In this paper
AbstractIntroductionElliptic rings in the kinematic pictureModel and MethodsNumerical ResultsAxis switchingTension and drift of circular ringsOscillation periodShape modesTwo further modelsConclusionsSupplementary MaterialAcknowledgmentsAuthor DeclarationsConflict of InterestAuthor ContributionsData AvailabilityReferencesWorksPhysics and Astronomy
Axis-switching oscillations of elliptic scroll rings in excitable media
Abstract
Scroll waves in three-dimensional excitable media rotate around a filament that moves according to its local curvature. Using simulations of the Barkley model, we show that initially planar, elliptic scroll rings perform the axis-switching oscillation known from elliptic vortex rings in fluids: while the ring drifts along its axis, it buckles into a saddle, becomes planar again with interchanged axes, and repeats this cycle. The oscillation occurs for vanishing, positive, and negative filament tension. As predicted by a kinematic filament equation, its period scales approximately with the square of the ring radius and inversely with the drift coefficient. For our main model parameters, however, the oscillation is about 1.45 times slower than the linear prediction based on the drift of circular rings of the same size. A curvature-dependent drift coefficient accounts for a factor of about 1.25, while the remaining slowdown results mainly from a weaker coupling of out-of-plane to in-plane deformations, which a fourth-order rigidity term does not describe. In contrast, the Panfilov-Rudenko model and the Atri model of Ca waves follow the linear prediction within about 6%, showing that the slowdown depends on the kinetics.
Rotating spiral waves of chemical or electrical activity occur in systems as different as the Belousov–Zhabotinsky reaction and the human heart, where they are associated with dangerous arrhythmias. In three-dimensional media, a spiral extends into a scroll wave that rotates around a curve called the filament. When the filament forms a closed loop, it travels along its axis, much like a vortex ring in a fluid. For these fluid rings, J. J. Thomson showed in 1883 that a slight elliptic deformation causes an oscillation in which the long and short axes repeatedly exchange places. Using numerical simulations of excitable reaction–diffusion models, we show here that elliptic scroll rings undergo the same oscillation. Its period scales with ring size approximately as predicted by the equation of motion that also describes vortex filaments in fluids, but in one of the models it is markedly longer than this equation predicts. Elliptic scroll rings therefore expose effects beyond the standard description of filament motion. is a phase singularity as all phases of the local excitation cycle meet at this point. In three dimensions, these singular points form a continuous line, the filament, around which the pattern rotates as a scroll wave.³ The filament organizes the surrounding three-dimensional wave field, and its motion is hence central to understanding how scroll waves drift, collide, and break up.⁶ In the heart, such rotating waves are associated with tachycardia and fibrillation.⁷,⁸
Introduction
Excitable media relay a signal by local excitation followed by a period of recovery. The best-known examples are nerve fibers and heart tissue, and similar excitation waves occur in the Belousov–Zhabotinsky (BZ) reaction,¹–³ as calcium waves inside living cells,⁴ and in populations of signaling slime mold cells.⁵ In two-dimensional excitable media, a wave with a free end typically curls into a rotating spiral. The tip of the spiral is a phase singularity as all phases of the local excitation cycle meet at this point.
For filaments whose radius of curvature is large compared with the size of the spiral core—more precisely, with the width of the response functions of the spiral—this motion follows a simple kinematic law.⁹,¹⁰ Each point of the filament moves with a velocity that is proportional to the local curvature and that has two components,
where is the position on the filament, and and are the unit normal and binormal vectors [xref?]. The normal vector points toward the local center of curvature, and the binormal vector is perpendicular to the plane in which the filament curves. The coefficient is the filament tension. For , a circular filament of radius contracts as , while for , rings expand and the medium can develop scroll wave turbulence.¹¹–¹⁴ The coefficient moves a circular ring along its symmetry axis with the velocity ; however, it vanishes if all reacting species diffuse equally fast.⁶,¹⁰
Equation (1) is closely related to a classical result of fluid dynamics. A thin vortex filament in an ideal fluid—for example, the core of a smoke ring—moves in the local induction approximation (LIA) with the velocity , that is, Eq. (1) with .¹⁵,¹⁶ J. J. Thomson calculated the vibrations of a slightly deformed vortex ring in 1883 and found that an elliptic deformation oscillates, so that the long and short axes of the ring repeatedly exchange places.¹⁷ Arms and Hama followed this oscillation for initially planar, elliptic rings in the LIA through the saddle shape and back to a planar ellipse with interchanged axes.[18] Dhanak and De Bernardinis computed it with the cut-off approximation of the Biot–Savart law[19] and observed it in experiments with air puffed through elliptic orifices; the motion was periodic for slightly elliptic rings, and strongly elongated rings split into two.[20] Viscous elliptic vortex rings and elliptic jets show a similar exchange of axes, known as axis switching.[21, 22, 23, 24]
For scroll waves, the tension term adds dissipation to this picture. A similar combination of binormal and dissipative normal motion describes vortices in superfluid helium.[25, 26] In the complex Ginzburg–Landau equation for oscillatory media, however, untwisted filaments have no binormal motion at lowest order and simply contract.[27, 28]
