Introduction
Low-mode, long wavelength internal gravity waves in oceans can travel thousands of kilometers from their generation site without dissipation (Zhao et al. 2016). The energy in these long waves can cascade to small scales through a variety of mechanisms, such as nonlinear interactions among the waves ([27 , 26 ] ), scattering through interaction with the seafloor topography ([22 ] ), and scattering through interaction with turbulent quasigeostrophic flows ([16 ] ). This transfer of energy to small scales will eventually lead to turbulence and mixing, which is essential for maintaining the meridional overturning circulation [31 ] . In weakly nonlinear wave-wave interactions, an internal gravity wave can become unstable via resonant triad interactions if it has the largest frequency in the triad [14 ] ; through this mechanism, energy is irreversibly transferred from a high frequency and low wavenumber primary wave to lower frequency and higher wavenumber secondary waves. In a resonant internal wave triad, a wave of angular frequency ω 3 \omega_3 ω 3 and wavevector k 3 \mathbf{k}_3 k 3 can resonantly transfer its energy to two ‘daughter’ waves when both the conditions k 3 = k 1 + k 2 \mathbf{k}_3 = \mathbf{k}_1 + \mathbf{k}_2 k 3 = k 1 + k 2 and ω 3 = ω 1 + ω 2 \omega_3 = \omega_1 + \omega_2 ω 3 = ω 1 + ω 2 are met [8 , 14 , 33 ] . The former condition is a consequence of the quadratic nonlinearity of the Navier-Stokes equations.
Since ocean’s density stratification is non-uniform, recent efforts have been directed towards understanding energy transfer in non-uniformly stratified fluids. In Varma & Mathur [36 ] , the conditions for the existence of resonant weakly nonlinear wave-wave interactions in a non-uniform stratification was studied. They proved that resonant triads and self-interactions can exist if (i) they satisfy the horizontal wavenumber condition, and (ii) each wave’s functional form in z z z −direction is non-orthogonal to the nonlinear forcing terms. Wunsch [39 ] studied self-interaction of an internal wave mode in the presence of a non-uniform stratification, the latter was simplified using a 3-layer model. It was shown that amplitude of the superharmonic wave, which is forced by the self-interaction of a parent wave, can be highly sensitive to changes in the stratification profile characteristics such as pycnocline depth and strength. Moreover, Liang et al. (2017) showed that self-interaction also occurs in the presence of uniform stratification, provided the nonlinear terms in the free surface boundary condition are taken into account. Self-interaction was also numerically studied by Sutherland [34 ] , and it was observed that, in the presence of non-uniform stratification, self-interaction of an internal wave mode was more dominant than triadic interactions for low Coriolis frequency. Furthermore, Baker & Sutherland (2020) studied self-interaction of a mode under angular frequency mismatch, and found that the daughter wave (superharmonic wave) can return its energy to the parent wave.
Apart from the weakly nonlinear wave-wave interactions, wave-topography interactions (where the topography is of small amplitude) have also been extensively studied. In Buhler & Holmes-Cerfon [3 ] , the decay of a mode-1 internal tide due to its interaction with a small amplitude sea floor topography was studied using ray-tracing. It was shown that if the bottom bathymetry is ‘resonant’ (see §6 for more detail), the internal mode-1 interacts with the bathymetry and resonantly gives its energy to the higher modes. The topography in this case acts like a stationary wave with zero angular frequency. In Couston et al. [5 ] , this scattering process was explored in a 3-dimensional setting, where the mode-1 internal wave was obliquely incident on a small amplitude bottom topography. Buhler & Holmes-Cerfon [3 ] and Li & Mei (2014) also focused on scattering of an internal gravity wave (IGW) by a small amplitude, stationary, zero mean random topography. Li & Mei (2014) consider topographies that vary in zonal and meridional directions. Both studies, under realistic parameters, estimate a decay length scale of about 500 − 1000kms for the mode-1 wave.
In Mathur et al. [28 ] , Green’s function approach along with numerical simulations was used to study IGW scattering under the linear, inviscid limit in the presence of constant and non-constant buoyancy frequency in a 2-dimensional setting for large amplitude topographies. Height of the topography and criticality (C r C_r C r ) were the two main factors that influence IGW scattering. In general, subcritical (C r < 1 C_r < 1 C r < 1 ) topographies were found to scatter the incoming wave lesser than supercritical topographies (C r > 1 C_r > 1 C r > 1 ). Critical topographies (C r ≈ 1 C_r \approx1 C r ≈ 1 ) were the most proficient in scattering the incoming wave. Scattering of large amplitude waves, breaking, and the ensuing kinetic energy dissipation is a very important quantity to study since it provides an estimate for local diffusivity. Internal wave breaking due to different types of topographies was focused in Legg [21 ] . In particular, a condition for internal wave breaking was given using the incoming wave’s Froude number. The Froude number for a mode-1 wave is defined as:
F r = U π α H ω (1) Fr = \frac{U\pi\alpha}{H\omega} \tag*{(1)} F r = H ω U π α (1)
where U U U and ω \omega ω are respectively the peak horizontal velocity and frequency of the wave, and H H H is the depth of the domain. α \alpha α is the slope of the wave. As the mode-1 wave shoals up a large amplitude topography its F r Fr F r increases. It is empirically determined that if wave’s F r Fr F r reaches a range of (0.3 − 1) due to shoaling, then the wave is prone to breaking. Wave’s local Froude number can also increase due to reflection from a topography. Highly nonlinear features such as bores were observed in regions of the topography where the local Froude number was greater than 1 [23 ] . Scattering and dissipation due to large amplitude highly supercritical topographies were focused in Klymak et al. (2013). Interestingly, it was observed that a mode-1 wave loses a maximum ∼ 20% of its energy at an isolated tall supercritical topography.
This paper is the ‘Part-2’ of Gururaj & Guha [12 ] , in which the effect of nonuniform (albeit slowly varying) stratification on internal wave triads in an unbounded domain was theoretically and numerically studied. It was shown that the variation in stratification profile may significantly affect the nonlinear coupling coefficients, (vertical wavenumber) detuning, and group speed of the wave packets constituting a triad, and hence the ensuing energy transfer. Different triads were also observed to undergo different amounts of detuning for the same change in the background stratification. The present paper extends the paradigm explored in Gururaj & Guha [12 ] to a vertically bounded domain with a mild slope bathymetry . To the best of our knowledge, this is the first work that considers the effect of the variation of the ocean depth on internal wave triads. A simplified schematic of the setup is given in Figure 1 . The motivation behind this study stems from the simple fact that ocean depth varies spatially (hence consideration of constant depth might be an over-simplification), hence waves can move from one depth to another while they are interacting in a medium of varying stratification. While the effect of change in fluid depth on resonant and near-resonant interactions between three distinct waves is the primary focus of this paper, we have also studied higher order self-interactions among the internal waves in the presence of a small amplitude topography. As an analogy, such higher order interactions have been studied in Alam et al. (2009) for surface waves. Wave-wave interactions in nonuniformly stratified vertically bounded domains have been previously considered in various studies Refs. (Baker & Sutherland 2020; Varma et al. 2020; Young et al. 2008). In this paper, reduced order model for wave-wave interactions in a region of varying h h h are derived, and hence spatially varying nonlinear coupling coefficients, group speed and detuning are all involved. Moreover, the equations can model wave-wave interactions of wave trains or finite width wave packets in a region of varying h h h . In § 6 , we have also derived (and validated numerically) reduced order equations which model higher order self interactions in the presence of a small amplitude topography. Equations in § 6 can also be used to model standard resonant self interactions in the presence of slowly varying large amplitude topographies.
Figure 1. A general schematic of the problem to be studied. The figure shows the streamfunction field of three wave packets interacting in the presence of a varying bathymetry h ( x ) h(x) h ( x ) . H H H is the mean depth (equivalent to the depth in flat bathymetry situation), while H b ( x ) H_b(x) H b ( x ) denotes the submarine topography shape. (b) The stratification profile used in constructing the modes. The same non-uniform stratification model is used throughout the paper.
The paper is organized as follows. In § 2 , we derive the amplitude evolution equations of the constituent waves of a triad in the presence of a slowly varying bathymetry using the Boussinesq Navier-Stokes equations in the f f f −plane. To derive these equations, the streamfunction, buoyancy perturbation, and meridional velocity due to each wave are assumed to be a product of a slowly varying amplitude and a rapidly varying phase that are functions of space and time. In § 3 and § 4 , the effect on the horizontal wavenumber condition when waves interact in a region of varying ocean depth in the presence of uniform and non-uniform stratification are respectively studied. In § 5 , we have studied the effect of ocean depth variation on the rate of energy transfer in triadic interactions and self-interactions in the presence of a non-uniform stratification. In § 6 , we analyze higher order self-interactions of a wave in the presence of small-amplitude monochromatic topography. In § 7 , the reduced order equations derived in this paper are validated by solving the full Boussinesq equations using an open source code Dedalus (Burns et al. [4 ] ). The paper has been summarized in § 8 .
Derivation of the governing equations in terrain-following coordinates
The incompressible, inviscid, 2D (in the x x x –z z z plane) Navier–Stokes equations on the f f f –plane under the Boussinesq approximation, hereafter referred to as the Boussinesq equations, can be expressed in terms of the perturbation streamfunction ψ \psi ψ , meridional velocity v v v (along y y y -direction), and the perturbation buoyancy b b b as follows:
∂ ∂ t ( ∇ 2 ψ ) + ∂ b ∂ x − f ∂ v ∂ z = − { ∇ 2 ψ , ψ } , (2) \frac{\partial}{\partial t}\left(\nabla^2\psi\right)+\frac{\partial b}{\partial x}-f\frac{\partial v}{\partial z}=-\{\nabla^2\psi,\psi\}, \tag*{(2)} ∂ t ∂ ( ∇ 2 ψ ) + ∂ x ∂ b − f ∂ z ∂ v = − { ∇ 2 ψ , ψ } , (2)
∂ v ∂ t + f ∂ ψ ∂ z = − { v , ψ } , (3) \frac{\partial v}{\partial t}+f\frac{\partial\psi}{\partial z}=-\{v,\psi\}, \tag*{(3)} ∂ t ∂ v + f ∂ z ∂ ψ = − { v , ψ } , (3)
∂ b ∂ t − N 2 ∂ ψ ∂ x = − { b , ψ } . (4) \frac{\partial b}{\partial t}-N^2\frac{\partial\psi}{\partial x}=-\{b,\psi\}. \tag*{(4)} ∂ t ∂ b − N 2 ∂ x ∂ ψ = − { b , ψ } . (4)
Here N 2 ( z ) ≡ − ( g / ρ ∗ ) ( d ρ ˉ / d z ) N^2(z)\equiv-(g/\rho^*)(d\bar{\rho}/dz) N 2 ( z ) ≡ − ( g / ρ ∗ ) ( d ρ ˉ / d z ) is the squared buoyancy frequency, ρ ˉ \bar{\rho} ρ ˉ is the base density profile, ρ ∗ \rho^* ρ ∗ is the reference density, and g g g is the acceleration due to gravity (directed along − z -z − z ). The perturbation buoyancy is defined as b ≡ − g ρ / ρ ∗ b\equiv-g\rho/\rho^* b ≡ − g ρ / ρ ∗ , where ρ \rho ρ is the perturbation density.
The operator { G 1 , G 2 } ≡ ( ∂ G 1 / ∂ x ) ( ∂ G 2 / ∂ z ) − ( ∂ G 1 / ∂ z ) ( ∂ G 2 / ∂ x ) \{G_1,G_2\}\equiv(\partial G_1/\partial x)(\partial G_2/\partial z)-(\partial G_1/\partial z)(\partial G_2/\partial x) { G 1 , G 2 } ≡ ( ∂ G 1 / ∂ x ) ( ∂ G 2 / ∂ z ) − ( ∂ G 1 / ∂ z ) ( ∂ G 2 / ∂ x ) denotes the Poisson bracket, and f f f is the Coriolis frequency. Viscous effects have been neglected owing to the fact that we consider waves with long wavelengths.
The fluid domain is bounded at the top (z = 0 z=0 z = 0 ) by a rigid-lid (i.e., zero vertical velocity, leading to the boundary condition ψ ( x , 0 ) = 0 \psi(x,0)=0 ψ ( x , 0 ) = 0 ). The bottom boundary at z = h ( x ) z=h(x) z = h ( x ) satisfies the impenetrable boundary condition ψ ( x , h ( x ) ) = 0 \psi(x,h(x))=0 ψ ( x , h ( x )) = 0 .
Instead of solving the fully nonlinear equations (2) –(4) numerically, we combine (2) –(4) into a single equation and employ a multiple-scale analysis. To this end, we perform ∂ \partial ∂ (2.1a)/ ∂ t + f ∂ /\partial t+f\,\partial / ∂ t + f ∂ (2.1b)/ ∂ z − ∂ /\partial z-\partial / ∂ z − ∂ (2.1c)/ ∂ x /\partial x / ∂ x , which results in
∂ 2 ∂ t 2 ( ∇ 2 ψ ) + N 2 ∂ 2 ψ ∂ x 2 + f 2 ∂ 2 ψ ∂ z 2 = − ∂ ∂ t ( { ∇ 2 ψ , ψ } ) + ∂ ∂ x ( { b , ψ } ) − f ∂ ∂ z ( { v , ψ } ) . (5) \frac{\partial^2}{\partial t^2}\left(\nabla^2\psi\right)+N^2\frac{\partial^2\psi}{\partial x^2}+f^2\frac{\partial^2\psi}{\partial z^2}
=-\frac{\partial}{\partial t}\left(\{\nabla^2\psi,\psi\}\right)+\frac{\partial}{\partial x}\left(\{b,\psi\}\right)-f\frac{\partial}{\partial z}\left(\{v,\psi\}\right). \tag*{(5)} ∂ t 2 ∂ 2 ( ∇ 2 ψ ) + N 2 ∂ x 2 ∂ 2 ψ + f 2 ∂ z 2 ∂ 2 ψ = − ∂ t ∂ ( { ∇ 2 ψ , ψ } ) + ∂ x ∂ ( { b , ψ } ) − f ∂ z ∂ ( { v , ψ } ) . (5)
Following the approach of Maugé & Gerkema [29 ] , we now change the governing equations to terrain following coordinates, where a new variable (η \eta η ) is defined as:
η ≡ − z h ( x ) . (6) \eta\equiv-\frac{z}{h(x)}. \tag*{(6)} η ≡ − h ( x ) z . (6)
According to the definition (6) , the bottom boundary condition at z = h ( x ) z=h(x) z = h ( x ) would now be enforced at η = − 1 \eta=-1 η = − 1 , while the surface boundary condition at z = 0 z=0 z = 0 remains unaltered, except that it is now at η = 0 \eta=0 η = 0 . The governing equations, which are in the x x x –z z z coordinates, need to be transformed into the x x x –η \eta η coordinates. The correspondence between the variables in the x x x –z z z and x x x –η \eta η coordinate systems are as follows:
ψ ( x , z , t ) ⇒ Ψ ( x , η , t ) , b ( x , z , t ) ⇒ B ( x , η , t ) , v ( x , z , t ) ⇒ V ( x , η , t ) . (7) \psi(x,z,t)\Rightarrow\Psi(x,\eta,t),\qquad b(x,z,t)\Rightarrow B(x,\eta,t),\qquad v(x,z,t)\Rightarrow V(x,\eta,t). \tag*{(7)} ψ ( x , z , t ) ⇒ Ψ ( x , η , t ) , b ( x , z , t ) ⇒ B ( x , η , t ) , v ( x , z , t ) ⇒ V ( x , η , t ) . (7)
On transforming the differential operators from x x x –z z z coordinates to x x x –η \eta η coordinates and substituting the transformed variables in (5) , we arrive at
[ ∂ 2 ∂ t 2 ( L x x + L η η ) + N 2 ( − h ( x ) η ) L x x + f 2 L η η ] Ψ = − ∂ ∂ t [ J { ( L x x + L η η ) Ψ , Ψ } ] + L x ( J { B , Ψ } ) − f L η ( J { V , Ψ } ) . (8) \left[\frac{\partial^2}{\partial t^2}(L_{xx}+L_{\eta\eta})+N^2(-h(x)\eta)L_{xx}+f^2L_{\eta\eta}\right]\Psi
=-\frac{\partial}{\partial t}\left[J\{(L_{xx}+L_{\eta\eta})\Psi,\Psi\}\right]+L_x\left(J\{B,\Psi\}\right)-fL_\eta\left(J\{V,\Psi\}\right). \tag*{(8)} [ ∂ t 2 ∂ 2 ( L xx + L η η ) + N 2 ( − h ( x ) η ) L xx + f 2 L η η ] Ψ = − ∂ t ∂ [ J {( L xx + L η η ) Ψ , Ψ } ] + L x ( J { B , Ψ } ) − f L η ( J { V , Ψ } ) . (8)
where the operators L x L_x L x , L η L_\eta L η , L x x L_{xx} L xx , L η η L_{\eta\eta} L η η , and J { G 1 , G 2 } J\{G_1,G_2\} J { G 1 , G 2 } have the following definitions:
L x ≡ ∂ ∂ x + η h ∂ h ∂ x ∂ ∂ η , L η ≡ − 1 h ∂ ∂ η , L η η ≡ 1 h 2 ∂ 2 ∂ η 2 , (9) L_x\equiv\frac{\partial}{\partial x}+\frac{\eta}{h}\frac{\partial h}{\partial x}\frac{\partial}{\partial\eta},\qquad
L_\eta\equiv-\frac{1}{h}\frac{\partial}{\partial\eta},\qquad
L_{\eta\eta}\equiv\frac{1}{h^2}\frac{\partial^2}{\partial\eta^2}, \tag*{(9)} L x ≡ ∂ x ∂ + h η ∂ x ∂ h ∂ η ∂ , L η ≡ − h 1 ∂ η ∂ , L η η ≡ h 2 1 ∂ η 2 ∂ 2 , (9)
L x x ≡ ∂ 2 ∂ x 2 + η 2 h 2 ( ∂ h ∂ x ) 2 ∂ 2 ∂ η 2 − 2 η h ( ∂ h ∂ x ) ∂ 2 ∂ η ∂ x + η h [ 2 h ( ∂ h ∂ x ) 2 − ∂ 2 h ∂ x 2 ] ∂ ∂ η , (10) L_{xx}\equiv\frac{\partial^2}{\partial x^2}+\frac{\eta^2}{h^2}\left(\frac{\partial h}{\partial x}\right)^2\frac{\partial^2}{\partial\eta^2}-2\frac{\eta}{h}\left(\frac{\partial h}{\partial x}\right)\frac{\partial^2}{\partial\eta\partial x}+\frac{\eta}{h}\left[\frac{2}{h}\left(\frac{\partial h}{\partial x}\right)^2-\frac{\partial^2h}{\partial x^2}\right]\frac{\partial}{\partial\eta}, \tag*{(10)} L xx ≡ ∂ x 2 ∂ 2 + h 2 η 2 ( ∂ x ∂ h ) 2 ∂ η 2 ∂ 2 − 2 h η ( ∂ x ∂ h ) ∂ η ∂ x ∂ 2 + h η [ h 2 ( ∂ x ∂ h ) 2 − ∂ x 2 ∂ 2 h ] ∂ η ∂ , (10)
J { G 1 , G 2 } = L x ( G 1 ) L η ( G 2 ) − L η ( G 1 ) L x ( G 2 ) . (11) J\{G_1,G_2\}=L_x(G_1)L_\eta(G_2)-L_\eta(G_1)L_x(G_2). \tag*{(11)} J { G 1 , G 2 } = L x ( G 1 ) L η ( G 2 ) − L η ( G 1 ) L x ( G 2 ) . (11)
For performing multiple-scale analysis, we assume wavelike perturbations, and the streamfunction due to the j j j -th wave (j = 1 , 2 , 3 j=1,2,3 j = 1 , 2 , 3 ) is given according to the following ansatz:
Ψ j = a j ( ϵ x x , ϵ t t ) Ξ ‾ j ( x , η , t ) + c . c . , (12) \Psi_j=a_j(\epsilon_x x,\epsilon_t t)\overline{\Xi}_j(x,\eta,t)+\mathrm{c.c.}, \tag*{(12)} Ψ j = a j ( ϵ x x , ϵ t t ) Ξ j ( x , η , t ) + c.c. , (12)
where ‘c.c.’ denotes the complex conjugate, a j a_j a j is the slowly varying complex amplitude, and Ξ ‾ j ( x , η , t ) \overline{\Xi}_j(x,\eta,t) Ξ j ( x , η , t ) is the rapidly varying phase part of the j j j -th wave. The small parameters ϵ t \epsilon_t ϵ t and ϵ x \epsilon_x ϵ x are respectively used to denote the weak variation of the amplitude function with time and streamwise (x x x ) direction. The amplitude is assumed to be an O ( ϵ a ) O(\epsilon_a) O ( ϵ a ) quantity, where ϵ a \epsilon_a ϵ a is a small parameter. The bathymetry (h h h ), which is simply the negative of the fluid depth, is assumed to be of the form:
h = − H + ϵ h H b ( k b x ) , (13) h=-H+\epsilon_h H_b(k_bx), \tag*{(13)} h = − H + ϵ h H b ( k b x ) , (13)
where H H H represents the mean depth of the fluid domain, H b H_b H b denotes the submarine topography shape, ϵ h \epsilon_h ϵ h is its amplitude, and k b − 1 k_b^{-1} k b − 1 represents the length scale of the bathymetry. We always assume the bathymetry to have a ‘mild slope’; for this we use an analog condition of that used for surface gravity waves [30 , 17 ] :
1 K j ∂ h ∂ x = O ( ϵ h ϵ k ) ≪ O ( 1 ) , (14) \frac{1}{\mathcal{K}_j}\frac{\partial h}{\partial x}=\mathcal{O}(\epsilon_h\epsilon_k)\ll\mathcal{O}(1), \tag*{(14)} K j 1 ∂ x ∂ h = O ( ϵ h ϵ k ) ≪ O ( 1 ) , (14)
where K j ≡ k j h \mathcal{K}_j\equiv k_jh K j ≡ k j h is the nondimensional horizontal wavenumber (k j k_j k j being the horizontal wavenumber) of the j j j –th internal wave. Moreover, the relation k j − 1 = ϵ k k b − 1 k_j^{-1}=\epsilon_k k_b^{-1} k j − 1 = ϵ k k b − 1 is used in (2.9), which implies that either of the parameters, ϵ h \epsilon_h ϵ h or ϵ k \epsilon_k ϵ k , could be a small quantity while the other could potentially be an O ( 1 ) \mathcal{O}(1) O ( 1 ) quantity. We note in passing that the mild slope condition in our case can still lead to internal gravity wave scattering 2 . Scaling analysis to find the relations between these small parameters is given in appendix B.
Leading order analysis
Next we substitute (2.7) in (2.5). At the leading order (O ( ϵ a ) \mathcal{O}(\epsilon_a) O ( ϵ a ) ), the governing equation (2.5) reduces to:
( ∂ 2 ∂ x 2 + 1 h 2 ∂ 2 ∂ η 2 ) ∂ 2 Ξ ‾ j ∂ t 2 + N 2 ( − h ( x ) η ) ∂ 2 Ξ ‾ j ∂ x 2 + f 2 h 2 ∂ 2 Ξ ‾ j ∂ η 2 = 0. (15) \left(\frac{\partial^2}{\partial x^2}+\frac{1}{h^2}\frac{\partial^2}{\partial\eta^2}\right)\frac{\partial^2\overline{\Xi}_j}{\partial t^2}+N^2(-h(x)\eta)\frac{\partial^2\overline{\Xi}_j}{\partial x^2}+\frac{f^2}{h^2}\frac{\partial^2\overline{\Xi}_j}{\partial\eta^2}=0. \tag*{(15)} ( ∂ x 2 ∂ 2 + h 2 1 ∂ η 2 ∂ 2 ) ∂ t 2 ∂ 2 Ξ j + N 2 ( − h ( x ) η ) ∂ x 2 ∂ 2 Ξ j + h 2 f 2 ∂ η 2 ∂ 2 Ξ j = 0. (15)
Hereafter we drop the argument of N 2 N^2 N 2 , assuming it is implied. Furthermore assuming Ξ ‾ j = Ξ ^ j ( x , η ) e − i ω j t \overline{\Xi}_j=\widehat{\Xi}_j(x,\eta)e^{-i\omega_jt} Ξ j = Ξ j ( x , η ) e − i ω j t , where ω j ∈ R + \omega_j\in\mathbb{R}^+ ω j ∈ R + is the angular frequency of the j j j -th internal wave, (2.10) simplifies to
[ ( N 2 − ω j 2 ) ∂ 2 ∂ x 2 − ω j 2 − f 2 h 2 ∂ 2 ∂ η 2 ] Ξ ^ j = 0. (16) \left[(N^2-\omega_j^2)\frac{\partial^2}{\partial x^2}-\frac{\omega_j^2-f^2}{h^2}\frac{\partial^2}{\partial\eta^2}\right]\widehat{\Xi}_j=0. \tag*{(16)} [ ( N 2 − ω j 2 ) ∂ x 2 ∂ 2 − h 2 ω j 2 − f 2 ∂ η 2 ∂ 2 ] Ξ j = 0. (16)
For a mild slope bathymetry (see appendix B for details), we can use variable separation to solve (2.11) at the leading order. To this end we assume Ξ ^ j = ϕ j ( η ; x ) P j ( x ) \widehat{\Xi}_j=\phi_j(\eta;x)P_j(x) Ξ j = ϕ j ( η ; x ) P j ( x ) , which leads to
h 2 P j ∂ 2 P j ∂ x 2 = ω j 2 − f 2 N 2 − ω j 2 1 ϕ j ∂ 2 ϕ j ∂ η 2 = − K j 2 , (17) \frac{h^2}{P_j}\frac{\partial^2P_j}{\partial x^2}=\frac{\omega_j^2-f^2}{N^2-\omega_j^2}\frac{1}{\phi_j}\frac{\partial^2\phi_j}{\partial\eta^2}=-\mathcal{K}_j^2, \tag*{(17)} P j h 2 ∂ x 2 ∂ 2 P j = N 2 − ω j 2 ω j 2 − f 2 ϕ j 1 ∂ η 2 ∂ 2 ϕ j = − K j 2 , (17)
where ϕ j \phi_j ϕ j parametrically depends on x x x via h h h . We emphasize that in the x x x –η \eta η coordinates, the presence of bathymetry makes N N N to also be a function of x x x ; see Figure 2 for clarity. Two separate equations, one for P j P_j P j and the other for ϕ j \phi_j ϕ j , can be formed from (2.12):
Figure 2. The effective change in the stratification profile N ( z ) N(z) N ( z ) when the coordinates are changed from (a) x x x –z z z to (b) x x x –η \eta η in the presence of bathymetry. For the latter case, if N N N is a function of z z z in x x x –z z z , it becomes a function of both η \eta η and x x x in x x x –η \eta η . N N N profiles corresponding to the top of a seamount and an abyssal plain region have been respectively denoted by blue and green colors.
