuthal mode orders ($n = 6$ and $9$, respectively) between the DNS and stability analyses while the actual maximum growth rates $\alpha$ easily match up to two decimal places. Growth rates for several azimuthal modes for the above thick and thin rings are shown in figure 6, which re-emphasizes the excellent match for the respective fastest growing modes (as expected), while also showing the lesser agreements with the other modes. For the thinner vortex in figure 6(b), the range of unstable modes is higher as $n=16$ is also unstable. Note that the modes $n = 5, 6$ in figure 6(b) are actually rotating modes, while the rest are stationary.

Figure 6.

Comparison of growth rates from DNS and stability calculations for the cases of table 1 showing (a) for $\epsilon ^{\ast } = 0.315$: $\circ$, $D1$; $\blacktriangle$, $E_V1$ and (b) for $\epsilon ^{\ast } = 0.213$: $\circ$, $D2$; $\blacktriangle$, $E_V2$. In both figures the $n$ corresponding to the maximum $\alpha$ is indicated via a vertical line.

3.3. Comparison with previous calculations: viscous growth rate

In this section we summarize our most important stability results where we directly compare these against other relevant data for vortex rings (see table 2). Here, we include the results of Shariff et al. (Reference Shariff, Verzicco and Orlandi1994) and Archer et al. (Reference Archer, Thomas and Coleman2008), who extracted growth rates from their DNS (viscous) calculations, prior to breakdown to turbulence. We also compare against the work of Gargan-Shingles et al. (Reference Gargan-Shingles, Rudman and Ryan2016), who report growth rates from solutions of linearised Navier–Stokes equations for $Re=10\,000$, while our inviscid results are matched against classical analytical solutions of Widnall & Tsai (Reference Widnall and Tsai1977) and also with the recent calculations of Blanco-Rodríguez & Le Dizés (Reference Blanco-Rodríguez and Le Dizés2016). Table 2 lists a few of these selected cases, all with initial $\epsilon = 0.4131$ since the DNS data of Shariff et al. (Reference Shariff, Verzicco and Orlandi1994) and Archer et al. (Reference Archer, Thomas and Coleman2008) are available only at this slenderness ratio.

Table 2.

Growth rates and azimuthal modes of the fastest growing modes for viscous and inviscid vortex rings with $\epsilon = 0.4131$. The base flow types are either Gaussian ‘$G$’ or equilibrated ‘$E$’, where for the latter type growth rates are extracted at $Re^\ast$ and $\epsilon ^\ast$ for stability calculations. For cases from the literature, numbers in parenthesis indicate the original case designations. For cases 1 and 2, the equilibrated relations of (3.6) and (3.7), respectively, are applied and $n = \kappa /\epsilon$ with $\kappa = 2.261$ (see appendix C).

At the inviscid limit, the growth rates from our stability calculations with equilibrated rings appear to be about 1 % smaller and 9 % larger than the asymptotic theory results of Widnall & Tsai (Reference Widnall and Tsai1977) (corrected for equilibrated core via 3.6) and Blanco-Rodríguez & Le Dizés (Reference Blanco-Rodríguez and Le Dizés2016) (compare cases 1 and 2, respectively, with our case 3 in table 2). This difference, explored in the next section, reduces as $\epsilon \rightarrow 0$, which is where the theories of Widnall & Tsai (Reference Widnall and Tsai1977) and Blanco-Rodríguez & Le Dizés (Reference Blanco-Rodríguez and Le Dizés2016) are strictly valid. Note here that cases 3 and 6 of table 2 are, respectively, inviscid and viscous stability results from the same base flow, obtained by equilibrating the Gaussian ring of $\epsilon = 0.4131$ at $Re = 10\,000$, using procedures discussed in § 2.2.

