betaFeedback

WorksPhysics and Astronomy

Nonlinear long extensional waves in an elastic layer with irregular free surfaces

Abstract

The propagation of nonlinear long extensional waves in a compressible hyperelastic layer with irregular free surfaces is studied within the framework of finite elasticity. The free surfaces are traction-free and symmetric with respect to the middle plane, and their irregularity changes the local thickness of the layer by a finite amount over distances long compared with the wavelength. In a distinguished limit in which weak nonlinearity, dispersion and the irregularity of the free surfaces appear at the same order, a multiple-scales analysis of the two-dimensional boundary-value problem leads to a Korteweg–de Vries equation with variable coefficients for the longitudinal strain. The irregularity acts on the wave only through the traction-free boundary conditions and changes the balance between nonlinearity and dispersion along the layer. After normalisation, the combined effect of the material and of the geometry is carried by a single variable nonlinear coefficient, whose sign decides whether compressive or tensile solitary waves exist. For special choices of the surface irregularity, the evolution equation reduces to forms known to be integrable, namely the Korteweg–de Vries and cylindrical Korteweg–de Vries equations. For periodically irregular layers of iron and copper and for model materials, numerical solutions show that a localized pulse propagates as a solitary wave, splits into a train of solitary waves or disperses into an oscillatory wave train, depending on the material, while the irregularity of the free surfaces redistributes the amplitudes of the emerging solitary waves.

Highlights

  • Long extensional waves in an elastic layer with irregular free surfaces are studied.

  • Surface irregularity alters the balance of nonlinearity and dispersion.

  • Special irregularity profiles reduce the model to known integrable forms.

  • Iron and copper layers support compressive strain solitary waves.

  • Pulses propagate as solitary waves, split into soliton trains or disperse.

Introduction

Elastic waveguides such as rods, plates and layers carry long longitudinal waves over large distances with little attenuation, and the way these waves change as they travel is a central question in the mechanics of solids. When the amplitude of the strain is small but finite, two effects compete. Nonlinearity steepens the wave profile. It has a geometric part, which comes from the finite-strain kinematics, and a physical part, which comes from the higher-order elastic constants. Dispersion spreads the profile. In a waveguide it arises mainly from the inertia of the transverse motion that the traction-free lateral surfaces allow. When these two effects balance, the strain can travel as a solitary wave of permanent form. It was recognised early that this balance makes the Korteweg–de Vries (KdV) equation and its generalisations the natural long-wave models of nonlinear elastic waveguides [1, 2, 3]. Strain solitary waves in rods and plates have since been studied in detail, including the influence of a surrounding medium [4], of two-dimensional effects [5] and of cubic and higher-order terms [6, 7]. Strain solitary waves have also been generated and recorded in polymer rods [8], and radiating solitary waves have been found in bi-layers with delamination [9]. These studies show that solitary waves in solids are not a mathematical idealisation. They are mechanical objects whose existence and polarity are fixed by the nonlinear constitutive response of the material, and their potential use in non-destructive evaluation follows from this.

For a layer with plane, traction-free surfaces, long extensional waves can be studied by an asymptotic reduction of the two-dimensional equations of finite elasticity, which leads to a KdV equation whose nonlinear coefficient depends on the second- and third-order elastic constants and whose dispersive coefficient reflects the transverse structure of the fundamental symmetric mode [10]. Boussinesq- and KdV-type models have also been obtained for compressible hyperelastic plates [11]. In such models dispersion is mainly geometric, being associated with the finite thickness of the waveguide. A different group of studies considers unbounded media in which dispersion is introduced through the constitutive law, for instance by higher-gradient, nonlocal or strain-limiting terms; in this setting, KdV-, Camassa–Holm- and Kadomtsev–Petviashvili-type equations have been derived for longitudinal and transverse waves [12, 13, 14, 15, 16][12, 13, 14, 15, 16], and, more recently, the KP equation of plane elastodynamics [17]. The present work belongs to the studies on plates and layers mentioned first, in which dispersion is of geometric origin: here the free surfaces of the layer give rise to dispersion, so that a change in their shape may be expected to affect the balance between nonlinearity and dispersion.

Real waveguides are rarely uniform. Their thickness changes because of manufacturing tolerances, wear and corrosion, deliberate tapering or the natural irregularity of geological layers. The linear theory of waves in layers with irregular boundaries has a long history in seismology and elastodynamics. Love waves in layers with irregular boundaries and in slowly varying layered media were among the first problems treated [18, 19, 20]. Later work considered irregular, corrugated and inhomogeneous layers and half-spaces in increasingly general settings [21, 22, 23, 24], as well as the scattering of shear waves by irregular surface topographies [25]. For plate waves, dispersion relations of Rayleigh–Lamb and shear-horizontal waves have been obtained for corrugated waveguides [26]. The propagation of Lamb waves in a layer with irregular free surfaces and the effect of surface roughness on generalised Rayleigh waves have also been examined [27, 28]. The common conclusion of these studies is that the geometry of the boundaries changes the dispersive properties of the waveguide. Much less is known when nonlinearity is also present. Nonlinear surface SH waves in a half-space covered by an irregular layer obey an evolution equation with variable coefficients [29]. The focusing of strain solitary waves in a narrowing rod has been demonstrated experimentally and explained by the variation of the coefficients of the governing equation along the rod [8]. The nonlinear S0 Lamb mode in a plate of linearly varying thickness [30] and solitary waves in rods made of an inhomogeneous hyperelastic material [31] show the same sensitivity to longitudinal variations of the waveguide.