[ ∂ 2 ∂ x 2 + K j 2 h 2 ] P j = 0 , (18) \left[\frac{\partial^2}{\partial x^2}+\frac{\mathcal{K}_j^2}{h^2}\right]P_j=0, \tag*{(18)} [ ∂ x 2 ∂ 2 + h 2 K j 2 ] P j = 0 , (18)
L j ϕ j ≡ [ ∂ 2 ∂ η 2 + K j 2 χ j 2 ] ϕ j = 0 , (19) \mathcal{L}_j\phi_j\equiv\left[\frac{\partial^2}{\partial\eta^2}+\mathcal{K}_j^2\chi_j^2\right]\phi_j=0, \tag*{(19)} L j ϕ j ≡ [ ∂ η 2 ∂ 2 + K j 2 χ j 2 ] ϕ j = 0 , (19)
where χ j ≡ ( N 2 − ω j 2 ) / ( ω j 2 − f 2 ) \chi_j\equiv\sqrt{(N^2-\omega_j^2)/(\omega_j^2-f^2)} χ j ≡ ( N 2 − ω j 2 ) / ( ω j 2 − f 2 ) is defined for convenience. The boundary conditions for (2.13b) are ϕ j = 0 \phi_j=0 ϕ j = 0 at η = 0 , − 1 \eta=0,-1 η = 0 , − 1 . The nondimensional horizontal wavenumber of the j j j –th wave, i.e. K j \mathcal{K}_j K j , is the set of eigenvalues obtained from (19) , which can vary in x x x when N N N is a function of x x x in x x x –η \eta η coordinates. An important point to note is the convention used in our study. While positive (negative) k j k_j k j implies waves propagating along + x +x + x (− x -x − x ), owing to the fact that h h h is negative, K j \mathcal{K}_j K j follows the exactly opposite convention. This means that a negative (positive) K j \mathcal{K}_j K j implies that the wave is traveling along + x +x + x (− x -x − x ) direction. Moreover we notice that (19) does not explicitly depend on h h h . The only way (19) can be influenced by h h h is through N N N when the latter varies in the z z z –direction (in x x x –z z z coordinates). However for a uniform stratification, i.e. N = constant N = \text{constant} N = constant , eigenvalues of (19) are independent of h h h . In this case the eigenvalues are given by
M j ≡ K j χ j = n π , (20) \mathcal{M}_j \equiv\mathcal{K}_j\chi_j = n\pi, \tag*{(20)} M j ≡ K j χ j = nπ , (20)
where n ∈ Z + n \in\mathbb{Z}^{+} n ∈ Z + . We also observe that the quantity M j \mathcal{M}_j M j behaves like the vertical wavenumber of the wave that is nondimensionalised by the local bathymetry h h h .
Meanwhile, P j P_j P j at the leading order of the WKB approximation is given by
P j = exp { i ∫ 0 x K j ( x ′ ) h ( x ′ ) d x ′ } . (21) P_j = \exp\left\{i\int_{0}^{x}\frac{\mathcal{K}_j(x')}{h(x')}\,\mathrm{d}x'\right\}. \tag*{(21)} P j = exp { i ∫ 0 x h ( x ′ ) K j ( x ′ ) d x ′ } . (21)
We introduce a function β j ( ϵ k x ) \beta_j(\epsilon kx) β j ( ϵ k x ) such that P j P_j P j is corrected to P j / β j P_j/\beta_j P j / β j . This slow varying function β j ( ϵ k x ) \beta_j(\epsilon kx) β j ( ϵ k x ) , which acts as a correction to the first order WKB solution (21) , is given in (2.22). We note in passing that P j / β j P_j/\beta_j P j / β j is still a solution of (18) in the leading order even after the above-mentioned correction. To normalize the eigenfunction of the waves obtained from (19) , every wave’s ϕ j \phi_j ϕ j is constrained to satisfy:
1 2 ∫ − 1 0 1 h 2 [ K j 2 ϕ j 2 + ( ∂ ϕ j ∂ η ) 2 ] d η = 1. (22) \frac{1}{2}\int_{-1}^{0}\frac{1}{h^2}\left[\mathcal{K}_j^2\phi_j^2+\left(\frac{\partial\phi_j}{\partial\eta}\right)^2\right]\mathrm{d}\eta=1. \tag*{(22)} 2 1 ∫ − 1 0 h 2 1 [ K j 2 ϕ j 2 + ( ∂ η ∂ ϕ j ) 2 ] d η = 1. (22)
After this normalisation, waves having the same amplitude (a j a_j a j ) will also have the same energy density at a given h h h , provided β j = 1 \beta_j = 1 β j = 1 .
The meridional velocity and the buoyancy perturbation at the leading order can be obtained by respectively converting (2.1b) and (2.1c) into the x x x –η \eta η coordinates and then substituting the streamfunction ansatz (2.7):
V j = i f h ω j a j β j ∂ ϕ j ∂ η P j e − i ω j t + c . c . , (23) \mathcal{V}_j = i\frac{f}{h\omega_j}\frac{a_j}{\beta_j}\frac{\partial\phi_j}{\partial\eta}\mathcal{P}_j e^{-i\omega_j t}+\mathrm{c.c.}, \tag*{(23)} V j = i h ω j f β j a j ∂ η ∂ ϕ j P j e − i ω j t + c.c. , (23)
B j = i N 2 ω j a j β j ∂ P j ∂ x ϕ j e − i ω j t + c . c . (24) B_j = i\frac{N^2}{\omega_j}\frac{a_j}{\beta_j}\frac{\partial\mathcal{P}_j}{\partial x}\phi_j e^{-i\omega_j t}+\mathrm{c.c.} \tag*{(24)} B j = i ω j N 2 β j a j ∂ x ∂ P j ϕ j e − i ω j t + c.c. (24)
Second order analysis
Amplitude evolution equations for a resonant triad in non-uniform stratification
Triad interaction between three internal waves occurs at O ( ϵ 2 ) O(\epsilon^2) O ( ϵ 2 ) . Below we describe the detailed derivation that finally leads to the amplitude evolution equations (2.24a)–(2.24c) of the waves constituting a triad.
After substituting the streamfunction (2.7), meridional velocity (2.17), and buoyancy perturbation (2.18) in (2.5), the equation for the j j j -th wave can be written as:
a j β j P j L j ϕ j = − F j , (25) \frac{a_j}{\beta_j}\mathcal{P}_j\mathcal{L}_j\phi_j=-\mathcal{F}_j, \tag*{(25)} β j a j P j L j ϕ j = − F j , (25)
where P j ≡ P j e − i ω j t \mathcal{P}_j\equiv P_j e^{-i\omega_jt} P j ≡ P j e − i ω j t , L j \mathcal{L}_j L j has been defined in (2.13b), and
F j ≡ i ∂ a j ∂ t ( ϕ j K j 2 − ∂ 2 ϕ j ∂ η 2 ) 2 ω j h 2 ( P j β j ) + 2 i ( N 2 − ω j 2 ) ( K j h ϕ j ∂ a j ∂ x ) ( P j β j ) + i ( N 2 − ω j 2 ) K j h [ 2 ∂ ϕ j ∂ x + ϕ j h K j ∂ ∂ x ( K j h ) − 2 η h ∂ h ∂ x ∂ ϕ j ∂ η − 2 ϕ j β j d ( β j ) d x ] ( a j P j β j ) − N L j . (26) \begin{aligned}
\mathcal{F}_j \equiv{}& i\frac{\partial a_j}{\partial t}\left(\phi_j\mathcal{K}_j^2-\frac{\partial^2\phi_j}{\partial\eta^2}\right)\frac{2\omega_j}{h^2}\left(\frac{\mathcal{P}_j}{\beta_j}\right)
+2i(N^2-\omega_j^2)\left(\frac{\mathcal{K}_j}{h}\phi_j\frac{\partial a_j}{\partial x}\right)\left(\frac{\mathcal{P}_j}{\beta_j}\right)\\
&+i(N^2-\omega_j^2)\frac{\mathcal{K}_j}{h}\left[2\frac{\partial\phi_j}{\partial x}+\frac{\phi_j h}{\mathcal{K}_j}\frac{\partial}{\partial x}\left(\frac{\mathcal{K}_j}{h}\right)-\frac{2\eta}{h}\frac{\partial h}{\partial x}\frac{\partial\phi_j}{\partial\eta}-\frac{2\phi_j}{\beta_j}\frac{d(\beta_j)}{dx}\right]\left(a_j\frac{\mathcal{P}_j}{\beta_j}\right)-\mathrm{NL}_j. \tag*{(26)}
\end{aligned} F j ≡ i ∂ t ∂ a j ( ϕ j K j 2 − ∂ η 2 ∂ 2 ϕ j ) h 2 2 ω j ( β j P j ) + 2 i ( N 2 − ω j 2 ) ( h K j ϕ j ∂ x ∂ a j ) ( β j P j ) + i ( N 2 − ω j 2 ) h K j [ 2 ∂ x ∂ ϕ j + K j ϕ j h ∂ x ∂ ( h K j ) − h 2 η ∂ x ∂ h ∂ η ∂ ϕ j − β j 2 ϕ j d x d ( β j ) ] ( a j β j P j ) − NL j . (26)
F j \mathcal{F}_j F j is the collection of all the linear and nonlinear (N L j \mathrm{NL}_j NL j ) terms at O ( ϵ 2 ) O(\epsilon^2) O ( ϵ 2 ) which have the phase of the j j j -th wave. Equation (2.19) can have a non-trivial solution when F j \mathcal{F}_j F j is orthogonal to the adjoint solutions of the linear operator L j \mathcal{L}_j L j , and this procedure is outlined in Craik [6 ] . The complete mathematical proof for using such condition is given in detail in Ince [15 ] . Following Craik [6 ] , F j \mathcal{F}_j F j is multiplied by ϕ j \phi_j ϕ j (since L j \mathcal{L}_j L j is a self-adjoint operator, ϕ j \phi_j ϕ j is also the solution of the adjoint of L j \mathcal{L}_j L j ) and then integrated in the η \eta η direction inside the boundary limits. This would result in:
2 [ i ω j ∂ a j ∂ t ( γ j ( 1 ) K j 2 − γ j ( 2 ) ) 1 h 2 + i γ j ( 3 ) ( K j h ∂ a j ∂ x ) ] P j β j + K j h [ 2 γ j ( 4 ) + h γ j ( 3 ) K j ∂ ∂ x ( K j h ) − γ j ( 5 ) 2 h ∂ h ∂ x − 2 γ j ( 3 ) β j d β j d x ] a j P j β j = ∫ − 1 0 N L j ϕ j d η , (27) \begin{aligned}
2\left[i\omega_j\frac{\partial a_j}{\partial t}\left(\gamma_j^{(1)}\mathcal{K}_j^2-\gamma_j^{(2)}\right)\frac{1}{h^2}+i\gamma_j^{(3)}\left(\frac{\mathcal{K}_j}{h}\frac{\partial a_j}{\partial x}\right)\right]\frac{\mathcal{P}_j}{\beta_j}\\
+\frac{\mathcal{K}_j}{h}\left[2\gamma_j^{(4)}+\frac{h\gamma_j^{(3)}}{\mathcal{K}_j}\frac{\partial}{\partial x}\left(\frac{\mathcal{K}_j}{h}\right)-\gamma_j^{(5)}\frac{2}{h}\frac{\partial h}{\partial x}-\frac{2\gamma_j^{(3)}}{\beta_j}\frac{d\beta_j}{dx}\right]a_j\frac{\mathcal{P}_j}{\beta_j}
=\int_{-1}^{0}\mathrm{NL}_j\phi_j\,d\eta, \tag*{(27)}
\end{aligned} 2 [ i ω j ∂ t ∂ a j ( γ j ( 1 ) K j 2 − γ j ( 2 ) ) h 2 1 + i γ j ( 3 ) ( h K j ∂ x ∂ a j ) ] β j P j + h K j [ 2 γ j ( 4 ) + K j h γ j ( 3 ) ∂ x ∂ ( h K j ) − γ j ( 5 ) h 2 ∂ x ∂ h − β j 2 γ j ( 3 ) d x d β j ] a j β j P j = ∫ − 1 0 NL j ϕ j d η , (27)
where γ j ( n ) \gamma_j^{(n)} γ j ( n ) are functions that vary in the x x x -direction, and are obtained after integration in the η \eta η direction. γ j ( n ) \gamma_j^{(n)} γ j ( n ) are provided in appendix A. Up to this point, β j \beta_j β j is an arbitrary function, and for convenience, we define β j \beta_j β j such that the second square-bracketed term in the LHS of (2.21) vanishes identically. It also implies that ‘Linear term-2’ in (2.20) also vanishes identically. In mathematical terms this means,
β j = exp { ∫ 0 x h 2 K j γ j ( 3 ) [ ( 2 γ j ( 4 ) − γ j ( 5 ) 2 h ∂ h ∂ x ) ( K j h ) + γ j ( 3 ) ∂ ∂ x ( K j h ) ] d x } . (28) \beta_j=\exp\left\{\int_{0}^{x}\frac{h}{2\mathcal{K}_j\gamma_j^{(3)}}\left[\left(2\gamma_j^{(4)}-\gamma_j^{(5)}\frac{2}{h}\frac{\partial h}{\partial x}\right)\left(\frac{\mathcal{K}_j}{h}\right)+\gamma_j^{(3)}\frac{\partial}{\partial x}\left(\frac{\mathcal{K}_j}{h}\right)\right]dx\right\}. \tag*{(28)} β j = exp { ∫ 0 x 2 K j γ j ( 3 ) h [ ( 2 γ j ( 4 ) − γ j ( 5 ) h 2 ∂ x ∂ h ) ( h K j ) + γ j ( 3 ) ∂ x ∂ ( h K j ) ] d x } . (28)
For constant N N N , β j \beta_j β j can be analytically simplified to β j = h ( x ) / h ( 0 ) = − h ( x ) / H \beta_j = h(x)/h(0) = -h(x)/H β j = h ( x ) / h ( 0 ) = − h ( x ) / H , where it is assumed that h ( 0 ) = − H h(0) = -H h ( 0 ) = − H . We note in passing that the equivalent of β j \beta_j β j functions was derived in Lahaye & Llewellyn Smith [20 ] using a different approach. For this particular choice of β j \beta_j β j , if the amplitudes a j a_j a j are x x x -invariant, the energy flux will be x x x -invariant as well, regardless of the modal shape, or depth. More importantly, a wave packet’s maximum amplitude does not change when β j \beta_j β j , given by (28) , is used in (27) . This invariance of the maximum value of the a j a_j a j with varying h h h is very useful in estimating wave growth rates in our study, in which a major focus is on wave interactions in a region of varying h h h .
Next we outline the procedure to obtain ∫ − 1 0 N L j ϕ j d η \int_{-1}^{0}\mathrm{NL}_j\phi_j\mathrm{d}\eta ∫ − 1 0 NL j ϕ j d η in (27) to complete the amplitude evolution equations. The streamfunction, meridional velocity and buoyancy frequency ansatz are substituted in the nonlinear terms of (8) . As a result, the resultant resonant nonlinear terms, after omitting non-resonant terms, can be written in a compact form as given below:
∫ − 1 0 N L 1 ϕ 1 d η = [ ∫ − 1 0 ( N L ^ ( Ψ , 1 ) + N L ^ ( B , 1 ) + N L ^ ( V , 1 ) ) ϕ 1 d η ] a 3 a ‾ 2 β 2 β 3 ϕ ^ 3 ϕ ^ 2 (29) \int_{-1}^{0}\mathrm{NL}_1\phi_1\mathrm{d}\eta=
\left[\int_{-1}^{0}\left(\widehat{\mathrm{NL}}_{(\Psi,1)}+\widehat{\mathrm{NL}}_{(B,1)}+\widehat{\mathrm{NL}}_{(V,1)}\right)\phi_1\mathrm{d}\eta\right]\frac{a_3\overline{a}_2}{\beta_2\beta_3}\widehat{\phi}_3\widehat{\phi}_2 \tag*{(29)} ∫ − 1 0 NL 1 ϕ 1 d η = [ ∫ − 1 0 ( NL ( Ψ , 1 ) + NL ( B , 1 ) + NL ( V , 1 ) ) ϕ 1 d η ] β 2 β 3 a 3 a 2 ϕ 3 ϕ 2 (29)
∫ − 1 0 N L 2 ϕ 2 d η = [ ∫ − 1 0 ( N L ^ ( Ψ , 2 ) + N L ^ ( B , 2 ) + N L ^ ( V , 2 ) ) ϕ 2 d η ] a 3 a ‾ 1 β 1 β 3 ϕ ^ 3 ϕ ^ 1 (30) \int_{-1}^{0}\mathrm{NL}_2\phi_2\mathrm{d}\eta=
\left[\int_{-1}^{0}\left(\widehat{\mathrm{NL}}_{(\Psi,2)}+\widehat{\mathrm{NL}}_{(B,2)}+\widehat{\mathrm{NL}}_{(V,2)}\right)\phi_2\mathrm{d}\eta\right]\frac{a_3\overline{a}_1}{\beta_1\beta_3}\widehat{\phi}_3\widehat{\phi}_1 \tag*{(30)} ∫ − 1 0 NL 2 ϕ 2 d η = [ ∫ − 1 0 ( NL ( Ψ , 2 ) + NL ( B , 2 ) + NL ( V , 2 ) ) ϕ 2 d η ] β 1 β 3 a 3 a 1 ϕ 3 ϕ 1 (30)
∫ − 1 0 N L 3 ϕ 3 d η = [ ∫ − 1 0 ( N L ^ ( Ψ , 3 ) + N L ^ ( B , 3 ) + N L ^ ( V , 3 ) ) ϕ 3 d η ] a 1 a 2 β 1 β 2 ϕ ^ 1 ϕ ^ 2 (31) \int_{-1}^{0}\mathrm{NL}_3\phi_3\mathrm{d}\eta=
\left[\int_{-1}^{0}\left(\widehat{\mathrm{NL}}_{(\Psi,3)}+\widehat{\mathrm{NL}}_{(B,3)}+\widehat{\mathrm{NL}}_{(V,3)}\right)\phi_3\mathrm{d}\eta\right]\frac{a_1a_2}{\beta_1\beta_2}\widehat{\phi}_1\widehat{\phi}_2 \tag*{(31)} ∫ − 1 0 NL 3 ϕ 3 d η = [ ∫ − 1 0 ( NL ( Ψ , 3 ) + NL ( B , 3 ) + NL ( V , 3 ) ) ϕ 3 d η ] β 1 β 2 a 1 a 2 ϕ 1 ϕ 2 (31)
We define N L ( ∗ , j ) ≡ ∫ − 1 0 N L ^ ( ∗ , j ) ϕ j d η \mathrm{NL}_{(\ast,j)} \equiv\int_{-1}^{0}\widehat{\mathrm{NL}}_{(\ast,j)}\phi_j\mathrm{d}\eta NL ( ∗ , j ) ≡ ∫ − 1 0 NL ( ∗ , j ) ϕ j d η for convenience, and their expressions are provided in appendix A. Note that N L ( ∗ , j ) \mathrm{NL}_{(\ast,j)} NL ( ∗ , j ) is directly used in amplitude evolution equations given just below in equation (36) .
The amplitude evolution equations for the three internal gravity waves are finally obtained after equating the LHS of (27) with its RHS; where the latter has been expressed in terms of (A 2)–(A 4):
∂ a 1 ∂ t + c ( x , 1 ) ( g ) ∂ a 1 ∂ x = N 1 a 3 a ‾ 2 exp { ∫ 0 x i ( K 3 − K 1 − K 2 ) / h d x ′ + i Δ ω t } (32) \frac{\partial a_1}{\partial t}+c_{(x,1)}^{(g)}\frac{\partial a_1}{\partial x}
=\mathfrak{N}_1a_3\overline{a}_2\exp\left\{\int_{0}^{x}\mathrm{i}(K_3-K_1-K_2)/h\ \mathrm{d}x' + \mathrm{i}\Delta\omega t\right\} \tag*{(32)} ∂ t ∂ a 1 + c ( x , 1 ) ( g ) ∂ x ∂ a 1 = N 1 a 3 a 2 exp { ∫ 0 x i ( K 3 − K 1 − K 2 ) / h d x ′ + i Δ ω t } (32)
∂ a 2 ∂ t + c ( x , 2 ) ( g ) ∂ a 2 ∂ x = N 2 a 3 a ‾ 1 exp { ∫ 0 x i ( K 3 − K 1 − K 2 ) / h d x ′ + i Δ ω t } (33) \frac{\partial a_2}{\partial t}+c_{(x,2)}^{(g)}\frac{\partial a_2}{\partial x}
=\mathfrak{N}_2a_3\overline{a}_1\exp\left\{\int_{0}^{x}\mathrm{i}(K_3-K_1-K_2)/h\ \mathrm{d}x' + \mathrm{i}\Delta\omega t\right\} \tag*{(33)} ∂ t ∂ a 2 + c ( x , 2 ) ( g ) ∂ x ∂ a 2 = N 2 a 3 a 1 exp { ∫ 0 x i ( K 3 − K 1 − K 2 ) / h d x ′ + i Δ ω t } (33)
∂ a 3 ∂ t + c ( x , 3 ) ( g ) ∂ a 3 ∂ x = N 3 a 1 a 2 exp { ∫ 0 x i ( K 1 + K 2 − K 3 ) / h d x ′ − i Δ ω t } (34) \frac{\partial a_3}{\partial t}+c_{(x,3)}^{(g)}\frac{\partial a_3}{\partial x}
=\mathfrak{N}_3a_1a_2\exp\left\{\int_{0}^{x}\mathrm{i}(K_1+K_2-K_3)/h\ \mathrm{d}x' - \mathrm{i}\Delta\omega t\right\} \tag*{(34)} ∂ t ∂ a 3 + c ( x , 3 ) ( g ) ∂ x ∂ a 3 = N 3 a 1 a 2 exp { ∫ 0 x i ( K 1 + K 2 − K 3 ) / h d x ′ − i Δ ω t } (34)
where
c ( x , j ) ( g ) = [ 2 i K j γ j ( 3 ) h D j ] , in which D j = 2 i ω j ( γ j ( 1 ) K j 2 − γ j ( 2 ) ) / h 2 (35) c_{(x,j)}^{(g)}=\left[\frac{2\mathrm{i}K_j\gamma_j^{(3)}}{h\mathcal{D}_j}\right],\qquad\text{in which}\qquad\mathcal{D}_j=2\mathrm{i}\omega_j\left(\gamma_j^{(1)}K_j^2-\gamma_j^{(2)}\right)/h^2 \tag*{(35)} c ( x , j ) ( g ) = [ h D j 2 i K j γ j ( 3 ) ] , in which D j = 2 i ω j ( γ j ( 1 ) K j 2 − γ j ( 2 ) ) / h 2 (35)
N j = 1 D j [ N L ( V , j ) + N L ( B , j ) + N L ( Ψ , j ) ] . (36) \mathfrak{N}_j=\frac{1}{\mathcal{D}_j}\left[\mathrm{NL}_{(V,j)}+\mathrm{NL}_{(B,j)}+\mathrm{NL}_{(\Psi,j)}\right]. \tag*{(36)} N j = D j 1 [ NL ( V , j ) + NL ( B , j ) + NL ( Ψ , j ) ] . (36)
In the above equation
D 1 = D 1 β 2 β 3 β 1 , D 2 = D 2 β 1 β 3 β 2 , D 3 = D 3 β 1 β 2 β 3 . (37) \mathcal{D}_1=\mathcal{D}_1\frac{\beta_2\beta_3}{\beta_1},\qquad
\mathcal{D}_2=\mathcal{D}_2\frac{\beta_1\beta_3}{\beta_2},\qquad
\mathcal{D}_3=\mathcal{D}_3\frac{\beta_1\beta_2}{\beta_3}. \tag*{(37)} D 1 = D 1 β 1 β 2 β 3 , D 2 = D 2 β 2 β 1 β 3 , D 3 = D 3 β 3 β 1 β 2 . (37)
The coefficient c ( x , j ) ( g ) c_{(x,j)}^{(g)} c ( x , j ) ( g ) denotes the (weakly varying) horizontal group speed and N j \mathfrak{N}_j N j denotes the nonlinear coupling coefficient of the j j j -th wave; N j \mathfrak{N}_j N j determines the rate of energy transfer between the waves. Δ ω ≡ ω 1 + ω 2 − ω 3 \Delta\omega\equiv\omega_1+\omega_2-\omega_3 Δ ω ≡ ω 1 + ω 2 − ω 3 denotes the detuning in the frequency. The argument of the exponential terms in (2.24a)–(2.24c) denote both the detuning in the horizontal wavenumber condition and the frequency condition. For a pure resonant triad, K 3 − K 1 − K 2 = 0 K_3-K_1-K_2=0 K 3 − K 1 − K 2 = 0 and Δ ω = 0 \Delta\omega=0 Δ ω = 0 . When K 3 − K 1 − K 2 ≠ 0 K_3-K_1-K_2\ne0 K 3 − K 1 − K 2 = 0 or Δ ω ≠ 0 \Delta\omega\ne0 Δ ω = 0 , the triad is said to be detuned. The equations are only valid when both Δ ω / ω j ≪ 1 \Delta\omega/\omega_j\ll1 Δ ω / ω j ≪ 1 and Δ K / K j ≪ 1 \Delta K/K_j\ll1 Δ K / K j ≪ 1 are satisfied, that is, the equations are valid only near the vicinity of resonance. Analytical methods have also been developed for studying wave-wave interactions in non-resonant regimes in the presence of a slowly varying background shear flow (for example: Voelker et al. [38 ] ; Grimshaw [9 , 10 ] ), where the wave train can pass through non-resonant regimes and resonant regimes. However, this is not in the scope of this paper. To summarize, amplitude (a j a_j a j ) in the wave amplitude equations can vary because of the group speed term, or the nonlinear term. The group speed term is responsible the advection of a wave packet, while the nonlinear term is responsible for energy transfer among the waves. Moreover, waves’ energy density changes because of its motion through a region of varying h h h . ϕ j \phi_j ϕ j and β j \beta_j β j are heavily involved in the change in energy density that occur in a wave due to its motion through a region of varying h h h . Note that the evolution of a j a_j a j does not provide complete information of the changes in a wave’s quantities. This is because Ψ j = a j ϕ j / β j P j e − i ω j t \Psi_j=a_j\phi_j/\beta_jP_j e^{-i\omega_j t} Ψ j = a j ϕ j / β j P j e − i ω j t , where ϕ j \phi_j ϕ j and β j \beta_j β j themselves are functions of x x x .
For a triad, the parent wave is always the wave-3, while the daughter waves (subharmonic waves) are wave-1 and wave-2. For self-interactions, we use a different convention; see § 2.2.2 . To determine how fast the daughter waves grow, a growth rate parameter (σ \sigma σ ) is defined as follows:
σ ≡ N 1 N 2 A 3 2 , (38) \sigma\equiv\sqrt{\mathcal{N}_1\mathcal{N}_2 A_3^2}, \tag*{(38)} σ ≡ N 1 N 2 A 3 2 , (38)
where A 3 A_3 A 3 is the parent wave’s amplitude, which is held constant. To obtain this expression, the pump wave approximation of Craik et al. [7 ] is used. Pump wave approximation is a strong assumption which is only valid at initial times where the parent wave has much more energy than the daughter waves. Equation (2.27) reveals that growth rate is directly dependent on the nonlinear coupling coefficients. If we ignore the nonlinear terms, equations (2.24a)–(2.24c) model the movements of internal wavepackets over a mild-slope bathymetry. We emphasize here that wave scattering is not included in these equations. The amplitude variation of internal waves was recently analyzed by Lahaye & Llewellyn Smith [20 ] (the authors focused on internal wave scattering, which is essentially a linear mechanism). While we have restricted our study to mild-slope conditions, we have extended the previous works by including the physics of (i) finite width wave packets, (ii) nonlinearity, and (iii) detuning in the horizontal wavenumber condition and hence investigation of both resonant (zero detuning) and near-resonant conditions. In this paper, we mainly focus on the variation of detuning, and growth rates (using pump wave approximation) with h / H h/H h / H for wave-wave interactions. Even though equations (2.24a)–(2.24c) allows finite width wave packets, we do not discuss it significantly since these have been studied in Gururaj & Guha [12 ] . The combined effect of nonlinear coupling coefficients, group speed and detuning have been discussed in Gururaj & Guha [12 ] .