The viscous growth rates of the equilibrated rings obtained from our stability calculations match quite well with the maximum growth rates of Archer et al. (Reference Archer, Thomas and Coleman2008) (compare case 4 with our case 6 for $Re = 10\,000$ and case 10 with our cases 11 and 12 for $Re = 5500$ in table 2). In fact, this was quite expected after our discussion in § 2.4 and observations in § 3.2 where our DNS calculations showed a good match with the corresponding equilibrated base flow stability results. At $Re = 10\,000$, the calculated equilibrated ring growth rate (case 6 in table 2) falls in between those of Archer et al. (Reference Archer, Thomas and Coleman2008) and Gargan-Shingles et al. (Reference Gargan-Shingles, Rudman and Ryan2016) (compare our case 6 with cases 4 and 5 in table 2). Here, Gargan-Shingles et al. (Reference Gargan-Shingles, Rudman and Ryan2016) evolved perturbations via linearised Navier–Stokes over an arbitrary time period while growth rates were found from changes in perturbation amplitudes over that period, so that their growth rates are slightly higher than Archer et al. (Reference Archer, Thomas and Coleman2008). On the other hand, maximum growth rates obtained by Shariff et al. (Reference Shariff, Verzicco and Orlandi1994) are larger (compare between cases 8 and 10 of table 2), which has been attributed by Archer et al. (Reference Archer, Thomas and Coleman2008) to the use of streamwise periodic conditions by Shariff et al. (Reference Shariff, Verzicco and Orlandi1994) that allowed vorticity shed from the adjacent ring to alter the flow around the original ring, especially at later times when (linear) growth rates were extracted. Moreover, even though Shariff et al. (Reference Shariff, Verzicco and Orlandi1994) started with a Gaussian core distribution in their rings, by the time their growth rates were extracted, the corresponding velocity profiles would have rather resembled some sort of an ‘equilibrated’ distribution. Our stability calculations with Gaussian core clearly shows this where the $\alpha$ obtained from our case 9 of table 2 does not match the corresponding case 8 of Shariff et al. (Reference Shariff, Verzicco and Orlandi1994). Note here that, in their work Shariff et al. (Reference Shariff, Verzicco and Orlandi1994) had obtained nearly identical growth rates for both Gaussian and equilibrated-core vortex rings with negligible differences in their energy evolution post the initial transients. At the higher $Re$, the viscous case 7 of table 2 is not allowed to equilibrate before its base flow is used for stability calculations, whose computed maximum $\alpha$ is again clearly different to those seen in cases 4–6 of table 2.

A comparison of eigenfunctions between equilibrated and Gaussian core vortex rings, shown in figure 7 for cases 12 and 7, respectively, of table 2, are qualitatively quite similar even though these are at different $Re$. The process of equilibration although yields a slightly smaller core for case 12 (see figure 7a). Note that these are typical eigenfunctions for vortex rings subjected to elliptic instability, which are also reported elsewhere (Gargan-Shingles et al. Reference Gargan-Shingles, Rudman and Ryan2016; Hattori et al. Reference Hattori, Blanco-Rodríguez and Le Dizés2019). A couple of eigenfunctions of the rotating modes that look qualitatively similar to the spiral modes of Hattori et al. (Reference Hattori, Blanco-Rodríguez and Le Dizés2019) are shown in figure 8. We could track these modes only for the viscous Gaussian cases (see table 4), while neither for the corresponding inviscid cases nor for any of the equilibrated cases were these found. In this work we make no attempt to study these modes any further, whereas a more detailed study can be found in Naveen (Reference Naveen2021). Eigenfunctions for case 9 of table 2 are shown in appendix C in the context of establishing a connection between the vortex ring and line vortex dynamics.

Figure 7.

Viscous eigenfunctions for (a) case 12 and (b) case 7 of table 2 showing eight equally spaced $\textrm {Re}(\hat {u}_{\theta })$ contours spanning $\pm 0.02$ (excluding zero). The negative contours are dashed. The grey disk is the corresponding vortex core, shown in figure 1(a).

Figure 8.

Eigenfunctions of a Gaussian vortex ring with $\epsilon = 0.2141$ for (a) $Re = 20\,000$, $n = 14$, $\omega = 0.3257 + 0.1175{\rm i}$ and (b) $Re = 10\,000$, $n = 14$, $\omega = 0.3317 + 0.1009{\rm i}$, both from table 6, showing eight equally spaced $\textrm {Re}(\hat {u}_{\theta })$ contours spanning ${\pm }0.005$ (excluding zero). The rest is the same as figure 7.

Table 3.

Details of global stability cases for the equilibrated vortex rings as used in figure 9 including several parameters during the time of equilibration listed with an $()^\ast$. The stability analyses results shown in figure 9 are performed at corresponding $Re^\ast$ values. Here $N_y = N_z$ for all cases.

Table 4.

