## Abstract

This paper derives the Whitham modulation equations for the Ostrovsky equation. The equations are then used to analyse localized cnoidal wavepacket solutions of the Ostrovsky equation in the weak rotation limit. The analysis is split into two main parameter regimes: the Ostrovsky equation with normal dispersion relevant to typical oceanic parameters and the Ostrovsky equation with anomalous dispersion relevant to strongly sheared oceanic flows and other physical systems. For anomalous dispersion a new steady, symmetric cnoidal wavepacket solution is presented. The new wavepacket can be represented as a solution of the modulation equations and an analytical solution for the outer solution of the wavepacket is given. For normal dispersion the modulation equations are used to describe the unsteady finite-amplitude wavepacket solutions produced from the rotation-induced decay of a Korteweg–de Vries solitary wave. Again, an analytical solution for the outer solution can be given. The centre of the wavepacket closely approximates a train of solitary waves with the results suggesting that the unsteady wavepacket is a localized, modulated cnoidal wavetrain. The formation of wavepackets from solitary wave initial conditions is considered, contrasting the rapid formation of the packets in anomalous dispersion with the slower formation of unsteady packets under normal dispersion.

## 1. Introduction

Oceanic internal waves are often assumed to have amplitudes small compared with the depth (weak nonlinearity) and wavelengths long compared with the depth (weak dispersion). Under these assumptions, the Korteweg–de Vries (KdV) equation
*γ* represents the relative strength of background rotation. It follows directly from (1.2) with *γ*≠0, that localized solutions of the Ostrovsky equation satisfy the zero-mass constraint

The effect of adding the non-local integral term to the right-hand side of the KdV equation is to introduce ‘large-scale’ dispersion into the system. For oceanic waves, the coefficient representing the strength of rotation is usually positive, *γ*>0, and this parameter regime has been described as the Ostrovsky equation with normal dispersion [7]. The opposite case, when *γ*<0, arises in strongly sheared ocean flows and acoustic problems [7] and is described as anomalous dispersion.

In the strong-rotation limit (*γ*≫1), it has been shown that the Ostrovsky equation supports localized wavepacket solutions for both cases of dispersion and several recent articles [7–10] have examined the dynamics of these wavepackets. In the weak-rotation limit (*γ*≪1), wavepackets have been observed for normal dispersion only. Helfrich [11], who was the first to discover these rotation-induced wavepackets, originally stated the solutions resemble modulated cnoidal wavetrain solutions of the KdV equation; however, these observations remain currently unexplained.

For the unperturbed KdV equation (1.1), it is possible to derive a set of equations, often referred to as the Whitham modulation equations [12], that describe the evolution of a modulated cnoidal wavetrain. The simplest derivation is based on the integrability of the KdV equation and thus does not extend directly to the Ostrovsky equation, which is not known to be integrable. Myint & Grimshaw [13] formulated a way of deriving the modulation equations for a KdV equation with a perturbation term,
*V* (*u*) is an arbitrary functional of *u* and *ϵ*≪1. Kamchatnov [14] has generalized their results but applying this work to the Ostrovsky equation is not direct because of the atypical nature of the non-local perturbation term.

This paper makes some minor changes to the method of Myint & Grimshaw [13] to derive a set of modulation equations for the Ostrovsky equation. Section 2 sketches briefly the derivation of the perturbed modulation equations which are verified against numerical integrations of the Ostrovsky equation. The derived equations are then used in §3 to describe wavepacket solutions of the Ostrovsky equation with §3a considering steady wavepacket solutions with anomalous dispersion and §3b unsteady wavepacket solutions with normal dispersion. Section 4 addresses the emergence of these wavepackets and in particular the rapid formation of packets under anomalous dispersion in contrast to their slower formation under normal dispersion.

## 2. Derivation of the modulation equations

Let the perturbative rotational term in the Ostrovsky equation be of order *ϵ*^{2} and take the derivative of equation (1.2), so that it is in the form
*ϵ*>0 and the positive and negative signs refer to normal and anomalous dispersion, respectively. If a solution of (2.1) has period 2*L*, is of compact support lying in |*x*|<*L* or decays sufficiently rapidly as *L* is infinite in the last case.

