Strato-Rotational Instability in Couette Flow

• Strato-Rotational Instability (SRI) is a hydrodynamic instability in Taylor–Couette flow where stable axial stratification enables growing non-axisymmetric spiral modes.
• It is governed by critical values of nondimensional numbers (Reynolds, Brunt–Väisälä, and Froude) which define a bounded instability window and favor an m=1 dominant mode.
• Experimental and numerical studies reveal amplitude modulations and transitions between upward and downward propagating spirals, underlining its relevance to astrophysical and geophysical flows.

Strato-Rotational Instability (SRI) is a purely hydrodynamic instability of Taylor–Couette flow with a stable axial density or temperature stratification. In its classical form, it occurs for non-axisymmetric perturbations in centrifugally stable anticyclonic shear, so that axial stratification and rotation, each stabilizing when considered separately, together permit growing spiral modes (Rüdiger et al., 2016). Across linear theory, laboratory observations, direct numerical simulation (DNS), and long-time computation, SRI has emerged as a canonical instability of rotating-stratified shear flow, with relevance to quasi-Keplerian configurations, accretion-disk analogues, and rotating geophysical fluids (Meletti et al., 2020).

1. Definition and theoretical setting

The defining feature of SRI is that it destabilizes a viscous Taylor–Couette flow between concentric cylinders in the presence of a stable axial stratification, even when the unstratified flow satisfies the Rayleigh stability condition. In the unstratified inviscid problem, centrifugal stability is determined by

$\frac{\partial (R^4\Omega^2)}{\partial R} >0,$

or, in the standard cylinder-ratio notation,

$\mu>\eta^2,$

with $\mu=\Omega_{\rm out}/\Omega_{\rm in}$ and $\eta=R_{\rm in}/R_{\rm out}$ (Rüdiger et al., 2016). Under stable stratification with

$$N^2=-\,\frac{g}{\rho_0}\,\frac{\mathrm{d}\rho_0}{\mathrm{d}z}>0,$$

non-axisymmetric modes with azimuthal wave number $m\ge 1$ can nevertheless grow (Rüdiger et al., 2016).

Under the Boussinesq approximation, the stratified Taylor–Couette system is governed by incompressibility, momentum balance, and temperature or density transport. In one formulation used for DNS,

$$\partial_tu + \frac12[(u\cdot\nabla)u + \nabla\cdot(u\,u)] = -\nabla p + \nu\nabla^2u + F,$$

$$\partial_tT + \frac12[(u\cdot\nabla)T + \nabla\cdot(u\,T)] = \kappa\nabla^2T, \qquad \nabla\cdot u = 0,$$

with $F=\alpha g T e_z$ (Meletti et al., 2020). In linear stability analyses, one typically adopts a normal-mode ansatz

$$\{u_R,u_\phi,u_z,p,\rho'\}(R,\phi,z,t) =F(R)\,e^{\,i\,(m\phi+k z-\omega t)},$$

which closes to an eigenvalue problem for the complex frequency $\mu>\eta^2,$0, with instability corresponding to $\mu>\eta^2,$1 (Rüdiger et al., 2016).

The principal nondimensional groups are the Reynolds number,

$\mu>\eta^2,$2

the Brunt–Väisälä number,

$\mu>\eta^2,$3

and the Froude number,

$\mu>\eta^2,$4

together with $\mu>\eta^2,$5 and $\mu>\eta^2,$6 (Rüdiger et al., 2016). These parameters organize both onset and mode structure. A recurrent result is that the dominant mode at moderate parameters is $\mu>\eta^2,$7, with axial scale largely controlled by $\mu>\eta^2,$8 rather than by $\mu>\eta^2,$9 and $\mu=\Omega_{\rm out}/\Omega_{\rm in}$0 separately (Rüdiger et al., 2016).