List of global stability cases for the Gaussian vortex rings with $\epsilon = 0.4131$ as used in figure 9 and in appendix B.2. Only the azimuthal mode $n$ that has the maximum growth rate $\alpha$ is listed. The dimensions and mesh along the $y$ and $z$ directions are identical. The last three cases are only used for the grid and domain convergence studies, discussed in appendix B.2.

Figure 9 includes more cases from Shariff et al. (Reference Shariff, Verzicco and Orlandi1994), Archer et al. (Reference Archer, Thomas and Coleman2008) and our stability calculations (listed in tables 3, 4 and 6), which additionally shows a set of viscous growth corrections. For the latter, Shariff et al. (Reference Shariff, Verzicco and Orlandi1994) proposed an empirical correction

(3.3)\begin{equation} \alpha = \alpha_{W}(\epsilon)\left [1 - \frac{\hat{\alpha}_1}{\widehat{Re}}\right ], \end{equation}

where $\alpha _{W}$ is the inviscid growth rate of the ring, due to Widnall & Tsai (Reference Widnall and Tsai1977) that include corrections due to a Gaussian core vorticity, explained later in §§ 3.4 and 3.5 where we discuss inviscid solutions (see 3.6 and 3.8, respectively, for equilibrated and Gaussian rings), with

(3.4)\begin{equation} \widehat{Re} = Re \frac{3 \epsilon_1^2}{16 {\rm \pi}}\left[\ln\left(\frac{8}{\epsilon_e}\right)-\frac{17}{12}\right], \end{equation}

based on strain rates proposed by Saffman (Reference Saffman1978). Here, for equilibrated rings, $\epsilon _1 = a^*_1/R^*_{\theta }$ and $\epsilon _e = a^*_e/R^*_{\theta }$ (see table 3), while for Gaussian rings $\epsilon _1 = 1.12141 \epsilon$ and $\epsilon _e = 1.3607 \epsilon$. Furthermore, in (3.3), Shariff et al. (Reference Shariff, Verzicco and Orlandi1994) used $\hat {\alpha }_1 =18$, as calculated by them by varying $\widehat {Re}$ and $\epsilon$ over the following ranges: $19 \le \widehat {Re} \le 110$ and $0.2066 \le \epsilon \le 0.4131$, while Archer et al. (Reference Archer, Thomas and Coleman2008) proposed a lower value $\hat {\alpha }_1=8$ (both corrections shown in figure 9). In contrast, our equilibrated core vortex ring calculations yield $\hat {\alpha }_1=11.66$ in (3.3), obtained using growth rate data from the corresponding viscous stability analysis, discussed later in § 3.4 and tabulated here in table 3.

Figure 9.

Growth rates of the most unstable mode for viscous vortex rings, showing the DNS cases of , Shariff et al. (Reference Shariff, Verzicco and Orlandi1994) and , Archer et al. (Reference Archer, Thomas and Coleman2008) compared with our stability calculations with equilibrated ($\bullet$, $\epsilon ^\ast = 0.2899$; , $\epsilon ^\ast = 0.1866$) and Gaussian ($\circ$, $\epsilon = 0.4131$; , $\epsilon = 0.4033$; , $\epsilon = 0.3$; , $\epsilon = 0.2141$) core rings. The viscous correction lines shown are (see 3.3) (thin), $1-18/\widehat {Re}$ (Shariff et al. Reference Shariff, Verzicco and Orlandi1994) and $\mbox {- - - -}$ (thin), $1-8/\widehat {Re}$ (Archer et al. Reference Archer, Thomas and Coleman2008) compared with (3.5), showing (thick), $1-11.66/\widehat {Re}$ (equilibrated) and $\mbox {- - - -}$ (thick), $0.709-12.61/\widehat {Re}$ (Gaussian). The critical Reynolds numbers ($\widehat {Re}_c$) for each of the viscous corrections are also indicated.