In excitable media, Keener and Tyson described noncircular filaments in thin BZ layers by purely normal, curvature-driven motion,[29] and Dutta and Steinbock showed that planar, hairpin-shaped filaments in the BZ reaction and in the Barkley model with equal diffusion coefficients translate with constant shape, as predicted by the curve-shortening flow.[30] In these cases, the binormal motion is negligible and the filament stays in its plane. In a BZ gel, Winfree and Jahnke photographed an initially elliptical scroll ring that became round and then broke at two opposite sides, which they attributed to the filament reaching the gel boundary where its curvature had been largest.[31]
Knotted and linked filaments, anticipated by Winfree and Strogatz,[32] preserve their topology and can persist for thousands of rotation periods,[33, 34] whereas tangled unknots relax toward circles.[35] Such structures recall Kelvin’s 1867 theory of vortex atoms, in which atoms are vortex rings in a frictionless cosmic fluid, possibly linked or knotted, with modes of vibration that give rise to spectral lines.[36]
Bánsági and Steinbock analyzed the growth of the shape modes of expanding BZ scroll rings under negative tension,[14] but the oscillation that the binormal term drives in such modes has not been studied. Moreover, Eq. (1) is only the lowest order of a more general theory that includes curvature-dependent coefficients, a fourth-order term known as filament rigidity, and couplings to the twist of the filament.[37, 38, 39]
Here, we use three-dimensional simulations of the Barkley model[40] to study scroll rings whose filaments start out as planar ellipses. We show that these rings perform the axis-switching oscillation of elliptic vortex rings for vanishing, positive, and negative filament tension, with a period that scales approximately as predicted by Eq. (1). At our main parameters, however, the oscillation is about 1.45 times slower than predicted from the drift of circular rings of the same size, which we trace to the curvature dependence of the drift coefficient and to a weaker coupling of out-of-plane to in-plane deformations. Two further models of excitable media show the same oscillation with frequencies that differ from the prediction by about 6% or less.
Elliptic rings in the kinematic picture
The binormal term moves each part of the filament perpendicular to its local plane of curvature, with a speed proportional to the curvature. For an initially planar ellipse, this motion is along the ring axis and is fastest at the two ends of the long axis. These ends therefore advance ahead of the flatter sides, deforming the ring into a saddle [Fig. 1(b,d)]. The leading ends now form crests, while the lagging sides form troughs. This out-of-plane bending tilts the binormal velocity inward at the leading ends and outward at the lagging sides. Accordingly, the original long axis shortens while the short axis lengthens, until the ring becomes planar again with its axes interchanged [Fig. 1(e)]. The cycle then repeats. The tension term damps the oscillation for and amplifies it for .
To make this picture quantitative, we consider a ring that deviates slightly from a circle of radius . At the polar angle , its radial deformation is and its vertical deformation is . We expand both in Fourier modes , with amplitudes and . The mode shifts or tilts the ring, and makes it elliptic. Linearizing Eq. (1) yields (see the supplementary material for this and the following derivations)
The binormal terms couple the in-plane and out-of-plane deformations; the resulting exchange is the axis-switching oscillation. To linear order, the deformations average to zero around the ring and hence do not affect the rates of change of the mean radius and the mean height. Accordingly, the ring drifts like a circle of the same mean radius, with the velocity , while its shape oscillates. For , the mode oscillates with the angular frequency
Successive axis switches of an ellipse () are separated by the half-period . An initially planar ellipse reaches its first switch after , when its projection onto the original plane is circular, and becomes planar with interchanged axes after . The full period equals times the time that the ring needs to travel its own circumference, in agreement with Thomson’s result.[17] Between successive switches, the ring hence drifts along its axis by , independent of .
The eigenvalues of the system of Eqs. (2) are ; hence an elliptic ring () oscillates only if . For , it then shrinks while it oscillates with decaying amplitude; for , it expands while the deformation grows.
Because both and the oscillation frequency scale with , a ring with constant coefficients oscillates at a constant rate in the logarithm of its radius, which means faster and faster in time if it shrinks. For the elliptic mode, the phase advances by , with , for every factor by which the radius changes, and the envelope of the relative ellipticity varies as . The number of axis switches per factor is hence , which depends only on . A shrinking ring with switches its axes many times before collapse, whereas a ring with switches about once per factor .
Equations (2) and (3) assume constant and . To discuss deviations from this lowest-order picture, we write the linearized mode equations more generally as
where converts vertical into radial deformations and converts radial into vertical ones, and and damp or amplify the two deformations. If the coefficients depend on the local curvature alone, the two couplings are affected differently: a radial deformation changes the local curvature, so that
whereas a vertical deformation only tilts and without changing the curvature to first order, so that equals the drift coefficient of the ring for every . Analogously, and in such a curvature-only law. Higher-order local nonlinearities give these relations: the rigidity terms of Dierckx et al.[37] add to and to , and analogous terms with to and (supplementary material).
An oscillating solution of Eqs. (4) decays at the rate and has the frequency , where is the difference of the two diagonal rates. The amplitudes of the two deformations are related by
where and are the amplitudes of the two damped sinusoids. Equation (6) holds for any diagonal rates; only shifts the radial and vertical oscillations away from quadrature, by an angle with . Frequency, amplitude ratio, decay rate, and phase therefore yield all four coefficients; numerical solutions of Eq. (1) with a curvature-dependent agree with the predicted frequencies and amplitude ratios within 0.6% and 1.5% (supplementary material).