The main results of scaling analysis, detailed in appendix B, is summarized here. The relation between the small parameters are given by:
ϵ t ∼ N ω ϵ a − c ^ g ϵ x (39) \epsilon_t \sim\frac{\mathcal{N}}{\omega}\epsilon_a-\widehat{c}_g\epsilon_x \tag*{(39)} ϵ t ∼ ω N ϵ a − c g ϵ x (39)
where c ^ g \widehat{c}_g c g is a non-dimensional term that gives a scale of the group speed. Equation (2.28) provides the scaling for ‘Linear term-1’ and N L j \mathrm{NL}_j NL j given in (2.20). These are also the final terms which are present in wave amplitude equations (2.24a)–(2.24c). The wave amplitude (a j a_j a j ) can evolve due to the group speed term or the nonlinear term. Note that if N \mathcal{N} N or ϵ a \epsilon_a ϵ a is reduced (implying that nonlinear coupling coefficients or amplitude is reduced), then we can expect nonlinear effects to decrease. However, if ϵ x \epsilon_x ϵ x is reduced (which means packet width is increased), then the effect of group speed, which advects the packets, is reduced.
Amplitude evolution equations for self-interaction in non-uniform stratification
Self-interactions can be considered as a special case of triad interactions. During resonant self-interactions, an internal wave spontaneously gives its energy to another internal wave which has twice its frequency and horizontal wavenumber [39 ] . In non-uniform stratification, a resonant self-interaction occurs when both ( ω , k ) (\omega, k) ( ω , k ) and ( 2 ω , 2 k ) (2\omega, 2k) ( 2 ω , 2 k ) satisfy the dispersion relation. The evolution equations for the self-interaction of a mode can be obtained from the set of equations (2.24a)–(2.24c) after some straightforward modifications. The complete set of governing equations for the self-interaction of a mode in the presence of a mild-slope bathymetry h h h is given below:
∂ a ( 3 , s ) ∂ t + c ( x , 3 ) ( g ) ∂ a ( 3 , s ) ∂ x = N 3 a ( 1 , s ) 2 exp { ∫ 0 x i ( 2 K 1 − K 3 ) / h d x ′ + i Δ ω s t } (40) \frac{\partial a_{(3,s)}}{\partial t}+c^{(g)}_{(x,3)}\frac{\partial a_{(3,s)}}{\partial x}=\mathcal{N}_3a_{(1,s)}^2\exp\left\{\int_0^x \mathrm{i}(2\mathcal{K}_1-\mathcal{K}_3)/h\,\mathrm{d}x'+\mathrm{i}\Delta\omega_s t\right\} \tag*{(40)} ∂ t ∂ a ( 3 , s ) + c ( x , 3 ) ( g ) ∂ x ∂ a ( 3 , s ) = N 3 a ( 1 , s ) 2 exp { ∫ 0 x i ( 2 K 1 − K 3 ) / h d x ′ + i Δ ω s t } (40)
∂ a ( 1 , s ) ∂ t + c ( x , 1 ) ( g ) ∂ a ( 1 , s ) ∂ x = N 1 a ( 3 , s ) a ˉ ( 1 , s ) exp { ∫ 0 x i ( K 3 − 2 K 1 ) / h d x ′ − i Δ ω s t } (41) \frac{\partial a_{(1,s)}}{\partial t}+c^{(g)}_{(x,1)}\frac{\partial a_{(1,s)}}{\partial x}=\mathcal{N}_1a_{(3,s)}\bar{a}_{(1,s)}\exp\left\{\int_0^x \mathrm{i}(\mathcal{K}_3-2\mathcal{K}_1)/h\,\mathrm{d}x'-\mathrm{i}\Delta\omega_s t\right\} \tag*{(41)} ∂ t ∂ a ( 1 , s ) + c ( x , 1 ) ( g ) ∂ x ∂ a ( 1 , s ) = N 1 a ( 3 , s ) a ˉ ( 1 , s ) exp { ∫ 0 x i ( K 3 − 2 K 1 ) / h d x ′ − i Δ ω s t } (41)
where the subscript ‘s’ denotes self-interaction. Moreover, Δ ω s = 2 ω 1 − ω 3 \Delta\omega_s=2\omega_1-\omega_3 Δ ω s = 2 ω 1 − ω 3 . Unlike the triad case, the parent wave for self-interaction is wave-1 while the daughter (superharmonic) wave is wave-3. The notation throughout this paper follows the convention that wave-3 always has the highest frequency (hence for triads, wave-3 becomes the parent wave). The functions c ( x , j ) ( g ) c^{(g)}_{(x,j)} c ( x , j ) ( g ) are the same as the expressions given in (35) . The functions N j \mathcal{N}_j N j , which are the nonlinear coupling coefficients for the self-interaction process, are given by:
N 1 = N 2 (42) \mathcal{N}_1=\mathcal{N}_2 \tag*{(42)} N 1 = N 2 (42)
N 3 = 2 K 1 3 h 4 D 3 ( Γ 1 ( 4 ) ω 1 ) − 2 f 2 h 4 D 3 ( Γ 1 ( 3 ) K 1 ω 1 ) + K 1 ω 3 h 4 D 3 ( ζ 1 ω 1 2 Γ 1 ( 1 ) − ζ 1 Γ 1 ( 2 ) − Γ 1 ( 3 ) ) (43) \mathcal{N}_3=\frac{2\mathcal{K}_1^3}{h^4D_3}\left(\frac{\Gamma_1^{(4)}}{\omega_1}\right)-\frac{2f^2}{h^4D_3}\left(\frac{\Gamma_1^{(3)}\mathcal{K}_1}{\omega_1}\right)+\frac{\mathcal{K}_1\omega_3}{h^4D_3}\left(\zeta_1\omega_1^2\Gamma_1^{(1)}-\zeta_1\Gamma_1^{(2)}-\Gamma_1^{(3)}\right) \tag*{(43)} N 3 = h 4 D 3 2 K 1 3 ( ω 1 Γ 1 ( 4 ) ) − h 4 D 3 2 f 2 ( ω 1 Γ 1 ( 3 ) K 1 ) + h 4 D 3 K 1 ω 3 ( ζ 1 ω 1 2 Γ 1 ( 1 ) − ζ 1 Γ 1 ( 2 ) − Γ 1 ( 3 ) ) (43)
where ζ j ≡ K j 2 / ( ω j 2 − f 2 ) \zeta_j\equiv\mathcal{K}_j^2/(\omega_j^2-f^2) ζ j ≡ K j 2 / ( ω j 2 − f 2 ) is defined for convenience. Here all Γ , D j \Gamma,D_j Γ , D j terms in equations (42) and (43) are evaluated using (A 5) and (37) by simply considering all ‘2’ subscripts as ‘1’. For example, substituting β 1 \beta_1 β 1 for β 2 \beta_2 β 2 in D j D_j D j and similarly substituting ϕ 1 \phi_1 ϕ 1 for ϕ 2 \phi_2 ϕ 2 in Γ \Gamma Γ expressions. This is because in self-interaction, wave-2 is the same as wave-1.
The equations (2.29a)–(2.29b) can predict the growth of the daughter wave and the consequent decay of the parent wave. For obtaining the growth rate of the daughter waves, we use the pump wave approximation and hence treat the parent wave’s amplitude a ( 1 , s ) a_{(1,s)} a ( 1 , s ) as constant. This yields (assuming plane waves in the x x x -direction):
a ( 3 , s ) = [ N 3 a ( 1 , s ) 2 ] t (44) a_{(3,s)}=\left[\mathcal{N}_3a_{(1,s)}^2\right]t \tag*{(44)} a ( 3 , s ) = [ N 3 a ( 1 , s ) 2 ] t (44)
where the term in square brackets denote the growth rate. From the above equation it is evident that N 3 \mathcal{N}_3 N 3 acts as a proxy to the growth rate.
Energy evaluation
The time average energy density for an internal gravity wave over its time period is given by:
⟨ T E j ⟩ = ω j 2 π ∫ 0 2 π / ω j ρ 0 2 [ ( ∂ ψ j ∂ z ) 2 + ( ∂ ψ j ∂ x ) 2 + v j 2 + b j 2 N 2 ] d t . (45) \langle\mathrm{TE}_j \rangle= \frac{\omega_j}{2\pi}\int_{0}^{2\pi/\omega_j} \frac{\rho_0}{2}\left[\left(\frac{\partial\psi_j}{\partial z}\right)^2+\left(\frac{\partial\psi_j}{\partial x}\right)^2+v_j^2+\frac{b_j^2}{N^2}\right]\mathrm{d}t. \tag*{(45)} ⟨ TE j ⟩ = 2 π ω j ∫ 0 2 π / ω j 2 ρ 0 [ ( ∂ z ∂ ψ j ) 2 + ( ∂ x ∂ ψ j ) 2 + v j 2 + N 2 b j 2 ] d t . (45)
The domain integrated total energy is given by:
T E ^ j = ∫ 0 D ∫ h 0 ⟨ T E j ⟩ d z d x = ∫ 0 D ∫ − 1 0 ⟨ T E j ⟩ ( − h ( x ) ) d η d x . (46) \widehat{\mathrm{TE}}_j=\int_{0}^{D}\int_{h}^{0}\langle\mathrm{TE}_j\rangle\mathrm{d}z\mathrm{d}x=\int_{0}^{D}\int_{-1}^{0}\langle\mathrm{TE}_j\rangle(-h(x))\mathrm{d}\eta\mathrm{d}x. \tag*{(46)} TE j = ∫ 0 D ∫ h 0 ⟨ TE j ⟩ d z d x = ∫ 0 D ∫ − 1 0 ⟨ TE j ⟩ ( − h ( x )) d η d x . (46)
After some simplification, we arrive at:
T E ^ j = ∫ 0 D ∫ − 1 0 − 2 h [ K j 2 ϕ j 2 + ( ∂ ϕ j ∂ η ) 2 ] ∣ a j ∣ 2 ρ 0 β j 2 d η d x , (47) \widehat{\mathrm{TE}}_j=\int_{0}^{D}\int_{-1}^{0}\frac{-2}{h}\left[\mathcal{K}_j^2\phi_j^2+\left(\frac{\partial\phi_j}{\partial\eta}\right)^2\right]\lvert a_j\rvert^2\frac{\rho_0}{\beta_j^2}\mathrm{d}\eta\mathrm{d}x, \tag*{(47)} TE j = ∫ 0 D ∫ − 1 0 h − 2 [ K j 2 ϕ j 2 + ( ∂ η ∂ ϕ j ) 2 ] ∣ a j ∣ 2 β j 2 ρ 0 d η d x , (47)
where D D D is the length of the domain in the x x x -direction. We non-dimensionalize T E ^ j \widehat{\mathrm{TE}}_j TE j with the initial energy of parent wave (abbreviated as ‘P w Pw P w ’) : E j = T E ^ j / T E ^ P w ∣ t = 0 E_j=\widehat{\mathrm{TE}}_j/\widehat{\mathrm{TE}}_{Pw}|_{t=0} E j = TE j / TE P w ∣ t = 0 . Note that P w = 3 Pw=3 P w = 3 (i.e. wave-3) for triads and P w = 1 Pw=1 P w = 1 (i.e. wave-1) for self-interactions.
Triad interactions in a uniform stratification in the presence of a mild-slope bathymetry
In this section, we consider resonant and near-resonant triads in a uniform background stratification in the presence of a mild-slope bathymetry. Here, we will briefly consider the horizontal wavenumber triad condition in uniform stratification as h h h is varied. Without any loss of generality, the triad condition for the horizontal wavenumber is:
K 3 = K 1 + K 2 , (48) \mathcal{K}_3=\mathcal{K}_1+\mathcal{K}_2, \tag*{(48)} K 3 = K 1 + K 2 , (48)
However, using (2.13b), it can be seen that K j \mathcal{K}_j K j are constants for uniform stratification. As a result, triad conditions are satisfied everywhere in the domain, provided the conditions are perfectly satisfied for any given domain height.
Effect of bathymetry on the nonlinear coupling coefficients of resonant triads
Here we focus on the nonlinear coupling coefficients in resonant triad interactions (i.e. no detuning) in the presence of a uniform stratification and a weakly varying bathymetry. (38) revealed that the growth rate of the daughter waves is dependent on the nonlinear coupling coefficients. For constant N N N , the nonlinear coupling coefficients (N j \mathcal{N}_j N j ) in (2.25b) can be further simplified:
N 1 = i H 2 h 2 ω 1 κ 1 κ 2 κ 3 [ N 2 ( κ 3 − κ 2 ) { ( κ 3 ω 3 − κ 2 ω 2 ) ( κ 2 M 3 − κ 3 M 2 ) } + ω 1 { ( κ 2 M 3 − κ 3 M 2 ) ( M 2 2 + κ 2 2 − κ 3 2 − M 3 2 ) } + f 2 ( M 3 − M 2 ) { ( κ 3 ω 3 + κ 2 ω 2 ) ( M 3 M 2 ) − ( κ 2 ω 2 + κ 3 ω 3 ) ( M 2 2 + M 3 2 ) } ] (49) \mathcal{N}_1=\frac{iH}{2h^2\omega_1\kappa_1\kappa_2\kappa_3}\left[
N^2(\kappa_3-\kappa_2)\left\{\left(\frac{\kappa_3}{\omega_3}-\frac{\kappa_2}{\omega_2}\right)(\kappa_2\mathcal{M}_3-\kappa_3\mathcal{M}_2)\right\}
+\omega_1\left\{(\kappa_2\mathcal{M}_3-\kappa_3\mathcal{M}_2)(\mathcal{M}_2^2+\kappa_2^2-\kappa_3^2-\mathcal{M}_3^2)\right\}
+f^2(\mathcal{M}_3-\mathcal{M}_2)\left\{\left(\frac{\kappa_3}{\omega_3}+\frac{\kappa_2}{\omega_2}\right)(\mathcal{M}_3\mathcal{M}_2)-\left(\frac{\kappa_2}{\omega_2}+\frac{\kappa_3}{\omega_3}\right)(\mathcal{M}_2^2+\mathcal{M}_3^2)\right\}
\right] \tag*{(49)} N 1 = 2 h 2 ω 1 κ 1 κ 2 κ 3 i H [ N 2 ( κ 3 − κ 2 ) { ( ω 3 κ 3 − ω 2 κ 2 ) ( κ 2 M 3 − κ 3 M 2 ) } + ω 1 { ( κ 2 M 3 − κ 3 M 2 ) ( M 2 2 + κ 2 2 − κ 3 2 − M 3 2 ) } + f 2 ( M 3 − M 2 ) { ( ω 3 κ 3 + ω 2 κ 2 ) ( M 3 M 2 ) − ( ω 2 κ 2 + ω 3 κ 3 ) ( M 2 2 + M 3 2 ) } ] (49)
N 2 = i H 2 h 2 ω 1 κ 1 κ 2 κ 3 [ N 2 ( K 3 − K 1 ) { ( K 3 ω 3 − K 1 ω 1 ) ( K 1 M 3 − K 3 M 1 ) } + ω 2 { ( K 1 M 3 − K 3 M 1 ) ( M 1 2 + K 1 2 − K 3 2 − M 3 2 ) } + f 2 ( M 3 − M 1 ) { ( K 1 ω 1 + K 3 ω 3 ) ( M 1 M 3 ) − ( K 1 ω 3 + K 3 ω 1 ) ( M 1 2 + M 3 2 ) } ] (50) \mathfrak{N}_2 = \frac{iH}{2h^2\omega_1\kappa_1\kappa_2\kappa_3}\left[N^2(\mathcal{K}_3-\mathcal{K}_1)\left\{\left(\frac{\mathcal{K}_3}{\omega_3}-\frac{\mathcal{K}_1}{\omega_1}\right)(\mathcal{K}_1\mathcal{M}_3-\mathcal{K}_3\mathcal{M}_1)\right\}+\omega_2\left\{(\mathcal{K}_1\mathcal{M}_3-\mathcal{K}_3\mathcal{M}_1)(\mathcal{M}_1^2+\mathcal{K}_1^2-\mathcal{K}_3^2-\mathcal{M}_3^2)\right\}+f^2(\mathcal{M}_3-\mathcal{M}_1)\left\{\left(\frac{\mathcal{K}_1}{\omega_1}+\frac{\mathcal{K}_3}{\omega_3}\right)(\mathcal{M}_1\mathcal{M}_3)-\left(\frac{\mathcal{K}_1}{\omega_3}+\frac{\mathcal{K}_3}{\omega_1}\right)(\mathcal{M}_1^2+\mathcal{M}_3^2)\right\}\right] \tag*{(50)} N 2 = 2 h 2 ω 1 κ 1 κ 2 κ 3 i H [ N 2 ( K 3 − K 1 ) { ( ω 3 K 3 − ω 1 K 1 ) ( K 1 M 3 − K 3 M 1 ) } + ω 2 { ( K 1 M 3 − K 3 M 1 ) ( M 1 2 + K 1 2 − K 3 2 − M 3 2 ) } + f 2 ( M 3 − M 1 ) { ( ω 1 K 1 + ω 3 K 3 ) ( M 1 M 3 ) − ( ω 3 K 1 + ω 1 K 3 ) ( M 1 2 + M 3 2 ) } ] (50)
N 3 = i H 2 h 2 ω 3 κ 1 κ 2 κ 3 [ N 2 ( K 1 + K 2 ) { ( K 1 ω 1 − K 2 ω 2 ) ( K 2 M 1 − K 1 M 2 ) } + ω 3 { ( K 2 M 1 − K 1 M 2 ) ( M 1 2 + K 1 2 − K 2 2 − M 2 2 ) } + f 2 ( M 1 + M 2 ) { ( K 1 ω 2 + K 2 ω 1 ) ( M 1 2 + M 2 2 ) − ( K 1 ω 1 + K 2 ω 2 ) ( M 1 M 2 ) } ] (51) \mathfrak{N}_3 = \frac{iH}{2h^2\omega_3\kappa_1\kappa_2\kappa_3}\left[N^2(\mathcal{K}_1+\mathcal{K}_2)\left\{\left(\frac{\mathcal{K}_1}{\omega_1}-\frac{\mathcal{K}_2}{\omega_2}\right)(\mathcal{K}_2\mathcal{M}_1-\mathcal{K}_1\mathcal{M}_2)\right\}+\omega_3\left\{(\mathcal{K}_2\mathcal{M}_1-\mathcal{K}_1\mathcal{M}_2)(\mathcal{M}_1^2+\mathcal{K}_1^2-\mathcal{K}_2^2-\mathcal{M}_2^2)\right\}+f^2(\mathcal{M}_1+\mathcal{M}_2)\left\{\left(\frac{\mathcal{K}_1}{\omega_2}+\frac{\mathcal{K}_2}{\omega_1}\right)(\mathcal{M}_1^2+\mathcal{M}_2^2)-\left(\frac{\mathcal{K}_1}{\omega_1}+\frac{\mathcal{K}_2}{\omega_2}\right)(\mathcal{M}_1\mathcal{M}_2)\right\}\right] \tag*{(51)} N 3 = 2 h 2 ω 3 κ 1 κ 2 κ 3 i H [ N 2 ( K 1 + K 2 ) { ( ω 1 K 1 − ω 2 K 2 ) ( K 2 M 1 − K 1 M 2 ) } + ω 3 { ( K 2 M 1 − K 1 M 2 ) ( M 1 2 + K 1 2 − K 2 2 − M 2 2 ) } + f 2 ( M 1 + M 2 ) { ( ω 2 K 1 + ω 1 K 2 ) ( M 1 2 + M 2 2 ) − ( ω 1 K 1 + ω 2 K 2 ) ( M 1 M 2 ) } ] (51)
where κ j = M j 2 + K j 2 \kappa_j=\sqrt{\mathcal{M}_j^2+\mathcal{K}_j^2} κ j = M j 2 + K j 2 . Note that the above expressions are obtained only when the vertical wavenumber condition is satisfied. The terms inside the square brackets are constant and hence do not vary with the bathymetry h h h (the fact that K j \mathcal{K}_j K j and M j \mathcal{M}_j M j are constants for a constant N N N is given in (20) ). For constant N N N , β j = − h ( x ) / H \beta_j=-h(x)/H β j = − h ( x ) / H , which has been used in (49) –(51) , and this finally results in N j ∝ 1 / h 2 \mathfrak{N}_j \propto1/h^2 N j ∝ 1/ h 2 . Hence for waves traveling from a given fluid depth to a lesser depth (i.e. as the waves climb up a seamount), the nonlinear coupling coefficients, and hence the growth rates, increase following the inverse square rule.
In summary, for a uniform stratification, if three modes satisfy the resonant triad condition at a particular domain height, then they would satisfy the resonant triad condition for any domain height. Moreover, we also showed that the nonlinear coupling coefficients increase (decrease) as the fluid depth decreases (increases) following an inverse square law.
Triad and self interactions in a non-uniform stratification in the presence of a mild-slope bathymetry: detuning effects
In § 3 , it was shown that in the presence of a uniform stratification, if the triad condition is satisfied between three modes at a particular h h h , then they are satisfied for all h h h . However in non-uniform stratification, such a simple outcome is not possible. In certain types of triads, there can be a heavy mismatch in the horizontal wavenumber condition as the waves involved in the triad interact in a region of varying domain height. This may affect the energy transfer between the waves.
In this section we study the factors that decide the detuning (or mismatch) between the horizontal wavenumber of the waves as h h h is varied in the presence of non-uniform stratification. Here as well as in the rest of this paper, we will consider a Gaussian function to represent the buoyancy frequency:
N ( z ) = N b + N max exp [ − { z − z c W p } 2 ] , (52) N(z)=N_b+N_{\max}\exp\left[-\left\{\frac{z-z_c}{W_p}\right\}^2\right], \tag*{(52)} N ( z ) = N b + N m a x exp [ − { W p z − z c } 2 ] , (52)
where the parameters N b N_b N b , N max N_{\max} N m a x , W p W_p W p , and z c z_c z c are varied. This kind of profile (see figure 1(b)) is a simplified representation of oceanic stratification and is widely used in the literature; see Grisouard et al. [11 ] , Mathur et al. [28 ] , and Varma & Mathur [36 ] . We choose stratification profiles such that the pycnocline is above the topography. If the topography cuts the pycnocline, internal wave scattering may be significant as shown in Hall et al. [13 ] .
Effect of varying h h h on the horizontal wavenumber condition for waves satisfying f ≪ ω j ≪ N b f \ll\omega_j \ll N_b f ≪ ω j ≪ N b
First we study the class of triads for which the angular frequencies of the constituent waves obey the condition f ≪ ω j ≪ N b f \ll\omega_j \ll N_b f ≪ ω j ≪ N b . It is assumed that the parent wave (angular frequency ω 3 \omega_3 ω 3 ) gives its energy to two subharmonic daughter waves of angular frequencies ω 1 \omega_1 ω 1 and ω 2 \omega_2 ω 2 respectively, that is, the condition ω 1 < ω 3 \omega_1 < \omega_3 ω 1 < ω 3 and ω 2 < ω 3 \omega_2 < \omega_3 ω 2 < ω 3 is always assumed. A parameter α ∈ ( 0 , 1 ) \alpha\in(0,1) α ∈ ( 0 , 1 ) is defined such that ω 1 = α ω 3 \omega_1 = \alpha\omega_3 ω 1 = α ω 3 and ω 2 = ( 1 − α ) ω 3 \omega_2 = (1-\alpha)\omega_3 ω 2 = ( 1 − α ) ω 3 . Two different types of interactions, Class-1 and Class-2, are defined for which a parent wave can form a triad with the subharmonic daughter waves.
Class-1 interactions
We consider three waves with angular frequencies ( ω 3 , ω 1 , ω 2 ) (\omega_3,\omega_1,\omega_2) ( ω 3 , ω 1 , ω 2 ) such that ω 3 = ω 1 + ω 2 \omega_3 = \omega_1 + \omega_2 ω 3 = ω 1 + ω 2 . Furthermore we assume that at a particular h h h , the horizontal wavenumber condition is satisfied between mode i i i of wave-1, mode j j j of wave-2, mode k k k of wave-3, i.e.
K 3 ( k ) = K 1 ( i ) + K 2 ( j ) , (53) \mathcal{K}_{3(k)} = \mathcal{K}_{1(i)} + \mathcal{K}_{2(j)}, \tag*{(53)} K 3 ( k ) = K 1 ( i ) + K 2 ( j ) , (53)
where i i i , j j j and k k k are not all equal. This constitutes a Class-1 interaction. Now if the stratification profile changes (the stratification profile will change in x x x –η \eta η coordinates provided h h h is varying), then the wavenumbers K 1 ( i ) \mathcal{K}_{1(i)} K 1 ( i ) , K 2 ( j ) \mathcal{K}_{2(j)} K 2 ( j ) , K 3 ( k ) \mathcal{K}_{3(k)} K 3 ( k ) will also change. However, for a given change in h h h , all the wavenumbers need not change in a way such that the condition (53) is satisfied. For example, if K 1 ( i ) = func ( h ) \mathcal{K}_{1(i)} = \operatorname{func}(h) K 1 ( i ) = func ( h ) , then it is possible that K 2 ( j ) ≠ c func ( h ) \mathcal{K}_{2(j)} \ne c\operatorname{func}(h) K 2 ( j ) = c func ( h ) , where c c c and func ( ) \operatorname{func}() func ( ) denote an arbitrary constant and function respectively. Therefore, even though the triad condition may be satisfied at a particular h h h , it may not be satisfied for all h h h . Hence Class-1 triads might get detuned as they interact in a region of varying h h h .
To measure the detuning (or mismatch) in the horizontal wavenumber, we define a new variable Δ K \Delta\mathcal{K} Δ K :
Δ K ≡ K 3 ( k ) − K 1 ( i ) − K 2 ( j ) K min , (54) \Delta\mathcal{K} \equiv\frac{\mathcal{K}_{3(k)}-\mathcal{K}_{1(i)}-\mathcal{K}_{2(j)}}{\mathcal{K}_{\min}}, \tag*{(54)} Δ K ≡ K m i n K 3 ( k ) − K 1 ( i ) − K 2 ( j ) , (54)
where K min \mathcal{K}_{\min} K m i n is the minimum wavenumber of the three wavenumbers at a particular x x x -coordinate. Δ K \Delta\mathcal{K} Δ K basically acts as a non-dimensional measure of the detuning between the waves, and for a resonant triad, Δ K = 0 \Delta\mathcal{K} = 0 Δ K = 0 .