2. Linear stability, onset windows, and mode selection

A central result of moderate-$\mu=\Omega_{\rm out}/\Omega_{\rm in}$1 SRI theory is that instability exists only within a bounded Reynolds-number interval for fixed stratification: $\mu=\Omega_{\rm out}/\Omega_{\rm in}$2 If $\mu=\Omega_{\rm out}/\Omega_{\rm in}$3, rotation is too weak; if $\mu=\Omega_{\rm out}/\Omega_{\rm in}$4, rotation is too strong and epicyclic restoring effects suppress the non-axisymmetric mode (Rüdiger et al., 2016). For the potential flow $\mu=\Omega_{\rm out}/\Omega_{\rm in}$5, the lower and upper neutral branches are nearly straight in the $\mu=\Omega_{\rm out}/\Omega_{\rm in}$6-plane, with leading-order scalings

$\mu=\Omega_{\rm out}/\Omega_{\rm in}$7

where $\mu=\Omega_{\rm out}/\Omega_{\rm in}$8 and $\mu=\Omega_{\rm out}/\Omega_{\rm in}$9 for wide gaps $\eta=R_{\rm in}/R_{\rm out}$0 (Rüdiger et al., 2016).

For flows beyond the Rayleigh limit, the SRI band occupies the quasi-Keplerian interval

$\eta=R_{\rm in}/R_{\rm out}$1

in the inviscid, strongly stratified limit, but finite viscosity complicates this classification (Leclercq et al., 2016). At finite $\eta=R_{\rm in}/R_{\rm out}$2, centrifugal instability (CI), SRI, and radiative instability (RI) are continuously connected in parameter space, so that the boundary at $\eta=R_{\rm in}/R_{\rm out}$3 is no longer a sharp cutoff for non-axisymmetric modes (Leclercq et al., 2016). This continuity explains why experimental and numerical onset behavior can deviate from the simplest inviscid criteria.

Another important constraint is the maximal accessible shear for moderate-$\eta=R_{\rm in}/R_{\rm out}$4 SRI. As $\eta=R_{\rm in}/R_{\rm out}$5 is increased above the Rayleigh line, the instability region shrinks; for $\eta=R_{\rm in}/R_{\rm out}$6,

$\eta=R_{\rm in}/R_{\rm out}$7

and for $\eta=R_{\rm in}/R_{\rm out}$8 only isolated “islands” of instability remain at very large $\eta=R_{\rm in}/R_{\rm out}$9 (Rüdiger et al., 2016). This makes moderately stratified, flatter rotation profiles increasingly difficult to destabilize through classical SRI.

The axial structure of the unstable pattern is likewise highly constrained. Along neutral curves, for the dominant $$N^2=-\,\frac{g}{\rho_0}\,\frac{\mathrm{d}\rho_0}{\mathrm{d}z}>0,$$0 mode and $$N^2=-\,\frac{g}{\rho_0}\,\frac{\mathrm{d}\rho_0}{\mathrm{d}z}>0,$$1,

$$N^2=-\,\frac{g}{\rho_0}\,\frac{\mathrm{d}\rho_0}{\mathrm{d}z}>0,$$2

so that the axial cell length grows linearly with $$N^2=-\,\frac{g}{\rho_0}\,\frac{\mathrm{d}\rho_0}{\mathrm{d}z}>0,$$3 (Rüdiger et al., 2016). For $$N^2=-\,\frac{g}{\rho_0}\,\frac{\mathrm{d}\rho_0}{\mathrm{d}z}>0,$$4, the cells are highly prolate, whereas for $$N^2=-\,\frac{g}{\rho_0}\,\frac{\mathrm{d}\rho_0}{\mathrm{d}z}>0,$$5 they are flattened (Rüdiger et al., 2016). This scaling is consistent with laboratory measurements showing that the axial wavelength $$N^2=-\,\frac{g}{\rho_0}\,\frac{\mathrm{d}\rho_0}{\mathrm{d}z}>0,$$6 grows nearly linearly with Froude number: $$N^2=-\,\frac{g}{\rho_0}\,\frac{\mathrm{d}\rho_0}{\mathrm{d}z}>0,$$7 for $$N^2=-\,\frac{g}{\rho_0}\,\frac{\mathrm{d}\rho_0}{\mathrm{d}z}>0,$$8 (Ibanez et al., 2016).