When the properties of a waveguide vary slowly compared with the wavelength, the natural long-wave model is a KdV equation with variable coefficients. Its theory has been developed mainly for water waves and internal waves over variable topography. A solitary wave in a slowly varying medium adjusts adiabatically: its amplitude follows from the conservation of the wave action, or energy flux, and a trailing shelf of small amplitude is generated behind it. If the medium changes rapidly, or too much, the solitary wave fissions into several solitary waves together with a dispersive tail [32, 33]. Whether elevation or depression waves emerge is decided by the sign of the nonlinear coefficient, which may change along the path of propagation [34, 35]. The variable-coefficient KdV equation reduces to forms known to be integrable only for particular choices of its coefficients; the Painlevé analysis yields conditions on the coefficients that are satisfied, for instance, by the constant-coefficient and cylindrical KdV equations [36, 37, 38]. A layer of variable thickness is the elastic counterpart of fluid flow over variable topography. In the solid, however, the coefficients of the evolution equation are not free parameters. They are fixed by the finite-elastic constitutive law and by the traction-free condition on the irregular free surfaces, and how the geometry enters each coefficient has to be found from the boundary-value problem itself.

The purpose of the present paper is to answer this question for long extensional waves in a compressible hyperelastic layer whose traction-free surfaces are irregular and symmetric with respect to the middle plane. The thickness is allowed to change by a finite amount on a scale long compared with the wavelength. We adopt a distinguished limit in which weak nonlinearity, dispersion and the slow variation of the thickness appear at the same order. The method of multiple scales applied to the full two-dimensional problem of finite elasticity then gives a KdV equation with variable coefficients for the longitudinal strain. The analysis shows that the irregularity enters this equation in only two ways. The first is a linear term that expresses the conservation of the energy flux through a cross-section. The second is a dispersive coefficient proportional to the square of the local thickness. The nonlinear coefficient is that of the uniaxial response of a layer with free surfaces, and the dispersive coefficient depends on Poisson’s ratio alone. After normalisation, a single variable coefficient governs the polarity of the solitary waves. For special choices of the surface irregularity, namely a uniform layer and a cube-root thickness profile, the equation reduces to forms known to be integrable, the KdV and the cylindrical KdV equations. Bounded irregularities do not lead to such forms, and the evolution is then computed numerically. The evolution of localized pulses in periodically irregular layers of iron and copper, and of model materials, is computed and interpreted in terms of the material and of the geometry.

The paper is organised as follows. The boundary-value problem is formulated in § 3. The asymptotic analysis leading to the variable-coefficient KdV equation, and the mechanical interpretation of its coefficients, are given in § 4. § 5 deals with the normalised equation, the polarity of the solitary waves, integrability and the numerical results. Conclusions and possible extensions are given in § 6. The stress components and the operators of the perturbation problems are listed in the Appendices.

Formulation of the problem

In this section the boundary-value problem governing the plane motion of an elastic layer with irregular free surfaces is set up within the theory of finite elasticity. The geometry of the layer, its motion and the balance equations are introduced first, in the material description. The constitutive equations appropriate to deformations of small but finite amplitude are given next. Finally, the problem is written in dimensionless form, and the small parameters that characterise long waves, weak nonlinearity and the slow variation of the thickness are identified.

Geometry, motion and balance equations

Consider a homogeneous, isotropic and compressible hyperelastic layer referred to a fixed rectangular Cartesian frame. The material coordinates of a particle are denoted by XK=(X,Y,Z)X_K=(X,Y,Z) and its spatial coordinates at time tt by xkx_k. In the undeformed configuration the layer occupies the region

B0={(X,Y,Z):−H(X)<Y<H(X),−∞<X,Z<∞},(1)\mathcal{B}_0=\{(X,Y,Z):-H(X)<Y<H(X),\quad-\infty<X,Z<\infty\}, \tag*{(1)}

bounded by the two material surfaces

S0±:Y=±H(X),H(X)=h+af1.(2)\mathcal{S}_0^{\pm}: \quad Y=\pm H(X),\qquad H(X)=h+af_1. \tag*{(2)}

Here 2h2h is the thickness of the corresponding layer with plane free surfaces, aa is the amplitude of the surface irregularity and f1f_1 is a bounded, sufficiently smooth function of XX whose longitudinal scale is specified in Section 2.3. The two free surfaces are mirror images of each other with respect to the middle plane Y=0Y=0. This restriction is essential for what follows: it allows the symmetric (extensional) mode of the uniform layer to be continued into a waveguide whose local thickness 2H(X)2H(X) varies along the direction of propagation.