We now study how different (nondimensional) wavenumbers K 3 ( n ) \mathcal{K}_{3(n)} K 3 ( n ) of frequency ω 3 \omega_3 ω 3 change as h h h is varied in the presence of a non-uniform stratification. To obtain K 3 ( n ) \mathcal{K}_{3(n)} K 3 ( n ) for a given stratification profile, we solve (2.13b) for h / H ∈ [ − 1 , − 0.2 ] h/H \in[-1,-0.2] h / H ∈ [ − 1 , − 0.2 ] . The functional form of h h h , as long it is mildly varying, does not influence the wavenumbers or detuning at a particular h h h . The non-uniform stratification profile, given by (4.1) is used throughout this paper. The stratification profiles are chosen such that N max = ( 2 N b , 4 N b , … 12 N b ) N_{\max} = (2N_b,4N_b,\ldots12N_b) N m a x = ( 2 N b , 4 N b , … 12 N b ) , W p = ( H / 200 , 2 H / 200 … 5 H / 200 ) W_p = (H/200,2H/200\ldots5H/200) W p = ( H /200 , 2 H /200 … 5 H /200 ) , and z c = ( H / 80 , H / 40 , H / 20 , H / 10 ) z_c = (H/80,H/40,H/20,H/10) z c = ( H /80 , H /40 , H /20 , H /10 ) ; and we consider all possible (120) combinations. Moreover, ω 3 = 0.1 N b \omega_3 = 0.1N_b ω 3 = 0.1 N b and f = 0 f = 0 f = 0 is used consistently for all combinations. Figure 3 show the variation of K ^ 3 ( n ) ≡ K 3 ( n ) ( h ) / K 3 ( n ) ( H ) \widehat{\mathcal{K}}_{3(n)} \equiv\mathcal{K}_{3(n)}(h)/\mathcal{K}_{3(n)}(H) K 3 ( n ) ≡ K 3 ( n ) ( h ) / K 3 ( n ) ( H ) with h / H h/H h / H for different modes n n n . Figure 3 (a)–Figure 3 (c) uses the stratification profile N ( 1 ) N^{(1)} N ( 1 ) with the following parameters: N max = 2 N b N_{\max} = 2N_b N m a x = 2 N b , W p = H / 200 W_p = H/200 W p = H /200 , and z c = H / 80 z_c = H/80 z c = H /80 . Moreover, Figure 3 (d)–Figure 3 (f) uses the profile N ( 2 ) N^{(2)} N ( 2 ) given by: N max = 10 N b N_{\max} = 10N_b N m a x = 10 N b , W p = H / 50 W_p = H/50 W p = H /50 , and z c = H / 10 z_c = H/10 z c = H /10 . Note that N ( 1 ) N^{(1)} N ( 1 ) has a sharp pycnocline, while N ( 2 ) N^{(2)} N ( 2 ) has a larger W p W_p W p resulting in a wider pycnocline. For profiles where all three parameters are low (e.g. N ( 1 ) N^{(1)} N ( 1 ) ), K ^ 3 ( 1 ) \widehat{\mathcal{K}}_{3(1)} K 3 ( 1 ) is nearly constant for some range of h / H h/H h / H and then starts decreasing. This can be seen in figure 3(a), where the first five modes exhibit this behaviour. Moreover, for profiles where all z c z_c z c , W p W_p W p , N max N_{\max} N m a x are high (e.g. N ( 2 ) N^{(2)} N ( 2 ) ), K ^ 3 ( 1 ) \widehat{\mathcal{K}}_{3(1)} K 3 ( 1 ) decreases almost linearly with h / H h/H h / H , as can be clearly seen in Figure 3 (d). For any profile, K ^ 3 ( 1 ) \widehat{K}_{3(1)} K 3 ( 1 ) always decreases as the fluid depth is reduced for h / H ∈ [ − 1 , − 0.2 ] h/H \in[-1,-0.2] h / H ∈ [ − 1 , − 0.2 ] . However this behaviour does not hold for any mode other than mode 1. For example, for z c = H / 10 z_c = H/10 z c = H /10 (regardless of W p W_p W p , N max N_{\max} N m a x ), K ^ 3 ( 2 ) \widehat{K}_{3(2)} K 3 ( 2 ) increases for some h / H h/H h / H as fluid depth is reduced, see Figure 3 (d) (blue curve). Similar behaviour is also observed for modes 3, 4 and 5 when W p W_p W p is low. In summary, the variation of K ^ 3 ( 1 ) \widehat{K}_{3(1)} K 3 ( 1 ) with h / H h/H h / H can be different from that of the higher modes’ wavenumber, which can result in detuning.
Figure 3. Variation of K ^ 3 ( n ) \widehat{K}_{3(n)} K 3 ( n ) with h / H h/H h / H for two different stratification profiles. For stratification profile N ( 1 ) N^{(1)} N ( 1 ) : (a) modes 1-5, (b) modes 6-10 and (c) modes 11-15. For stratification profile N ( 2 ) N^{(2)} N ( 2 ) : (d) modes 1-5, (e) modes 6-10 and (f) modes 11-15.
For profiles with high W p W_p W p , K ^ 3 ( n ) \widehat{K}_{3(n)} K 3 ( n ) for n > 10 n > 10 n > 10 starts to collapse on each other, see Figure 3 (f). In such kind of scenarios, since K ^ 3 ( n ) \widehat{K}_{3(n)} K 3 ( n ) remains nearly the same, Δ K \Delta K Δ K will not be induced by the difference in higher modes’ K ^ 3 ( n ) \widehat{K}_{3(n)} K 3 ( n ) . In general it was observed that as W p W_p W p is reduced, n n n has to be higher for the modes to collapse on each other.
The interaction of mode-1 internal wave (wave-3) with different modes in the presence of two different non-uniform stratification profiles is considered next. These profiles are a part of the 120 profiles that we already mentioned. Sample results are shown in Figure 4 in which the frequencies and stratification profile parameters are as follows:
Figure 4. Variation of detuning (Δ K \Delta K Δ K ) with h / H h/H h / H for various modal interactions pertaining to the stratification profiles (a) N ( 3 ) N^{(3)} N ( 3 ) , and (b) N ( 4 ) N^{(4)} N ( 4 ) . The legends indicate what daughter waves were involved in the triad interactions. α ≡ ω 1 / ω 3 \alpha\equiv\omega_1/\omega_3 α ≡ ω 1 / ω 3 .
N ( 3 ) N^{(3)} N ( 3 ) : ω 3 = 0.1 N b \omega_3 = 0.1N_b ω 3 = 0.1 N b , f = 0 f = 0 f = 0 , N max = 10 N b N_{\max} = 10N_b N m a x = 10 N b , W p = H / 100 W_p = H/100 W p = H /100 and z c = H / 10 z_c = H/10 z c = H /10 .
N ( 4 ) N^{(4)} N ( 4 ) : ω 3 = 0.1 N b \omega_3 = 0.1N_b ω 3 = 0.1 N b , f = 0 f = 0 f = 0 , N max = 10 N b N_{\max} = 10N_b N m a x = 10 N b , W p = H / 50 W_p = H/50 W p = H /50 and z c = H / 20 z_c = H/20 z c = H /20 .
For each profile, we have shown 5 different modal interactions. Figure 4 clearly reveals that the detuning can be quite sensitive to the changes in the domain height.
Class-2 interactions: A special case of triad interactions
The interaction in Class-2 is between the n n n -th modes (where n ∈ Z + n \in\mathbb{Z}^{+} n ∈ Z + ) of different waves constituting a triad. For example, if mode 1 with frequency ω 1 \omega_1 ω 1 , mode 1 with frequency ω 2 \omega_2 ω 2 and mode 1 with frequency ω 3 \omega_3 ω 3 form a triad, it is classified as a Class-2 interaction. This kind of triad is possible when f ≪ ω j ≪ N b f \ll\omega_j \ll N_b f ≪ ω j ≪ N b . To show how this interaction is possible, we consider the eigen-problem concerning the n n n -th mode of the j j j -th wave:
∂ 2 ϕ j ( n ) ∂ η 2 + K j ( n ) 2 χ j 2 ϕ j ( n ) ≈ ∂ 2 ϕ j ( n ) ∂ η 2 + ( K j ( n ) ω j ) 2 N 2 ϕ j ( n ) = 0 , (55) \frac{\partial^2 \phi_{j(n)}}{\partial\eta^2} + K_{j(n)}^2 \chi_j^2 \phi_{j(n)} \approx\frac{\partial^2 \phi_{j(n)}}{\partial\eta^2} + \left(\frac{K_{j(n)}}{\omega_j}\right)^2 N^2 \phi_{j(n)} = 0, \tag*{(55)} ∂ η 2 ∂ 2 ϕ j ( n ) + K j ( n ) 2 χ j 2 ϕ j ( n ) ≈ ∂ η 2 ∂ 2 ϕ j ( n ) + ( ω j K j ( n ) ) 2 N 2 ϕ j ( n ) = 0 , (55)
where we used χ j ≈ N / ω j \chi_j \approx N/\omega_j χ j ≈ N / ω j (under the approximation f ≪ ω j ≪ N b f \ll\omega_j \ll N_b f ≪ ω j ≪ N b ), and the system is solved using the boundary conditions: ϕ j ( n ) = 0 \phi_{j(n)} = 0 ϕ j ( n ) = 0 at η = 0 \eta= 0 η = 0 and η = − 1 \eta= -1 η = − 1 . However, by Sturm Liouville theory, for a given operator (here ∂ 2 / ∂ η 2 \partial^2/\partial\eta^2 ∂ 2 / ∂ η 2 ) and weight function (here N ( z ) 2 N(z)^2 N ( z ) 2 ), the n n n -th eigenvalue (here K j ( n ) / ω j K_{j(n)}/\omega_j K j ( n ) / ω j ) is unique, i.e., K j ( n ) / ω j = constant ∀ j K_{j(n)}/\omega_j = \text{constant }\forall j K j ( n ) / ω j = constant ∀ j . Therefore, if the triad condition for frequency: ω 3 = ω 2 + ω 1 \omega_3 = \omega_2 + \omega_1 ω 3 = ω 2 + ω 1 is valid, this automatically implies validity of the wavenumber condition K 3 ( n ) = K 1 ( n ) + K 2 ( n ) K_{3(n)} = K_{1(n)} + K_{2(n)} K 3 ( n ) = K 1 ( n ) + K 2 ( n ) .
The situation mentioned above is true for all stratification profiles satisfying f ≪ ω j ≪ N ( z ) f \ll\omega_j \ll N(z) f ≪ ω j ≪ N ( z ) (at all z z z locations). This is especially important because in the presence of a bathymetry, the stratification profile changes in the x x x direction in x − η x-\eta x − η coordinates. However, for Class-2 interaction, all three non-dimensional wavenumbers (eigenvalues) are the same functions of h h h since they are the same eigenvalues divided by their frequency. Thus the resonant triad condition will be still be satisfied even if h h h is varied significantly (i.e. variation of h h h will not cause detuning). Note that, in the parameter regime of f ≪ ω j ≪ N ( z ) f \ll\omega_j \ll N(z) f ≪ ω j ≪ N ( z ) , only class-2 self interactions were observed in numerical experiments of Sutherland [34 ] , hence class-2 triads may always be dominated by self interactions, resulting in parent wave’s energy transfer to the superharmonics instead of subharmonics. As a result, class-2 triads may not be practically as relevant as class-2 self interactions.
Effect of bathymetry on horizontal wavenumber condition for Class-1 self-interaction
Detuning can also be introduced during a self-interaction process as h h h is varied. Following the same terminology as before, we classify self-interactions as Class-1 and Class-2. As shown by Wunsch [39 ] , Class-2 self-interactions will always be slightly detuned, where the detuning increases as f f f increases. This is due to the fact that in a non-uniform stratification, if the n n n -th mode of ( ω ) (\omega) ( ω ) satisfies the dispersion relation, then the n n n -th mode of ( 2 ω ) (2\omega) ( 2 ω ) will be able to satisfy it only approximately.
Following (54) , the detuning for a self-interaction process is defined as
Δ K s = K 3 ( k ) − 2 K 1 ( i ) K 1 ( i ) , (56) \Delta K_s = \frac{K_{3(k)} - 2K_{1(i)}}{K_{1(i)}}, \tag*{(56)} Δ K s = K 1 ( i ) K 3 ( k ) − 2 K 1 ( i ) , (56)
where wave-3 is the superharmonic (daughter) wave, while wave-1 is the parent wave, i.e. ω 1 = ω 3 / 2 \omega_1 = \omega_3/2 ω 1 = ω 3 /2 (following the convention used throughout this paper that wave-3 has the highest frequency). The amplitude evolution equations for a self-interaction process is discussed in § 2.2.2 .
For the range f ≪ ω j ≪ N ( z ) f \ll\omega_j \ll N(z) f ≪ ω j ≪ N ( z ) , Class-2 self-interaction process will follow similar principles outlined in § 4.2 . As mentioned in § 4.2 , significant variations in h h h for this frequency range will not introduce detuning in Class-2 self-interactions. We now study the other end of the parameter space where ( N 2 − ω j 2 ) ≈ N 2 (N^2-\omega_j^2) \approx N^2 ( N 2 − ω j 2 ) ≈ N 2 , which was the basic approximation used in much of our analysis is §4.1, is no longer valid. Hence, out of Class-1 and Class-2 self-interactions, only the latter is possible. This would mean that as the domain height changes, the detuning introduced could be significant. Interestingly though, if the wavenumbers involved in the self-interaction change with h / H h/H h / H in a similar way, the detuning is insignificant, see Figure 5 . The frequencies and the stratification profile parameters used here are:
Figure 5. (a) Variations of K 3 ( 2 ) \mathcal{K}_{3(2)} K 3 ( 2 ) and K 1 ( 3 ) \mathcal{K}_{1(3)} K 1 ( 3 ) with h / H h/H h / H for the parameter set N ( 6 ) \mathcal{N}^{(6)} N ( 6 ) . (b) Variation of detuning with h / H h/H h / H for the same case. (c) The detuning for three different self-interaction combinations for the parameter set N ( 7 ) \mathcal{N}^{(7)} N ( 7 ) . Here the notation (Pa)(Db) implies ‘Parent wave’ with ‘mode-a’ and ‘Daughter wave’ with ‘mode-b’.
N ( 6 ) N^{(6)} N ( 6 ) : N max = 10 N b N_{\max}=10N_b N m a x = 10 N b , W p = H / 100 W_p=H/100 W p = H /100 and z c = H / 10 z_c=H/10 z c = H /10 .
N ( 7 ) N^{(7)} N ( 7 ) : N max = 10 N b N_{\max}=10N_b N m a x = 10 N b , W p = H / 100 W_p=H/100 W p = H /100 and z c = H / 20 z_c=H/20 z c = H /20 ,
and f = 0 f=0 f = 0 always. Figure 5 (a) uses the set N ( 6 ) N^{(6)} N ( 6 ) , and shows the variation of the horizontal wavenumber of mode-2 (of ω 3 = 0.89 N b \omega_3=0.89N_b ω 3 = 0.89 N b ) and mode-3 (of ω 1 = ω 3 / 2 \omega_1=\omega_3/2 ω 1 = ω 3 /2 ). These modes satisfy the condition for resonant self-interaction. We observe that these two wavenumbers behave quite similarly for a wide range of h / H h/H h / H , and hence the detuning, shown in Figure 5 (b), is small (and constant for an appreciable range), in spite of the fact that it is a Class-1 interaction. The same phenomenon is also shown for several other self-interaction combinations in Figure 5 (c), where the parameter set N ( 7 ) N^{(7)} N ( 7 ) is used.
The detuning for all the combinations shown stays constant for a certain range of h / H h/H h / H . We note in passing that Class-1 triad interactions may also give rise to a small detuning for a range of h / H h/H h / H , provided all the modes involved behave in a similar way. However, this is a more stringent condition than a self-interaction process, where only two waves are involved. Even though equations derived in § 2 are only valid when Δ K ≪ 1 \Delta K \ll1 Δ K ≪ 1 (or Δ K s ≪ 1 \Delta K_s \ll1 Δ K s ≪ 1 ), there are significant number of interactions where Δ K \Delta K Δ K , or Δ K s \Delta K_s Δ K s , is a small quantity even for O ( 1 ) O(1) O ( 1 ) changes in depth and the wavenumber. For example, for interactions shown in this section, and for class-2 interactions, detuning can stay as a small quantity even for O ( 1 ) O(1) O ( 1 ) changes in depth. However, we do note that in several triad interactions detuning can be sensitive to h h h and in those cases O ( 1 ) O(1) O ( 1 ) changes in depth cannot be accurately modeled by the wave-amplitude equations.
To summarize, in the presence of a non-uniform stratification, we divide triad and self-interactions into two classes: Class-1 and Class-2. Class-1 interactions contain waves whose mode numbers are all not the same, while Class-2 interactions contain waves which are the n n n –th modes of their respective frequencies. Class-1 interactions, may undergo detuning with the variation in h h h , irrespective of the frequency. However, interestingly, certain Class-1 self-interactions do not undergo detuning as h h h is varied inside a certain range. For both triads and self-interactions, Class-2 interactions can only exist for f ≪ ω j ≪ N ( z ) f \ll\omega_j \ll N(z) f ≪ ω j ≪ N ( z ) , and does not get detuned as h h h is varied.
Variation of growth rates and nonlinear coupling coefficients with depth for non-uniform stratification
In this section, we focus on the effects of domain height variation on the growth rate (σ \sigma σ ) of triads, and the nonlinear coupling coefficient N 3 \mathcal{N}_3 N 3 , which provides a measure of the growth of the daughter wave in a self-interaction. The non-uniform stratification profile (4.1) will be used in this section.
Variation of growth rates with domain height for triads
Triad interactions are important for the decay of internal waves near the 28.9° latitude [27 , 26 ] , specifically the mode-1 wave, which is the most energy containing mode [37 ] . Here we study this phenomena in the high latitude region (f / ω 3 > 0.3 f/\omega_{3} > 0.3 f / ω 3 > 0.3 ) for varying N ( z ) N(z) N ( z ) and h h h . The mode-1 wave (which, being the parent wave, is wave-3) can decay forming various triad combinations; we restrict the subharmonic daughter waves (wave-1 and wave-2) up to mode-50. Moreover, for studying growth rates in the presence of varying h h h , the triads are identified separately at different h / H h/H h / H . This is because a triad combination at a particular h / H h/H h / H value may not satisfy the horizontal wavenumber condition at a different h / H h/H h / H (as explained in § 4 ).
Three main branches of triads are considered here for the mode-1 internal wave:
∣ K 3 ∣ = ∣ K 2 ∣ − ∣ K 1 ∣ ⏟ Branch-1 or ∣ K 3 ∣ = ∣ K 1 ∣ − ∣ K 2 ∣ ⏟ Branch-2 or ∣ K 3 ∣ ≈ ∣ K 1 ∣ + ∣ K 2 ∣ ⏟ Branch-3 . (57) \underbrace{|\mathcal{K}_{3}| = |\mathcal{K}_{2}| - |\mathcal{K}_{1}|}_{\text{Branch-1}}
\quad\text{or}\quad
\underbrace{|\mathcal{K}_{3}| = |\mathcal{K}_{1}| - |\mathcal{K}_{2}|}_{\text{Branch-2}}
\quad\text{or}\quad
\underbrace{|\mathcal{K}_{3}| \approx|\mathcal{K}_{1}| + |\mathcal{K}_{2}|}_{\text{Branch-3}} . \tag*{(57)} Branch-1 ∣ K 3 ∣ = ∣ K 2 ∣ − ∣ K 1 ∣ or Branch-2 ∣ K 3 ∣ = ∣ K 1 ∣ − ∣ K 2 ∣ or Branch-3 ∣ K 3 ∣ ≈ ∣ K 1 ∣ + ∣ K 2 ∣ . (57)
For Branch-1(2) triads, wavenumber of wave-2(1) is larger in magnitude than that of wave-1(2). The only possible Branch-3 interaction is a Class-2 interaction, where both the daughter waves are also mode-1 of their respective frequencies. However this interaction, like the Class-2 self-interaction, also undergoes heavy detuning for high f f f values. Therefore Branch-3 being an inefficient energy transfer pathway, we restrict our focus to Branch-1 and Branch-2. Triads are studied for f / ω 3 = ( 0.3 , 0.4 , 0.45 ) f/\omega_{3} = (0.3, 0.4, 0.45) f / ω 3 = ( 0.3 , 0.4 , 0.45 ) in the presence of various stratification profiles. The triads are computed for α ∈ [ 0.31 , 0.5 ] \alpha\in[0.31, 0.5] α ∈ [ 0.31 , 0.5 ] , α ∈ [ 0.41 , 0.5 ] \alpha\in[0.41, 0.5] α ∈ [ 0.41 , 0.5 ] and α ∈ [ 0.455 , 0.5 ] \alpha\in[0.455, 0.5] α ∈ [ 0.455 , 0.5 ] for f / ω 3 = 0.3 f/\omega_{3} = 0.3 f / ω 3 = 0.3 , 0.4 and 0.45 respectively (see §4.1 for the definition of α \alpha α ).
Figure 6 shows the non-dimensionalised growth rate contour for a mode-1 wave. All growth rates σ \sigma σ are non-dimensionalised with a reference growth rate value σ r e f \sigma_{\mathrm{ref}} σ ref , where the latter denotes the maximum growth rate for all Branch-1 triads at h = − H h = -H h = − H (hence the value of A 3 A_{3} A 3 does not impact the results shown). The frequency of the mode-1 wave is ω 3 / N b = 0.2 \omega_{3}/N_{b} = 0.2 ω 3 / N b = 0.2 , while f / ω 3 = 0.4 f/\omega_{3} = 0.4 f / ω 3 = 0.4 is taken. The stratification profile is given by
Figure 6. Contours of non-dimensional growth rate (σ / σ r e f \sigma/\sigma_{\mathrm{ref}} σ / σ ref ) of triads formed between mode-n n n , mode-m m m , and mode-1 (i.e., wave-3). Blue and red colors respectively represent Branch-1 and Branch-2 triads, and both colors represent positive values. For Branch-1 triads, mode-m ( n ) m(n) m ( n ) is wave-1(2), while for Branch-2 triads, mode-n ( m ) n(m) n ( m ) is wave-1(2).
Branch-1(2) triads have the higher(lower) frequency daughter wave propagating in the same direction as the parent wave. Figure 6 reveals that from both branches, the highest growth rates are centered around n ≈ m n \approx m n ≈ m . However, majority of the white region contains resonant triads, but their growth rates are significantly lower in comparison to that clustered around n ≈ m n \approx m n ≈ m . Note that the central region is asymmetric between Branch-1 and Branch-2 triads, and this is purely a consequence of internal wave’s dispersion relation. When the lower frequency daughter wave (wave-1) travels in the same direction as the parent wave (i.e. Branch-2), wave-1’s modenumber (n n n ) should always be higher than wave-2’s modenumber (m m m ) for the triad condition to be satisfied. However for Branch-1, where wave-2 travels in the same direction as the parent wave, the modenumber of wave-2 (n n n ) need not be higher than wave-1’s modenumber (m m m ).
The clustering around n ≈ m n \approx m n ≈ m is consistently observed for any setting or stratification profile considered in our study. As a result instead of focusing on all possible triads, we choose specific lines of interaction near the central region and plot the growth rate along that line of interaction. For example, the interaction lines ( n , n ) (n,n) ( n , n ) , ( n + 1 , n ) (n+1,n) ( n + 1 , n ) and ( n , n + 1 ) (n,n+1) ( n , n + 1 ) are plotted for n ∈ ( 1 , 50 ) n \in(1,50) n ∈ ( 1 , 50 ) in Figure 7 (a)–Figure 7 (c) for Branch-1 triads, and ( n + 4 , n ) (n+4,n) ( n + 4 , n ) in Figure 7 (d) for Branch-2 triads. The notation (a,b) means wave-1(2) is mode-a(b). The notation is same for both branches. The dominant nature of the interaction lines ( n , n ) (n,n) ( n , n ) , ( n , n + 1 ) (n,n+1) ( n , n + 1 ) , ( n + 1 , n ) (n+1,n) ( n + 1 , n ) has also been observed in Young et al. [40 ] while studying the stability of mode-1 internal wave in the presence of near inertial daughter waves (with frequency f f f ). Furthermore, Figure 7 also reveals that the different lines are sensitive to h h h . For completeness, we explore another stratification profile given by:
Figure 7. Plots of non-dimensional growth rates of mode-1 triads for profile N ( 8 ) N^{(8)} N ( 8 ) with f / ω 3 = 0.4 f/\omega_3 = 0.4 f / ω 3 = 0.4 and ω 3 / N b = 0.2 \omega_3/N_b = 0.2 ω 3 / N b = 0.2 . (a) Line ( n , n ) (n,n) ( n , n ) of Branch-1, (b) line ( n + 1 , n ) (n+1,n) ( n + 1 , n ) of Branch-1, (c) line ( n , n + 1 ) (n,n+1) ( n , n + 1 ) of Branch-1, and (d) line ( n + 4 , n ) (n+4,n) ( n + 4 , n ) of Branch-2.
N ( 9 ) N^{(9)} N ( 9 ) : N max = 10 N b N_{\max}=10N_b N m a x = 10 N b , W p = H / 50 W_p=H/50 W p = H /50 , z c = H / 80 z_c=H/80 z c = H /80 ,
and the corresponding plots are in Figure 8 . Both figures 7 and 8 show that the growth rates along different lines of interaction have a significant oscillatory nature with n n n . In general, line ( n , n ) (n,n) ( n , n ) has the largest amplitude of oscillations. More importantly, the growth rate of a modal combination can significantly change as h h h changes. For example, Figure 7 (a) shows that the most unstable modal combination at h = − H h=-H h = − H is ( 5 , 5 ) (5,5) ( 5 , 5 ) . However, for h = − 0.8 H h=-0.8H h = − 0.8 H , the most unstable triad is the modal combination ( 4 , 4 ) (4,4) ( 4 , 4 ) . Moreover, the combination ( 5 , 5 ) (5,5) ( 5 , 5 ) has approximately 0.25 times of ( 4 , 4 ) (4,4) ( 4 , 4 ) growth rate at h = − 0.8 H h=-0.8H h = − 0.8 H . This behavior can be seen for the line ( n , n ) (n,n) ( n , n ) in both Figure 7 and Figure 8 . This effectively means that the growth rate of certain daughter wave combinations can be sensitive to changes in h h h (especially the combinations which involve lower modes). Such combinations may not be effective in a region of varying h h h because of the significant drop in the growth rates. However, sensitivity to h h h is slowly reduced as the modenumber is increased for both the branches. Even though Branch-2 triads have considerably less growth rates for the profiles N ( 8 ) N^{(8)} N ( 8 ) and N ( 9 ) N^{(9)} N ( 9 ) , for different profiles (not displayed here) Branch-2 can have σ \sigma σ comparable to that of the Branch-1 triads.
Figure 8. Plots of non-dimensional growth rates of mode-1 triads for profile N ( 9 ) N^{(9)} N ( 9 ) with f / ω 3 = 0.4 f/\omega_3 = 0.4 f / ω 3 = 0.4 and ω 3 / N b = 0.2 \omega_3/N_b = 0.2 ω 3 / N b = 0.2 . (a) Line ( n , n ) (n,n) ( n , n ) of Branch-1, (b) line ( n + 1 , n ) (n+1,n) ( n + 1 , n ) of Branch-1, (c) line ( n , n + 1 ) (n,n+1) ( n , n + 1 ) of Branch-1, and (d) line ( n + 6 , n ) (n+6,n) ( n + 6 , n ) of Branch-2.