In fact, our calculated viscous growth rate data, shown in figure 9, supports a new two-parameter formula

(3.5)\begin{equation} \alpha = \alpha_{W}(\epsilon)\left [\hat{\alpha}_2 - \frac{\hat{\alpha}_1}{\widehat{Re}}\right ],\end{equation}

with $\hat {\alpha }_1 = 11.66$, $\hat {\alpha }_2 = 1.0$ for equilibrated rings and $\hat {\alpha }_1 = 12.61$, $\hat {\alpha }_2 = 0.709$ for Gaussian rings. For the Gaussian ring data in figure 9, we use the first three cases of table 4 and other viscous cases corresponding to vortex rings of $\epsilon = 0.4033$, $0.3$ and $0.2141$, all from table 6 in § 3.5, where for each $\epsilon$, stability analyses are performed for $Re = 20\,000$, $10\,000$ and $5500$. Note here that for all these viscous cases, the maximum growth azimuthal mode $n$ matches the corresponding inviscid $n$ listed in table 6. The value of $\hat {\alpha }_2 = 0.709$ for the Gaussian rings implies a $\sim$30 % difference when compared with the theory. Again, this growth rate mismatch for Gaussian core rings was expected as the asymptotic theories effectively yield equilibrated base rings and so there is a significant error for the Gaussian rings (recall the discussion in § 2.4), which is confirmed here from their respective growth rates in table 2. As we investigate in the next two sections, this difference is reconfirmed to be due to the differences in their respective velocity profiles (see § 2.4). Note that $\hat {\alpha }_1$ (i.e. the slopes of 3.3 and 3.5) from our global stability results are almost identical across equilibrated and Gaussian rings, while they are in between the values obtained by Archer et al. (Reference Archer, Thomas and Coleman2008) and Shariff et al. (Reference Shariff, Verzicco and Orlandi1994). The slope for our cases also matches reasonably well with the small viscosity extensions of Widnall & Tsai (Reference Widnall and Tsai1977) theory, as proposed by Fukumoto & Hattori (Reference Fukumoto and Hattori2005). In their version of (3.3), $\hat {\alpha }_1$ is a weak function of $\epsilon$, so that for $\epsilon = 0.4131$ and $0.2141$, we get $\hat {\alpha }_1 = 12.63$ and $13.39$, respectively, for a ring corrected for Gaussian vorticity distribution. The critical Reynolds number based on the strain $\widehat {Re}_{c}$, calculated from our viscous corrections (also marked in figure 9), lies in between those obtained by Shariff et al. (Reference Shariff, Verzicco and Orlandi1994) and Archer et al. (Reference Archer, Thomas and Coleman2008).

3.4. Sensitivity to base flows

We may expect vortex rings equilibrated at different $Re$ for a given $\epsilon$ to be close to the asymptotic solution (though inviscid), but they prove to be sufficiently different to effect significant differences in growth rates, as we show in this section. Here, vortex rings with Gaussian core distributions are evolved via procedures described in § 2.2 at $Re = 5500$, $10\,000$ and $20\,000$ till a stationary field with an equilibrated vortex is obtained. Table 3 in § 3.3 shows two sets of such base flow data corresponding to a thicker $\epsilon ^\ast = 0.2899$ and a thinner $\epsilon ^\ast = 0.1866$ equilibrated ring. In each of these cases the initial value of $\epsilon$ is suitably chosen to reach one of these $\epsilon ^\ast (= a^*_{\theta }/R^*_{\theta })$ values at time $t^\ast$, irrespective of the initial $Re$. Inviscid stability analysis is then performed with the equilibrated base flows, which for the same slenderness ratios $\epsilon ^\ast$ are only slightly different, nearly steady solutions of the Navier–Stokes equations (shown in figure 10). The small differences in the vorticity fields are because they have evolved and become stationary at different $Re^\ast$ (see table 3). Table 3 also lists other geometric parameters of the rings, including $a^*_1$ and $a^*_e$, which are extracted from our simulations using methods suggested by Archer et al. (Reference Archer, Thomas and Coleman2008) and used in the viscous growth correction expressions of § 3.3. Furthermore, $R^*_c$ is the distance from the centre of the ring to the location where $\bar u_x(r,x_c) = 0$, the stagnation point in axial velocity, used in the theory of Blanco-Rodríguez et al. (Reference Blanco-Rodríguez, Le Dizés, Selçuk, Delbende and Rossi2015) as an estimate for the equilibrated ring radius.

Figure 10.

Base flow of equilibrated vortex rings showing five equally spaced azimuthal vorticity contours for (a) $\epsilon ^\ast = 0.2899$, spanning from $-0.2$ to $-2$ (outside to inside) for , ${E_I2}$; $\mbox {- - - -}$, ${E_I3}$; ${\mathinner {\cdotp \cdotp \cdotp \cdotp \cdotp \cdotp }}$, ${E_I4}$ and (b) $\epsilon ^\ast = 0.1866$, spanning from $-0.5$ to $-4.5$ (outside to inside) for , ${E_I5}$; $\mbox {- - - -}$, ${E_I6}$; ${\mathinner {\cdotp \cdotp \cdotp \cdotp \cdotp \cdotp }}$, ${E_I7}$ cases of table 3. In each figure the grey disk is the vortex core after equilibration.