Following Myint & Grimshaw [13], the equations governing modulated nonlinear wavetrain solutions of (2.1) can be derived straightforwardly by and seeking a solution of the form
*θ* and the slow space and time variables *X* and *T* are defined as
*ω*(*X*,*T*), local wavenumber *k*(*X*,*T*) and local phase velocity *c*(*X*,*T*) are defined by

Let the modulations be periodic with period 2*L*, or of compact support lying in |*X*|<*L*, or decay sufficiently rapidly as *v*(*X*),
*L* infinite in the last case. If, further, *v* satisfies (2.2), then
*ϵ* to give
*C*(*T*). Let the period in *θ* of the carrier waves making up the wavetrain be 2*P*. Then, integrating (2.12) over a period 2*P* in *θ* and using the periodicity of *u*_{0} shows that *C*(*T*) is identically zero. Hence, (2.11) can be written
*θ* over a period 2*P*, and using the periodicity of *u*_{1}, gives
*P*-periodic function *v*(*θ*) is defined as
*u*_{0}, integrating with respect to *θ* over a period 2*P*, and using periodicity and the fact that *u*_{0} satisfies (2.13), gives
*v*(*θ*,*X*) of the slowly varying *X* and rapidly varying *θ*, of period 2*P* in *θ*,
*P* in *θ* from the construction of 〈*v*〉. Thus,

Combined with the consistency relation (2.6), equations (2.16) and (2.18) give the modulation equations for the leading-order solution *u*_{0}. The remainder of the analysis is confined to this leading-order solution and because the modulation equations form a closed system the subscript zero can be dropped without ambiguity. Substituting for *f*_{1} in (2.16) and (2.18) then using (2.20) gives
*a*
and
*b*
These equations are precisely those of Myint & Grimshaw [13] with the operator *V* (*u*) replaced by integration with respect to *X*, the slow-*x* variable.

Equation (2.13) has the exact periodic cnoidal solution,
*cn* is the Jacobi elliptic function with parameter *m* (described in [13] as the modulus, usually used for *m*^{1/2}). Choose *b*, so that *d* is the mean value of *u*, i.e. *d*=〈*u*〉, then
*a*
*b*
*c*
*d*
and all parameters are slowly varying functions of the form *a*(*X*,*T*), *b*(*X*,*T*), *c*(*X*,*T*), *d*(*X*,*T*), *k*(*X*,*T*), *m*(*X*,*T*) and *β*(*X*,*T*). The phase shift *θ*_{0}(*X*,*T*) can be determined only by considering the second-order terms in the expansion (2.3), but as its value does not affect the other parameters it is ignored in the subsequent analysis.

Integrating (2.13) twice with respect to *θ* gives
*a*
and
*b*
Differentiating (2.24) with respect to *θ* and dividing by 2*u*_{θ} gives
*u* and averaging gives
*a*
*b*
*c*
using *a*–*d*) between the seven parameters *a*, *b*, *c*, *d*, *k*, *m*, *β* only three are independent and hence equations (2.32*a*–*c*) are sufficient to determine the solution.

Following Whitham [12], equations (2.32*a*–*c*) can be greatly simplified introducing the variables
*p*, *q*, *r*, (*r*≤*q*≤*p*) given by
*a*
*b*
*c*
are the three roots of the polynomial on the right-hand side of (2.24). The modulation equations (2.32) can be written
*r*_{i} is a Riemann invariant that propagates with characteristic velocity *Q*_{i}, given by
*a*
*b*
*c*
The change of variables required to derive equations (2.35) from equations (2.32) is laborious (details can be found in [13,12]).