Model and Methods
For most of our numerical experiments, we use the Barkley model[40]
in which is a fast activator and a slow inhibitor that does not diffuse. We fix . Our main parameter set is , , for which two-dimensional spirals rotate rigidly with a period of 4.44 around a circular core of radius 0.49 and emit waves of wavelength 12.0. Additional simulations use with , 0.16, and 0.18 (also rigid, with periods of 4.10 to 4.69, core radii of 0.37 to 0.58, and wavelengths of 11.2 to 12.7; supplementary material), with (rigid), and with , for which spirals meander.
Equations (7) are solved with an explicit Euler scheme on a cubic grid of spacing , a seven-point Laplacian, and the time step , in periodic domains in which the ring stays at least one wavelength from its periodic images and is kept at the center by vertical shifts of the fields. Larger domains, a halved time step, and a seed turned by relative to the grid change the mean frequency ratio of Sec. IV C by less than 1%; grid spacings of 0.15 and 0.1 change it by 2 to 3% (supplementary material).
Scroll rings are initiated by a broken wave confined to the interior of an ellipse with semi-axes and (written below): inside the ellipse, a thin excited layer () is placed below a refractory layer (). The free edge of this wave curls up and forms a single scroll ring within about three time units. Rings with an -fold deformation are prepared with the boundary , where to 0.15, and tilted rings by rotating the plane of a circular seed by . We also simulate straight scroll waves whose filament is initially displaced by in a domain that is periodic in with height (supplementary material).
The filament is computed as the intersection of the isosurfaces and .[40] This curve follows the spiral tip, which circles the center of rotation once per rotation period; we remove this wobble by averaging all ring measures over one period. We analyze only frames in which the extracted filament is a single closed curve (more than 99% of all frames). From the filament we compute its length and ; the arc-length-weighted mean horizontal distance of the filament from its center; the half-widths and of the ring along and ; the axial position of its center; and the Fourier amplitudes and of its horizontal distance from the center and of its height, as functions of the polar angle and weighted by angle. An axis switch occurs whenever changes sign, and is the saddle amplitude; a half-cycle is the interval between two switches. We denote by and the coefficients of circular rings of radius (Sec. IV B) and by the drift coefficient that a deformed ring has according to its own drift speed ; for the weakly deformed rings of Sec. IV D, we use , which differs from by at most 1.4%. Alternative filament definitions and the numerical solution of Eq. (1) are described in the supplementary material.
Use of generative AI. This study made extensive use of the large language model Claude Opus 5.5 (Anthropic), accessed in October and November 2026 through the Claude Code desktop application. Under the author’s direction, the model wrote and debugged the simulation and analysis code, executed the simulations, analyzed the results and contributed to their interpretation, derived the linear mode analysis of § 2, produced the figures with plotting scripts, searched the literature, and drafted the manuscript text. To identify errors, Claude agents, ChatGPT 6.1-Sol, and GPT-6 Astra Pro (both OpenAI) reviewed the deposited code, the manuscript, and the supplementary material. All data were produced by the deposited, deterministic simulation code, and all figures are plots generated from these data by deposited scripts. The author critically reviewed all AI-generated content.
Numerical Results
Axis switching
Figure 1 (Multimedia available online) shows a scroll ring that starts from an ellipse at , . After its formation, the filament is a nearly planar ellipse with half-widths of 6.8 and 4.2 at [Figure 1(c)]. The ends of the long axis, where the filament is most strongly curved, immediately move ahead of the rest of the ring, and the ring buckles into a saddle [, Figure 1(b,Figure 1d)]. At , the ring is planar again, but now its long axis points in the -direction; at , it has returned to its original shape. Meanwhile, the ring travels steadily along its axis by about 0.33 space units per time unit.
Figure 1. Axis switching of an elliptic scroll ring. (a) Planar elliptic filament (black) with its tangent , normal , and binormal at the end of the long axis, where the curvature is largest. The scroll wave rotates around the filament in the plane spanned by and , as indicated at the opposite end of the long axis (green arrow). The binormal motion , averaged around the ring, makes the ring drift along its axis (gray arrow). The filament is oriented by the right-hand rule (fingers along the rotation, thumb along ); with this orientation, the ring drifts along , as in all our simulations (). (b)–(f) Simulation (Barkley model, , , ; initial ellipse with semi-axes 8 and 5). (b) Excitation wave (surface ) and filament (tube) in the saddle phase at ; the quadrant of the wave facing the viewer is cut away to reveal the filament. (c)–(f) Filament over one axis-switching cycle, seen from the same direction: planar ellipse with its long axis along (), saddle (), planar ellipse with its long axis along (), and saddle with reversed orientation (). The filament is colored by its height relative to the ring center, in space units (red: above, blue: below); the gray curve is its projection onto the horizontal plane. The bar in (f) is 5 space units long and applies to (c)–(f). Panels (b)–(f) are shown in a frame that moves with the ring center. The movie shows panel (b) from to 40, with the filament projected onto a horizontal plane below the wave. (Multimedia available online.)
As shown in Figure 2, the oscillation is long-lived. Over 120 time units, the axes switch seven times while the amplitude decreases by a factor of 3 to 4. We compare the simulated ring with what we refer to as the lowest-order law, namely Eq. (1) with the coefficients of circular rings of the same size. For this ring, we set , start from the measured shape at , and let the model ring follow the measured filament length of the simulated ring, so that both rings always have the same size (dashed). This law captures the shape of the cycle, but its half-cycles are 1.37 to 1.46 times shorter than those of the simulated ring, and its oscillation does not decay. Following the measured length matches the size of the two rings, not their shape, so that the law keeps its relative deformation whereas the simulated ring becomes rounder. We quantify the difference in period half-cycle by half-cycle in Sec. IV C.