Effect of variation of f / ω 3 f/\omega_3 f / ω 3 and ω 3 / N b \omega_3/N_b ω 3 / N b on different Branches
For both stratification profiles used in § 5.1 , f / ω 3 = ( 0.3 , 0.45 ) f/\omega_3=(0.3, 0.45) f / ω 3 = ( 0.3 , 0.45 ) for ω 3 / N b = ( 0.2 , 0.7 ) \omega_3/N_b=(0.2, 0.7) ω 3 / N b = ( 0.2 , 0.7 ) is explored (hence total of 4 different combinations). For N ( 8 ) N^{(8)} N ( 8 ) , in all 4 cases, the qualitative behaviour of all Branch-1 lines are similar to figure 7. However, the maximum growth rate has a significant increase from h = − H h=-H h = − H to h = − 0.8 H h=-0.8H h = − 0.8 H for ω 3 / N b = 0.7 \omega_3/N_b=0.7 ω 3 / N b = 0.7 . Moreover, Branch-2 triads’ maximum growth rate significantly increases in 2 cases of f / ω 3 = 0.3 f/\omega_3=0.3 f / ω 3 = 0.3 in comparison to f / ω 3 = 0.4 f/\omega_3=0.4 f / ω 3 = 0.4 . For N ( 9 ) N^{(9)} N ( 9 ) , the qualitative behaviour of line ( n , n ) (n,n) ( n , n ) is similar to what was observed in f / ω 3 = 0.4 f/\omega_3=0.4 f / ω 3 = 0.4 . In general, the maximum growth rate is the ( n , n ) (n,n) ( n , n ) modal combination. Interestingly, it is found that the maximum growth rate among all triads increased nearly twice from h = − H h=-H h = − H to h = − 0.8 H h=-0.8H h = − 0.8 H for ( f / ω 3 = 0.45 , ω 3 / N b = 0.7 ) (f/\omega_3=0.45,\omega_3/N_b=0.7) ( f / ω 3 = 0.45 , ω 3 / N b = 0.7 ) . Significant increase in maximum growth rate is also found for ω 3 / N b = 0.2 \omega_3/N_b=0.2 ω 3 / N b = 0.2 for the same f f f . For ( f / ω 3 = 0.3 , ω 3 / N b = 0.7 ) (f/\omega_3=0.3,\omega_3/N_b=0.7) ( f / ω 3 = 0.3 , ω 3 / N b = 0.7 ) the behaviour of line ( n + 1 , n ) (n+1,n) ( n + 1 , n ) has a significant oscillation with n n n , similar to line ( n , n ) (n,n) ( n , n ) . Note that this is different from the line ( n + 1 , n ) (n+1,n) ( n + 1 , n ) shown in figure 8. In general, it is also observed that reducing the fluid depth increases the maximum growth rate of all possible triads even without considering the β \beta β term of the parent wave amplitude.
Variation of nonlinear coupling coefficient with domain height for self-interaction process
Here we restrict to self-interaction of internal gravity waves that do not experience significant detuning Δ K \Delta K Δ K with changes in h h h . In this subsection, we mainly focus on the superharmonic wave’s nonlinear coupling coefficient N 3 \mathcal{N}_3 N 3 given in (43) .
Class-1 interactions
As previously mentioned in § 4.3 , some Class-1 self-interactions can have negligible detuning even for a finite range of h / H h/H h / H . We study the variation of N ~ 3 \widetilde{\mathcal{N}}_3 N 3 under such circumstances; the different interactions considered (denoted by I p I_p I p ) are given below:
∣ I 1 − [ P 3 , D 2 ] ∣ ∣ I 2 − [ P 4 , D 2 ] ∣ ∣ I 3 − [ P 4 , D 3 ] ∣ ∣ I 4 − [ P 5 , D 3 ] ∣ ∣ I 5 − [ P 5 , D 4 ] ∣ ∣ I 6 − [ P 6 , D 3 ] ∣ ∣ I 7 − [ P 6 , D 4 ] ∣ ∣ I 8 − [ P 6 , D 5 ] ∣ ∣ I 9 − [ P 7 , D 4 ] ∣ ∣ I 10 − [ P 7 , D 5 ] ∣ \begin{aligned}
&\left|I_1-[P_3,D_2]\right|\left|I_2-[P_4,D_2]\right|\left|I_3-[P_4,D_3]\right|\left|I_4-[P_5,D_3]\right|\left|I_5-[P_5,D_4]\right| \\
&\left|I_6-[P_6,D_3]\right|\left|I_7-[P_6,D_4]\right|\left|I_8-[P_6,D_5]\right|\left|I_9-[P_7,D_4]\right|\left|I_{10}-[P_7,D_5]\right|
\end{aligned} ∣ I 1 − [ P 3 , D 2 ] ∣ ∣ I 2 − [ P 4 , D 2 ] ∣ ∣ I 3 − [ P 4 , D 3 ] ∣ ∣ I 4 − [ P 5 , D 3 ] ∣ ∣ I 5 − [ P 5 , D 4 ] ∣ ∣ I 6 − [ P 6 , D 3 ] ∣ ∣ I 7 − [ P 6 , D 4 ] ∣ ∣ I 8 − [ P 6 , D 5 ] ∣ ∣ I 9 − [ P 7 , D 4 ] ∣ ∣ I 10 − [ P 7 , D 5 ] ∣
Here the notation [ P m , D n ] [P_m,D_n] [ P m , D n ] denotes that the parent wave is the m m m –th mode and daughter (superharmonic) wave is the n n n –th mode.
The stratification profiles are chosen such that N max = ( 2 N b , 5 N b , 10 N b ) N_{\max}=(2N_b,5N_b,10N_b) N m a x = ( 2 N b , 5 N b , 10 N b ) , W p = ( H / 200 , H / 100 , H / 50 ) W_p=(H/200,H/100,H/50) W p = ( H /200 , H /100 , H /50 ) , and z c = ( H / 40 , H / 20 , H / 10 ) z_c=(H/40,H/20,H/10) z c = ( H /40 , H /20 , H /10 ) . For the profiles considered, we study variations of N ~ 3 \widetilde{\mathcal{N}}_3 N 3 for interactions that strictly satisfy ∣ Δ K s ∣ < 0.01 |\Delta K_s|<0.01 ∣Δ K s ∣ < 0.01 for h / H ∈ [ − 1 , − 0.8 ] h/H\in[-1,-0.8] h / H ∈ [ − 1 , − 0.8 ] . Figure 9 shows variations of N ~ 3 ≡ ∣ N 3 ∣ / max ( ∣ N 3 ∣ ) \widetilde{\mathcal{N}}_3\equiv|\mathcal{N}_3|/\max(|\mathcal{N}_3|) N 3 ≡ ∣ N 3 ∣/ max ( ∣ N 3 ∣ ) for two Class-1 self-interactions. Figure 9 reveals that interactions can have a non-monotonic variation of N ~ 3 \widetilde{\mathcal{N}}_3 N 3 with h / H h/H h / H . Moreover, the figure reveals that even relatively small changes in h / H h/H h / H can lead to significant variations in N ~ 3 \widetilde{\mathcal{N}}_3 N 3 . For all interactions considered in table 5.2.1, small changes in h / H h/H h / H can cause significant change in N ~ 3 \widetilde{\mathcal{N}}_3 N 3 . Some generic features are summarised below. Sensitivity of N ~ 3 \widetilde{\mathcal{N}}_3 N 3 to h / H h/H h / H increases as z c z_c z c is increased for a given W p , N max W_p,N_{\max} W p , N m a x .2 Moreover, increasing W p W_p W p also increases the sensitivity of N ~ 3 \widetilde{\mathcal{N}}_3 N 3 to changes in h / H h/H h / H . Increasing N max N_{\max} N m a x for a given ( z c , W p ) (z_c,W_p) ( z c , W p ) also increases the sensitivity of N ~ 3 \widetilde{\mathcal{N}}_3 N 3 to h / H h/H h / H .
Figure 9. Variation of N ~ 3 \widetilde{N}_3 N 3 with h h h for Class-1 self-interactions (a) I 1 \mathcal{I}_1 I 1 and (b) I 10 \mathcal{I}_{10} I 10 . Each sub-figure is plotted for two different stratification profiles (for details see legend).
Figure 10. The variation of non-dimensionalised nonlinear coupling coefficient (N ~ 3 \widetilde{N}_3 N 3 ) of wave-3 (the superharmonic wave) as h h h is varied. Altogether there are 9 blocks, each consisting of 5 units, and each unit represents the modenumber (given by n n n ). For each block, (N max , W p , z c N_{\max}, W_p, z_c N m a x , W p , z c ) is fixed. The figure is further subdivided into three horizontal panels (each panel consisting of three blocks): (a) N max = 2 N b N_{\max}=2N_b N m a x = 2 N b , W p = H / 200 W_p=H/200 W p = H /200 , and z c z_c z c is varied, (b) N max = 5 N b N_{\max}=5N_b N m a x = 5 N b , z c = H / 20 z_c=H/20 z c = H /20 and W p W_p W p is varied, and (c) z c = H / 10 z_c=H/10 z c = H /10 , W p = H / 50 W_p=H/50 W p = H /50 and N max N_{\max} N m a x is varied.
Class-2 interactions
We initially study the variation of N 3 N_3 N 3 with h h h for Class-2 self interactions. To this end, we consider N 3 N_3 N 3 of the first 5 modes for 27 different stratification profiles. Similar to the case of Class-1 self-interactions in §5.2.1, the stratification profiles are chosen such that N max = ( 2 N b , 5 N b , 10 N b ) N_{\max}=(2N_b,5N_b,10N_b) N m a x = ( 2 N b , 5 N b , 10 N b ) , W p = ( H / 200 , H / 100 , H / 50 ) W_p=(H/200,H/100,H/50) W p = ( H /200 , H /100 , H /50 ) , and z c = ( H / 40 , H / 20 , H / 10 ) z_c=(H/40,H/20,H/10) z c = ( H /40 , H /20 , H /10 ) , where we consider all possible (3 3 3^3 3 3 ) combinations. Out of the 3 3 = 27 3^3 = 27 3 3 = 27 combinations, 9 profiles are chosen for plotting figure 10 and thereby elucidating the effect of each individual parameter in the stratification profile. For all cases, ω 3 = 0.1 N b \omega_3 = 0.1N_b ω 3 = 0.1 N b and f = 0 f = 0 f = 0 . For some higher modes, the nonlinear coupling coefficient has a band like structure; there exists some range of h / H h/H h / H where N ~ 3 \tilde{N}_3 N ~ 3 is significantly higher in magnitude than that corresponding to other values of h / H h/H h / H . For example, the mode n = 5 n = 5 n = 5 corresponding to N max = 5 N b N_{\max} = 5N_b N m a x = 5 N b in figure 10(c) reveals a large increase in N ~ 3 \tilde{N}_3 N ~ 3 near h / H ≈ − 0.75 h/H \approx-0.75 h / H ≈ − 0.75 , while it is much lower at either ends. Wunsch [39 ] also observed such banded structure in the self-interaction of different modes as the stratification profile was changed. The reason behind the direct analogy between our observations and that of Wunsch [39 ] is as straightforward – when an internal wave travels to a different domain height, it essentially travels to a different stratification profile.
For mode-1, when z c z_c z c , W p W_p W p , N max N_{\max} N m a x are all on the lower side, we observe that N ~ 3 ∝ 1 / h 4 \tilde{N}_3 \propto1/h^4 N ~ 3 ∝ 1/ h 4 . For higher N max N_{\max} N m a x , even lower values of z c z_c z c and W p W_p W p do not have the property of N ~ 3 ∝ 1 / h 4 \tilde{N}_3 \propto1/h^4 N ~ 3 ∝ 1/ h 4 . In general for modes > 1 > 1 > 1 , proportionality to 1 / h 4 1/h^4 1/ h 4 is lost faster as z c z_c z c , W p W_p W p , N max N_{\max} N m a x is increased. In several profiles, N ~ 3 \tilde{N}_3 N ~ 3 of higher modes is also more sensitive to changes in h h h than N ~ 3 ∝ 1 / h 4 \tilde{N}_3 \propto1/h^4 N ~ 3 ∝ 1/ h 4 .
Higher order self-interactions in the presence of a small amplitude monochromatic topography
The focus of this section is on higher order self-interactions between a parent wave of frequency ω 1 \omega_1 ω 1 and a superharmonic daughter wave of frequency ω 3 = 2 ω 1 \omega_3 = 2\omega_1 ω 3 = 2 ω 1 in the presence of a small amplitude monochromatic topography. In such kind of scenarios, the topography can act as a ‘zero frequency wave’, which can lead to resonant higher order interactions, and is similar to the Class-2 studied in Alam et al. [1 ] for surface gravity waves. Such kind of higher order self-interactions might be important for mode-1 internal waves propagating in regions where f > ω d / 2 f > \omega_d/2 f > ω d /2 (ω d \omega_d ω d is the semidiurnal frequency), since triad interactions involving two (subharmonic) daughter waves is not possible. Moreover, resonant self-interaction for mode-1 internal wave of frequency ω d \omega_d ω d is also not possible when f > ω d / 2 f > \omega_d/2 f > ω d /2 for any stratification profile as a consequence of its dispersion relation (Wunsch [39 ] ). This arises from the fact that a mode-1 parent wave with frequency ω d \omega_d ω d and a superharmonic daughter wave with frequency 2 ω d 2\omega_d 2 ω d fail to satisfy the horizontal wavenumber condition for a self-interaction process. Note that higher order interactions are different from Bragg resonance focused in Buhler & Holmes-Cerfon [3 ] , which is also a mechanism via which a parent mode-1 wave can decay by transferring its energy to the higher modes. In a standard Bragg resonance, resonant wave-topography interaction occurs if the bottom topography has a wavenumber k b k_b k b such that ( ω 1 , k b ± k 1 ) (\omega_1,k_b \pm k_1) ( ω 1 , k b ± k 1 ) satisfies the dispersion relation.
To study higher order self-interactions, we follow the streamfunction ansatz used in Couston et al. [5 ] for studying internal wave Bragg resonance, and in Lahaye & Llewellyn Smith [20 ] for studying internal wave scattering due to interaction with a large amplitude topography. This ansatz for the streamfunction of the j j j -th wave is as follows:
Ψ j = A j ( x ) ϕ j ( n ; x ) e − i ω j t + c . c . , (58) \Psi_j = A_j(x)\phi_j(n;x)e^{-i\omega_j t} + \mathrm{c.c.}, \tag*{(58)} Ψ j = A j ( x ) ϕ j ( n ; x ) e − i ω j t + c.c. , (58)
The corresponding buoyancy frequency and meridional velocity is given by
B j = i N 2 ω j ∂ A j ∂ x ϕ j e − i ω j t + c . c . , (59) B_j = \frac{iN^2}{\omega_j}\frac{\partial A_j}{\partial x}\phi_j e^{-i\omega_j t} + \mathrm{c.c.}, \tag*{(59)} B j = ω j i N 2 ∂ x ∂ A j ϕ j e − i ω j t + c.c. , (59)
V j = i f A j ω j h ∂ ϕ j ∂ η e − i ω j t + c . c . (60) \mathcal{V}_{j}=\frac{\mathrm{i}f A_{j}}{\omega_{j}h}\frac{\partial\phi_{j}}{\partial\eta}e^{-\mathrm{i}\omega_{j}t}+\mathrm{c.c.} \tag*{(60)} V j = ω j h i f A j ∂ η ∂ ϕ j e − i ω j t + c.c. (60)
The above-mentioned ansatz can also be used to study systems where the detuning Δ K s ∼ O ( 1 ) \Delta\mathcal{K}_{s}\sim O(1) Δ K s ∼ O ( 1 ) in the presence of a flat or slowly varying bathymetry. The functions ϕ j \phi_{j} ϕ j are same as the functions used in § 2 and are given by solving (19) . Here we only consider small amplitude topography whose wavenumber is comparable to the parent wave, i.e., ϵ h ≪ O ( 1 ) \epsilon_{h}\ll O(1) ϵ h ≪ O ( 1 ) and ϵ k ∼ O ( 1 ) \epsilon_{k}\sim O(1) ϵ k ∼ O ( 1 ) .
To study higher order interactions of a parent wave propagating in the presence of a small amplitude topography, we also consider the linear scattering of the parent wave. Note that the linear scattering of the parent wave on its own is not resonant (we do not consider topography wavenumbers which allow resonant Bragg scattering) and hence over a long distance has a negligible effect on the parent wave’s amplitude. However, even the non-resonant linear interaction of the parent wave with the topography, which leads to higher modes with ω 1 \omega_{1} ω 1 frequency, can significantly impact the growth of the superharmonic wave. To derive the linear scattering of the parent wave as it moves through a topography, we assume the streamfunction of the waves to be:
Ψ 1 = ∑ n = 1 M n A ( 1 , n ) ( x ) ϕ ( 1 , n ) ( η ; x ) e − i ω 1 t + c . c . (61) \Psi_{1}=\sum_{n=1}^{M_{n}} A_{(1,n)}(x)\phi_{(1,n)}(\eta;x)e^{-\mathrm{i}\omega_{1}t}+\mathrm{c.c.} \tag*{(61)} Ψ 1 = n = 1 ∑ M n A ( 1 , n ) ( x ) ϕ ( 1 , n ) ( η ; x ) e − i ω 1 t + c.c. (61)
where M n M_{n} M n is the maximum mode number after which the series is truncated, and ϕ ( 1 , n ) \phi_{(1,n)} ϕ ( 1 , n ) is the n n n –th eigenfunction of ω 1 \omega_{1} ω 1 frequency. The streamfunction ansatz (61) is substituted in (8) , and similar to § 2 , the linear terms of (8) is multiplied by ϕ ( 1 , n ) \phi_{(1,n)} ϕ ( 1 , n ) and integrated in the η \eta η direction. This leads to M n M_{n} M n ordinary differential equations, where the n n n –th differential equation is given by:
γ n ( 3 ) [ ∂ 2 A ( 1 , n ) ∂ x 2 + K ( 1 , n ) 2 A ( 1 , n ) h 2 ] = − ∑ m = 1 M n [ 2 γ ( m , n ) ( 5 ) + γ ( m , n ) ( 6 ) h 2 ( ∂ h ∂ x ) 2 − 2 γ ( m , n ) ( 7 ) h ∂ h ∂ x ] A ( 1 , m ) − ∑ m = 1 M n [ γ ( m , n ) ( 8 ) − γ ( m , n ) ( 5 ) h ( ∂ 2 h ∂ x 2 ) ] A ( 1 , m ) − ∑ m = 1 M n 2 [ γ ( m , n ) ( 4 ) − γ ( m , n ) ( 5 ) h ∂ h ∂ x ] ∂ A ( 1 , m ) ∂ x , (62) \gamma_{n}^{(3)}\left[\frac{\partial^{2}A_{(1,n)}}{\partial x^{2}}+\mathcal{K}_{(1,n)}^{2}\frac{A_{(1,n)}}{h^{2}}\right]
=
-\sum_{m=1}^{M_{n}}\left[\frac{2\gamma_{(m,n)}^{(5)}+\gamma_{(m,n)}^{(6)}}{h^{2}}\left(\frac{\partial h}{\partial x}\right)^{2}-\frac{2\gamma_{(m,n)}^{(7)}}{h}\frac{\partial h}{\partial x}\right]A_{(1,m)}
-\sum_{m=1}^{M_{n}}\left[\gamma_{(m,n)}^{(8)}-\frac{\gamma_{(m,n)}^{(5)}}{h}\left(\frac{\partial^{2}h}{\partial x^{2}}\right)\right]A_{(1,m)}
-\sum_{m=1}^{M_{n}}2\left[\gamma_{(m,n)}^{(4)}-\frac{\gamma_{(m,n)}^{(5)}}{h}\frac{\partial h}{\partial x}\right]\frac{\partial A_{(1,m)}}{\partial x}, \tag*{(62)} γ n ( 3 ) [ ∂ x 2 ∂ 2 A ( 1 , n ) + K ( 1 , n ) 2 h 2 A ( 1 , n ) ] = − m = 1 ∑ M n h 2 2 γ ( m , n ) ( 5 ) + γ ( m , n ) ( 6 ) ( ∂ x ∂ h ) 2 − h 2 γ ( m , n ) ( 7 ) ∂ x ∂ h A ( 1 , m ) − m = 1 ∑ M n γ ( m , n ) ( 8 ) − h γ ( m , n ) ( 5 ) ( ∂ x 2 ∂ 2 h ) A ( 1 , m ) − m = 1 ∑ M n 2 γ ( m , n ) ( 4 ) − h γ ( m , n ) ( 5 ) ∂ x ∂ h ∂ x ∂ A ( 1 , m ) , (62)
where K ( 1 , n ) \mathcal{K}_{(1,n)} K ( 1 , n ) is the corresponding eigenvalue of ϕ ( 1 , n ) \phi_{(1,n)} ϕ ( 1 , n ) . Moreover γ ( m , n ) ( ∗ ) \gamma_{(m,n)}^{(*)} γ ( m , n ) ( ∗ ) are evaluated using the expressions given in appendix § 9 . The above set of equations are similar to the equations derived in Lahaye & Llewellyn Smith [20 ] , except that we do not consider waves that travel in the direction opposite to the parent wave since they are assumed to be negligible. Now that we have the full wave spectrum with ω 1 \omega_{1} ω 1 frequency by solving (62) , we model the evolution of the superharmonic wave. For simplicity, the feedback to the parent wave is neglected, which is analogous to the pump-wave approximation used in § 2 .
The streamfunction of the superharmonic wave Ψ 3 \Psi_{3} Ψ 3 is substituted in (8) , and the linear terms are multiplied by ϕ 3 \phi_3 ϕ 3 and integrated in the η \eta η direction. This leads to:
L I N 3 ≡ [ ( γ 3 ( 3 ) ∂ 2 A 3 ∂ x 2 + κ 3 2 ( A 3 h 2 γ 3 ( 3 ) ) ) e − i ω 3 t + 2 [ γ 3 ( 5 ) h 2 ( ∂ h ∂ x ) 2 − γ 3 ( 7 ) h ∂ h ∂ x ] A 3 e − i ω 3 t + [ γ 3 ( 6 ) h 2 ( ∂ h ∂ x ) 2 − γ 3 ( 5 ) h ( ∂ 2 h ∂ x 2 ) + γ 3 ( 8 ) ] A 3 e − i ω 3 t + 2 ( γ 3 ( 4 ) − γ 3 ( 5 ) h ∂ h ∂ x ) ∂ A 3 ∂ x e − i ω 3 t ] . (63) \mathrm{LIN}_3 \equiv\left[\left(\gamma_3^{(3)}\frac{\partial^2\mathcal{A}_3}{\partial x^2}+\kappa_3^2\left(\frac{\mathcal{A}_3}{h^2}\gamma_3^{(3)}\right)\right)e^{-i\omega_3t}+2\left[\frac{\gamma_3^{(5)}}{h^2}\left(\frac{\partial h}{\partial x}\right)^2-\frac{\gamma_3^{(7)}}{h}\frac{\partial h}{\partial x}\right]\mathcal{A}_3e^{-i\omega_3t}
+\left[\frac{\gamma_3^{(6)}}{h^2}\left(\frac{\partial h}{\partial x}\right)^2-\frac{\gamma_3^{(5)}}{h}\left(\frac{\partial^2h}{\partial x^2}\right)+\gamma_3^{(8)}\right]\mathcal{A}_3e^{-i\omega_3t}+2\left(\gamma_3^{(4)}-\frac{\gamma_3^{(5)}}{h}\frac{\partial h}{\partial x}\right)\frac{\partial\mathcal{A}_3}{\partial x}e^{-i\omega_3t}\right]. \tag*{(63)} LIN 3 ≡ [ ( γ 3 ( 3 ) ∂ x 2 ∂ 2 A 3 + κ 3 2 ( h 2 A 3 γ 3 ( 3 ) ) ) e − i ω 3 t + 2 [ h 2 γ 3 ( 5 ) ( ∂ x ∂ h ) 2 − h γ 3 ( 7 ) ∂ x ∂ h ] A 3 e − i ω 3 t + [ h 2 γ 3 ( 6 ) ( ∂ x ∂ h ) 2 − h γ 3 ( 5 ) ( ∂ x 2 ∂ 2 h ) + γ 3 ( 8 ) ] A 3 e − i ω 3 t + 2 ( γ 3 ( 4 ) − h γ 3 ( 5 ) ∂ x ∂ h ) ∂ x ∂ A 3 e − i ω 3 t ] . (63)
L I N 3 \mathrm{LIN}_3 LIN 3 only contains linear terms, and models the propagation of superharmonic wave in the presence of a topography. Note that superharmonic wave cannot exchange energy with higher modes of ω 3 \omega_3 ω 3 . Now we move on to deriving the nonlinear terms which force the superharmonic wave. Since we are focusing on higher order interactions, all nonlinear terms (including terms containing x x x -direction derivatives of ϕ \phi ϕ and h h h ), which have the same angular frequency as the superharmonic wave, are retained. In the terrain following coordinates, this would however lead to a large number of terms that need to be evaluated. This issue can be circumvented following the procedure outlined below. The nonlinear terms in the terrain following coordinates are given by right hand side of (8) .
We assume that the superharmonic wave is forced nonlinearly by the ω 1 \omega_1 ω 1 spectrum. To model this, we substitute Ψ 1 \Psi_1 Ψ 1 , B 1 B_1 B 1 , V 1 \mathcal{V}_1 V 1 into the nonlinear terms of (8) . Note that we can obtain B 1 B_1 B 1 and V 1 \mathcal{V}_1 V 1 from (59) and (60) respectively. After the substitution, similar to the linear terms of wave-3, the nonlinear terms are multiplied by ϕ 3 \phi_3 ϕ 3 and integrated in η \eta η direction within the domain limits. The resultant expression obtained is as follows:
⟨ N L 3 ⟩ = ∫ − 1 0 ϕ 3 [ i ω 3 J { ( L x x + L η η ) Ψ 1 , Ψ 1 } + L x ( J { B 1 , Ψ 1 } ) − f L η ( J { V 1 , Ψ 1 } ) ] d η . (64) \langle\mathrm{NL}_3\rangle=\int_{-1}^{0}\phi_3\left[i\omega_3\mathcal{J}\left\{\left(L_{xx}+L_{\eta\eta}\right)\Psi_1,\Psi_1\right\}+L_x\left(\mathcal{J}\left\{B_1,\Psi_1\right\}\right)-fL_\eta\left(\mathcal{J}\left\{\mathcal{V}_1,\Psi_1\right\}\right)\right]\mathrm{d}\eta. \tag*{(64)} ⟨ NL 3 ⟩ = ∫ − 1 0 ϕ 3 [ i ω 3 J { ( L xx + L η η ) Ψ 1 , Ψ 1 } + L x ( J { B 1 , Ψ 1 } ) − f L η ( J { V 1 , Ψ 1 } ) ] d η . (64)
Therefore the final superharmonic wave equation can be written in a compact form:
L I N 3 = ⟨ N L 3 ⟩ . (65) \mathrm{LIN}_3=\langle\mathrm{NL}_3\rangle. \tag*{(65)} LIN 3 = ⟨ NL 3 ⟩ . (65)
In equations (63) and (65) , instead of splitting A j \mathcal{A}_j A j into a product of slowly varying amplitude and rapidly varying phase part, we simply solve the equations numerically by retaining A j \mathcal{A}_j A j as it is. This is mainly because, as mentioned above, the number of nonlinear terms would be significantly high in terrain following coordinates. For high ratios of f / ω 1 f/\omega_1 f / ω 1 (for example, north of critical latitude), the parent wave cannot resonantly self interact with the superharmonic wave in the presence of a flat bottom. However, a resonant higher order self-interaction can occur provided the topography has a wavenumber k b k_b k b such that:
k b = k 3 − 2 k 1 , (66) k_b=k_3-2k_1, \tag*{(66)} k b = k 3 − 2 k 1 , (66)
where k 3 k_3 k 3 is the wavenumber of the superharmonic wave and the k 1 k_1 k 1 is the wavenumber of the parent wave. In such scenarios, the daughter wave’s amplitude will consistently grow. However this being a higher order interaction, the growth rate of daughter wave (consequently, the decay of the parent wave) can be expected to be slower than a resonant self interaction.