The growth rate results for these cases tabulated in table 5 show a ${\sim }9\,\%$ difference in growth rates for the thicker $\epsilon ^\ast =0.2899$ ring and ${\sim }3.5\,\%$ for the thinner ring when evolved from different initial conditions and at different $Re^\ast$. Growth rates from the asymptotic theory found using (3.8) with $\epsilon ^\ast$ are still ${\sim }14$ %–25 % higher than the growth rates obtained from our global stability analysis. However, in order to use (3.8) for the equilibrated rings two corrections are incorporated: $(a)$ the whole expression is multiplied with $\varGamma ^*/R^{*2}_{\theta }$ and $(b)$ different values of correction factor $f^\ast = a^*_1/a^*_{\theta }$ are used in contrast to the constant $f$ of (3.8). The first correction takes into account the change in circulation and radius of the ring, while the second one accounts for the change in core shape. This corrected growth rate is then

(3.6)\begin{equation} \alpha^*_{W} = \frac{\varGamma^*}{2 {\rm \pi}R^{*2}_{\theta}} \sqrt{\left(0.428 \ln{\left(\frac{8}{f^\ast \epsilon^\ast}\right)} + 0.0788 - 0.534\right)^2 - 0.3367^2}, \end{equation}

with $f^\ast$ tabulated in table 3 for different rings. This corrected theoretical growth rate is now at the worst about 3.5 % higher than the growth rates obtained from our global stability analysis with the thinner ring while at the best just about 0.1 % lower for the thicker rings (see table 5). The corresponding corrected growth rate for (3.9) is

(3.7)\begin{equation} \alpha^*_{B} = \frac{\varGamma^*}{2 {\rm \pi}R^{*2}_c} \left(0.5171 \ln{\left(\frac{8 R^*_c}{a^*_{\theta}}\right)} - 0.9285\right), \end{equation}

which yields growth rates from global stability analysis to be within ${\sim }3$ % of the theoretical values of Blanco-Rodríguez & Le Dizés (Reference Blanco-Rodríguez and Le Dizés2016) for the thinner equilibrated rings (see table 5). Although the differences with (3.7) are slightly higher than when using (3.6), the errors largely reduce for the thinner ring considered here. Note that unlike in § 3.5, we do not attempt to obtain $\alpha ^{ m}$ (see figure 11 and table 6) over a range of $\epsilon$ for the equilibrated rings here, as there is a better match with both the theories for the chosen values of $\epsilon ^\ast$.

Figure 11.

Sensitivity of inviscid growth rates over varying ranges of $\epsilon$ corresponding to several azimuthal mode numbers $n$ (maximum growths) shown for , $n = 6$; , $n = 8$; , $n = 10$; , $n = 11$; , $n = 15$; , $n = 22$; , $n = 35$; , $n = 45$, where $\mbox {- - - -}$ in each indicate extrapolations by fitting on a parabola. The inviscid growth rates from asymptotic theories are indicated by thick grey lines including , Widnall & Tsai (Reference Widnall and Tsai1977) (see 3.8) and $\mbox {- - - -}$, Blanco-Rodríguez & Le Dizés (Reference Blanco-Rodríguez and Le Dizés2016) (see 3.9). Insets A and B, as labelled, are magnified views for $n = 45$ and $n = 35$ cases.

Table 5.

Sensitivity of the calculated inviscid growth rates for the cases of table 3 compared with the asymptotic theories of Widnall & Tsai (Reference Widnall and Tsai1977) and Blanco-Rodríguez & Le Dizés (Reference Blanco-Rodríguez and Le Dizés2016).

Table 6.

Peak growth rates $\alpha ^{ m}$ for different azimuthal mode numbers $n$, as shown in figure 11 compared with the $\alpha _{W}$ and $\alpha _{B}$ of the asymptotic theories of Widnall & Tsai (Reference Widnall and Tsai1977) and Blanco-Rodríguez & Le Dizés (Reference Blanco-Rodríguez and Le Dizés2016), respectively. Here, $\kappa = 2.261$ at the intersection point ${ L_1}$ in figure 17(a), but for an inviscid line vortex. Also, $N_y = N_z$.