As a function of the Riemann invariants, the solution (2.22) takes the form
*ξ*_{0} is a phase shift. The wavelength *λ*=2*P*/*k* is given by
*u* are
*u* are given by

To test the results derived above, both the Ostrovsky equation (2.1) and equations (2.35) were integrated numerically with equivalent initial conditions. To integrate both the modulation equations (2.35) and the Ostrovsky equation (2.1) a pseudo-spectral Fourier discretization on a periodic domain in *x* and an adaptive fourth-order Runge–Kutta time-stepping in *t* was used. The initial condition was chosen as a cnoidal wavetrain (2.22) with *d*=0 and *m*=0.97. Figure 1 shows the results of the integration. The solutions obtained from integrating the Ostrovsky equation are shown by the solid lines and the envelope solutions obtained from integrating the modulation equations are shown by the dashed lines. Figure 1*a* shows the solution used for both integrations at *t*=0. Figure 1*b* shows a comparison of the integrations at *t*=325 when no large-scale dispersion is present. This case corresponds to the KdV equation and unperturbed Whitham modulation equations. Figure 1*c* shows a comparison of the integrations at *t*=325 for normal dispersion of magnitude *ϵ*=0.1 and Figure 1*d* shows a comparison of the integrations at *t*=300 for anomalous dispersion of magnitude *ϵ*=0.224.

The solutions of the perturbed modulation equations agree qualitatively with the corresponding integrations of the Ostrovsky equation. The modulation equations, however, are only valid for sufficiently small *ϵ* (≈0.1, in practice). For larger *ϵ*, the KdV equation cnoidal wave (2.22) is no longer a close approximation to a solution of the Ostrovsky equation, and the theory becomes less accurate. This is apparent in Figure 1*d,* where *ϵ*=0.224 and the wavetrain to the left of the disturbance no longer appears cnoidal.

A striking feature of the integrations shown in figure 1 is that the modulation equations appear to correctly capture the nonlinear steepening behaviour exhibited when either normal or anomalous dispersion are present (figure 1*c*–*d*). Furthermore, for the normal dispersion case in figure 1*c,* the three peaked structure located between *x*=−200 and −300 resembles that of the rotation-induced wavepackets seen in [8,9,11].

## 3. Application to wavepackets

### (a) Anomalous dispersion

For anomalous dispersion, it is known steady localized solutions exist. In the strong-rotation limit (*ϵ*≫1), these solutions originate from the point in wavenumber space where the linear phase and group velocities are equal [7], but as the strength of rotation is decreased the wavepacket structure of this family of solutions is eventually replaced by a soliton solution (figure 2*a*) that is described asymptotically by a KdV soliton with pedestal of order *ϵ*≪1 [15].

Section 4 describes how these single-peak solitary wave solutions arise naturally from the evolution of an initial KdV soliton in the presence of weak rotation. These are not, however, the only steadily propagating solutions of compact support. Figure 2*b*,*c* shows new steady wavepacket solutions found by a spectral Newton–Kantorovich iteration on a periodic grid.

The packet in figure 2*b* has the form of three adjacent KdV solitons. The packet in figure 2*c* resembles neither figure 2*a* nor figure 2*b*. As more peaks are added to the solution, the leading and trailing edges retain the form of solitary waves but the centre of the packet becomes more sinusoidal. Solutions with more peaks can be obtained only by decreasing the strength of the rotation, with the packet speed and rotation strength of the solutions in figure 2 related through the normalization here so that *c*=20*ϵ*/3. An approximation for the upper and lower envelopes of the multi-peak solutions was found by numerical interpolation through the peaks and troughs of the packets. It was observed that the maximum of the lower envelope occurred at the centre of the packet and its value approached zero as the number of peaks increased, a result that is derived below from the modulation theory. The small peak-number (strong rotation) limit of the solution family shown here coincides with the first member of the solution family in figure 4 of Obregon & Stepanyants [16], found, for stronger rotation than considered here, by placing two individual solitons, each similar to that in figure 2*a*, so that the maximum of one soliton coincides with a local minimum of the other.

For an anomalous wavepacket moving at constant velocity *c*_{0}, the time dependence of the modulation equations (2.35) can be removed by introducing the variables *r*_{i}=*r*_{i}′(*X*′), *Q*_{i}=*Q*_{i}′(*X*′), *d*=*d*′(*X*′), where *X*′=*X*−*c*_{0}*T*. Substituting the new variables into (2.35) and dropping dashes gives
*c*=*c*_{0}, one of the Riemann invariants, *r*_{i}, can be eliminated using (2.38). Further if the packet is assumed to be symmetric, forcing *D*(0)=0, then integrating equation (2.32*c*) after transforming into the frame, *X*′=*X*−*c*_{0}*T*, shows that
*A*_{0}=*A*(0) and *sgn*(*X*) is the signum function. The sign of the root in (3.2) has been chosen to agree with packets like that those shown in figure 2, where the ‘mass’ is initially negative when approaching the packet from negative *x*. Hence, for a steady, localized, symmetric wavepacket the modulation equations are reduced to a set of two ordinary differential equations.