The motion is restricted to the (X,Y)(X,Y)-plane,

x1=X+u1(X,Y,t),x2=Y+u2(X,Y,t),x3=Z,(3)x_1=X+u_1(X,Y,t),\qquad x_2=Y+u_2(X,Y,t),\qquad x_3=Z, \tag*{(3)}

where uku_k are the components of the displacement vector. In the absence of body forces, the balance of linear momentum referred to the undeformed configuration reads [39]

TKk,K=ρ0u¨k,k=1,2,(4)T_{Kk,K}=\rho_0\ddot{u}_k,\qquad k=1,2, \tag*{(4)}

where TKkT_{Kk} is the first Piola–Kirchhoff stress tensor, ρ0\rho_0 is the mass density in the undeformed configuration, a comma followed by a majuscule index denotes partial differentiation with respect to the corresponding material coordinate, a superposed dot denotes the material time derivative, and summation over repeated indices is understood.

The surfaces S0±\mathcal{S}_0^{\pm} are free of traction. Denoting by NK±N_K^{\pm} the components of the exterior unit normals of S0±\mathcal{S}_0^{\pm},

(N1±,N2±)=(∓H′(X),±1)1+H′(X)2,(5)(N_1^{\pm},N_2^{\pm})=\frac{(\mp H'(X),\pm1)}{\sqrt{1+H'(X)^2}}, \tag*{(5)}

the boundary conditions are [39]

NK±TKk=0,k=1,2,on S0±.(6)N_K^{\pm}T_{Kk}=0,\qquad k=1,2,\qquad\text{on }\mathcal{S}_0^{\pm}. \tag*{(6)}

In contrast to a layer of uniform thickness, (6) involves a longitudinally varying normal. The irregular geometry therefore enters the boundary-value problem through the traction vector itself, without any external surface loading, and this is the only mechanism by which the variable thickness acts on the wave field.

Constitutive equations

The balance equations (4) and the boundary conditions (6) hold for any material; the mechanical response of the layer is specified by the constitutive equations. For a hyperelastic material, the stress is derived from a strain energy function Σ\Sigma, measured per unit volume of the undeformed configuration, so that [39]

TKk=∂Σ∂xk,K.(7)T_{Kk}=\frac{\partial\Sigma}{\partial x_{k,K}}. \tag*{(7)}

For an isotropic solid, Σ\Sigma depends on the deformation through the principal invariants IEI_E, IIEII_E and IIIEIII_E of the Lagrangian strain tensor

EKL=12(uK,L+uL,K+uM,KuM,L),uK=ukδkK.(8)E_{KL} = \frac{1}{2}\left(u_{K,L} + u_{L,K} + u_{M,K}u_{M,L}\right), \qquad u_K = u_k\delta_{kK}. \tag*{(8)}

We consider deformations of small but finite amplitude and accordingly expand Σ\Sigma in powers of the strain invariants IEI_E, IIEII_E and IIIEIII_E, retaining all terms up to the third degree in the strain components [39, 10]. The resulting expression involves the Lamé constants λ\lambda and μ\mu and the third-order elastic constants ll, mm and nn. Since IIIEIII_E vanishes identically for the plane motion (3), the constant nn does not enter the problem.

Substitution of (8) and of the strain energy function described above into (7) gives the stress components as polynomials in the displacement gradients. Consistently with the degree of the strain energy function, they are truncated at the second degree; the resulting expressions are listed in Appendix § 11. They involve the two linear wave speeds

cL2=λ+2μρ0,cT2=μρ0,(9)c_L^2 = \frac{\lambda+ 2\mu}{\rho_0}, \qquad c_T^2 = \frac{\mu}{\rho_0}, \tag*{(9)}

and four nonlinear material coefficients

ηi=βiρ0,i=1,…,4,(10)\eta_i = \frac{\beta_i}{\rho_0}, \qquad i = 1,\ldots,4, \tag*{(10)}

with

β1=32λ+3μ+3l,β2=12λ+3l+m,β3=μ−m2,β4=λ2+μ−m4.(11)\beta_1 = \frac{3}{2}\lambda+ 3\mu+ 3l, \qquad\beta_2 = \frac{1}{2}\lambda+ 3l + m, \qquad\beta_3 = \mu- \frac{m}{2}, \qquad\beta_4 = \frac{\lambda}{2} + \mu- \frac{m}{4}. \tag*{(11)}

Each βi\beta_i combines a contribution of the geometric nonlinearity, carried by the Lamé constants, with a contribution of the physical nonlinearity, carried by the third-order constants. The coefficients ηi\eta_i have the dimension of a squared velocity. It will be seen that the linear constants and the nonlinear coefficients enter the long-wave model in clearly separated ways.

Dimensionless boundary-value problem

Equations (4) and (6), together with the stress components of Appendix § 11, constitute a nonlinear boundary-value problem for u1u_1 and u2u_2 in a domain of variable width. Its asymptotic treatment requires a scaling that distinguishes the direction of propagation from the transverse direction. Let LL be a characteristic wavelength in the direction of propagation and cc a characteristic velocity, to be fixed in the course of the analysis. We introduce the dimensionless quantities

x=XL,y=Yh,τ=ctL,u=u1L,v=u2h,A=ah,δ=hL.(12)x = \frac{X}{L}, \qquad y = \frac{Y}{h}, \qquad\tau= \frac{ct}{L}, \qquad u = \frac{u_1}{L}, \qquad v = \frac{u_2}{h}, \qquad A = \frac{a}{h}, \qquad\delta= \frac{h}{L}. \tag*{(12)}