3. Experimental realizations and observed onset behavior

A representative experimental configuration is the classical Taylor–Couette apparatus with independently rotating cylinders under stable axial thermal stratification. In the combined experimental and DNS study of long-time SRI dynamics, the geometry was $$N^2=-\,\frac{g}{\rho_0}\,\frac{\mathrm{d}\rho_0}{\mathrm{d}z}>0,$$9, $m\ge 1$0, $m\ge 1$1, $m\ge 1$2, with aspect ratio $m\ge 1$3 and radius ratio $m\ge 1$4 (Meletti et al., 2020). A top lid heated via Peltier elements and a bottom lid cooled by an external chiller established a nearly linear $m\ge 1$5–$m\ge 1$6, corresponding to $m\ge 1$7–$m\ge 1$8 and $m\ge 1$9–$$\partial_tu + \frac12[(u\cdot\nabla)u + \nabla\cdot(u\,u)] = -\nabla p + \nu\nabla^2u + F,$$0 (Meletti et al., 2020). The working fluid was silicone oil with $$\partial_tu + \frac12[(u\cdot\nabla)u + \nabla\cdot(u\,u)] = -\nabla p + \nu\nabla^2u + F,$$1, $$\partial_tu + \frac12[(u\cdot\nabla)u + \nabla\cdot(u\,u)] = -\nabla p + \nu\nabla^2u + F,$$2, and $$\partial_tu + \frac12[(u\cdot\nabla)u + \nabla\cdot(u\,u)] = -\nabla p + \nu\nabla^2u + F,$$3 (Meletti et al., 2020).

Low-frequency particle image velocimetry (PIV) was performed in the $$\partial_tu + \frac12[(u\cdot\nabla)u + \nabla\cdot(u\,u)] = -\nabla p + \nu\nabla^2u + F,$$4 plane at mid-height. The imaging used a GoPro Hero4 camera at $$\partial_tu + \frac12[(u\cdot\nabla)u + \nabla\cdot(u\,u)] = -\nabla p + \nu\nabla^2u + F,$$5, $$\partial_tu + \frac12[(u\cdot\nabla)u + \nabla\cdot(u\,u)] = -\nabla p + \nu\nabla^2u + F,$$6, synchronized with a $$\partial_tu + \frac12[(u\cdot\nabla)u + \nabla\cdot(u\,u)] = -\nabla p + \nu\nabla^2u + F,$$7-thick green laser sheet at $$\partial_tu + \frac12[(u\cdot\nabla)u + \nabla\cdot(u\,u)] = -\nabla p + \nu\nabla^2u + F,$$8, covering approximately $$\partial_tu + \frac12[(u\cdot\nabla)u + \nabla\cdot(u\,u)] = -\nabla p + \nu\nabla^2u + F,$$9 in azimuth and $$\partial_tT + \frac12[(u\cdot\nabla)T + \nabla\cdot(u\,T)] = \kappa\nabla^2T, \qquad \nabla\cdot u = 0,$$0 radially (Meletti et al., 2020). Interrogation windows of $$\partial_tT + \frac12[(u\cdot\nabla)T + \nabla\cdot(u\,T)] = \kappa\nabla^2T, \qquad \nabla\cdot u = 0,$$1 with $$\partial_tT + \frac12[(u\cdot\nabla)T + \nabla\cdot(u\,T)] = \kappa\nabla^2T, \qquad \nabla\cdot u = 0,$$2 overlap yielded $$\partial_tT + \frac12[(u\cdot\nabla)T + \nabla\cdot(u\,T)] = \kappa\nabla^2T, \qquad \nabla\cdot u = 0,$$3 spatial resolution, and time series at fixed $$\partial_tT + \frac12[(u\cdot\nabla)T + \nabla\cdot(u\,T)] = \kappa\nabla^2T, \qquad \nabla\cdot u = 0,$$4 were acquired for 1000 radial points over $$\partial_tT + \frac12[(u\cdot\nabla)T + \nabla\cdot(u\,T)] = \kappa\nabla^2T, \qquad \nabla\cdot u = 0,$$5–$$\partial_tT + \frac12[(u\cdot\nabla)T + \nabla\cdot(u\,T)] = \kappa\nabla^2T, \qquad \nabla\cdot u = 0,$$6 per parameter set (Meletti et al., 2020).