3.5. Sensitivity to slenderness ratio

As the slenderness ratio $\epsilon$ decreases, the azimuthal mode number $n$ with the largest growth rate increases (see table 1), which is the azimuthal variation that soon dominates the dynamics. For a line vortex in a strain field, Tsai & Widnall (Reference Tsai and Widnall1976) had found instability to occur for a band of axial wavenumbers, whose width is proportional to the strength of the strain field $\tau$, about the wavenumber at the intersection of the respective dispersion curves (see also the discussion in appendix C). The width of this band does not vary with $\epsilon$, but as $\epsilon$ becomes smaller the increased $n$ ensures resonance over smaller ranges of $\epsilon$. In table 6, results from inviscid, global stability analyses for a large number of vortex rings confirm this expectation of $(a)$ instability with the same azimuthal mode $n$ for a range of $\epsilon$ with a local peak $\alpha ^{ m}$ in the growth rate, and $(b)$ that this range diminishes monotonically as $\epsilon \rightarrow 0$ (visually depicted in figure 11). For example, it suffices to obtain growth rates for $n=6$ when $\epsilon = 0.4131$, though there are other modes with non-negative but smaller growth rates. Such fastest growing modes are tracked with varying $\epsilon$ in figure 11. The domain size for base flow computations in all these cases are identical to case ${ {G_I1}}$ of table 4, while the corresponding grid sizes are as in table 6 with $N_z = N_y$. Note that for very thin rings much finer grids are needed in our Poisson solver to obtain the base flow. For the corresponding stability analyses, e.g. for the $n = 6$ cases, $N_x \times N_r = 120 \times 120$ points are used, while for the others $N_x \times N_r = 100 \times 100$ points suffice since the growth rates converged much faster as $\epsilon$ is reduced (see also figure 12).

Figure 12.

Inviscid eigenfunction for (a) $n = 6, \epsilon = 0.4033$, showing eight equally spaced contours spanning ${\pm }0.0008$ (excluding zero) and (b) $n = 45, \epsilon = 0.0504$, showing eight equally spaced contours spanning ${\pm }0.015$ (excluding zero), both from table 6. The rest are the same as in figure 7.

As $\epsilon$ is varied, growth rates for the several azimuthal modes $6 \le n \le 45$ shown in figure 11, clearly indicate $\alpha ^{ m}$ changes little for thicker rings ($\epsilon \approx 0.4$). We observe the sensitivity to increase monotonically as $\epsilon \rightarrow 0$. If the line vortex results of appendix C are used for the rings, we may expect the value of $\epsilon$ at which vortex ring perturbations have the maximum growth rate to be close to $\kappa /n$, where $\kappa = 2.261$ is the relevant inviscid resonance wavenumber. From the values of $\kappa /n$ listed in table 6, we observe the difference $|\epsilon - \kappa /n|$ to be about 7 % of $\epsilon$ for $n=6$, but only 0.3 % for $n=45$. Such observations lead to expectations that growth rate estimates from the asymptotic theory should be better for thinner rings. But, our computed values from table 6 clearly show that, for all ring thicknesses, growth rates from our stability analyses $\alpha ^{ m}$ are about 24 % to 33 % smaller than the corresponding $\alpha _{W}$ from the theoretical prediction (Widnall & Tsai Reference Widnall and Tsai1977) with

(3.8)\begin{equation} \alpha_{W} = \frac{1}{2 {\rm \pi}} \sqrt{\left(0.428 \ln{\left(\frac{8}{f \epsilon}\right)} + 0.0788 - 0.534\right)^2 - 0.3367^2}, \end{equation}

where $f=1.12141$ corrects for the Gaussian core (proposed by Shariff et al. (Reference Shariff, Verzicco and Orlandi1994)). Furthermore, when our growth rates are compared with the theoretical growth rates of Blanco-Rodríguez & Le Dizés (Reference Blanco-Rodríguez and Le Dizés2016),

(3.9)\begin{equation} \alpha_{B} = \frac{1}{2 {\rm \pi}} \left(0.5171 \ln{\left(\frac{8}{\epsilon}\right)} - 0.9285\right), \end{equation}

our $\alpha ^{ m}$ are still about 14 % to 33 % smaller (see table 6). Obviously, this is due to the absence of the critical $O(\epsilon ^2)$ part in the Gaussian vorticity distribution, as discussed in § 2.4. Even though as $\epsilon \rightarrow 0$, the differences between the Gaussian and asymptotic base profiles reduce (see figure 4) for all the cases discussed here and listed in table 6 the corresponding velocity profiles are qualitatively similar to that shown in figure 3. As these calculated growth rates are shown to be extremely sensitive to the exact details of the velocity profiles (see § 3.4), it is not surprising to observe the huge differences in growth rates, even for rather small-$\epsilon$ Gaussian rings, for the cases in table 6.