The longitudinal and transverse coordinates are scaled differently, and this anisotropy must be kept when the displacement gradients are transformed:

u1,X=∂u∂x,u1,Y=1δ∂u∂y,u2,X=δ∂v∂x,u2,Y=∂v∂y.(13)u_{1,X} = \frac{\partial u}{\partial x}, \qquad u_{1,Y} = \frac{1}{\delta}\frac{\partial u}{\partial y}, \qquad u_{2,X} = \delta\frac{\partial v}{\partial x}, \qquad u_{2,Y} = \frac{\partial v}{\partial y}. \tag*{(13)}

Since the dimensional displacement components u1u_1 and u2u_2 do not appear again, subscripts attached to uu and vv will from now on indicate the order of a term in a perturbation expansion. We are concerned with waves that are long compared with the thickness of the layer, δ≪1\delta\ll1, and of small but finite amplitude. Let ε≪1\varepsilon\ll1 measure the magnitude of the displacement gradients. The effects of nonlinearity and dispersion are of the same order in the distinguished limit δ2=O(ε)\delta^2 = O(\varepsilon) [10], and without loss of generality we set

δ2=ε.(14)\delta^2 = \varepsilon. \tag*{(14)}

The surface irregularity is assumed to vary slowly on the scale of the wavelength; specifically, f1f_1 depends on XX only through the slow variable

x1=εx.(15)x_1 = \varepsilon x. \tag*{(15)}

In dimensionless form, the free surfaces of the layer are y=±P(x1)y = \pm P(x_1), with

P(x1)=Hh=1+Af1(x1)>0,P′(x1)=dPdx1=Adf1dx1,(16)P(x_1) = \frac{H}{h} = 1 + A f_1(x_1) > 0, \qquad P'(x_1) = \frac{dP}{dx_1} = A \frac{df_1}{dx_1}, \tag*{(16)}

so that

∂P∂x=εP′(x1),P=O(1),P′=O(1).(17)\frac{\partial P}{\partial x} = \varepsilon P'(x_1), \qquad P = O(1), \qquad P' = O(1). \tag*{(17)}

No smallness is imposed on the amplitude AA of the irregularity: the asymptotic assumption concerns the rate at which the local thickness changes along the layer, not the magnitude of the change. The geometry can therefore modify the waveguide by a finite amount, while its effect on the wave accumulates gradually over the long propagation distance x=O(ε−1)x = O(\varepsilon^{-1}).

With (12)–(14) and the stress components of Appendix § 11, the balance equations (4) take the form

L1(u,v)=N1(u,v),L2(u,v)=N2(u,v),−P(x1)<y<P(x1),(18)\mathcal{L}_1(u,v) = \mathcal{N}_1(u,v), \qquad\mathcal{L}_2(u,v) = \mathcal{N}_2(u,v), \qquad-P(x_1) < y < P(x_1), \tag*{(18)}

where Li\mathcal{L}_i are linear and Ni\mathcal{N}_i are quadratic differential operators. The normals (5) become, up to a positive factor,

(N1±,N2±)∥(∓δ∂P∂x,±1),(19)(N_1^\pm,N_2^\pm) \parallel\left(\mp\delta\frac{\partial P}{\partial x},\pm1\right), \tag*{(19)}

so that the inclination of the free surfaces is weighted by the slenderness parameter, and the traction conditions (6) are written as

B1±(u,v)=0,B2±(u,v)=0on y=±P(x1),(20)\mathcal{B}_1^\pm(u,v) = 0, \qquad\mathcal{B}_2^\pm(u,v) = 0 \qquad\text{on } y = \pm P(x_1), \tag*{(20)}

where the subscripts 1 and 2 refer to the xx- and yy-components of the traction. The explicit forms of Li\mathcal{L}_i, Ni\mathcal{N}_i and Bi±\mathcal{B}_i^\pm are given in Appendix § 12.

Because the geometry is symmetric with respect to y=0y = 0, the boundary-value problem (18)–(20) admits solutions with

u(x,−y,τ)=u(x,y,τ),v(x,−y,τ)=−v(x,y,τ),(21)u(x,-y,\tau) = u(x,y,\tau), \qquad v(x,-y,\tau) = -v(x,y,\tau), \tag*{(21)}

which constitute the symmetric (extensional) family considered in this work. For such solutions the conditions on y=−Py = -P are satisfied whenever those on y=Py = P are, and only the latter need be imposed. In the limit A→0A \to0 the problem reduces to that of the symmetric long waves in a layer of uniform thickness.

Asymptotic analysis

The dimensionless boundary-value problem of § 3.3 is now solved asymptotically for relatively small values of ε\varepsilon, 0<ε≪10 < \varepsilon\ll1. The expansion is introduced first. The perturbation problems are then solved successively: the first two orders determine the transverse structure of the wave and its velocity of propagation, whereas the third order, at which nonlinearity, dispersion and the variation of the thickness appear simultaneously, provides the equation governing the evolution of the wave amplitude. The coefficients of this equation are discussed at the end of the section.