To test if the modulation equations have wavepacket solutions, equations (3.1) were integrated numerically, using an initial condition corresponding to the wavepacket solution shown in figure 2*c*. An approximation for the envelope of the solution in figure 2*c* was found using numerical interpolation of the peaks and troughs. Using the known velocity *c*_{0} and equations (2.40) and (2.41), approximations for *r*_{1}(0) and *r*_{2}(0) were then found. Rather than use the analytical expression for *D* given in equation (3.2), the equation *dD*/*dX*=*d* was used, to avoid the difficulty of integrating from a stationary point.

The results of the integration are shown in figure 3. The dashed lines in figure 3*a* for *x*<180 show the envelope obtained from integrating equations (3.1), and the solid lines show the exact wavepacket solution from figure 2*c*. The values for the Riemann invariants obtained from the integration are also shown in figure 3*b*. The numerical integration failed for |*r*_{2}−*r*_{3}|<10^{−8} which corresponds to the *m*=1 solitary wave limit and at the point *r*_{2}=*r*_{3} the solution is likely to be discontinuous. Whitham [12] found that the non-rotating equations can have discontinuities across characteristics and these discontinuities can take the form of a jump from a solitary wave to a state of no waves, exactly as seen here in front of the leading solitary wave at *x*≈180.

In the limit *a*=0, and *r*_{2} and *r*_{3} must jump to the *r*_{1} solution at the leading edge of the solitary wave. Therefore, only *r*_{1} is continuous across this front. Additionally, in the limit *A*_{0} in (3.2) must be zero because of the zero mass constraint *X*_{0} is a constant. The solution (3.4) is plotted for *x*>180 in figure 3 where the constant *X*_{0} was fixed using the solution from figure 2*c*. Note in figure 3*b* that while *d* is continuous across the edge of the leading solitary wave, its derivative is not. At the point *r*_{2}=*r*_{3} (depicted by the vertical dotted line), *d* has infinite gradient.

For a steady solution with *c*=*c*_{0}, the identity _{r}=(∂/∂*r*_{1},∂/∂*r*_{2},∂/∂*r*_{3}) and *r*_{1},*r*_{2},*r*_{3}) space. Because *X* is constrained to the plane *r*_{1}+*r*_{2}+*r*_{3}=6*c*_{0} and so has only two degrees of freedom. For a localized solution, *A*=(3*c*_{0}−*r*_{1})(3*c*_{0}−*r*_{2})(3*c*_{0}−*r*_{3}). The symmetry condition *D*(0)=0, then implies that one of the Riemann invariants equals 3*c*_{0} at *X*=0. For *c*_{0}>0 and distinct *r*_{i}(0), the restriction 0≤*m*≤1 means that the only possibility is *r*_{2}(0)=3*c*_{0}. Either side of *X*=0 the function *D* is single signed and bounded away from zero and thus affects simply the speed at which a point moves along the flow lines. Figure 4 shows the trajectories (for *X*>0, with arrowheads in the direction of increasing *X*) projected onto the plane *r*_{2}=0. The flow is bounded by the lines *r*_{3}=(6*c*_{0}−*r*_{1})/2 and *r*_{3}=3*c*_{0}−*r*_{1} corresponding to *m*=1 and *r*_{2}=3*c*_{0}, respectively. Figure 4 shows that any trajectory starting in this sector moves to *m*=1 and hence the wavetrain part of the solution terminates in solitons. The condition *r*_{2}(0)=3*c*_{0} also means that solutions satisfy *X*=0 as observed in the numerical solutions. The bold (red) line in figure 4 shows the trajectory of the solution in figure 3.