Figure 2. Axis switching of the elliptic ring of Figure 1. (a) Half-widths (blue) and (orange) of the ring. (b) Saddle amplitude . Curves: reaction–diffusion simulation (solid) and the lowest-order law (dashed; see text), both averaged over one rotation period.
Axis switching is not restricted to rings with vanishing tension. At , the tension is clearly positive, and an initially ellipse shrinks from a mean radius of 9 to collapse within 92 time units while switching its axes four times (Figure 3). The intervals between switches shorten from 26 to about 9 time units, but they stay nearly constant on a logarithmic radius scale [0.39 and 0.33 in , Figure 3(c)], as expected from § 2. The lowest-order law with the coefficients of circular rings of the same size places the first two switches at similar values of but reaches them earlier, because it shrinks the ring faster; the positions in test only . For , , where the shrinkage and drift of the elliptic ring give and (), the ring completes only a single, strongly damped switch before it collapses, again in line with the linear theory. At , the tension is negative; the ring grows from a mean radius of 8.5 to 12 and switches its axes twice within 90 time units while its relative ellipticity grows to about 0.4. The switching also occurs for the meandering
Figure 3. Axis switching of a shrinking ring under positive tension (, , initial ellipse). (a) Half-widths and (b) saddle amplitude; solid: simulation; dashed: lowest-order law, Eq. (1) with the tension and drift coefficient of circular rings of the same instantaneous size (Fig. S1 of the supplementary material), started from the measured shape at and shown until the model ring radius drops below 2.5. (c) Number of axis switches (circles) and planar moments midway between them (diamonds), from the crossings in (a) and (b), versus the logarithmic decrease of the ring radius, , for the simulation (filled) and the lowest-order law (open), shown while .
spiral at , .
Tension and drift of circular rings
To predict the oscillation period independently of the elliptic rings, we measure the tension and drift coefficient of circular rings from their contraction, , and their drift, with (Figure 4). Both coefficients have to be measured, because large-excitability asymptotics predicts vanishing and at leading order for the Barkley model [Ref. 41, Secs. III A and IV D; Ref. 42, Sec. VI]. For , , rings of very different sizes collapse onto single curves (Figure 4(b,Figure 4c)), and both coefficients depend strongly on curvature. As increases from 0.04 to 0.29, the tension falls from 0.63 through zero at to , while the drift coefficient falls from 2.9 to 1.6. At small , falls steeply, more like a logarithm than like the polynomial in expected from a local gradient expansion[38] (supplementary material). Polynomial fits to these curves provide and for the analysis below, and we calibrate , 0.16, and 0.18 in the same way (supplementary material).
Figure 4. Circular scroll rings at , . (a) Ring radius versus time for seed radii between 4 and 13; the gray band marks the final radii of the rings that approach –4.9 from below and above. (b) Tension and (c) drift coefficient versus curvature , measured in 10-unit time windows after on 14 rings with seed radii of 4.5 to 26; colors encode the seed radius, and the lines are the polynomial fits used in the analysis. Open circles: early windows of a ring with seed radius 38, which develops undulations and is not used.
Because the tension changes sign [Figure 4(b)], large rings shrink and small rings grow: rings seeded with radii of 4.5 to 6.5 approach a nearly stationary radius from both sides and end between and 4.82 while still converging slowly [gray band in Figure 4(a)], whereas a ring seeded with radius 4 collapses. We regard this range, –4.9, as a long-lived, nearly stationary radius rather than as an established attractor (our runs cover at most 300 time units, and moves to about 4.5 at ); here, it provides a background state of nearly constant size for the mode analysis of Sec. IV D. At , rings settle at , whereas at all rings collapse and at all rings expand. Scroll rings whose contraction stops at a finite radius were reported earlier in simulations;[43, 44, 45] whether this reflects higher-order terms of the filament law or an interaction between opposite parts of the ring is an open question.[6]
Oscillation period
Next, we test whether the lowest-order law predicts the axis-switching period quantitatively. The coefficients of circular rings (Sec. IV B) leave this comparison without any adjustable parameter. For every half-cycle between two axis switches, we integrate the angular frequency that Eqs. (2) predict with the coefficients and of circular rings of the measured radius over the observed duration and divide by . As rings shrink by up to 32% within one half-cycle, we integrate over rather than use a single radius. The result is the ratio of the predicted to the observed mean frequency. This frequency ratio states how many times slower the ring oscillates than predicted. The prediction includes two small corrections: the nonlinear Eq. (1) changes the half-period by at most 2%, and the tension changes the frequency by less than 0.2%. Figure 5(a) shows the result for 22 half-cycles from five simulations (all values in the supplementary material).

Figure 5. (a) Observed versus predicted half-period of the axis-switching oscillation for 22 half-cycles of five simulations at ( small: seed , Figure 2; large and , 0.16, 0.18: seeds ), predicted with the coefficients of circular rings of the measured radius (Sec. IV C). Solid line: equality; dashed line: mean ratio at . (b) Frequency ratio (filled) and the same ratio divided by the slowdown that the curvature-only law, Eq. (5), predicts (open; the two coincide for the Atri model) for half-cycles between planar moments, versus the ring diameter in spiral wavelengths: Barkley rings of (a) except (colors as in a), weakly deformed Barkley rings at and at a diameter of 2.0 wavelengths (squares; the latter from the mode fit of Sec. IV D), the Panfilov–Rudenko model (diamonds), and the Atri model (triangles). The two Barkley half-cycles with , where is poorly constrained, are omitted.