Multiple-scales expansion

The propagation of long waves of small but finite amplitude is studied by the method of multiple scales [40]. In addition to the slow variable (15) we write x0=xx_0 = x and regard the field quantities as functions of (x0,x1,y,τ)(x_0,x_1,y,\tau), so that

∂∂x⟶∂∂x0+ε∂∂x1.(22)\frac{\partial}{\partial x} \longrightarrow\frac{\partial}{\partial x_0} + \varepsilon\frac{\partial}{\partial x_1}. \tag*{(22)}

The displacement components are expanded as

u=∑n=1∞εnun(x0,x1,y,τ),v=∑n=1∞εnvn(x0,x1,y,τ).(23)u = \sum_{n=1}^{\infty} \varepsilon^n u_n(x_0,x_1,y,\tau), \qquad v = \sum_{n=1}^{\infty} \varepsilon^n v_n(x_0,x_1,y,\tau). \tag*{(23)}

Substituting (22) and (23) into (18)–(20) and collecting the coefficients of like powers of ε\varepsilon yields a hierarchy of boundary-value problems on the strip −P(x1)<y<P(x1)-P(x_1) < y < P(x_1), in which x1x_1 enters only as a parameter. With the operators written as in Appendix § 12, the problem of order εn\varepsilon^n determines the dependence of unu_n and vn−1v_{n-1} on yy. The operators Lnj\mathcal{L}^{nj}, Nnj\mathcal{N}^{nj} and Bnj\mathcal{B}^{nj} (j=1,2j = 1,2) appearing below are listed in Appendix § 13. In view of (21), unu_n is even and vnv_n is odd in yy, and the boundary conditions are imposed on y=P(x1)y = P(x_1).

First- and second-order problems

We begin with the two lowest orders of the hierarchy. Their solutions describe a wave that is uniform across the thickness to leading order, and their compatibility fixes the characteristic velocity cc.

O(ε)O(\varepsilon): The first-order problem is

cT2∂2u1∂y2=0,cT2∂u1∂y=0on y=P(x1),(24)c_T^2 \frac{\partial^2 u_1}{\partial y^2}=0,\qquad c_T^2 \frac{\partial u_1}{\partial y}=0 \quad\text{on } y=P(x_1), \tag*{(24)}

whose even solution is

u1=A∗(x0,x1,τ),(25)u_1=A^*(x_0,x_1,\tau), \tag*{(25)}

where A∗A^* is an as yet arbitrary function.

O(ε2)O(\varepsilon^2): The second-order problem is

−cT2∂2u2∂y2+L21(u1,v1)=N21(u1,v1),(26)-c_T^2\frac{\partial^2 u_2}{\partial y^2}+\mathcal{L}^{21}(u_1,v_1)=\mathcal{N}^{21}(u_1,v_1), \tag*{(26)}
L22(u1,v1)=N22(u1,v1),(27)\mathcal{L}^{22}(u_1,v_1)=\mathcal{N}^{22}(u_1,v_1), \tag*{(27)}

with

cT2∂u2∂y+B21(u1,v1)=0,(28)c_T^2\frac{\partial u_2}{\partial y}+\mathcal{B}^{21}(u_1,v_1)=0, \tag*{(28)}
B22(u1,v1)=0(29)\mathcal{B}^{22}(u_1,v_1)=0 \tag*{(29)}

on y=P(x1)y=P(x_1). By (25), equation (27) reduces to cL2∂2v1/∂y2=0c_L^2\partial^2v_1/\partial y^2=0, whose odd solution is v1=γ1(x0,x1,τ)yv_1=\gamma_1(x_0,x_1,\tau)y. The condition (29) then gives cL2γ1+(cL2−2cT2)∂A∗/∂x0=0c_L^2\gamma_1+(c_L^2-2c_T^2)\partial A^*/\partial x_0=0, that is,

v1=(2cT2cL2−1)∂A∗∂x0y.(30)v_1=\left(\frac{2c_T^2}{c_L^2}-1\right)\frac{\partial A^*}{\partial x_0}y. \tag*{(30)}

Thus, to leading order, each cross-section contracts or expands uniformly in the transverse direction in proportion to the local longitudinal strain, in such a way that the normal traction on the free surfaces vanishes.

Inserting (25) and (30) into (26) we obtain

∂2u2∂y2=V∗(x0,x1,τ),V∗=c2cT2∂2A∗∂τ2+(2cT2cL2−3)∂2A∗∂x02,(31)\frac{\partial^2u_2}{\partial y^2}=V^*(x_0,x_1,\tau),\qquad V^*=\frac{c^2}{c_T^2}\frac{\partial^2A^*}{\partial\tau^2}+\left(\frac{2c_T^2}{c_L^2}-3\right)\frac{\partial^2A^*}{\partial x_0^2}, \tag*{(31)}

and hence the even solution

u2=V∗(x0,x1,τ)y22+B1(x0,x1,τ),(32)u_2=V^*(x_0,x_1,\tau)\frac{y^2}{2}+B_1(x_0,x_1,\tau), \tag*{(32)}

where B1B_1 is a new arbitrary function. The remaining boundary condition (28) becomes