To elucidate and validate the higher order self interaction process, we perform numerical simulations by solving the complete 2D Boussinesq equations and comparing the output with the results of the reduced order model derived in this section. We run three simulations where the parent and daughter waves’ frequencies are held fixed. They are denoted by Case-1, Case-2, and Case-3. For all the simulations, the parent wave frequency is ω 1 / N b = 0.2 \omega_1/N_b=0.2 ω 1 / N b = 0.2 , where ω 1 \omega_1 ω 1 is the semi-diurnal frequency, i.e. ω 1 = 1.4 × 10 − 4 s − 1 \omega_1=1.4\times10^{-4}\,\mathrm{s}^{-1} ω 1 = 1.4 × 1 0 − 4 s − 1 . Both the parent and daughter waves are mode-1 of their respective frequencies (ω 1 \omega_1 ω 1 and 2 ω 1 2\omega_1 2 ω 1 ).
These parameters would result in a significant detuning between the two waves at high f f f values. The bathymetry profile is given by:
h = − H + [ ϵ h H sin ( k b ( x − x c ) ) ] × 1 ( 1 + ( x − x c ) 32 / W T 32 ) , (67) h = -H + [\epsilon_h H \sin(k_b(x-x_c))] \times\frac{1}{(1+(x-x_c)^{32}/W_T^{32})}, \tag*{(67)} h = − H + [ ϵ h H sin ( k b ( x − x c ))] × ( 1 + ( x − x c ) 32 / W T 32 ) 1 , (67)
where ϵ h = 0.01 \epsilon_h = 0.01 ϵ h = 0.01 , H = 3000 m H = 3000\mathrm{m} H = 3000 m , domain length L = 540 H L = 540H L = 540 H , and x c = L / 2 x_c = L/2 x c = L /2 are held fixed across all three simulations. The stratification profile parameters, f f f value, incoming maximum velocity of the parent mode (u max u_{\max} u m a x ), and W T W_T W T for the three simulations are given in Table 1 . The results after solving the reduced order model and 2D Boussinesq equations for the above mentioned parameters are shown in Figure 11 . The 2D Boussinesq equations are solved using Dedalus. More details on the simulations are given in the end of § 7 . In all three sub-figures, the amplitude of the daughter wave is observed to be slowly increasing due to the higher order self-interaction. Moreover, the daughter wave’s amplitude also rapidly oscillates because of the non-resonant standard self-interaction process between the parent wave and the daughter wave. In the absence of a varying bathymetry, only the rapid non-resonant interaction would be present without any consistent growth in the daughter wave’s amplitude. Therefore we have shown that for scenarios where Bragg resonances are not resonant, higher order interaction might be a possible mechanism that can scatter the energy of the mode–1 internal wave.
Figure 11. Higher order self-interaction of mode-1 internal wave in the presence of a monochromatic bathymetry for three different cases: (a) Case-1, (b) Case-2, and (c) Case-3. The topography profile (not to scale) is shown for all three cases.
Table 1. The stratification profile parameters, Coriolis frequency, W T W_T W T , and the velocity amplitude of the three waves for Case-1, Case-2, and Case-3.
Numerical Validation
In this section, we provide numerical validations for the reduced–order equations (2.29a)–(2.29b) derived through multiple-scale analysis for two different cases. This is done by solving the 2D Boussinesq equations in terrain-following coordinates using an open–source, pseudo–spectral code Dedalus [4 ] . The above mentioned equations in primitive variables along with viscous and hyperviscous terms (the latter terms damping much smaller scales than the former) are given below:
∂ u ∂ t + L x ( P ) + u L x ( u ) + w L η ( u ) = ν L η η ( u ) + ( ν 6 z H 6 ∂ 6 u ∂ η 6 + ν 6 x ∂ 6 u ∂ x 6 ) (68) \frac{\partial u}{\partial t} + L_x(P) + uL_x(u) + wL_\eta(u) = \nu L_{\eta\eta}(u) + \left(\frac{\nu_{6z}}{H^6}\frac{\partial^6u}{\partial\eta^6} + \nu_{6x}\frac{\partial^6u}{\partial x^6}\right) \tag*{(68)} ∂ t ∂ u + L x ( P ) + u L x ( u ) + w L η ( u ) = ν L η η ( u ) + ( H 6 ν 6 z ∂ η 6 ∂ 6 u + ν 6 x ∂ x 6 ∂ 6 u ) (68)
∂ w ∂ t + L η ( P ) + u L x ( w ) + w L η ( w ) = ν L η η ( w ) + B (69) \frac{\partial w}{\partial t} + L_\eta(P) + uL_x(w) + wL_\eta(w) = \nu L_{\eta\eta}(w) + B \tag*{(69)} ∂ t ∂ w + L η ( P ) + u L x ( w ) + w L η ( w ) = ν L η η ( w ) + B (69)
∂ B ∂ t + N 2 w + u L x ( B ) + w L η ( B ) = ν L η η ( B ) + ( ν 6 z H 6 ∂ 6 B ∂ η 6 + ν 6 x ∂ 6 B ∂ x 6 ) (70) \frac{\partial B}{\partial t} + N^2w + uL_x(B) + wL_\eta(B) = \nu L_{\eta\eta}(B) + \left(\frac{\nu_{6z}}{H^6}\frac{\partial^6B}{\partial\eta^6} + \nu_{6x}\frac{\partial^6B}{\partial x^6}\right) \tag*{(70)} ∂ t ∂ B + N 2 w + u L x ( B ) + w L η ( B ) = ν L η η ( B ) + ( H 6 ν 6 z ∂ η 6 ∂ 6 B + ν 6 x ∂ x 6 ∂ 6 B ) (70)
L x ( u ) + L η ( w ) = 0. (71) L_x(u) + L_\eta(w) = 0. \tag*{(71)} L x ( u ) + L η ( w ) = 0. (71)
Here ( u , w ) = ( L η ( Ψ ) , − L x ( Ψ ) ) (u,w) = (L_\eta(\Psi),-L_x(\Psi)) ( u , w ) = ( L η ( Ψ ) , − L x ( Ψ )) , meaning that here the velocity field is defined in x x x –η \eta η instead of x x x –z z z coordinates. In all our simulations, ν = 10 − 5 m 2 s − 1 \nu= 10^{-5}\,\mathrm{m}^2\mathrm{s}^{-1} ν = 1 0 − 5 m 2 s − 1 , ν 6 x = 10 8 m 6 s − 1 \nu_{6x} = 10^8\,\mathrm{m}^6\mathrm{s}^{-1} ν 6 x = 1 0 8 m 6 s − 1 and ν 6 z = 81 m 6 s − 1 \nu_{6z} = 81\,\mathrm{m}^6\mathrm{s}^{-1} ν 6 z = 81 m 6 s − 1 . Equations (2.29a)–(2.29b) are solved using RK4 method for timestepping and second order accurate discretization scheme for the term ∂ a j / ∂ x \partial a_j/\partial x ∂ a j / ∂ x , where the scheme is forward or backward depending on the group speed direction of the particular wave. We estimate the validity of two different cases.
For Case 1, self interaction of a plane wave in the presence of a constant h h h is simulated. The parameters of the simulation are as follows: H = 3000 m H = 3000\,\mathrm{m} H = 3000 m and N b = 10 − 3 s − 1 N_b = 10^{-3}\,\mathrm{s}^{-1} N b = 1 0 − 3 s − 1 . The frequency of the parent wave is taken as ω 1 / N b = 0.447 \omega_1/N_b = 0.447 ω 1 / N b = 0.447 , while f = 0 f = 0 f = 0 is chosen. The stratification profile (4.1) is considered and the parameters used are:
Mode-3 of parent wave frequency (ω 1 \omega_1 ω 1 ) in this scenario resonantly self interacts with mode-2 of 2 ω 1 2\omega_1 2 ω 1 [36 ] . The parent wave streamfunction input (initial condition) to the full numerical simulation is given by:
Ψ = A 1 ϕ 1 sin ( k 1 x ) exp ( − ( x − x c ) 2 / W 1 2 ) , (72) \Psi= A_1\phi_1\sin(k_1x)\exp\left(-(x-x_c)^2/W_1^2\right), \tag*{(72)} Ψ = A 1 ϕ 1 sin ( k 1 x ) exp ( − ( x − x c ) 2 / W 1 2 ) , (72)
where ∣ A 1 ∣ = 0.06 |A_1| = 0.06 ∣ A 1 ∣ = 0.06 , W 1 → ∞ W_1 \to\infty W 1 → ∞ is chosen (note that x c x_c x c can be any value since W 1 → ∞ W_1 \to\infty W 1 → ∞ ). Ψ \Psi Ψ is used to obtain ( u , w ) (u,w) ( u , w ) . The numerical code is also initialised with the corresponding buoyancy frequency for the streamfunction given in (7.2).
For estimating the energy of parent and daughter waves from the numerical simulations, only the potential energy of the waves is considered. This is valid because when f = 0 f = 0 f = 0 , energy is equally partitioned between potential energy and kinetic energy. For evaluating the potential energy of the two waves, we take the Fourier transform of B B B in the x x x –direction. Then by simply isolating the k 3 k_3 k 3 and k 1 k_1 k 1 wavenumbers, the respective fields due to wave-3 and wave-1 can be obtained for all time. The resulting energy evolution of the waves for Case 1 is shown in Figure 12 (a). At the end of the simulation, the parent wave energy was observed to be 86% of the total energy in the Boussinesq equations simulation, while the reduced order model predicted that 85.5% of the total energy will be contained in the parent wave at the specified time interval. Moreover, at the end of the simulation, the daughter wave’s energy in Boussinesq equations simulation and the reduced order model are 13.0% and 14.4% respectively.
Figure 12. (a) The energy evolution of waves for Case-1 from reduced-order equations and numerical simulations. The superscript ( N ) (N) ( N ) denotes the results from numerical simulation of 2D Boussinesq equations. (b) The energy evolution of waves for Case-2 from reduced-order equations and numerical simulations in the time span t ∗ = 30 t^{*}=30 t ∗ = 30 to t ∗ ≈ 40 t^{*}\approx40 t ∗ ≈ 40 , where t ∗ ≡ ω 1 t / 2 π t^{*}\equiv\omega_{1}t/2\pi t ∗ ≡ ω 1 t /2 π . (c) Fourier transform of B B B at η = − 0.42 \eta=-0.42 η = − 0.42 and t ∗ = 40 t^{*}=40 t ∗ = 40 .
In Case 2, we consider the self interaction of a parent wave packet travelling in the presence of a slowly varying bathymetry. The parameters considered are as follows: H = 3000 m H=3000\mathrm{m} H = 3000 m and N b = 2.5 × 10 − 3 s − 1 N_b=2.5\times10^{-3}\mathrm{s}^{-1} N b = 2.5 × 1 0 − 3 s − 1 . The following stratification profile is used:
ω 1 / N b = 0.447 \omega_1/N_b=0.447 ω 1 / N b = 0.447 with f = 0 f=0 f = 0 is chosen. Mode-3 of ω 1 \omega_1 ω 1 resonantly self interacts with mode-2 of 2 ω 1 2\omega_1 2 ω 1 for h / H ∈ [ − 1 , − 0.8 ] h/H\in[-1,-0.8] h / H ∈ [ − 1 , − 0.8 ] . The bathymetry is given by:
h = − H + 0.1 H [ tanh ( ( x − x t 1 ) / W t 1 ) + tanh ( ( x t 2 − x ) / W t 2 ) ] , (73) h=-H+0.1H\left[\tanh\left((x-x_{t1})/W_{t1}\right)+\tanh\left((x_{t2}-x)/W_{t2}\right)\right], \tag*{(73)} h = − H + 0.1 H [ tanh ( ( x − x t 1 ) / W t 1 ) + tanh ( ( x t 2 − x ) / W t 2 ) ] , (73)
where W t 1 = 2.7 H W_{t1}=2.7H W t 1 = 2.7 H , W t 2 = W t 1 / 1.3 W_{t2}=W_{t1}/1.3 W t 2 = W t 1 /1.3 , x t 1 = 25 H x_{t1}=25H x t 1 = 25 H and x t 2 = 83 H x_{t2}=83H x t 2 = 83 H was considered, where 100 H 100H 100 H is the domain length in x x x -direction. The bathymetry shape can be visualised in Figure 13 .
Figure 13. Horizontal velocity plot (u u u ) from Case-2 simulation at (a) t ∗ = 0 t^* = 0 t ∗ = 0 and (b) t ∗ = 30 t^* = 30 t ∗ = 30 . The stratification profile shape used is also shown for visual purposes. A faint but clear mode-2 trail left by the mode-3 parent wave can be seen at (b). The shape of the stratification profile used in Case-2 is given by green curves in (b).
The parent wave streamfunction of the form (7.2) is used with ∣ A 1 ∣ = 0.022 |A_1|=0.022 ∣ A 1 ∣ = 0.022 , W 1 = 3.57 H W_1=3.57H W 1 = 3.57 H and x c = 10.2 H x_c=10.2H x c = 10.2 H is chosen. Here k 1 k_1 k 1 is evaluated at h = − H h=-H h = − H for the initial conditions. The x c x_c x c value is chosen such that the wave packet is just at the bottom of the ‘plateau-like’ topography at t = 0 t=0 t = 0 (as shown in Figure 13 (a)). The bathymetry is considered to be slowly varying so that the wave packet scattering by the bathymetry is negligible. Here the same procedure of energy evaluation as Case 1 is followed, except the energy is evaluated from t ∗ = 30 t^{*}=30 t ∗ = 30 , when the entire energy of both wave packets is almost confined to the top of the plateau region (where h = − 0.8 H h=-0.8H h = − 0.8 H ). This makes the energy evaluation straightforward. We only consider B B B in the range x / H ∈ [ 33 , 68 ] x/H\in[33,68] x / H ∈ [ 33 , 68 ] (where most of the energy is contained), and then again perform Fourier transform to separate the energy of the daughter and parent wave packets.
In Case-2, since wave packets are considered, the Fourier transform of B B B would not have a sharp peak at k 1 k_1 k 1 and k 3 k_3 k 3 . Instead, a smoother peak in k k k -space would be produced as shown in Figure 12 (c). We define a nondimensional wavenumber k ~ \tilde{k} k ~ as k ~ ≡ k / k 1 \tilde{k}\equiv k/k_1 k ~ ≡ k / k 1 . For evaluating the energy of both wave packets, amplitude ( ∣ a k ~ ∣ ) (|a_{\tilde{k}}|) ( ∣ a k ~ ∣ ) in a finite range of k ~ \tilde{k} k ~ is considered. The energy contained in (k ~ ∈ [ 0.6 , 1.4 ] \tilde{k} \in[0.6,1.4] k ~ ∈ [ 0.6 , 1.4 ] ) is considered as the energy of the parent wave packet, while the energy in (k ~ ∈ [ 1.6 , 2.4 ] \tilde{k} \in[1.6,2.4] k ~ ∈ [ 1.6 , 2.4 ] ) is considered as the daughter wave’s energy. For example, the k ~ \tilde{k} k ~ range considered for the parent and daughter waves are highlighted in Figure 12 (c) using colored dotted lines for a specific η \eta η and t ∗ t^* t ∗ . For the parent wave, ∣ a k ~ ∣ |a_{\tilde{k}}| ∣ a k ~ ∣ between the blue dotted lines is considered. Similarly for evaluating the energy of the daughter wave packet, we consider the amplitude (∣ a k ~ ∣ |a_{\tilde{k}}| ∣ a k ~ ∣ ) between the red dotted lines. The energy evolution of the wave packets are shown in 12(b). The Parent wave packet energy in Boussinesq equations simulation and the reduced order model was observed to be 88.1% and 88.9% respectively at the end of the simulations. At the same time, the daughter wave’s energy in Boussinesq equations simulation and the reduced order model are 8.95% and 10.6% respectively.
To obtain the numerical results in Figure 11 , equations (7.1a)–(7.1d) along with equation for meridional velocity (v v v ) were solved. Vertical hyperviscous term was not used, while a different horizontal hyperviscous term was used: ν 12 x ∂ 12 / ∂ x 12 \nu_{12x}\partial^{12}/\partial x^{12} ν 12 x ∂ 12 / ∂ x 12 , where ν 12 x = 1.4 × 10 25 m 12 s − 1 \nu_{12x} = 1.4 \times10^{25}\,\mathrm{m}^{12}\mathrm{s}^{-1} ν 12 x = 1.4 × 1 0 25 m 12 s − 1 . The kinematic viscosity was chosen to be ν = 10 − 3 m 2 s − 1 \nu= 10^{-3}\,\mathrm{m}^2\mathrm{s}^{-1} ν = 1 0 − 3 m 2 s − 1 . The primary wave was forced by using a forcing function in the u u u –momentum equation, which sends a constant amplitude mode-1 wave train onto the small amplitude topography.
Summary and Conclusion
Weakly nonlinear wave-wave interactions is one of the mechanisms through which internal gravity waves’ energy cascade from large length scales (hundreds of kilometers) to small scales (centimeters to meters). At small length scales, internal waves can give rise to convective or shear instabilities [19 ] and cause mixing, thus resulting in increased diffusion in oceans. The 2D Boussinesq equations are written in terrain following coordinates (x x x –η \eta η ). Using multiple-scale analysis, we derive the amplitude evolution equations for internal gravity waves undergoing weakly nonlinear wave-wave interactions in the presence of varying density stratification (resembling that of actual oceanic scenarios) as well as mild slope bathymetry in a vertically bounded domain. If the stratification varies with z z z in the x x x –z z z coordinates, then it becomes a function of both x x x and η \eta η in the x x x –η \eta η coordinates when bathymetry, h h h , varies with x x x . In other words, the effective stratification profile varies with the ocean depth. Both triads and self-interactions are studied, and both pure resonant conditions as well as systems with wavenumber detuning are analyzed. The main results of this paper are given in a brief format in Figure 14 .
Figure 14. A summary diagram shows how different factors such as detuning, nonlinear coefficients can vary for different classes of interaction in the presence of non-uniform stratification. It also provides a brief picture of the higher order self interaction studied in §6.
In the presence of uniform stratification, we show that the horizontal wavenumber triad condition, given by k ( 1 , a ) + k ( 2 , b ) + k ( 3 , c ) = 0 k_{(1,a)} + k_{(2,b)} + k_{(3,c)} = 0 k ( 1 , a ) + k ( 2 , b ) + k ( 3 , c ) = 0 , is not violated due to changes in h h h . Here ( a , b , c ) (a,b,c) ( a , b , c ) are the modenumbers of waves 1, 2, and 3 respectively. Moreover, in the presence of uniform stratification, the nonlinear coupling coefficients are inversely proportional to the square of the fluid depth (∝ 1 / h 2 \propto1/h^2 ∝ 1/ h 2 ).
For non-uniform stratifications, we define two classes of interaction for both triads and self-interactions. Class-1 involves weakly nonlinear interactions of waves that do not have the same mode number. Class-2 is a special situation that involves interactions of waves with the same modenumber, i.e. a = b = c a=b=c a = b = c . Class-2 triad interactions can exist only in the parameter regime of f ≪ ω j ≪ N f \ll\omega_j \ll N f ≪ ω j ≪ N . Moreover, in the same parameter regime, near-resonant Class-2 self-interactions can exist with very low detuning even as h h h is varied. This is because the wavenumbers involved in a self-interaction, change in the same way as h h h changes. For Class-1 interactions, detuning may be induced in triads and self-interactions if the waves interact in a region of varying h h h . This is because in a vertically bounded domain, the horizontal wavenumbers are not only a function of h h h but is also a function of the modenumber. Moreover, the functional dependence of the wavenumber on h h h may change as the modenumber changes. Therefore, in a weakly nonlinear interaction where different modenumbers are involved, there is no constraint for the wavenumbers to satisfy the triad condition in a given range of h h h .
The variation of the growth rate of the daughter waves in both triadic- and self-interactions is studied when h h h is varied. For both Class-1 and Class-2 self-interactions, it is observed that small changes in h h h may result in large changes in the growth rate of the daughter waves. This characteristic is especially observed for Class-1 self-interactions. Variation of growth rates with h h h is studied for triads of a mode-1 parent wave in the presence of non-uniform stratification. Triads were identified such that the daughter waves can be up to mode-50. For relatively small changes in h h h , the growth rates can vary significantly for triads that involve only lower modes. Moreover, the most unstable daughter wave combination for the same parent wave can also change for relatively small changes in h h h . Unlike uniform stratification, in non-uniform stratification, the growth rates do not have a monotonic behavior with h h h . This was observed for both triadic- and self-interactions.
Reduced-order equations for higher-order self-interactions of an internal wave in the presence of a small amplitude, monochromatic topography is also derived. In the higher-order self-interaction process, the small amplitude topography behaves as a zero frequency wave. It is shown that such higher-order interactions can cause resonant growth of the superharmonic wave. Such higher-order interactions can play a crucial role in the decay of the mode-1 internal wave at latitudes greater than 28.9°. This is because sum-type triad interactions are not possible (Olbers et al. (2020)) and a mode-1 internal wave cannot resonantly self-interact for high values of f f f (Wunsch (2017)).
Declaration of interests
The authors report no conflict of interest.
Appendix A
Expressions for γ j \gamma_j γ j , Γ j \Gamma_j Γ j and N L ( ∗ , j ) \mathrm{NL}_{(*,j)} NL ( ∗ , j )
The expressions for γ ( j , i ) ( n ) \gamma^{(n)}_{(j,i)} γ ( j , i ) ( n ) , which are used in § 2.2.1 and § 6 , are provided below:
γ ( j , i ) ( 1 ) = ∫ − 1 0 [ ϕ j ] ϕ i d η , γ ( j , i ) ( 2 ) = ∫ − 1 0 [ ∂ 2 ϕ j ∂ η 2 ] ϕ i d η , γ ( j , i ) ( 3 ) = ∫ − 1 0 [ ( N 2 − ω j 2 ) ϕ j ] ϕ i d η , γ ( j , i ) ( 4 ) = ∫ − 1 0 [ ( N 2 − ω j 2 ) ∂ ϕ j ∂ x ] ϕ i d η , γ ( j , i ) ( 5 ) = ∫ − 1 0 [ η ( N 2 − ω j 2 ) ∂ ϕ j ∂ η ] ϕ i d η , γ ( j , i ) ( 6 ) = ∫ − 1 0 [ η 2 ( N 2 − ω j 2 ) ∂ 2 ϕ j ∂ η 2 ] ϕ i d η , γ ( j , i ) ( 7 ) = ∫ − 1 0 [ η ( N 2 − ω j 2 ) ∂ 2 ϕ j ∂ x ∂ η ] ϕ i d η , γ ( j , i ) ( 8 ) = ∫ − 1 0 [ ( N 2 − ω j 2 ) ∂ 2 ϕ j ∂ x 2 ] ϕ i d η . (74) \begin{aligned}
\gamma^{(1)}_{(j,i)} &= \int_{-1}^{0} [\phi_j]\phi_i\,\mathrm{d}\eta, &
\gamma^{(2)}_{(j,i)} &= \int_{-1}^{0} \left[\frac{\partial^2\phi_j}{\partial\eta^2}\right]\phi_i\,\mathrm{d}\eta, \\
\gamma^{(3)}_{(j,i)} &= \int_{-1}^{0} \left[(N^2-\omega_j^2)\phi_j\right]\phi_i\,\mathrm{d}\eta, &
\gamma^{(4)}_{(j,i)} &= \int_{-1}^{0} \left[(N^2-\omega_j^2)\frac{\partial\phi_j}{\partial x}\right]\phi_i\,\mathrm{d}\eta, \\
\gamma^{(5)}_{(j,i)} &= \int_{-1}^{0} \left[\eta(N^2-\omega_j^2)\frac{\partial\phi_j}{\partial\eta}\right]\phi_i\,\mathrm{d}\eta, &
\gamma^{(6)}_{(j,i)} &= \int_{-1}^{0} \left[\eta^2(N^2-\omega_j^2)\frac{\partial^2\phi_j}{\partial\eta^2}\right]\phi_i\,\mathrm{d}\eta, \\
\gamma^{(7)}_{(j,i)} &= \int_{-1}^{0} \left[\eta(N^2-\omega_j^2)\frac{\partial^2\phi_j}{\partial x\partial\eta}\right]\phi_i\,\mathrm{d}\eta, &
\gamma^{(8)}_{(j,i)} &= \int_{-1}^{0} \left[(N^2-\omega_j^2)\frac{\partial^2\phi_j}{\partial x^2}\right]\phi_i\,\mathrm{d}\eta. \tag*{(74)}
\end{aligned} γ ( j , i ) ( 1 ) γ ( j , i ) ( 3 ) γ ( j , i ) ( 5 ) γ ( j , i ) ( 7 ) = ∫ − 1 0 [ ϕ j ] ϕ i d η , = ∫ − 1 0 [ ( N 2 − ω j 2 ) ϕ j ] ϕ i d η , = ∫ − 1 0 [ η ( N 2 − ω j 2 ) ∂ η ∂ ϕ j ] ϕ i d η , = ∫ − 1 0 [ η ( N 2 − ω j 2 ) ∂ x ∂ η ∂ 2 ϕ j ] ϕ i d η , γ ( j , i ) ( 2 ) γ ( j , i ) ( 4 ) γ ( j , i ) ( 6 ) γ ( j , i ) ( 8 ) = ∫ − 1 0 [ ∂ η 2 ∂ 2 ϕ j ] ϕ i d η , = ∫ − 1 0 [ ( N 2 − ω j 2 ) ∂ x ∂ ϕ j ] ϕ i d η , = ∫ − 1 0 [ η 2 ( N 2 − ω j 2 ) ∂ η 2 ∂ 2 ϕ j ] ϕ i d η , = ∫ − 1 0 [ ( N 2 − ω j 2 ) ∂ x 2 ∂ 2 ϕ j ] ϕ i d η . (74)
Throughout the paper, γ ( j , j ) \gamma_{(j,j)} γ ( j , j ) is simply denoted by γ j \gamma_j γ j for convenience.
The expressions for N L ( ∗ , j ) \mathrm{NL}_{(*,j)} NL ( ∗ , j ) used in amplitude evolution equations (2.24a)–(2.24c) in §2.2.1 are provided below. Note that N L ( ∗ , j ) \mathrm{NL}_{(*,j)} NL ( ∗ , j ) is used in (2.25b).