At , 13 half-cycles of two strongly elliptic rings with to 8.0 give frequency ratios of 1.41 to 1.49 (mean 1.450), without a trend within either run, and weakly deformed rings at (Sec. IV D) give . The ratio hence depends on the amplitude of the deformation by only a few percent. At and 0.16, the ratio is 1.20 to 1.43, and at , the single analyzed half-cycle gives 1.47. We use rather than the ring’s own, smaller drift coefficient (Sec. III and supplementary material), because is measured independently of the elliptic rings. With , the ratios would be smaller ( at ) and would vary within each run. For the meandering spiral,[46] averaging over one rotation period does not remove the meander, and we therefore do not analyze its period.
The curvature dependence of the drift coefficient accounts for much, but not all, of the observed slowdown. In Eqs. (4), the oscillation frequency depends mainly on . A curvature-only law keeps , whereas Eq. (5) with the measured gives at , which lengthens the predicted period by a factor of . For the weakly deformed rings at the nearly stationary radius , where linear theory applies, this factor accounts for most of the observed slowdown of 1.48 and leaves a residual of about 1.17 that no curvature-only law explains. For the strongly elliptic rings, the factor is 1.19 to 1.27 and the residual 1.09 to 1.23, depending on how the factor is evaluated (supplementary material). At and 0.16, where depends more weakly on curvature, the factor is 1.0 to 1.25 and the residual 1.1 to 1.4. The observed half-periods range from 9.4 to 52 and approximately follow the scaling, with a prefactor that depends on .
Shape modes
The remaining slowdown comes mainly from the out-of-plane coupling , which depends on the wavenumber of the deformation rather than on the curvature of the ring. To show this, we prepared weakly deformed rings with twofold, threefold, and fourfold modes at and on larger rings with radii of up to about 15. Damped sinusoids fitted to the radial and vertical mode amplitudes give the four coefficients of Eqs. (4) (supplementary material, tbl:S5, S6, and S11). A curvature-only law requires , but falls with increasing , to for at [Figure 6(a)]. Rings with different and but equal nevertheless have nearly the same : 1.31 for both rings at , and 1.62 and 1.64 at [Figure 6(b)]. In contrast, lies within 19% of Eq. (5), and within 4% for and 3 at . The amplitude ratio of Eq. (6) separates the two couplings directly: for at , it is 0.81, against 0.69 for the curvature-only law.
Figure 6. Response of scroll wave filaments to deformations of wavenumber (, ). (a) Couplings (vertical to radial) and (radial to vertical) of ring modes relative to of circular rings of the same radius, versus : twofold (squares), threefold (diamonds), and fourfold (triangles) modes at (filled) and on larger rings (open), with the two couplings offset by in for visibility. The gray line marks of the curvature-only law, the shaded band the range of that this law predicts through Eq. (5), and the gray circle of the tilted ring (), which equals 1 by symmetry. (b) Absolute of the ring modes (bars: ranges of Table S5); gray circle: tilted ring; of circular rings with (black line; dashed: extrapolated); the binormal coefficient of straight filaments with a small sinusoidal deformation (blue; bars: uncertainty of the extrapolation to zero amplitude); and the coefficient obtained from the frequencies of the growing undulations of a ring with initial radius 38 (; green). Fig. S3 also shows the data at and 0.16.
At , falls close to the drift coefficient of a circular ring whose curvature equals ; all values lie within 14% of , including three rings that we prepared to test this relation. Note that for , this comparison extrapolates beyond its calibrated range. For at , the reduction slows the oscillation by , matching the residual of 1.17 (Sec. IV C). However, this relation does not carry over to other parameters. At , varies little with curvature, yet still falls with and stays within about 16% of its values at (Fig. S3); the same holds at . It is hence likely that the agreement with at is partly coincidental. At and 0.16, is about 0.7 to 0.85 of , which accounts for a residual of about 1.1 to 1.2, the lower part of the observed range.
Straight filaments with a small sinusoidal deformation show the same dependence. Their binormal coefficient falls from 2.29 at to 1.04 at $k = 0.79, far from linear in as a fourth-order law would require [Figure 6(b)], and it barely changes between and 0.17 (Fig. S3). At equal , of the ring modes lies about 20% below these values, whereas the growing undulations of a large ring () agree with them, so that the ring curvature matters as well. The effect is not tied to the instability of straight filaments at the main parameters,[37] because straight filaments are stable at .
because straight filaments are stable at . The results at hold for all five filament definitions and at , and a fourfold mode rotated by relative to the grid agrees within about 1%. The tilted ring, for which under any dynamics, gives 1.01. Neither the twist that the deformation induces nor a single rigidity coefficient accounts for the reduction, although the rigidity term also depends on . The fitted to grows in magnitude from 1.8 at to 10–13 at –0.28, and the local slope of versus grows in the same way, so that no single applies even at the smallest studied. Moreover, an matched to the frequency moves the amplitude ratio of the mode at away from the measured value (supplementary material). On small rings, the modes are also damped more strongly than the curvature-only law allows: at , this law lets the and 3 modes grow at rates of about 0.01 and 0.03, whereas they decay at 0.011 and 0.058, and the mode at 0.12. On our largest rings, this difference reverses: the and 4 modes grow slowly, at about 0.002 to 0.005, where this law predicts decay.