P(x1)cT2{c2cT2∂2A∗∂τ2−4(1−cT2cL2)∂2A∗∂x02}=0.(33)P(x_1)c_T^2\left\{\frac{c^2}{c_T^2}\frac{\partial^2A^*}{\partial\tau^2}-4\left(1-\frac{c_T^2}{c_L^2}\right)\frac{\partial^2A^*}{\partial x_0^2}\right\}=0. \tag*{(33)}

Since P>0P>0, the expression in braces must vanish. This is the solvability condition of the second-order problem. We now fix the characteristic velocity as

c2=4cT2(1−cT2cL2)=E(1−ν2)ρ0(34)c^2=4c_T^2\left(1-\frac{c_T^2}{c_L^2}\right)=\frac{E}{(1-\nu^2)\rho_0} \tag*{(34)}

where EE is Young’s modulus and ν\nu is Poisson’s ratio; cc is the velocity of long extensional waves in a plate [39]. The solvability condition then reduces to

∂2A∗∂τ2−∂2A∗∂x02=0,(35)\frac{\partial^2 A^*}{\partial\tau^2}-\frac{\partial^2 A^*}{\partial x_0^2}=0, \tag*{(35)}

and for a wave travelling in the positive x0x_0-direction we take

A∗=A∗(η,ξ),η=τ−x0,ξ=x1.(36)A^*=A^*(\eta,\xi), \qquad\eta=\tau-x_0,\qquad\xi=x_1. \tag*{(36)}

It is worth noting that the local thickness PP factors out of (33): the velocity of the long wave is not affected by the surface irregularity, whose influence is deferred to the next order. The general forms (25)–(32), in which the derivatives with respect to τ\tau and x0x_0 are kept separate, are used in the following subsection; the relations (34) and (36) are imposed only in the final solvability condition.

Third-order problem

The amplitude A∗A^* is left undetermined by the first two orders apart from its dependence on τ−x0\tau-x_0. Its dependence on the slow variable x1x_1 follows from the next order, to which we now turn.

O(ε3)O(\varepsilon^3): The third-order problem is

−cT2∂2u3∂y2+L31(u1,u2,v1,v2)=N31(u1,u2,v1,v2),(37)-c_T^2\frac{\partial^2 u_3}{\partial y^2}+\mathcal{L}^{31}(u_1,u_2,v_1,v_2)=\mathcal{N}^{31}(u_1,u_2,v_1,v_2), \tag*{(37)}
L32(u1,u2,v1,v2)=N32(u1,u2,v1,v2),(38)\mathcal{L}^{32}(u_1,u_2,v_1,v_2)=\mathcal{N}^{32}(u_1,u_2,v_1,v_2), \tag*{(38)}

with

cT2∂u3∂y+B31(u1,u2,v1,v2)=0,(39)c_T^2\frac{\partial u_3}{\partial y}+\mathcal{B}^{31}(u_1,u_2,v_1,v_2)=0, \tag*{(39)}
B32(u1,u2,v1,v2)=0(40)\mathcal{B}^{32}(u_1,u_2,v_1,v_2)=0 \tag*{(40)}

on y=P(x1)y=P(x_1). Using the lower-order solutions in (38) we find

∂2v2∂y2=χ(x0,x1,τ)y,χ=cT2cL2(2cT2cL2−cL2cT2)∂3A∗∂x0∂τ2+(3−4cT2cL2)∂3A∗∂x03,(41)\frac{\partial^2 v_2}{\partial y^2}=\chi(x_0,x_1,\tau)y,\qquad \chi=\frac{c_T^2}{c_L^2}\left(\frac{2c_T^2}{c_L^2}-\frac{c_L^2}{c_T^2}\right)\frac{\partial^3 A^*}{\partial x_0\partial\tau^2} +\left(3-\frac{4c_T^2}{c_L^2}\right)\frac{\partial^3 A^*}{\partial x_0^3}, \tag*{(41)}

and therefore the odd solution

v2=χ(x0,x1,τ)y36+K1(x0,x1,τ)y.(42)v_2=\chi(x_0,x_1,\tau)\frac{y^3}{6}+K_1(x_0,x_1,\tau)y. \tag*{(42)}

The function K1K_1 is determined by the boundary condition (40):

K1=P22(1−cT2cL2){2c2cL2∂3A∗∂x0∂τ2−4cT2cL2∂3A∗∂x03}−1cL2{η1(2cT2cL2−1)2+2η2(2cT2cL2−1)+η2}{(∂A∗∂x0)2+(2cT2cL2−1)(∂A∗∂x1+∂B1∂x0)}.(43)K_1=\frac{P^2}{2}\left(1-\frac{c_T^2}{c_L^2}\right)\left\{\frac{2c^2}{c_L^2}\frac{\partial^3 A^*}{\partial x_0\partial\tau^2}-\frac{4c_T^2}{c_L^2}\frac{\partial^3 A^*}{\partial x_0^3}\right\} -\frac{1}{c_L^2}\left\{\eta_1\left(\frac{2c_T^2}{c_L^2}-1\right)^2+2\eta_2\left(\frac{2c_T^2}{c_L^2}-1\right)+\eta_2\right\}\left\{\left(\frac{\partial A^*}{\partial x_0}\right)^2+\left(\frac{2c_T^2}{c_L^2}-1\right)\left(\frac{\partial A^*}{\partial x_1}+\frac{\partial B_1}{\partial x_0}\right)\right\}. \tag*{(43)}

The three groups of terms in (43) represent, respectively, the correction of the transverse contraction due to the non-uniform distribution of the displacement across the thickness, the second-order (nonlinear) part of the transverse contraction, and the contraction associated with the slow modulation and with the second-order longitudinal displacement.