The comparison above shows that the modulation equations accurately describe the wavepacket solution of figure 2*c*. Similar direct comparison, not reproduced here, of modulation solutions with the wavepacket solution of 2*b*, show similar accuracy.

### (b) Normal dispersion

In contrast to the anomalous case, for normal dispersion, there are no known steady wavepacket solutions. Although the wavepacket envelope propagates at a constant velocity, the carrier waves propagate at a velocity different to that of the envelope making the solution intrinsically unsteady. It has also been observed that the phase velocities are non-constant [9]. In the strong-rotation limit, the wavepacket dynamics are described asymptotically by a nonlinear Schrödinger equation bright soliton envelope solution with frequency-shifted linear carrier waves [9]. However, in the weak-rotation limit, despite being the most physically relevant parameter range, very little is known about the dynamics of the normal dispersion wavepackets.

For a normal dispersion wavepacket moving at constant velocity, *s*_{0}, the time dependence of the modulation equations can be eliminated by moving into the frame *X*−*s*_{0}*T*. Therefore, similar to the anomalous case, the solution in the translating frame is described by the following equations
*s*_{0} replacing *c*_{0} and an opposite sign on the right-hand side representing the change in dispersion. If the packet is assumed to be symmetric, forcing *D*(0)=0, then integrating equation (2.32c) after transforming into the frame, *X*−*s*_{0}*T*, gives
*ϕ*=2*A*+2(*s*_{0}−*c*)(*cd*+*B*) and *ϕ*_{0} is the value of *ϕ* at *X*=0. The sign of the root in (3.7) has been chosen to describe a solution where the ‘mass’ is initially positive when approaching the packet from negative *x* and hence is opposite in sign to (3.2). For a localized, symmetric wavepacket with normal dispersion, the modulation equations are therefore reduced to a set of three ordinary differential equations. There is one equation more than for the anomalous case as it is not possible to eliminate a Riemann invariant, using the expression for *c* in (2.38).

For steady wavepackets with anomalous dispersion, it was shown in figure 4 that all solution trajectories travel towards the limit *m*=1 and terminate in solitary waves. The direction of the trajectories in figure 4 is determined by sign of the dispersion. For wavepackets with normal dispersion, it is plausible to predict solution trajectories would move in the opposite direction and travel towards the limit *m*=0, where they terminate in infinitesimal linear waves. This hypothesis is in agreement with the wavepackets found by Whitfield & Johnson [9], where it was noted that the wavepackets had exponentially small radiating tails at their edges. Under this assumption, a solution similar to (3.4) can be found for the outer limit of the wavepacket.

Following Myint & Grimshaw [13], we consider the limit *a*/*m* fixed and approximately given by 12*k*^{2}*β*^{2}. In this limit, to leading order in *a*, it can be shown *D*^{2}=2*ϕ*_{0}−*s*_{0}*d*^{2}+2*d*^{3}/3 and hence for a localized solution *ϕ*_{0} in (3.7) must be zero because of the zero mass constraint *r*_{3} gives
*X*_{0} is a constant. In the limit *a* is given by *u*=*d* and hence it is expected that (3.9) would be the solution at the wavepacket edge.

In Whitfield & Johnson [9], numerical solutions were obtained for unsteady wavepackets with normal dispersion. Therefore, to test if the modulation equations have wavepacket solutions, equations (3.6) were integrated numerically using an initial condition corresponding to the wavepacket with the weakest rotation in [9]. Using the measured envelope values from [9] approximations for *r*_{1}(0), *r*_{2}(0) and *r*_{3}(0) could be parametrized as a function of *c*(0). The constant *ϕ*_{0} in (3.7) was assumed to be zero for the reasons outlined in the preceding paragraph and hence imposing the symmetry condition, *D*(0)=0, fixed the value of *s*_{0} for a given *c*(0). The lack of data made it impossible to determine *c*(0) and therefore it was left as a parameter. As was done for anomalous dispersion, the equation d*D*/d*X*=*d* was used to avoid the difficulty of integrating from a stationary point.