Two further models
To test whether axis switching and its period depend on the specific kinetics, we use the piecewise-linear model of Panfilov and Rudenko[47] (supplementary material). Note that, as in the Barkley model, its recovery variable does not diffuse, so that need not vanish. The model suits our purpose because its scroll rings change from contraction to expansion as the excitability parameter decreases, while they keep drifting along their axis.[47] Near this transition, the tension vanishes but the drift does not.
For and 0.8, spirals rotate rigidly with wavelengths of 37 and 34, circular rings have slightly negative and weakly positive tension ( and ), and their drift coefficient changes little with curvature (supplementary material). Elliptic rings perform the same cycle as in the Barkley model (Fig. S4). Over 1600 time units, a ring at switches its axes four times without changing its size much, and a ring at switches six times while it shrinks to half its radius. To compare the three models on equal terms, we time their half-cycles by the planar moments () and use as the radius (supplementary material). The frequency ratio is then at (three half-cycles) and at (seven half-cycles of four runs at grid spacings of 0.6 to 0.3), compared with 1.41 to 1.48 for the Barkley rings at analyzed in the same way. The small shortfall at is consistent with the curvature dependence of , for which Eq. (5) predicts 0.970, although this factor depends on the grid spacing by a few percent (supplementary material). The damping again departs from the constant-coefficient law: at , the relative ellipticity at successive planar moments decreases from 0.21 to 0.12 while the ring grows slightly, whereas the slightly negative tension of circular rings would let it grow by about 20%. Axis switching is hence not a peculiarity of one model, whereas the size of the departure from the lowest-order law depends on the kinetics: at comparable ring sizes in units of the spiral
wavelength (Figure 5(b)), the Panfilov–Rudenko rings are consistent with the curvature-only law, and the Barkley rings are not. Note that the larger relative size of the Atri rings does not explain their agreement with the prediction, because a weakly deformed Barkley ring of similar size still oscillates about 1.3 times slower than predicted (Figure 5(b)). One might suspect an interaction of opposite sides of the ring through the emitted waves. However, Panfilov–Rudenko rings of similar diameter show no residual, and the Barkley residual does not change between diameters of 0.8 and 1.3 wavelengths. Nandapurkar and Winfree simulated a scroll ring in this model at that became elliptic and periodically interchanged its axes (their Figure 3), an effect they attributed to the walls of their box;[48] we suspect that it was the binormal axis switching described here, which our rings show without walls.
We also tested a model with a direct physiological counterpart, the model of Atri et al. for waves in Xenopus oocytes,[49] in which diffuses with and the fraction of receptors that are not inactivated by does not diffuse. At a constant concentration of 0.1 , spirals rotate rigidly with a wavelength of 120 , and circular rings expand slowly with a drift coefficient that is nearly independent of the radius (supplementary material). Elliptic rings with mean radii between 93 and 146 again switch their axes (Fig. S5) and oscillate more slowly than predicted (13 half-cycles of five rings), or for the five half-cycles that start from nearly circular rings. In this respect, the Atri model is close to the Panfilov–Rudenko kinetics rather than to the Barkley model. Rings of 190 to 290 diameter would switch their axes every 5 to 12 min, although such rings may be hard to form and observe in oocytes (supplementary material).
We also considered simulations with the two-variable Oregonator model of the BZ reaction, for immobilized catalysts and for diffusing ones such as Ce(III/IV). However, the regimes we examined did not allow a test of axis switching. With an immobilized catalyst, spirals mostly meandered. With a diffusing catalyst, spirals rotated rigidly, but rings contracted too quickly to switch their axes. Near the transition to negative tension,[50] contraction should be slow, but the large spiral core would require very large rings and long runs, which we did not attempt (supplementary material).
Conclusions
One and a half centuries ago, Thomson showed that elliptic vortex rings in fluid dynamics oscillate as they travel, with their long and short axes repeatedly exchanging places.[17] We have found that elliptic scroll rings in excitable media perform the same oscillation, although their motion involves no macroscopic mass transport and arises from reaction and diffusion instead of fluid inertia. The oscillation occurred for positive, negative, and vanishing filament tension, and between two axis switches, the ring must travel about one radius.
The filament law that also describes vortex rings predicts the period of this oscillation without any adjustable parameter, and two of our three models kept this time within about 6%. The Barkley model, however, ran about 1.45 times slower at our main parameters. Elliptic scroll rings hence act as sensitive clocks whose rate reflects the kinetics of the medium. In a modest way, this
echoes Kelvin’s hope that the vibrations of vortex rings would identify atoms through their spectra,[36] a hope that had faded by the time J. J. Thomson, who had himself calculated these vibrations, discovered the electron.
Part of the delay of the Barkley rings is simple to understand, because their drift speed grows more slowly than their curvature. The remaining part surprised us. When we bent a ring out of its plane, its response depended mainly on the wavelength of the deformation, so that rings of different sizes and shapes responded alike if this wavelength was the same. Thin vortex filaments show a similar dependence, known since Kelvin’s work[16, 51] on their bending waves,[16, 51] and there it arises from the long-range flow that the vortex induces. For scroll rings, the origin remains open, and candidates are the finite size of the spiral core and the waves that the filament emits.