Next, the solutions for u1u_1, v1v_1, u2u_2 and v2v_2 are substituted into (37). This gives

∂2u3∂y2=y22[cL4cT4∂4A∗∂τ4−2cC2(2cL4−cT4)cL4cT2∂4A∗∂x02∂τ2+(5−4cT2cL2)∂4A∗∂x04]+(4cT2cL2−6)∂2A∗∂x0∂x1+P2(cL2−cT2)2cL4cT2(−cT2∂4A∗∂x02∂τ2+2cT2∂4A∗∂x04)\begin{aligned} \frac{\partial^2 u_3}{\partial y^2}=\frac{y^2}{2}\left[\frac{c_L^4}{c_T^4}\frac{\partial^4 A^*}{\partial\tau^4}-\frac{2c_C^2(2c_L^4-c_T^4)}{c_L^4c_T^2}\frac{\partial^4 A^*}{\partial x_0^2\partial\tau^2}+\left(5-\frac{4c_T^2}{c_L^2}\right)\frac{\partial^4 A^*}{\partial x_0^4}\right] \\ &+\left(\frac{4c_T^2}{c_L^2}-6\right)\frac{\partial^2 A^*}{\partial x_0\partial x_1}+P^2\frac{(c_L^2-c_T^2)^2}{c_L^4c_T^2}\left(-c_T^2\frac{\partial^4 A^*}{\partial x_0^2\partial\tau^2}+2c_T^2\frac{\partial^4 A^*}{\partial x_0^4}\right) \end{aligned}
−4cT2η4cL2cT2∂A∗∂x0∂2A∗∂τ2+Z∗∂A∗∂x0∂2A∗∂x02+c2cT2∂2B1∂τ2+(2cT2cL2−3)∂2B1∂x02.(44)-\frac{4c_T^2\eta_4}{c_L^2c_T^2}\frac{\partial A^*}{\partial x_0}\frac{\partial^2 A^*}{\partial\tau^2}+Z^*\frac{\partial A^*}{\partial x_0}\frac{\partial^2 A^*}{\partial x_0^2}+\frac{c^2}{c_T^2}\frac{\partial^2 B_1}{\partial\tau^2}+\left(\frac{2c_T^2}{c_L^2}-3\right)\frac{\partial^2 B_1}{\partial x_0^2}. \tag*{(44)}

where the coefficient of the remaining quadratic terms is

Z∗=2η1(8cL2cT2−5cL4−4cT4)cL6+2η2(5cL2−8cT2)cL4+2η3(cL2−2cT2)cL4+4η4(3cL2−2cT2)cL4.(45)Z^*=\frac{2\eta_1(8c_L^2c_T^2-5c_L^4-4c_T^4)}{c_L^6}+\frac{2\eta_2(5c_L^2-8c_T^2)}{c_L^4}+\frac{2\eta_3(c_L^2-2c_T^2)}{c_L^4}+\frac{4\eta_4(3c_L^2-2c_T^2)}{c_L^4}. \tag*{(45)}

An explicit expression for u3u_3 is not needed. Since u3u_3 is even in yy, ∂u3/∂y\partial u_3/\partial y vanishes on y=0y=0; integrating (44) with respect to yy from 0 to P(x1)P(x_1) and eliminating ∂u3/∂y\partial u_3/\partial y on y=Py=P by means of the boundary condition (39), we obtain the solvability condition of the third-order problem,

P[c2cT2∂2B1∂τ2+4(cT2−cL2)cL2∂2B1∂x02+8(cT2cL2−1)∂2A∗∂x0∂x1+−16cL4η1+24cL2cT2(η1−η2)+12cL4(η2−η1)cL6∂A∗∂x0∂2A∗∂x02]+P3{cL46cT4∂4A∗∂τ4−c2(11cL4−18cL2cT2+8cT4)6cL4cT2∂4A∗∂x02∂τ2+2(cL2−cT2)(5cL2−6cT2)3cL4∂4A∗∂x04}+4(cT2cL2−1)P′∂A∗∂x0=0.(46)\begin{aligned} P\left[\frac{c^2}{c_T^2}\frac{\partial^2 B_1}{\partial\tau^2}+\frac{4(c_T^2-c_L^2)}{c_L^2}\frac{\partial^2 B_1}{\partial x_0^2}+8\left(\frac{c_T^2}{c_L^2}-1\right)\frac{\partial^2 A^*}{\partial x_0\partial x_1}+\frac{-16c_L^4\eta_1+24c_L^2c_T^2(\eta_1-\eta_2)+12c_L^4(\eta_2-\eta_1)}{c_L^6}\frac{\partial A^*}{\partial x_0}\frac{\partial^2 A^*}{\partial x_0^2}\right] \\ &+P^3\left\{\frac{c_L^4}{6c_T^4}\frac{\partial^4 A^*}{\partial\tau^4}-\frac{c^2(11c_L^4-18c_L^2c_T^2+8c_T^4)}{6c_L^4c_T^2}\frac{\partial^4 A^*}{\partial x_0^2\partial\tau^2}+\frac{2(c_L^2-c_T^2)(5c_L^2-6c_T^2)}{3c_L^4}\frac{\partial^4 A^*}{\partial x_0^4}\right\} \\ &+4\left(\frac{c_T^2}{c_L^2}-1\right)P'\frac{\partial A^*}{\partial x_0}=0. \tag*{(46)} \end{aligned}