N L ( ψ , 1 ) = ω 1 h 4 [ K 3 ( ζ 3 ω 3 2 Γ 2 ( 1 ) − ζ 3 Γ 2 ( 2 ) − Γ 3 ( 3 ) ) − K 2 ( ζ 2 ω 2 2 Γ 3 ( 1 ) − ζ 2 Γ 3 ( 2 ) − Γ 2 ( 3 ) ) ] + ω 1 h 4 [ ( K 2 2 − K 3 2 ) ( K 2 Γ 3 ( 1 ) + K 3 Γ 1 ( 1 ) ) ] , N L ( ψ , 2 ) = ω 2 h 4 [ K 3 ( ζ 3 ω 3 2 Γ 1 ( 1 ) − ζ 3 Γ 1 ( 2 ) − Γ 3 ( 3 ) ) − K 1 ( ζ 1 ω 1 2 Γ 3 ( 1 ) − ζ 1 Γ 3 ( 2 ) − Γ 1 ( 3 ) ) ] + ω 2 h 4 [ ( K 1 2 − K 3 2 ) ( K 1 Γ 3 ( 1 ) + K 3 Γ 1 ( 1 ) ) ] , N L ( ψ , 3 ) = ω 3 h 4 [ K 1 ( ζ 2 ω 2 2 Γ 1 ( 1 ) − ζ 2 Γ 1 ( 2 ) − Γ 2 ( 3 ) ) + K 2 ( ζ 1 ω 1 2 Γ 2 ( 1 ) − ζ 1 Γ 2 ( 2 ) − Γ 1 ( 3 ) ) ] + ω 3 h 4 [ ( K 2 2 − K 1 2 ) ( K 1 Γ 2 ( 1 ) − K 2 Γ 1 ( 1 ) ) ] . (75) \begin{aligned}
\mathrm{NL}_{(\psi,1)}={}&\frac{\omega_1}{h^4}\left[\mathcal{K}_3\left(\zeta_3\omega_3^2\Gamma_2^{(1)}-\zeta_3\Gamma_2^{(2)}-\Gamma_3^{(3)}\right)-\mathcal{K}_2\left(\zeta_2\omega_2^2\Gamma_3^{(1)}-\zeta_2\Gamma_3^{(2)}-\Gamma_2^{(3)}\right)\right]\\
&+\frac{\omega_1}{h^4}\left[(\mathcal{K}_2^2-\mathcal{K}_3^2)\left(\mathcal{K}_2\Gamma_3^{(1)}+\mathcal{K}_3\Gamma_1^{(1)}\right)\right],\\
\mathrm{NL}_{(\psi,2)}={}&\frac{\omega_2}{h^4}\left[\mathcal{K}_3\left(\zeta_3\omega_3^2\Gamma_1^{(1)}-\zeta_3\Gamma_1^{(2)}-\Gamma_3^{(3)}\right)-\mathcal{K}_1\left(\zeta_1\omega_1^2\Gamma_3^{(1)}-\zeta_1\Gamma_3^{(2)}-\Gamma_1^{(3)}\right)\right]\\
&+\frac{\omega_2}{h^4}\left[(\mathcal{K}_1^2-\mathcal{K}_3^2)\left(\mathcal{K}_1\Gamma_3^{(1)}+\mathcal{K}_3\Gamma_1^{(1)}\right)\right],\\
\mathrm{NL}_{(\psi,3)}={}&\frac{\omega_3}{h^4}\left[\mathcal{K}_1\left(\zeta_2\omega_2^2\Gamma_1^{(1)}-\zeta_2\Gamma_1^{(2)}-\Gamma_2^{(3)}\right)+\mathcal{K}_2\left(\zeta_1\omega_1^2\Gamma_2^{(1)}-\zeta_1\Gamma_2^{(2)}-\Gamma_1^{(3)}\right)\right]\\
&+\frac{\omega_3}{h^4}\left[(\mathcal{K}_2^2-\mathcal{K}_1^2)\left(\mathcal{K}_1\Gamma_2^{(1)}-\mathcal{K}_2\Gamma_1^{(1)}\right)\right]. \tag*{(75)}
\end{aligned} NL ( ψ , 1 ) = NL ( ψ , 2 ) = NL ( ψ , 3 ) = h 4 ω 1 [ K 3 ( ζ 3 ω 3 2 Γ 2 ( 1 ) − ζ 3 Γ 2 ( 2 ) − Γ 3 ( 3 ) ) − K 2 ( ζ 2 ω 2 2 Γ 3 ( 1 ) − ζ 2 Γ 3 ( 2 ) − Γ 2 ( 3 ) ) ] + h 4 ω 1 [ ( K 2 2 − K 3 2 ) ( K 2 Γ 3 ( 1 ) + K 3 Γ 1 ( 1 ) ) ] , h 4 ω 2 [ K 3 ( ζ 3 ω 3 2 Γ 1 ( 1 ) − ζ 3 Γ 1 ( 2 ) − Γ 3 ( 3 ) ) − K 1 ( ζ 1 ω 1 2 Γ 3 ( 1 ) − ζ 1 Γ 3 ( 2 ) − Γ 1 ( 3 ) ) ] + h 4 ω 2 [ ( K 1 2 − K 3 2 ) ( K 1 Γ 3 ( 1 ) + K 3 Γ 1 ( 1 ) ) ] , h 4 ω 3 [ K 1 ( ζ 2 ω 2 2 Γ 1 ( 1 ) − ζ 2 Γ 1 ( 2 ) − Γ 2 ( 3 ) ) + K 2 ( ζ 1 ω 1 2 Γ 2 ( 1 ) − ζ 1 Γ 2 ( 2 ) − Γ 1 ( 3 ) ) ] + h 4 ω 3 [ ( K 2 2 − K 1 2 ) ( K 1 Γ 2 ( 1 ) − K 2 Γ 1 ( 1 ) ) ] . (75)
N L ( B , 1 ) = ( K 3 − K 2 ) h 4 [ K 2 K 3 ( 1 ω 2 − 1 ω 3 ) Γ ( 4 ) + ( K 3 ω 3 − K 2 ω 2 ) ( K 3 Γ 2 ( 2 ) − K 2 Γ 3 ( 2 ) ) ] , N L ( B , 2 ) = ( K 3 − K 1 ) h 4 [ K 1 K 3 ( 1 ω 1 − 1 ω 3 ) Γ ( 4 ) + ( K 3 ω 3 − K 1 ω 1 ) ( K 3 Γ 1 ( 2 ) − K 1 Γ 3 ( 2 ) ) ] , N L ( B , 3 ) = ( K 1 + K 2 ) h 4 [ K 1 K 2 ( 1 ω 1 + 1 ω 2 ) Γ ( 4 ) + ( K 2 ω 2 − K 1 ω 1 ) ( K 2 Γ 2 ( 2 ) − K 1 Γ 1 ( 2 ) ) ] . (76) \begin{aligned}
\mathrm{NL}_{(B,1)}={}&\frac{(\mathcal{K}_3-\mathcal{K}_2)}{h^4}\left[\mathcal{K}_2\mathcal{K}_3\left(\frac{1}{\omega_2}-\frac{1}{\omega_3}\right)\Gamma^{(4)}+\left(\frac{\mathcal{K}_3}{\omega_3}-\frac{\mathcal{K}_2}{\omega_2}\right)\left(\mathcal{K}_3\Gamma_2^{(2)}-\mathcal{K}_2\Gamma_3^{(2)}\right)\right],\\
\mathrm{NL}_{(B,2)}={}&\frac{(\mathcal{K}_3-\mathcal{K}_1)}{h^4}\left[\mathcal{K}_1\mathcal{K}_3\left(\frac{1}{\omega_1}-\frac{1}{\omega_3}\right)\Gamma^{(4)}+\left(\frac{\mathcal{K}_3}{\omega_3}-\frac{\mathcal{K}_1}{\omega_1}\right)\left(\mathcal{K}_3\Gamma_1^{(2)}-\mathcal{K}_1\Gamma_3^{(2)}\right)\right],\\
\mathrm{NL}_{(B,3)}={}&\frac{(\mathcal{K}_1+\mathcal{K}_2)}{h^4}\left[\mathcal{K}_1\mathcal{K}_2\left(\frac{1}{\omega_1}+\frac{1}{\omega_2}\right)\Gamma^{(4)}+\left(\frac{\mathcal{K}_2}{\omega_2}-\frac{\mathcal{K}_1}{\omega_1}\right)\left(\mathcal{K}_2\Gamma_2^{(2)}-\mathcal{K}_1\Gamma_1^{(2)}\right)\right]. \tag*{(76)}
\end{aligned} NL ( B , 1 ) = NL ( B , 2 ) = NL ( B , 3 ) = h 4 ( K 3 − K 2 ) [ K 2 K 3 ( ω 2 1 − ω 3 1 ) Γ ( 4 ) + ( ω 3 K 3 − ω 2 K 2 ) ( K 3 Γ 2 ( 2 ) − K 2 Γ 3 ( 2 ) ) ] , h 4 ( K 3 − K 1 ) [ K 1 K 3 ( ω 1 1 − ω 3 1 ) Γ ( 4 ) + ( ω 3 K 3 − ω 1 K 1 ) ( K 3 Γ 1 ( 2 ) − K 1 Γ 3 ( 2 ) ) ] , h 4 ( K 1 + K 2 ) [ K 1 K 2 ( ω 1 1 + ω 2 1 ) Γ ( 4 ) + ( ω 2 K 2 − ω 1 K 1 ) ( K 2 Γ 2 ( 2 ) − K 1 Γ 1 ( 2 ) ) ] . (76)
N L ( ν , 1 ) = f 2 h 4 [ ( 1 ω 3 + 1 ω 2 ) ( K 2 + K 3 ) ( ζ 3 ω 3 2 Γ 2 ( 1 ) − ζ 3 Γ 2 ( 2 ) + ζ 2 ω 2 2 Γ 3 ( 1 ) − ζ 2 Γ 3 ( 2 ) ) ] + f 2 h 4 [ ( K 2 ω 3 + K 3 ω 2 ) ( Γ 2 ( 3 ) + Γ 3 ( 3 ) ) ] , N L ( ν , 2 ) = f 2 h 4 [ ( 1 ω 3 + 1 ω 1 ) ( K 1 + K 3 ) ( ζ 3 ω 3 2 Γ 1 ( 1 ) − ζ 3 Γ 1 ( 2 ) + ζ 1 ω 1 2 Γ 3 ( 1 ) − ζ 1 Γ 3 ( 2 ) ) ] + f 2 h 4 [ ( K 1 ω 3 + K 3 ω 1 ) ( Γ 1 ( 3 ) + Γ 3 ( 3 ) ) ] , N L ( ν , 3 ) = f 2 h 4 [ ( 1 ω 1 − 1 ω 2 ) ( K 1 − K 2 ) ( ζ 1 ω 1 2 Γ 2 ( 1 ) − ζ 1 Γ 2 ( 2 ) + ζ 2 ω 2 2 Γ 1 ( 1 ) − ζ 2 Γ 1 ( 2 ) ) ] − f 2 h 4 [ ( K 2 ω 1 + K 1 ω 2 ) ( Γ 2 ( 3 ) + Γ 1 ( 3 ) ) ] . (77) \begin{aligned}
\mathrm{NL}_{(\nu,1)}={}&\frac{f^2}{h^4}\left[\left(\frac{1}{\omega_3}+\frac{1}{\omega_2}\right)(\mathcal{K}_2+\mathcal{K}_3)\left(\zeta_3\omega_3^2\Gamma_2^{(1)}-\zeta_3\Gamma_2^{(2)}+\zeta_2\omega_2^2\Gamma_3^{(1)}-\zeta_2\Gamma_3^{(2)}\right)\right]\\
&+\frac{f^2}{h^4}\left[\left(\frac{\mathcal{K}_2}{\omega_3}+\frac{\mathcal{K}_3}{\omega_2}\right)\left(\Gamma_2^{(3)}+\Gamma_3^{(3)}\right)\right],\\
\mathrm{NL}_{(\nu,2)}={}&\frac{f^2}{h^4}\left[\left(\frac{1}{\omega_3}+\frac{1}{\omega_1}\right)(\mathcal{K}_1+\mathcal{K}_3)\left(\zeta_3\omega_3^2\Gamma_1^{(1)}-\zeta_3\Gamma_1^{(2)}+\zeta_1\omega_1^2\Gamma_3^{(1)}-\zeta_1\Gamma_3^{(2)}\right)\right]\\
&+\frac{f^2}{h^4}\left[\left(\frac{\mathcal{K}_1}{\omega_3}+\frac{\mathcal{K}_3}{\omega_1}\right)\left(\Gamma_1^{(3)}+\Gamma_3^{(3)}\right)\right],\\
\mathrm{NL}_{(\nu,3)}={}&\frac{f^2}{h^4}\left[\left(\frac{1}{\omega_1}-\frac{1}{\omega_2}\right)(\mathcal{K}_1-\mathcal{K}_2)\left(\zeta_1\omega_1^2\Gamma_2^{(1)}-\zeta_1\Gamma_2^{(2)}+\zeta_2\omega_2^2\Gamma_1^{(1)}-\zeta_2\Gamma_1^{(2)}\right)\right]\\
&-\frac{f^2}{h^4}\left[\left(\frac{\mathcal{K}_2}{\omega_1}+\frac{\mathcal{K}_1}{\omega_2}\right)\left(\Gamma_2^{(3)}+\Gamma_1^{(3)}\right)\right]. \tag*{(77)}
\end{aligned} NL ( ν , 1 ) = NL ( ν , 2 ) = NL ( ν , 3 ) = h 4 f 2 [ ( ω 3 1 + ω 2 1 ) ( K 2 + K 3 ) ( ζ 3 ω 3 2 Γ 2 ( 1 ) − ζ 3 Γ 2 ( 2 ) + ζ 2 ω 2 2 Γ 3 ( 1 ) − ζ 2 Γ 3 ( 2 ) ) ] + h 4 f 2 [ ( ω 3 K 2 + ω 2 K 3 ) ( Γ 2 ( 3 ) + Γ 3 ( 3 ) ) ] , h 4 f 2 [ ( ω 3 1 + ω 1 1 ) ( K 1 + K 3 ) ( ζ 3 ω 3 2 Γ 1 ( 1 ) − ζ 3 Γ 1 ( 2 ) + ζ 1 ω 1 2 Γ 3 ( 1 ) − ζ 1 Γ 3 ( 2 ) ) ] + h 4 f 2 [ ( ω 3 K 1 + ω 1 K 3 ) ( Γ 1 ( 3 ) + Γ 3 ( 3 ) ) ] , h 4 f 2 [ ( ω 1 1 − ω 2 1 ) ( K 1 − K 2 ) ( ζ 1 ω 1 2 Γ 2 ( 1 ) − ζ 1 Γ 2 ( 2 ) + ζ 2 ω 2 2 Γ 1 ( 1 ) − ζ 2 Γ 1 ( 2 ) ) ] − h 4 f 2 [ ( ω 1 K 2 + ω 2 K 1 ) ( Γ 2 ( 3 ) + Γ 1 ( 3 ) ) ] . (77)
Moreover, Γ j ( n ) \Gamma_j^{(n)} Γ j ( n ) are defined as follows:
Γ 1 ( 1 ) = ∫ − 1 0 ϕ 2 ϕ 3 ∂ ϕ 1 ∂ η d η , Γ 2 ( 1 ) = ∫ − 1 0 ϕ 1 ϕ 3 ∂ ϕ 2 ∂ η d η , Γ 3 ( 1 ) = ∫ − 1 0 ϕ 2 ϕ 1 ∂ ϕ 3 ∂ η d η , \Gamma_1^{(1)}=\int_{-1}^{0}\phi_2\phi_3\frac{\partial\phi_1}{\partial\eta}\,d\eta,\qquad
\Gamma_2^{(1)}=\int_{-1}^{0}\phi_1\phi_3\frac{\partial\phi_2}{\partial\eta}\,d\eta,\qquad
\Gamma_3^{(1)}=\int_{-1}^{0}\phi_2\phi_1\frac{\partial\phi_3}{\partial\eta}\,d\eta, Γ 1 ( 1 ) = ∫ − 1 0 ϕ 2 ϕ 3 ∂ η ∂ ϕ 1 d η , Γ 2 ( 1 ) = ∫ − 1 0 ϕ 1 ϕ 3 ∂ η ∂ ϕ 2 d η , Γ 3 ( 1 ) = ∫ − 1 0 ϕ 2 ϕ 1 ∂ η ∂ ϕ 3 d η ,
Γ 1 ( 2 ) = ∫ − 1 0 N 2 ϕ 2 ϕ 3 ∂ ϕ 1 ∂ η d η , Γ 2 ( 2 ) = ∫ − 1 0 N 2 ϕ 1 ϕ 3 ∂ ϕ 2 ∂ η d η , Γ 3 ( 2 ) = ∫ − 1 0 N 2 ϕ 2 ϕ 1 ∂ ϕ 3 ∂ η d η , \Gamma_1^{(2)}=\int_{-1}^{0}N^2\phi_2\phi_3\frac{\partial\phi_1}{\partial\eta}\,d\eta,\qquad
\Gamma_2^{(2)}=\int_{-1}^{0}N^2\phi_1\phi_3\frac{\partial\phi_2}{\partial\eta}\,d\eta,\qquad
\Gamma_3^{(2)}=\int_{-1}^{0}N^2\phi_2\phi_1\frac{\partial\phi_3}{\partial\eta}\,d\eta, Γ 1 ( 2 ) = ∫ − 1 0 N 2 ϕ 2 ϕ 3 ∂ η ∂ ϕ 1 d η , Γ 2 ( 2 ) = ∫ − 1 0 N 2 ϕ 1 ϕ 3 ∂ η ∂ ϕ 2 d η , Γ 3 ( 2 ) = ∫ − 1 0 N 2 ϕ 2 ϕ 1 ∂ η ∂ ϕ 3 d η ,
Γ 1 ( 3 ) = ∫ − 1 0 ϕ 2 ϕ 3 ∂ 3 ϕ 1 ∂ η 3 d η , Γ 2 ( 3 ) = ∫ − 1 0 ϕ 1 ϕ 3 ∂ 3 ϕ 2 ∂ η 3 d η , Γ 3 ( 3 ) = ∫ − 1 0 ϕ 2 ϕ 1 ∂ 3 ϕ 3 ∂ η 3 d η , \Gamma_1^{(3)}=\int_{-1}^{0}\phi_2\phi_3\frac{\partial^3\phi_1}{\partial\eta^3}\,d\eta,\qquad
\Gamma_2^{(3)}=\int_{-1}^{0}\phi_1\phi_3\frac{\partial^3\phi_2}{\partial\eta^3}\,d\eta,\qquad
\Gamma_3^{(3)}=\int_{-1}^{0}\phi_2\phi_1\frac{\partial^3\phi_3}{\partial\eta^3}\,d\eta, Γ 1 ( 3 ) = ∫ − 1 0 ϕ 2 ϕ 3 ∂ η 3 ∂ 3 ϕ 1 d η , Γ 2 ( 3 ) = ∫ − 1 0 ϕ 1 ϕ 3 ∂ η 3 ∂ 3 ϕ 2 d η , Γ 3 ( 3 ) = ∫ − 1 0 ϕ 2 ϕ 1 ∂ η 3 ∂ 3 ϕ 3 d η ,
Γ ( 4 ) = ∫ − 1 0 ∂ N 2 ∂ η ϕ 1 ϕ 2 ϕ 3 d η . (78) \Gamma^{(4)}=\int_{-1}^{0}\frac{\partial N^2}{\partial\eta}\phi_1\phi_2\phi_3\,d\eta. \tag*{(78)} Γ ( 4 ) = ∫ − 1 0 ∂ η ∂ N 2 ϕ 1 ϕ 2 ϕ 3 d η . (78)
Appendix B
Here we perform a scaling analysis for all the terms appearing in (27) . Equation (27) is chosen here so that scaling analysis can be also done for the different terms that compose the β j \beta_j β j function (28) . Integrals (γ j \gamma_j γ j ) in (27) (γ j \gamma_j γ j expressions are given in (A1)) cannot be analytically simplified for non-uniform stratification profiles. Hence, we adopt a numerical approach where we study how different integrals scale in an ensemble of stratification profiles that resemble the profiles used throughout the paper. Using this information, we scale the different terms. To this end, the stratification profiles are chosen such that N max = ( 5 N b , 10 N b , 15 N b ) N_{\max}=(5N_b,10N_b,15N_b) N m a x = ( 5 N b , 10 N b , 15 N b ) , W p = ( H / 100 , 2 H / 100 , 3 H / 100 ) W_p=(H/100,2H/100,3H/100) W p = ( H /100 , 2 H /100 , 3 H /100 ) , and z c = ( H / 80 , H / 40 , H / 20 , H / 10 ) z_c=(H/80,H/40,H/20,H/10) z c = ( H /80 , H /40 , H /20 , H /10 ) ; and we consider all possible (36) combinations.
The analysis provides a relation between the time scale of the amplitude’s temporal evolution (ϵ t \epsilon_t ϵ t ), length scale of the amplitude function (ϵ x \epsilon_x ϵ x ), and the magnitude of the waves’ amplitude (ϵ a a j \epsilon_a a_j ϵ a a j ). Small parameters (ϵ h , ϵ k \epsilon_h,\epsilon_k ϵ h , ϵ k ) represent the bathymetry and they also influence the wave amplitude evolution. Equation (27) , after some simplifications to the nonlinear term, is given below:
∂ a j ∂ t + 1 D j [ 2 i γ j ( 3 ) ( K j h ∂ a j ∂ x ) + γ j ( 6 ) h 2 ( d h d x ) 2 a j ] + 1 2 D j [ 2 K j h γ j ( 4 ) + γ j ( 3 ) ∂ ∂ x ( K j h ) − γ j ( 5 ) 2 h ∂ h ∂ x ( K j h ) − γ j ( 3 ) β j 2 K j h d ( β j ) d x ] a j = N ^ j a 2 (79) \frac{\partial a_j}{\partial t}+\frac{1}{\mathcal{D}_j}\left[2\mathrm{i}\gamma_j^{(3)}\left(\frac{\mathcal{K}_j}{h}\frac{\partial a_j}{\partial x}\right)+\frac{\gamma_j^{(6)}}{h^2}\left(\frac{dh}{dx}\right)^2a_j\right]
+\frac{1}{2\mathcal{D}_j}\left[\frac{2\mathcal{K}_j}{h}\gamma_j^{(4)}+\gamma_j^{(3)}\frac{\partial}{\partial x}\left(\frac{\mathcal{K}_j}{h}\right)-\gamma_j^{(5)}\frac{2}{h}\frac{\partial h}{\partial x}\left(\frac{\mathcal{K}_j}{h}\right)-\frac{\gamma_j^{(3)}}{\beta_j}\frac{2\mathcal{K}_j}{h}\frac{d(\beta_j)}{dx}\right]a_j=\widehat{\mathcal{N}}_j a^2 \tag*{(79)} ∂ t ∂ a j + D j 1 [ 2 i γ j ( 3 ) ( h K j ∂ x ∂ a j ) + h 2 γ j ( 6 ) ( d x d h ) 2 a j ] + 2 D j 1 [ h 2 K j γ j ( 4 ) + γ j ( 3 ) ∂ x ∂ ( h K j ) − γ j ( 5 ) h 2 ∂ x ∂ h ( h K j ) − β j γ j ( 3 ) h 2 K j d x d ( β j ) ] a j = N j a 2 (79)
where N ^ j \widehat{\mathcal{N}}_j N j is defined as:
N ^ j = 1 D j [ N L ( V , j ) + N L ( B , j ) + N L ( Ψ , j ) ] . (80) \widehat{\mathcal{N}}_j=\frac{1}{\mathcal{D}_j}\left[\mathrm{NL}_{(\mathcal{V},j)}+\mathrm{NL}_{(\mathcal{B},j)}+\mathrm{NL}_{(\Psi,j)}\right]. \tag*{(80)} N j = D j 1 [ NL ( V , j ) + NL ( B , j ) + NL ( Ψ , j ) ] . (80)
The analysis is similar for all three waves, hence from here on all subscripts j j j (denoting the j j j −th wave) are dropped for convenience. Moreover, a term containing γ ( 6 ) \gamma^{(6)} γ ( 6 ) is also included in the above equation. It will be proved in this section that this term is an order of magnitude smaller than the other terms for the parameter regime we consider.
The time scale of wave amplitude’s evolution is assumed to be at least an order of magnitude larger than the time period of the wave. Therefore ∂ a / ∂ t \partial a/\partial t ∂ a / ∂ t will approximately scale as: ∂ a / ∂ t ∼ ϵ t ϵ a ω \partial a/\partial t\sim\epsilon_t\epsilon_a\omega ∂ a / ∂ t ∼ ϵ t ϵ a ω . The amplitude’s length scale is assumed to be much larger than the wavelength of the wave. Hence ∂ a / ∂ x \partial a/\partial x ∂ a / ∂ x will scale as ∂ a / ∂ x ∼ ϵ x ϵ a K / h \partial a/\partial x\sim\epsilon_x\epsilon_a\mathcal{K}/h ∂ a / ∂ x ∼ ϵ x ϵ a K / h . Using the above scaling, the ∂ a / ∂ x \partial a/\partial x ∂ a / ∂ x term in (B1) (including its coefficients) will scale as:
2 γ ( 3 ) D ( K h ∂ a ∂ x ) ∼ 1 ω γ ( 3 ) K 2 ω γ ( 1 ) K 2 − γ ( 2 ) ϵ x ϵ a ∼ ( c ^ g ) ω ϵ a (81) 2\frac{\gamma^{(3)}}{\mathcal{D}}\left(\frac{\mathcal{K}}{h}\frac{\partial a}{\partial x}\right)\sim\frac{1}{\omega}\frac{\gamma^{(3)}\mathcal{K}^2}{\omega\gamma^{(1)}\mathcal{K}^2-\gamma^{(2)}}\epsilon_x\epsilon_a\sim(\widehat{c}_g)\omega\epsilon_a \tag*{(81)} 2 D γ ( 3 ) ( h K ∂ x ∂ a ) ∼ ω 1 ω γ ( 1 ) K 2 − γ ( 2 ) γ ( 3 ) K 2 ϵ x ϵ a ∼ ( c g ) ω ϵ a (81)
where c ^ g \widehat{c}_g c g represents the scale of group speed term for the packet, and is given by:
c ^ g ≡ ( ω 2 − f 2 ) ω 2 [ γ ( 3 ) ( ω 2 − f 2 ) γ ( 1 ) + γ ( 3 ) ] . (82) \widehat{c}_g\equiv\frac{(\omega^2-f^2)}{\omega^2}\left[\frac{\gamma^{(3)}}{(\omega^2-f^2)\gamma^{(1)}+\gamma^{(3)}}\right]. \tag*{(82)} c g ≡ ω 2 ( ω 2 − f 2 ) [ ( ω 2 − f 2 ) γ ( 1 ) + γ ( 3 ) γ ( 3 ) ] . (82)
It can be noticed that as ϵ x \epsilon_x ϵ x is reduced, the effect of group speed diminishes as expected since a decrease in ϵ x \epsilon_x ϵ x means the length scale of the packet is increased. Here we also emphasize that for ω ≈ N \omega\approx N ω ≈ N : γ ( 3 ) ≪ ω 2 γ ( 1 ) \gamma^{(3)}\ll\omega^2\gamma^{(1)} γ ( 3 ) ≪ ω 2 γ ( 1 ) . In such kind of parameter regime, c ^ g ≪ 1 \widehat{c}_g\ll1 c g ≪ 1 , hence ∂ a / ∂ x \partial a/\partial x ∂ a / ∂ x term will have a reduced effect on the amplitude evolution. Moreover, for ω ≈ f \omega\approx f ω ≈ f , similar behavior is observed since c ^ g ≪ 1 \widehat{c}_g\ll1 c g ≪ 1 .