Observed onset behavior in such experiments illustrates several features not captured by a single threshold criterion. For $$\partial_tT + \frac12[(u\cdot\nabla)T + \nabla\cdot(u\,T)] = \kappa\nabla^2T, \qquad \nabla\cdot u = 0,$$7, the SRI boundary at moderate $$\partial_tT + \frac12[(u\cdot\nabla)T + \nabla\cdot(u\,T)] = \kappa\nabla^2T, \qquad \nabla\cdot u = 0,$$8 follows $$\partial_tT + \frac12[(u\cdot\nabla)T + \nabla\cdot(u\,T)] = \kappa\nabla^2T, \qquad \nabla\cdot u = 0,$$9, while the unstratified Taylor-vortex boundary is $F=\alpha g T e_z$0 (Ibanez et al., 2016). At $F=\alpha g T e_z$1, SRI appeared at $F=\alpha g T e_z$2 for $F=\alpha g T e_z$3, $F=\alpha g T e_z$4 for $F=\alpha g T e_z$5, and $F=\alpha g T e_z$6 for $F=\alpha g T e_z$7, whereas unstratified Taylor vortex flow appears at $F=\alpha g T e_z$8 (Ibanez et al., 2016). For $F=\alpha g T e_z$9, however, SRI can pre-empt Taylor vortex flow and even appear below the Rayleigh stability line for an inviscid fluid (Ibanez et al., 2016). The same experiments also found that for sufficiently large buoyancy frequency, $$\{u_R,u_\phi,u_z,p,\rho'\}(R,\phi,z,t) =F(R)\,e^{\,i\,(m\phi+k z-\omega t)},$$0, instability can arise even when $$\{u_R,u_\phi,u_z,p,\rho'\}(R,\phi,z,t) =F(R)\,e^{\,i\,(m\phi+k z-\omega t)},$$1 slightly exceeds $$\{u_R,u_\phi,u_z,p,\rho'\}(R,\phi,z,t) =F(R)\,e^{\,i\,(m\phi+k z-\omega t)},$$2, violating the criterion $$\{u_R,u_\phi,u_z,p,\rho'\}(R,\phi,z,t) =F(R)\,e^{\,i\,(m\phi+k z-\omega t)},$$3 proposed for the stratified inviscid laminar state (Ibanez et al., 2016).

At still larger Reynolds number, the primary instability need not be SRI at all. Above the gradient Reynolds number

$$\{u_R,u_\phi,u_z,p,\rho'\}(R,\phi,z,t) =F(R)\,e^{\,i\,(m\phi+k z-\omega t)},$$4

the flow was observed to transition directly to a nonperiodic state that rapidly erases the density stratification in the bulk fluid, without the classical SRI helicoidal modes (Ibanez et al., 2016). This establishes that SRI is a moderate-$$\{u_R,u_\phi,u_z,p,\rho'\}(R,\phi,z,t) =F(R)\,e^{\,i\,(m\phi+k z-\omega t)},$$5, stratification-sensitive instability rather than the universal high-$$\{u_R,u_\phi,u_z,p,\rho'\}(R,\phi,z,t) =F(R)\,e^{\,i\,(m\phi+k z-\omega t)},$$6 route to disorder in rotating stratified Couette flow.

4. Numerical methods and quantitative validation

DNS studies of SRI have used incompressible Boussinesq solvers tailored to long-time integration and modal fidelity. In the 2020 experiment–DNS comparison, temporal discretization used semi-implicit time stepping: Adams–Bashforth for convection and backward-Euler for diffusion (Meletti et al., 2020). A projection method improved by Hughes et al. required two Poisson solves for pressure prediction and correction and three Helmholtz solves for the momentum components (Meletti et al., 2020). Spatial discretization combined Fourier spectral representation in $$\{u_R,u_\phi,u_z,p,\rho'\}(R,\phi,z,t) =F(R)\,e^{\,i\,(m\phi+k z-\omega t)},$$7 with fourth-order compact finite differences in $$\{u_R,u_\phi,u_z,p,\rho'\}(R,\phi,z,t) =F(R)\,e^{\,i\,(m\phi+k z-\omega t)},$$8 and $$\{u_R,u_\phi,u_z,p,\rho'\}(R,\phi,z,t) =F(R)\,e^{\,i\,(m\phi+k z-\omega t)},$$9 (Meletti et al., 2020). Boundary conditions included no-slip cylinder walls, fixed temperatures at the lids, adiabatic sidewalls, and lids co-rotating with the outer cylinder (Meletti et al., 2020).