In this context it is interesting to compare the inviscid eigenfunctions for the two extreme cases of table 6 ($n = 6$ and $n = 45$, shown in figure 12a and b, respectively), which are still qualitatively similar to the viscous eigenfunctions of figure 7. For the thicker rings, there appears to be an asymmetry in the eigenfunctions about the respective core centres (see figures 7 and 12a), which is absent for the thinner ring of figure 12(b), more closely resembling that of a line vortex (compare figure 12b with figure 18b in appendix C). The small oscillations in the eigenfunction of figure 12(a) are due to relatively poor convergence.

3.6. Effects of viscosity

The effects of viscosity on the vortex rings are qualitatively similar for both types of base flow explored in this work, as our fitted viscous growth correction curves defined via (3.5) seem to attest, with similar slopes for the two ($\hat {\alpha }_1= 11.66$ for equilibrated and $12.61$ for Gaussian rings). Hence, for brevity, here we study the effects of viscosity on growth rates of the Gaussian rings only. Growth rates listed in table 4 show a monotonic rise with increasing Reynolds number for the same mode $n$. The effect of viscosity is similar for other unstable modes too. Growth rates for a range of $n$ is shown in figure 13 for cases ${G_V1}$, ${G_{V}2}$ and ${G_{V}3}$ of table 4. Unstable modes exist for $n \ge 5$, but the range of unstable modes decreases as $Re$ decreases. At $Re = 10\,000$ (case ${G_{V}3}$), the growth rate peaks at $n=6$, $10$ and $14$, while as $Re$ is lowered to $Re = 5500$, we find unstable modes only for $n \le 12$ with peaks at $n = 6$ and $10$ (case ${G_V1}$). At the still lower value of $Re = 2500$ (case ${G_{V}2}$), the corresponding $n \le 7$ with a peak at $5$. The distribution of growth rates with multiple peaks can be understood from the resonance mechanism discussed in Widnall & Tsai (Reference Widnall and Tsai1977), which is discussed in the context of stability results for line vortices in appendix C.

Figure 13.

Effects of viscosity on growth rates of the most unstable mode of the vortex ring with $\epsilon = 0.4131$, $\sigma =0$ for azimuthal modes $0 \le n \le 15$ shown for $\square$, case ${G_V1}$; $\vartriangle$, case ${G_{V}2}$ and $\circ$, case ${G_{V}3}$ of table 4. In addition, DNS data of $\blacklozenge$, case 1 of Shariff et al. (Reference Shariff, Verzicco and Orlandi1994) and $\blacktriangledown$, case ${ A1}$ of Archer et al. (Reference Archer, Thomas and Coleman2008) are shown (see table 2). Vertical lines indicate the peaks $n=6, 10$ and $14$.

Figure 13 also shows results from the DNS of Shariff et al. (Reference Shariff, Verzicco and Orlandi1994) and Archer et al. (Reference Archer, Thomas and Coleman2008) for the conditions of ${G_V1}$ of table 4. While the results of Archer et al. (Reference Archer, Thomas and Coleman2008) are closer to ours, growth rates of Shariff et al. (Reference Shariff, Verzicco and Orlandi1994) are consistently larger at all $n$ except $n=9$, but the peaks still occur at the same $n$. This is not unexpected after the discussions in § 3.3 where our stability calculations are shown to match better with the cases of Archer et al. (Reference Archer, Thomas and Coleman2008) for both Gaussian and equilibrated cases (see table 2). Note here that Shariff et al. (Reference Shariff, Verzicco and Orlandi1994) extracted unstable modes for $n = 1$ and $n = 4$, though not for $n=2$ or $n=3$, whereas our modal analysis yields stable stationary modes for $n < 5$, although we do find unstable rotating modes for $n = 3$ and $n =4$ with smaller growth rates.