The result of one integration is shown in figure 5. In figure 5*a,* the dashed lines for |*x*|<60 show the envelope obtained from numerically integrating equations (3.6) and the dashed lines for |*x*|>100 show the analytical solution (3.9) with *X*_{0}=−1.7756. The dotted lines in figure 5*a* show the measured envelope obtained by Whitfield & Johnson [9], and the solid lines show the exact unsteady solution from [9] at its maximum value. The Riemann invariant values obtained from the integration are also shown in figure 5*b*. The value of *c*(0) in figure 5 was chosen to produce the best agreement with the measured packet envelope, however, for a wide range of different *c*(0), it was found that the numerical integration failed before it could reach the edge of the wavepacket. This was because *dr*_{2}/*dX*| being extremely large. The reason for this behaviour is unclear. The approximate initial conditions are a possible source of error, as is the assumption of a symmetric solution, especially as it was shown conclusively in [9] that for *ϵ*≫1 wavepackets had asymmetry of order 1/*ϵ*.

Relaxing the localization condition by allowing *ϕ*_{0} to be non-zero allows initial conditions for the modulation equations to be chosen, so that their numerical integration can proceed indefinitely. Using the initial conditions of the wavepacket solution shown in figure 5 as a starting point, the values of *s* and *c*(0) were varied until a numerical integration was found, so that *Q*_{2} remained distinct from *s*_{0}. Figure 6 gives results from one such integration, showing a periodic solution of the modulation equations and suggesting the Ostrovsky equation possesses unsteady, periodically modulated cnoidal wave solutions with the modulation steady in a frame moving at speed *s*.

It has been suggested [17] that rotation-induced wavepackets, like that of figure 5*a*, could offer an explanation as to why non-rank-ordered internal solitary wavepackets are sometimes observed instead of the expected rank-ordered wavepacket structure of solutions of the (non-rotating) KdV equation. The modulated wavetrain of figure 6*a* provides further evidence that rotational effects influence internal wavetrains in such a way that they form non-rank-ordered structures. It remains to be shown, however, whether wavetrains of this form arise in the Ostrovsky equation with more general initial conditions.

## 4. Wavepacket emergence

The variables and analysis of §3 give a context for discussing the emergence of the wavepacket solutions above from KdV soliton initial conditions under weak rotation as observed by Grimshaw & Helfrich [8], Grimshaw *et al.* [7] and Whitfield & Johnson [9]. In terms of the variables (2.4), equation (2.1) has the inner solitary wave solution
*a*_{0}, *a*=*a*_{0} and *d*(0)=0. As *u*^{i}→*ϵd*(*T*) and so merges smoothly onto an outer solution *u*=*ϵu*^{o}(*X*,*T*), which satisfies, to leading order in *ϵ*
*c* can be taken to be constant with *c*=*c*_{o}=*a*/3>0. Moving to a frame translating to the right with speed *c*_{0} by writing *ξ*=*X*−*c*_{0}*T* gives
*a*
and
*b*
The problem is closed at this order by the mass constraint (1.3) which gives, to leading order,
*a*
*b*
using (4.3), where [ ]_{0} denotes the jump in the enclosed quantity across *ξ*=0 and *δ*(*ξ*) denotes the Dirac delta function. Writing
*ω*(*κ*)=−*c*_{0}*κ*±1/*κ*, with solution vanishing at *T*=0,
*κ*^{2}=±1/*c*_{0} and a superposition of dispersive waves of frequency *ω*(*κ*) and group velocity *c*_{g}(*κ*)=−*c*_{0}∓1/*κ*^{2}. The flow evolution and final state for normal and anomalous dispersion differ.

For normal dispersion, *c*_{g} is strictly negative for all *κ* and all waves propagate in the negative *ξ* direction. The poles in (4.7) and (4.8) lie on the real *κ* axis and thus contribute a steady downstream lee-wavetrain behind the soliton with the deviation of the solution from its final steady state decaying as *T*^{−1/2} at large *T*, so that as *a*. There is an order *ϵ*^{2} momentum flux in the negative *x*-direction in the lee of the soliton which exerts a drag on the soliton, leading to a decay in the soliton over the longer timescale *ϵT*=*ϵ*^{2}*t*. This longer-time decay of the soliton is discussed in greater detail by Grimshaw *et al.* [18,7].