The implementation relied on an rPDD tridiagonal solver and 2D-pencil decomposition for global transforms in the Poisson and Helmholtz solvers, achieving an order-of-magnitude speed-up over scalar codes, with wall-clock times reduced from months to hours for long runs (Meletti et al., 2020). This computational efficiency is not incidental: low-frequency SRI modulation and return-to-stability phenomena require integrations lasting many hours of physical time, far beyond the interval needed to establish primary linear growth.

Validation against experiment was performed at several levels. Radial-time Hovmöller diagrams of $\mu>\eta^2,$00 at $\mu>\eta^2,$01 showed identical $\mu>\eta^2,$02 oscillations in the co-rotating frame in experiment and DNS (Meletti et al., 2020). Mean azimuthal velocity profiles indicated that the stratified SRI flow is slower near the inner cylinder than the non-stratified Taylor–Couette reference, and slightly faster near the outer wall (Meletti et al., 2020). Spectra revealed dominant SRI frequencies $\mu>\eta^2,$03 at $\mu>\eta^2,$04 and $\mu>\eta^2,$05 at $\mu>\eta^2,$06, with azimuthal wave number $\mu>\eta^2,$07 confirmed by the Doppler-shift relation (Meletti et al., 2020).

A later numerical study, in the same $\mu>\eta^2,$08, $\mu>\eta^2,$09, $\mu>\eta^2,$10 geometry with $\mu>\eta^2,$11, used a spectral-Fourier / fourth-order compact finite-difference DNS on a $\mu>\eta^2,$12 grid in $\mu>\eta^2,$13 and reported at $\mu>\eta^2,$14, $\mu>\eta^2,$15 a finite-amplitude SRI spiral with fast oscillation $\mu>\eta^2,$16 and slow envelope period $\mu>\eta^2,$17 (Meletti et al., 8 Sep 2025). This suggests sensitivity of the reported fast frequency to setup, diagnostic, or normalization details across studies, while preserving the robust separation between fast spiral oscillation and slow envelope modulation.

5. Nonlinear dynamics, amplitude modulation, and pattern transitions

Long-time integrations reveal that the nonlinear SRI state is not merely a saturated single-frequency spiral. In DNS runs of at least $\mu>\eta^2,$18, strong low-frequency amplitude envelopes of $\mu>\eta^2,$19 were detected at approximately $\mu>\eta^2,$20, more than $\mu>\eta^2,$21 slower than the $\mu>\eta^2,$22 SRI oscillation (Meletti et al., 2020). Hilbert-envelope analysis yielded a regular modulation period $\mu>\eta^2,$23 after a transient of about $\mu>\eta^2,$24 (Meletti et al., 2020). Space–time diagrams and one-line time series displayed synchronous amplitude minima in DNS and PIV at $\mu>\eta^2,$25 and $\mu>\eta^2,$26, establishing that the modulation is not a numerical artifact (Meletti et al., 2020).

The modulation cycle corresponds to a recurring transition between two $\mu>\eta^2,$27 spiral states in the $\mu>\eta^2,$28 plane: a downward-inclined spiral propagating from top to bottom and an upward-inclined spiral propagating from bottom to top (Meletti et al., 2020). At amplitude minima, both spirals coexist and form a “chessboard” interference pattern (Meletti et al., 2020). The transitions are associated with small changes in spiral drift speed, but the mean axial speeds of the upward and downward modes are equal, and no phase-lock with the SRI frequency was found (Meletti et al., 2020). These observations define a genuinely secondary, slow dynamical process superposed on the primary SRI oscillation.

Dependence on stratification is explicit. The modulation disappears when the stratification is halved to $\mu>\eta^2,$29 and reappears when $\mu>\eta^2,$30 is increased to $\mu>\eta^2,$31, indicating direct sensitivity to $\mu>\eta^2,$32 rather than to Reynolds number alone (Meletti et al., 2020). A plausible implication is that the modulation is tied to wave competition enabled by sufficiently strong stable stratification.