Two approaches could settle this question. Computing the couplings from the response functions of the spiral[38, 52, 53, 54] would show how much a local theory can explain, and encasing a ring in a moving “glass tube”, which removes long-range wave interactions,[34] would test whether the emitted waves are responsible. Axis switching might also be observable in the BZ reaction if the drift coefficient is not small compared with the tension, for example near the transition to negative tension, using elliptic initiating wires and photomasks to prepare the rings and optical tomography[55, 56] to follow them. In the gelled reagent of Winfree and Jahnke, the estimated drift was less than half of the shrinkage,[31] so that a ring would shrink at least about threefold before its first switch. Finally, strongly elongated vortex rings in fluid dynamics split,[20, 21, 22, 57] and it will be interesting to see whether highly eccentric scroll rings do the same.
Supplementary Material
See the supplementary material for derivations of the results of Sec. II, including the rigidity extension; numerical controls, the filament definitions, and the filament solver; the half-cycle analysis with all Barkley half-cycles and the comparison of procedures; the circular-ring calibrations at to 0.18 and the two-dimensional spirals of all models; the shape-mode analysis with all coefficients and their ranges (Tables S5 and S6), fits, filament definitions, twist, the test of the rigidity terms, straight filaments (Table S10), the wavenumber dependence of (Table S11, Fig. S3), and the comparison with thin vortex rings; the definition and numerical treatment of the Panfilov–Rudenko model with its circular and elliptic rings (Tables S12–S13, Fig. S4); the Atri model of waves (Tables S14–S15, Fig. S5); and our simulations of the Oregonator model.
Acknowledgments
This work was supported by the National Science Foundation under Grant No. CHE-2515818.
Author Declarations
Conflict of Interest
The author has no conflicts to disclose.
Author Contributions
O.S.: Conceptualization; Funding acquisition; Investigation; Validation; Writing – review & editing.
Data Availability
The simulation and analysis code is openly available on GitHub at https://github.com/osteinbock/elliptic-scroll-rings. The data that support the findings of this study can be regenerated with this code.
References
- [6]V. N. Biktashev and I. V. Biktasheva, “Dynamics of filaments of scroll waves,” in Engineering of Chemical Complexity II, World Scientific Lecture Notes in Complex Systems, edited by A. S. Mikhailov and G. Ertl (World Scientific, Singapore, 2014) pp. 221–238.DOI
- [14]T. Bánsági and O. Steinbock, “Negative filament tension of scroll rings in an excitable system,” Phys. Rev. E 76, 045202(R) (2007).pubmed.ncbi.nlm.nih.gov/17995050
- [16]P. G. Saffman, Vortex Dynamics (Cambridge University Press, Cambridge, 1992).
- [17]J. J. Thomson, A Treatise on the Motion of Vortex Rings (Macmillan, London, 1883).
- [18]R. J. Arms and F. R. Hama, “Localized-induction concept on a curved vortex and motion of an elliptic vortex ring,” Phys. Fluids 8, 553–559 (1965).DOI
- [19]S. C. Crow, “Stability theory for a pair of trailing vortices,” AIAA J. 8, 2172–2179 (1970).
- [20]M. R. Dhanak and D. De Bernardinis, “The evolution of an elliptic vortex ring,” J. Fluid Mech. 109, 189–216 (1981).DOI
- [21]Y. Oshima, N. Izutsu, K. Oshima, and A. K. M. F. Hussain, “Bifurcation of an elliptic vortex ring,” Fluid Dyn. Res. 3, 133–139 (1988).DOI
- [22]M. Cheng, J. Lou, and T. T. Lim, “Evolution of an elliptic vortex ring in a viscous fluid,” Phys. Fluids 28, 037104 (2016).
- [23]J. Li and J. Hong, “Probing dynamics of elliptical vortex rings via direct vorticity measurements with digital inline holography,” Exp. Fluids 65, 37 (2024).arxiv.org/abs/2310.19730
- [24]F. Hussain and H. S. Husain, “Elliptic jets. Part 1. Characteristics of unexcited and excited jets,” J. Fluid Mech. 208, 257–320 (1989).DOI
- [25]K. W. Schwarz, “Three-dimensional vortex dynamics in superfluid 4He: Line-line and line-boundary interactions,” Phys. Rev. B 31, 5782–5804 (1985).pubmed.ncbi.nlm.nih.gov/9936573
- [26]R. A. Van Gorder, “Quantum Hashimoto transformation and nonlinear waves on a superfluid vortex filament under the quantum local induction approximation,” Phys. Rev. E 91, 053201 (2015).pubmed.ncbi.nlm.nih.gov/26066272
- [27]M. Gabbay, E. Ott, and P. N. Guzdar, “Motion of scroll wave filaments in the complex Ginzburg–Landau equation,” Phys. Rev. Lett. 78, 2012–2015 (1997).DOI
- [28]I. S. Aranson and L. Kramer, “The world of the complex Ginzburg–Landau equation,” Rev. Mod. Phys. 74, 99–143 (2002).DOI
- [29]J. P. Keener and J. J. Tyson, “The motion of untwisted untorted scroll waves in Belousov–Zhabotinsky reagent,” Science 239, 1284–1286 (1988).pubmed.ncbi.nlm.nih.gov/17833214
- [30]S. Dutta and O. Steinbock, “Steady motion of hairpin-shaped vortex filaments in excitable systems,” Phys. Rev. E 81, 055202(R) (2010).pubmed.ncbi.nlm.nih.gov/20866287
- [31]A. T. Winfree and W. Jahnke, “Three-dimensional scroll ring dynamics in the Belousov–Zhabotinskii reagent and in the two-variable Oregonator model,” J. Phys. Chem. 93, 2823–2832 (1989).DOI
- [32]A. T. Winfree and S. H. Strogatz, “Organizing centres for three-dimensional chemical waves,” Nature 311, 611–615 (1984).pubmed.ncbi.nlm.nih.gov/6548297
- [33]P. M. Sutcliffe and A. T. Winfree, “Stability of knots in excitable media,” Phys. Rev. E 68, 016218 (2003).pubmed.ncbi.nlm.nih.gov/12935236
- [34]J. Binysh, C. A. Whitfield, and G. P. Alexander, “Stable and unstable vortex knots in excitable media,” Phys. Rev. E 99, 012211 (2019).DOI
- [35]F. Maucher and P. Sutcliffe, “Untangling knots via reaction-diffusion dynamics of vortex strings,” Phys. Rev. Lett. 116, 178101 (2016).DOI
- [36]W. Thomson, “On vortex atoms,” Phil. Mag. 34, 15–24 (1867).