Now we focus on the term containing γ ( 6 ) \gamma^{(6)} γ ( 6 ) in (B1), which is given below (after some simplification):
( d h d x ) 2 W K 2 ω ϵ a 2 , (83) \left(\frac{dh}{dx}\right)^2\frac{\mathcal{W}}{\mathcal{K}^2}\frac{\omega\epsilon_a}{2}, \tag*{(83)} ( d x d h ) 2 K 2 W 2 ω ϵ a , (83)
where W \mathcal{W} W is a non-dimensional quantity defined as:
W = ω 2 − f 2 ω 2 γ ( 6 ) ( ω 2 − f 2 ) γ ( 1 ) + γ ( 3 ) . (84) \mathcal{W}=\frac{\omega^2-f^2}{\omega^2}\frac{\gamma^{(6)}}{(\omega^2-f^2)\gamma^{(1)}+\gamma^{(3)}}. \tag*{(84)} W = ω 2 ω 2 − f 2 ( ω 2 − f 2 ) γ ( 1 ) + γ ( 3 ) γ ( 6 ) . (84)
The integral γ ( 6 ) \gamma^{(6)} γ ( 6 ) is evaluated numerically to study its scaling. For uniform stratification, W \mathcal{W} W can be evaluated analytically, which is given below:
W u = − M 2 [ ω 2 − f 2 ω 2 N b 2 − ω 2 N b 2 − f 2 ] ( 1 3 − 1 2 M 2 ) (85) \mathcal{W}_u=-\mathcal{M}^2\left[\frac{\omega^2-f^2}{\omega^2}\frac{N_b^2-\omega^2}{N_b^2-f^2}\right]\left(\frac{1}{3}-\frac{1}{2\mathcal{M}^2}\right) \tag*{(85)} W u = − M 2 [ ω 2 ω 2 − f 2 N b 2 − f 2 N b 2 − ω 2 ] ( 3 1 − 2 M 2 1 ) (85)
where W u \mathcal{W}_u W u is used to denote W \mathcal{W} W in constant stratification N b N_b N b , and M = n π \mathcal{M}=n\pi M = nπ is the non-dimensionalised vertical wavenumber of the wave. Moreover, using ( d h / d x ) 2 ∼ ( ϵ h ϵ k ) 2 K 2 (dh/dx)^2\sim(\epsilon_h\epsilon_k)^2\mathcal{K}^2 ( d h / d x ) 2 ∼ ( ϵ h ϵ k ) 2 K 2 , the term given in equation (B5) will scale as
( ( ϵ h ϵ k ) 2 2 W ) ω ϵ a . (86) \left(\frac{(\epsilon_h\epsilon_k)^2}{2}\mathcal{W}\right)\omega\epsilon_a. \tag*{(86)} ( 2 ( ϵ h ϵ k ) 2 W ) ω ϵ a . (86)
Hence for the multiple-scale analysis to be consistent, W ( ( ϵ h ϵ k ) 2 / 2 ) \mathcal{W}((\epsilon_h\epsilon_k)^2/2) W (( ϵ h ϵ k ) 2 /2 ) has to be a small quantity. W \mathcal{W} W is plotted in figure 15 for nine stratification profiles, where f = 0 f=0 f = 0 and ω / N b = 0.4 \omega/N_b=0.4 ω / N b = 0.4 were used. In all subfigures, W u \mathcal{W}_u W u is also plotted for reference, where W u \mathcal{W}_u W u is evaluated with constant stratification N b N_b N b (hence W u \mathcal{W}_u W u in all subfigures is same). From figure 15, it can be seen that in general for any stratification profile, W \mathcal{W} W is almost proportional to the square of the modenumber n n n , similar to W u \mathcal{W}_u W u . Hence the bathymetry has to be more slowly varying (ϵ k \epsilon_k ϵ k has to be smaller) as the modenumber increases. Other pycnocline depths (z c = H / 20 , H / 40 , H / 80 z_c=H/20,H/40,H/80 z c = H /20 , H /40 , H /80 ) were also tested for different combinations of W p , N max W_p,N_{\max} W p , N m a x used in figure 15 that provided similar results.
The term which contains the γ ( 5 ) \gamma^{(5)} γ ( 5 ) integral is now analysed. For all non-uniform stratification profiles used in this appendix, it was observed that
γ ( 5 ) ≲ 1 2 γ ( 3 ) . (87) \gamma^{(5)}\lesssim\frac{1}{2}\gamma^{(3)}. \tag*{(87)} γ ( 5 ) ≲ 2 1 γ ( 3 ) . (87)
Using (B9), the term containing γ ( 5 ) \gamma^{(5)} γ ( 5 ) can be scaled to:
1 D j [ γ ( 5 ) 2 h ∂ h ∂ x ( K h ) ] a ∼ ( c ^ g 2 ϵ h ϵ k ) ω ϵ a . (88) \frac{1}{\mathcal{D}_j}\left[\gamma^{(5)}\frac{2}{h}\frac{\partial h}{\partial x}\left(\frac{\mathcal{K}}{h}\right)\right]a\sim\left(\frac{\hat{c}_g}{2}\epsilon_h\epsilon_k\right)\omega\epsilon_a. \tag*{(88)} D j 1 [ γ ( 5 ) h 2 ∂ x ∂ h ( h K ) ] a ∼ ( 2 c ^ g ϵ h ϵ k ) ω ϵ a . (88)
Now we analyse how the wavenumber of a mode changes as h h h changes. To this end, (2.13b), which provides the n n n -th eigenfunction, is differentiated in x x x –direction, yielding:
[ ∂ 2 ∂ η 2 + K n 2 χ 2 ] ∂ ϕ n ∂ x = − K n 2 ∂ χ 2 ∂ x ϕ n − d K n 2 d x χ 2 ϕ n , (89) \left[\frac{\partial^2}{\partial\eta^2}+\mathcal{K}_n^2\chi^2\right]\frac{\partial\phi_n}{\partial x}
=-\mathcal{K}_n^2\frac{\partial\chi^2}{\partial x}\phi_n
-\frac{d\mathcal{K}_n^2}{dx}\chi^2\phi_n, \tag*{(89)} [ ∂ η 2 ∂ 2 + K n 2 χ 2 ] ∂ x ∂ ϕ n = − K n 2 ∂ x ∂ χ 2 ϕ n − d x d K n 2 χ 2 ϕ n , (89)
where K n = k n h \mathcal{K}_n=k_nh K n = k n h . Equation (B11) can have a non-trivial solution only when the RHS is orthogonal to the solution of the self adjoint operator in the LHS. Hence multiplying RHS with ϕ n \phi_n ϕ n , and then integrating in the η \eta η –direction between the domain limits would result in:
1 k n d k n d x + 1 h d h d x = − 1 2 γ n ( 3 ) ∫ − 1 0 ∂ N 2 ∂ x ϕ n 2 d η . (90) \frac{1}{k_n}\frac{dk_n}{dx}+\frac{1}{h}\frac{dh}{dx}
=-\frac{1}{2\gamma_n^{(3)}}\int_{-1}^{0}\frac{\partial N^2}{\partial x}\phi_n^2d\eta. \tag*{(90)} k n 1 d x d k n + h 1 d x d h = − 2 γ n ( 3 ) 1 ∫ − 1 0 ∂ x ∂ N 2 ϕ n 2 d η . (90)
From (B12), we notice that the dimensional wavenumber k n k_n k n can change due to: (a) change in the domain height, and (b) change in the effective stratification profile. For uniform stratification, d K n / d x = 0 dK_n/dx=0 d K n / d x = 0 . For the lower modes (1–10) in profiles considered in this appendix, it was observed that
Figure 15. The variation in W \mathcal{W} W for modes 1–10 for different stratification profiles. (a) N max = 5 N b N_{\max}=5N_b N m a x = 5 N b is used with z c = H / 10 z_c=H/10 z c = H /10 and W p W_p W p is varied. (b) N max = 10 N b N_{\max}=10N_b N m a x = 10 N b is used with z c = H / 10 z_c=H/10 z c = H /10 and W p W_p W p is varied. (c) N max = 15 N b N_{\max}=15N_b N m a x = 15 N b is used with z c = H / 10 z_c=H/10 z c = H /10 and W p W_p W p is varied.
k h ≡ O ( 1 k n d k n d x ) / O ( 1 h d h d x ) ∼ O ( 1 ) . (91) k_h \equiv\mathcal{O}\left(\frac{1}{k_n}\frac{d k_n}{d x}\right)\bigg/\mathcal{O}\left(\frac{1}{h}\frac{d h}{d x}\right) \sim\mathcal{O}(1). \tag*{(91)} k h ≡ O ( k n 1 d x d k n ) / O ( h 1 d x d h ) ∼ O ( 1 ) . (91)
Moreover, in general it was observed that as the modenumber increases, k h k_h k h increases. Using (B13), the term containing the derivative of the wavenumber can be scaled as:
1 D j [ γ ( 3 ) d k d x a ] ∼ ( c ^ g 2 ϵ h ϵ k ) ω ϵ a . (92) \frac{1}{\mathcal{D}_j}\left[\gamma^{(3)}\frac{d k}{d x}a\right]\sim\left(\frac{\widehat{c}_g}{2}\epsilon_h\epsilon_k\right)\omega\epsilon_a. \tag*{(92)} D j 1 [ γ ( 3 ) d x d k a ] ∼ ( 2 c g ϵ h ϵ k ) ω ϵ a . (92)
We now evaluate γ ( 4 ) \gamma^{(4)} γ ( 4 ) for the stratification profiles considered in this appendix. For modes 1–10, we find
γ ( 4 ) γ ( 3 ) ∼ O ( 1 h d h d x ) . (93) \frac{\gamma^{(4)}}{\gamma^{(3)}}\sim\mathcal{O}\left(\frac{1}{h}\frac{d h}{d x}\right). \tag*{(93)} γ ( 3 ) γ ( 4 ) ∼ O ( h 1 d x d h ) . (93)
Using (B15), the term containing γ ( 4 ) \gamma^{(4)} γ ( 4 ) can be scaled to:
1 D j [ 2 K h γ ( 4 ) a ] ∼ ( c ^ g ϵ h ϵ k ) ω ϵ a . (94) \frac{1}{\mathcal{D}_j}\left[\frac{2\mathcal{K}}{h}\gamma^{(4)}a\right]\sim\left(\widehat{c}_g\epsilon_h\epsilon_k\right)\omega\epsilon_a. \tag*{(94)} D j 1 [ h 2 K γ ( 4 ) a ] ∼ ( c g ϵ h ϵ k ) ω ϵ a . (94)
Using (B10), (B14) and (B16), we observe that for the lower modes, the three terms that compose the β \beta β function can scale to a maximum value which is of the same order of magnitude. Hence they are all retained and are used in evaluating the β \beta β function (28) . Moreover, it can be seen that the topographic terms are all dependent on the magnitude of the group speed. This relation is naturally there because a wavepacket has to travel to different h h h fast enough to feel the effect of h h h variation. Scaling (B16) also holds for uniform stratification, where ϕ \phi ϕ still varies in the x x x –direction. This is because of the nature of the ϕ \phi ϕ normalisation, i.e. (22) , used in this paper.
The nonlinear coupling coefficient in the RHS cannot be further simplified, hence the nonlinear term scales as:
R H S ∼ N ^ ϵ a 2 . (95) \mathrm{RHS}\sim\widehat{\mathcal{N}}\epsilon_a^2. \tag*{(95)} RHS ∼ N ϵ a 2 . (95)
Hence the final scaling for (2.24a)–(2.24c), using all the scaling derived, and with the inclusion of the γ ( 6 ) \gamma^{(6)} γ ( 6 ) term, is given below after some simplification:
ϵ t ∼ N ω ϵ a − c g ^ ϵ e x − ( ϵ h ϵ k ) 2 2 W . (96) \epsilon_t \sim\frac{\mathcal{N}}{\omega}\epsilon_a-\widehat{c_g}\epsilon_{ex}-\frac{(\epsilon_h\epsilon_k)^2}{2}\mathcal{W}. \tag*{(96)} ϵ t ∼ ω N ϵ a − c g ϵ e x − 2 ( ϵ h ϵ k ) 2 W . (96)
.Here an important point to remember is that the multiple-scale analysis was derived with the assumption that internal waves do not scatter/exchange energy to different modes of the same angular frequency. Therefore the reduced order equations provide the most accurate results when the internal waves do not scatter significant amount of its energy as it passes over a bathymetry. Moreover, even when O ( ϵ h ϵ k ) ≪ O ( 1 ) \mathcal{O}(\epsilon_h\epsilon_k)\ll\mathcal{O}(1) O ( ϵ h ϵ k ) ≪ O ( 1 ) is satisfied, there could be special circumstances when waves may still get scattered significantly. An example of such a scenario is Bragg resonance of internal waves due to small amplitude, subcritical topographies [3 , 5 ] (Li & Mei 2014). Scattering/energy exchange can also occur for large amplitude, slowly varying topographies. However, it was observed that modes 1–8 are scattered very little for large amplitude topographies (ϵ h ≈ 0.5 \epsilon_h\approx0.5 ϵ h ≈ 0.5 ) with low criticality (≲ 0.1 \lesssim0.1 ≲ 0.1 ) in the presence of uniform stratification. Criticality is defined as the ratio of the maximum slope of the topography to the slope of the internal wave. Mode-8 has ≈ 8 % \approx8\% ≈ 8% variation in its amplitude as it propagates through a Gaussian topography with ϵ h = 0.5 \epsilon_h=0.5 ϵ h = 0.5 and criticality = 0.1 =0.1 = 0.1 . Low criticality topographies for mode n n n of any ω / N b \omega/N_b ω / N b is obtained when the condition n ϵ k ≪ O ( 1 ) n\epsilon_k\ll\mathcal{O}(1) n ϵ k ≪ O ( 1 ) is satisfied. Moreover for the condition n ϵ k ≪ O ( 1 ) n\epsilon_k\ll\mathcal{O}(1) n ϵ k ≪ O ( 1 ) , the last term in (B18) becomes an ϵ k 2 \epsilon_k^2 ϵ k 2 term even for large amplitude topographies. This can be seen by considering W u \mathcal{W}_u W u (which can also be used as a reference for non-uniform stratifications) given in (B7). Hence this term is neglected in the governing equations (2.24a)–(2.24c).
Appendix C
The scaling analysis for the governing equation (65) derived in § 6 is done with the help of results derived in appendix B. The above-mentioned governing equations are given below:
[ ( γ j ( 3 ) ∂ 2 A j ∂ x 2 ) + κ j 2 ( A j h 2 γ j ( 3 ) ) ] T j + 2 ( γ j ( 4 ) − γ j ( 5 ) h ∂ h ∂ x ) ∂ A j ∂ x T j + γ j ( 8 ) A j T j + [ γ j ( 6 ) h 2 ( ∂ h ∂ x ) 2 − γ j ( 5 ) h ( ∂ 2 h ∂ x 2 ) + 2 γ j ( 5 ) h 2 ( ∂ h ∂ x ) 2 − 2 γ j ( 7 ) h ∂ h ∂ x ] A j T j = ⟨ N L 3 ⟩ . (97) \left[\left(\gamma_j^{(3)}\frac{\partial^2 A_j}{\partial x^2}\right)+\kappa_j^2\left(\frac{A_j}{h^2}\gamma_j^{(3)}\right)\right]T_j+2\left(\gamma_j^{(4)}-\frac{\gamma_j^{(5)}}{h}\frac{\partial h}{\partial x}\right)\frac{\partial A_j}{\partial x}T_j+\gamma_j^{(8)}A_jT_j
+\left[\frac{\gamma_j^{(6)}}{h^2}\left(\frac{\partial h}{\partial x}\right)^2-\frac{\gamma_j^{(5)}}{h}\left(\frac{\partial^2h}{\partial x^2}\right)+\frac{2\gamma_j^{(5)}}{h^2}\left(\frac{\partial h}{\partial x}\right)^2-\frac{2\gamma_j^{(7)}}{h}\frac{\partial h}{\partial x}\right]A_jT_j=\langle\mathrm{NL}_3\rangle. \tag*{(97)} [ ( γ j ( 3 ) ∂ x 2 ∂ 2 A j ) + κ j 2 ( h 2 A j γ j ( 3 ) ) ] T j + 2 ( γ j ( 4 ) − h γ j ( 5 ) ∂ x ∂ h ) ∂ x ∂ A j T j + γ j ( 8 ) A j T j + [ h 2 γ j ( 6 ) ( ∂ x ∂ h ) 2 − h γ j ( 5 ) ( ∂ x 2 ∂ 2 h ) + h 2 2 γ j ( 5 ) ( ∂ x ∂ h ) 2 − h 2 γ j ( 7 ) ∂ x ∂ h ] A j T j = ⟨ NL 3 ⟩ . (97)
From here on the subscripts are omitted, since the analysis is similar for both the waves. The leading order terms scale as follows:
[ ∂ 2 A ∂ x 2 , κ 2 A h 2 ] ∼ ϵ a κ 2 h 2 . (98) \left[\frac{\partial^2\mathcal{A}}{\partial x^2},\kappa^2\frac{\mathcal{A}}{h^2}\right]\sim\epsilon_a\frac{\kappa^2}{h^2}. \tag*{(98)} [ ∂ x 2 ∂ 2 A , κ 2 h 2 A ] ∼ ϵ a h 2 κ 2 . (98)
Using (14) , the scalings derived in appendix B, and the small amplitude assumption for topography (ϵ h ≪ O ( 1 ) \epsilon_h\ll\mathcal{O}(1) ϵ h ≪ O ( 1 ) and ϵ k ∼ O ( 1 ) \epsilon_k\sim\mathcal{O}(1) ϵ k ∼ O ( 1 ) ), we obtain:
2 γ ( 4 ) γ ( 3 ) ∂ A ∂ x ∼ 2 ϵ h ϵ a κ 2 h 2 , 2 h ∂ h ∂ x γ ( 5 ) γ ( 3 ) ∂ A ∂ x ∼ ϵ h ϵ a κ 2 h 2 , 2\frac{\gamma^{(4)}}{\gamma^{(3)}}\frac{\partial\mathcal{A}}{\partial x}\sim2\epsilon_h\epsilon_a\frac{\kappa^2}{h^2},\qquad
\frac{2}{h}\frac{\partial h}{\partial x}\frac{\gamma^{(5)}}{\gamma^{(3)}}\frac{\partial\mathcal{A}}{\partial x}\sim\epsilon_h\epsilon_a\frac{\kappa^2}{h^2}, 2 γ ( 3 ) γ ( 4 ) ∂ x ∂ A ∼ 2 ϵ h ϵ a h 2 κ 2 , h 2 ∂ x ∂ h γ ( 3 ) γ ( 5 ) ∂ x ∂ A ∼ ϵ h ϵ a h 2 κ 2 ,
( 1 h ∂ 2 h ∂ x 2 ) γ ( 5 ) γ ( 3 ) A ∼ ϵ h 2 ϵ a κ 2 h 2 , 2 ( 1 h ∂ h ∂ x ) 2 γ ( 5 ) γ ( 3 ) A ∼ ϵ h 2 ϵ a κ 2 h 2 . (99) \left(\frac{1}{h}\frac{\partial^2h}{\partial x^2}\right)\frac{\gamma^{(5)}}{\gamma^{(3)}}\mathcal{A}\sim\frac{\epsilon_h}{2}\epsilon_a\frac{\kappa^2}{h^2},\qquad
2\left(\frac{1}{h}\frac{\partial h}{\partial x}\right)^2\frac{\gamma^{(5)}}{\gamma^{(3)}}\mathcal{A}\sim\epsilon_h^2\epsilon_a\frac{\kappa^2}{h^2}. \tag*{(99)} ( h 1 ∂ x 2 ∂ 2 h ) γ ( 3 ) γ ( 5 ) A ∼ 2 ϵ h ϵ a h 2 κ 2 , 2 ( h 1 ∂ x ∂ h ) 2 γ ( 3 ) γ ( 5 ) A ∼ ϵ h 2 ϵ a h 2 κ 2 . (99)
For the profiles and the parameters used in appendix B we observe that ( ω 2 − f 2 ) γ ( 1 ) + γ ( 3 ) ∼ γ ( 3 ) (\omega^2-f^2)\gamma^{(1)}+\gamma^{(3)}\sim\gamma^{(3)} ( ω 2 − f 2 ) γ ( 1 ) + γ ( 3 ) ∼ γ ( 3 ) . Hence the γ ( 6 ) \gamma^{(6)} γ ( 6 ) term can be scaled as:
γ ( 6 ) γ ( 3 ) 1 h 2 ( ∂ h ∂ x ) 2 A ∼ ( W ε h 2 ) ϵ a K 2 h 2 , (100) \frac{\gamma^{(6)}}{\gamma^{(3)}}\frac{1}{h^2}\left(\frac{\partial h}{\partial x}\right)^2 A\sim\left(\mathcal{W}\varepsilon_h^2\right)\epsilon_a\frac{\mathcal{K}^2}{h^2}, \tag*{(100)} γ ( 3 ) γ ( 6 ) h 2 1 ( ∂ x ∂ h ) 2 A ∼ ( W ε h 2 ) ϵ a h 2 K 2 , (100)
where W \mathcal{W} W is plotted in figure 15 for various stratification profiles. Therefore similar to appendix B, the term W ε h 2 \mathcal{W}\varepsilon_h^2 W ε h 2 has to be a small number for the multiple-scale analysis to be consistent. Furthermore, the γ j ( 7 ) \gamma_j^{(7)} γ j ( 7 ) term was observed to scale as:
( 2 h ∂ h ∂ x ) γ ( 7 ) γ ( 3 ) A ≲ K n 2 K 1 2 ( 1 h ∂ h ∂ x ) 2 2 ϵ a , (101) \left(\frac{2}{h}\frac{\partial h}{\partial x}\right)\frac{\gamma^{(7)}}{\gamma^{(3)}}A\lesssim\frac{\mathcal{K}_n^2}{\mathcal{K}_1^2}\left(\frac{1}{h}\frac{\partial h}{\partial x}\right)^2 2\epsilon_a, \tag*{(101)} ( h 2 ∂ x ∂ h ) γ ( 3 ) γ ( 7 ) A ≲ K 1 2 K n 2 ( h 1 ∂ x ∂ h ) 2 2 ϵ a , (101)
where K n \mathcal{K}_n K n is the nondimensional wavenumber of wave-1 (or wave-3), and n n n gives the wave’s modenumber. Note that this scaling has a similar behavior as W \mathcal{W} W , which is nearly proportional to n 2 n^2 n 2 . Now we focus on the scaling of the integral γ ( 8 ) \gamma^{(8)} γ ( 8 ) :
γ ( 8 ) = ( d h d x ) 2 ∫ − 1 0 ( N 2 − ω 2 ) ϕ ∂ 2 ϕ ∂ h 2 d η + d 2 h d x 2 ∫ − 1 0 ( N 2 − ω 2 ) ϕ ∂ ϕ ∂ h d η . (102) \gamma^{(8)}=\left(\frac{dh}{dx}\right)^2\int_{-1}^{0}(N^2-\omega^2)\phi\frac{\partial^2\phi}{\partial h^2}\,d\eta+\frac{d^2h}{dx^2}\int_{-1}^{0}(N^2-\omega^2)\phi\frac{\partial\phi}{\partial h}\,d\eta. \tag*{(102)} γ ( 8 ) = ( d x d h ) 2 ∫ − 1 0 ( N 2 − ω 2 ) ϕ ∂ h 2 ∂ 2 ϕ d η + d x 2 d 2 h ∫ − 1 0 ( N 2 − ω 2 ) ϕ ∂ h ∂ ϕ d η . (102)
For a uniform stratification, ∂ 2 ϕ / ∂ h 2 = 0 \partial^2\phi/\partial h^2=0 ∂ 2 ϕ / ∂ h 2 = 0 . Moreover, for the non-uniform stratification profiles used in appendix B, it was observed that:
∫ − 1 0 ( N 2 − ω 2 ) ϕ ∂ 2 ϕ ∂ h 2 d η ≲ K n 2 K 1 2 γ ( 3 ) h 2 , ∫ − 1 0 ( N 2 − ω 2 ) ϕ ∂ ϕ ∂ h d η ∼ γ ( 3 ) h . (103) \int_{-1}^{0}(N^2-\omega^2)\phi\frac{\partial^2\phi}{\partial h^2}\,d\eta\lesssim\frac{\mathcal{K}_n^2}{\mathcal{K}_1^2}\frac{\gamma^{(3)}}{h^2},\qquad\int_{-1}^{0}(N^2-\omega^2)\phi\frac{\partial\phi}{\partial h}\,d\eta\sim\frac{\gamma^{(3)}}{h}. \tag*{(103)} ∫ − 1 0 ( N 2 − ω 2 ) ϕ ∂ h 2 ∂ 2 ϕ d η ≲ K 1 2 K n 2 h 2 γ ( 3 ) , ∫ − 1 0 ( N 2 − ω 2 ) ϕ ∂ h ∂ ϕ d η ∼ h γ ( 3 ) . (103)
Hence using (C7), the scaling for γ ( 8 ) \gamma^{(8)} γ ( 8 ) can be given in a simpler form which is as follows:
γ ( 8 ) ∼ [ K n 2 K 1 2 ( 1 h d h d x ) 2 + 1 h d 2 h d x 2 ] γ ( 3 ) . (104) \gamma^{(8)}\sim\left[\frac{\mathcal{K}_n^2}{\mathcal{K}_1^2}\left(\frac{1}{h}\frac{dh}{dx}\right)^2+\frac{1}{h}\frac{d^2h}{dx^2}\right]\gamma^{(3)}. \tag*{(104)} γ ( 8 ) ∼ [ K 1 2 K n 2 ( h 1 d x d h ) 2 + h 1 d x 2 d 2 h ] γ ( 3 ) . (104)
For low modes in the presence of small amplitude topographies, the second term in RHS of (C8) would be significantly higher than the first term. For any mild-slope bathymetry, the nonlinear terms ⟨ N L 3 ⟩ \langle\mathrm{NL}_3\rangle ⟨ NL 3 ⟩ in (C1) can be scaled using the relation d n A / d x n ≈ ( K / h ) n A d^nA/dx^n\approx(\mathcal{K}/h)^nA d n A / d x n ≈ ( K / h ) n A , where n ∈ Z + n\in\mathbb{Z}^+ n ∈ Z + . Using this approximation, the nonlinear term can be scaled as:
⟨ N L 3 ⟩ ∼ 1 γ ( 3 ) [ N L ( V , 3 ) + N L ( B , 3 ) + N L ( Ψ , 3 ) ] ϵ a 2 . (105) \langle\mathrm{NL}_3\rangle\sim\frac{1}{\gamma^{(3)}}\left[\mathrm{NL}_{(\mathcal{V},3)}+\mathrm{NL}_{(\mathcal{B},3)}+\mathrm{NL}_{(\Psi,3)}\right]\epsilon_a^2. \tag*{(105)} ⟨ NL 3 ⟩ ∼ γ ( 3 ) 1 [ NL ( V , 3 ) + NL ( B , 3 ) + NL ( Ψ , 3 ) ] ϵ a 2 . (105)
The nonlinear coupling coefficients cannot be simplified further. Moreover the nonlinear terms have to be at least one order of magnitude lesser than the leading order terms (given in (C2)).