The 2025 numerical analysis sharpened this interpretation by separating upward and downward propagating spiral components with Radon transforms. Applied to $\mu>\eta^2,$33, the decomposition reconstructed upward and downward fields whose amplitudes

$\mu>\eta^2,$34

oscillate out of phase, with relative phase shift $\mu>\eta^2,$35 and a spectrum matching the slow modulation in the full signal (Meletti et al., 8 Sep 2025). On that basis, a toy model with two linearly coupled, weakly nonlinear spiral amplitudes of Ginzburg–Landau type was shown to reproduce the sequence “downward spiral $\mu>\eta^2,$36 chessboard transition $\mu>\eta^2,$37 upward spiral” (Meletti et al., 8 Sep 2025). The same study further proposed a QBO-like wave–mean-flow framework in which the alternating spiral branches feed momentum into a slowly oscillating axial mean flow, with spontaneous reversals over a period $\mu>\eta^2,$38 SRI-years (Meletti et al., 8 Sep 2025). This does not supersede the earlier observations; rather, it furnishes a reduced dynamical interpretation of the low-frequency modulation first isolated in long-time DNS and experiment.

6. Return to stability, related instabilities, and broader significance

One of the most consequential findings in moderate-$\mu>\eta^2,$39 SRI is that the instability can disappear again as Reynolds number increases. For $\mu>\eta^2,$40 and $\mu>\eta^2,$41, linear theory predicts instability roughly in the interval $\mu>\eta^2,$42 (Meletti et al., 2020). Consistent with this, both DNS and PIV show that beyond a critical $\mu>\eta^2,$43, the $\mu>\eta^2,$44 SRI oscillations and their harmonics vanish and the flow reverts to the laminar Taylor–Couette profile (Meletti et al., 2020). In the interpretation given there, at large $\mu>\eta^2,$45 the stratification becomes negligible because $\mu>\eta^2,$46 (Meletti et al., 2020). This return to stability is a defining characteristic of SRI at moderate Reynolds number and distinguishes it from monotonic centrifugal destabilization.

The relationship between SRI and other instability mechanisms is nontrivial. In the inviscid picture, CI occupies $\mu>\eta^2,$47 and SRI occupies $\mu>\eta^2,$48, but finite viscosity erodes this clean partition (Leclercq et al., 2016). Linear analyses demonstrate continuous connections among CI, SRI, and RI, including non-axisymmetric modes localized at the outer cylinder without stratification, modes propagating against inner-cylinder rotation with stratification, and a centrifugal-like helical mode beyond the Rayleigh line caused by the joint effects of stratification and viscosity (Leclercq et al., 2016). At the same time, the threshold $\mu>\eta^2,$49 appears to remain an impassable instability limit for axisymmetric modes, regardless of stratification, viscosity, and even disturbance amplitude (Leclercq et al., 2016). This sharp asymmetry between axisymmetric and non-axisymmetric behavior is central to the taxonomy of Taylor–Couette instabilities.

Astrophysical and geophysical significance follows from this synthesis. SRI provides a purely hydrodynamic route to non-axisymmetric spiral instability in Keplerian-like shear flows, complementing magnetohydrodynamic MRI-based theories of angular momentum transport in accretion disks (Meletti et al., 2020). The observed amplitude modulations and mode competition imply richer spatio-temporal dynamics than linear theory alone would predict, and may alter transport estimates in stratified disks (Meletti et al., 2020). In geophysical settings, similar modulated spiral patterns may occur in stratified planetary atmospheres or oceanic vortices where rotation and stable density gradients interact (Meletti et al., 2020). The QBO-like interpretation of alternating spiral branches and slowly reversing mean flow strengthens this cross-disciplinary connection by linking SRI to a broader class of wave–mean-flow interactions in rotating stratified media (Meletti et al., 8 Sep 2025).

A common misconception is that stable stratification simply suppresses shear-flow instability. The SRI literature shows instead that stable stratification can destabilize centrifugally stable anticyclonic flow by enabling non-axisymmetric resonant dynamics, while still suppressing other modes and eventually losing efficacy as $\mu>\eta^2,$50 becomes too large or mixing destroys the background density gradient (Rüdiger et al., 2016). Another misconception is that SRI has a single universal onset criterion. Experimental violation of $\mu>\eta^2,$51 at sufficiently large $\mu>\eta^2,$52, the bounded Reynolds-number window, the appearance of isolated instability islands at flatter rotation profiles, and the continuous finite-$\mu>\eta^2,$53 connection to CI and RI collectively show that SRI is controlled by an interplay of shear, stratification, viscosity, geometry, and mode selection rather than by one inviscid threshold alone (Ibanez et al., 2016).