For anomalous dispersion, the poles in (4.7) and (4.8) lie on the imaginary *κ* axis and thus contribute a symmetric steady contribution decaying exponentially away from a maximum at *ξ*=0. The group velocity vanishes when *κ*^{2}=*c*_{0}. Thus, transient long waves with *T*^{−1/2}, whereas transient short waves with *T*^{−1/2}. Waves with *dc*_{g}/*dκ*=2/*κ*^{3} vanishes nowhere, these waves decay as *T*^{−1/3}. Thus, as *b*. The final flow state is symmetric without waves and so suffers no wave drag. Unlike the case of normal dispersion, there is no further leading-order evolution over the longer timescale *ϵT*=*ϵ*^{2}*t*, in agreement with the numerical integrations of [7]. Note that although the *ϵ*^{2}, leading to a reduction in its amplitude. This higher-order effect is not captured by leading order analysis here. The departure from the analysis in [7], which leads to the prediction of a decaying soliton there, is that the far-field boundary condition on the leading order solution *u*^{(0)} of [7], compatible with (4.10), is *u*^{(1)}=constant is an allowable solution of the adjoint equation (3.9) of [7], giving conservation of mass to leading order in *γ*.

## 5. Discussion

The Whitham modulation equations for the Ostrovsky equation, derived using the method of Myint & Grimshaw [13], have been used to find wavepacket solutions of the Ostrovsky equation in the weak-rotation limit. Considering wavepacket solutions in the modulation equations offers two advantages: first, the anti-derivative representing large-scale dispersion is simplified to an algebraic term and, second, intrinsically unsteady wavepacket solutions of the Ostrovsky equation with normal dispersion, whose phase and group velocities differ, become steady solutions.

For anomalous dispersion, a new type of cnoidal wavepacket solution was obtained and shown to be representable as a solution of the modulation equations. The modulation equations showed, in agreement with numerical integrations, that the wavepacket has minimum zero at its centre and that the wavetrain section of the packet terminates in a KdV soliton at each edge. An analytical solution for the outer section of the wavepacket where no waves exist was also found.

In contrast to the anomalous dispersion case, for normal dispersion, it is argued that any localized symmetric wavetrain solution should approach infinitesimal linear waves at its edge and more solitary-like waves at its centre. Based on this assumption an analytical solution for the outer limit of the wavepacket was found. The outer solution was compared with data from a normal dispersion wavepacket found in [9], showing good agreement. The same packet data was also compared with solutions of the modulation equations for the inner wavetrain region of the packet and agreed well over the central portion of the packet. Towards the edge of the packet, the characteristic velocity of the intermediate Riemann invariant became equal to the packet propagation speed introducing an infinite gradient in the integrations across the packet. This singularity appears to be a robust feature of the equations when applied to these packets but its physical meaning has not been determined. The good agreement of the modulation solution with the data over the majority of the wavepacket suggests that the modulation equations do have wavepacket solutions for normal dispersion, supporting the observation of Helfrich [11] that the wavepackets under normal dispersion can be regarded as modulated cnoidal waves.

Section 4 describes the emergence of these wavepackets from KdV soliton initial conditions under weak rotation. For anomalous dispersion the flow adjusts over times of order *ϵ*^{−1} to a zero-drag symmetric state that undergoes no further evolution at this order. For normal dispersion, the flow evolves over times of order *ϵ*^{−1} to an asymmetric non-zero drag form that undergoes further slow decay over times of order *ϵ*^{−2}, as described by Grimshaw *et al.* [18].

Cnoidal wavepacket solutions are known [10] to arise from localized initial conditions in the Ostrovsky equation with normal dispersion. It remains to be shown whether the cnoidal wavepackets described above for the Ostrovsky equation with anomalous dispersion will arise from general initial conditions.

## Authors' contributions

The authors are equally responsible for the contributions to the paper and both authors gave final approval for publication.

## Competing interests

We declare we have no competing interests.

## Funding

The funding of the research rested solely on the authors’ being employed and supported by UCL.

- Received September 17, 2016.
- Accepted December 12, 2016.

- © 2017 The Author(s)

Published by the Royal Society. All rights reserved.