- [37]H. Dierckx, H. Verschelde, Ö. Selsil, and V. N. Biktashev, “Buckling of scroll waves,” Phys. Rev. Lett. 109, 174102 (2012).DOI
- [38]H. Dierckx and H. Verschelde, “Effective dynamics of twisted and curved scroll waves using virtual filaments,” Phys. Rev. E 88, 062907 (2013).DOI
- [39]E. Nakouzi, Z. A. Jiménez, V. N. Biktashev, and O. Steinbock, “Analysis of anchor-size effects on pinned scroll waves and measurement of filament rigidity,” Phys. Rev. E 89, 042902 (2014).DOI
- [40]D. Barkley, “A model for fast computer simulation of waves in excitable media,” Physica D 49, 61–70 (1991).DOI
- [43]P. K. Brazhnik, V. A. Davydov, V. S. Zykov, and A. S. Mikhailov, “Vortex rings in excitable media,” Sov. Phys. JETP 66, 984–990 (1987), [Zh. Eksp. Teor. Fiz. 93, 1725–1736 (1987)].
- [44]W. Skaggs, E. Lugosi, and A. T. Winfree, “Stable vortex rings of excitation in neuroelectric media,” IEEE Trans. Circuits Syst. 35, 784–787 (1988).DOI
- [45]M. Courtemanche, W. Skaggs, and A. T. Winfree, “Stable three-dimensional action potential circulation in the FitzHugh–Nagumo model,” Physica D 41, 173–182 (1990).DOI
- [46]H. Dierckx, I. V. Biktasheva, H. Verschelde, A. V. Panfilov, and V. N. Biktashev, “Filament tension and phase locking of meandering scroll waves,” Phys. Rev. Lett. 119, 258101 (2017).DOI
- [47]A. V. Panfilov and A. N. Rudenko, “Two regimes of the scroll ring drift in the three-dimensional active media,” Physica D 28, 215–218 (1987).DOI
- [48]P. J. Nandapurkar and A. T. Winfree, “Dynamical stability of untwisted scroll rings in excitable media,” Physica D 35, 277–288 (1989).DOI
- [49]A. Atri, J. Amundson, D. Clapham, and J. Sneyd, “A single-pool model for intracellular calcium oscillations and waves in the Xenopus laevis oocyte,” Biophys. J. 65, 1727–1739 (1993).pubmed.ncbi.nlm.nih.gov/8274661
- [50]S. Alonso, F. Sagüés, and A. S. Mikhailov, “Negative-tension instability of scroll waves and Winfree turbulence in the Oregonator model,” J. Phys. Chem. A 110, 12063–12071 (2006).pubmed.ncbi.nlm.nih.gov/17064196
- [51]W. Thomson, “Vibrations of a columnar vortex,” Phil. Mag. 10, 155–168 (1880).DOI
- [52]H. Henry and V. Hakim, “Scroll waves in isotropic excitable media: Linear instabilities, bifurcations, and restabilized states,” Phys. Rev. E 65, 046235 (2002).pubmed.ncbi.nlm.nih.gov/12006004
- [53]I. V. Biktasheva, D. Barkley, V. N. Biktashev, G. V. Bordyugov, and A. J. Foulkes, “Computation of the response functions of spiral waves in active media,” Phys. Rev. E 79, 056702 (2009).pubmed.ncbi.nlm.nih.gov/19518588
- [54]I. V. Biktasheva, D. Barkley, V. N. Biktashev, and A. J. Foulkes, “Computation of the drift velocity of spiral waves using response functions,” Phys. Rev. E 81, 066202 (2010).pubmed.ncbi.nlm.nih.gov/20866496
- [55]A. T. Winfree, S. Caudle, G. Chen, P. McGuire, and Z. Szilagyi, “Quantitative optical tomography of chemical waves and their organizing centers,” Chaos 6, 617–626 (1996).pubmed.ncbi.nlm.nih.gov/12780291
- [56]T. Bánsági, Jr. and O. Steinbock, “Three-dimensional spiral waves in an excitable reaction system: Initiation and dynamics of scroll rings and scroll ring pairs,” Chaos 18, 026102 (2008).DOI
- [57]T. Kambe and T. Takao, “Motion of distorted vortex rings,” J. Phys. Soc. Jpn. 31, 591–599 (1971).DOI