The coefficients η3\eta_3 and η4\eta_4 do not appear in (46): the terms proportional to them in (44) are balanced exactly by the corresponding terms of B13B_1^3. The last term of (46) is the only one in which the inclination of the free surfaces enters; it originates from the term of B13B_1^3 that is proportional to P′P'.

Variable-coefficient Korteweg–de Vries equation

The solvability condition (46) still contains the two unknown functions A∗A^* and B1B_1. They are separated by requiring the expansion to remain uniformly valid, which leads to a single evolution equation for A∗A^*. We now divide (46) by PP and use (34) and (36). Since ∂/∂τ=∂/∂η\partial/\partial\tau=\partial/\partial\eta and ∂/∂x0=−∂/∂η\partial/\partial x_0=-\partial/\partial\eta when acting on A∗A^*, the result is

c2cT2(∂2B1∂τ2−∂2B1∂x02)=8(cT2cL2−1)∂2A∗∂ξ∂η+4(cT2cL2−1)P′P∂A∗∂η+4(cL2−cT2)(cL2−2cT2)23cL6P2∂4A∗∂η4\begin{aligned} \frac{c^2}{c_T^2}\left(\frac{\partial^2 B_1}{\partial\tau^2}-\frac{\partial^2 B_1}{\partial x_0^2}\right)=8\left(\frac{c_T^2}{c_L^2}-1\right)\frac{\partial^2 A^*}{\partial\xi\partial\eta}+4\left(\frac{c_T^2}{c_L^2}-1\right)\frac{P'}{P}\frac{\partial A^*}{\partial\eta} \\ &+\frac{4(c_L^2-c_T^2)(c_L^2-2c_T^2)^2}{3c_L^6}P^2\frac{\partial^4 A^*}{\partial\eta^4} \end{aligned}
−4cL6[cT4η1+3cL4(η1−η2)+6cL2cT2(η2−η1)]∂A∗∂η∂2A∗∂η2.(47)-\frac{4}{c_L^6}\left[c_T^4\eta_1+3c_L^4(\eta_1-\eta_2)+6c_L^2c_T^2(\eta_2-\eta_1)\right]\frac{\partial A^*}{\partial\eta}\frac{\partial^2 A^*}{\partial\eta^2}. \tag*{(47)}

The right-hand side of (47) depends on x0x_0 and τ\tau only through η=τ−x0\eta=\tau-x_0 and is therefore a solution of the homogeneous wave equation on the left. Unless it vanishes, B1B_1 grows linearly with τ+x0\tau+x_0 and the expansion (23) ceases to be uniformly valid. Setting the right-hand side equal to zero, dividing by 8(cT2/cL2−1)8(c_T^2/c_L^2-1) and introducing

ϕ(η,ξ)=∂A∗∂η,(48)\phi(\eta,\xi)=\frac{\partial A^*}{\partial\eta}, \tag*{(48)}

we arrive at the variable-coefficient Korteweg–de Vries (KdV) equation

∂ϕ∂ξ+P′2Pϕ+Λ1ϕ∂ϕ∂η+Λ2P2∂3ϕ∂η3=0,(49)\frac{\partial\phi}{\partial\xi}+\frac{P'}{2P}\phi+\Lambda_1\phi\frac{\partial\phi}{\partial\eta}+\Lambda_2P^2\frac{\partial^3\phi}{\partial\eta^3}=0, \tag*{(49)}

with the constant coefficients

Λ1=4cT4η1+3cL4(η1−η2)+6cL2cT2(η2−η1)2cL4(cL2−cT2),(50)\Lambda_1=\frac{4c_T^4\eta_1+3c_L^4(\eta_1-\eta_2)+6c_L^2c_T^2(\eta_2-\eta_1)}{2c_L^4(c_L^2-c_T^2)}, \tag*{(50)}
Λ2=−(cL2−2cT2)26cL4.(51)\Lambda_2=-\frac{(c_L^2-2c_T^2)^2}{6c_L^4}. \tag*{(51)}

Because ∂u/∂x=−ϵϕ+O(ϵ2)\partial u/\partial x=-\epsilon\phi+O(\epsilon^2), the function −ϵϕ-\epsilon\phi is the leading-order longitudinal strain; positive values of ϕ\phi correspond to compression. Equation (49) describes the evolution of this strain along the layer, in a frame moving with the velocity cc. The surface irregularity enters in two ways: through the linear term proportional to P′/PP'/P, and through the factor P2P^2 multiplying the dispersive term. For P≡1P\equiv1 the classical KdV equation for the symmetric long waves in a uniform layer is recovered [10].

The linear geometric term is removed by the transformation

ϕ(η,ξ)=ψ(η,ξ)P(ξ),(52)\phi(\eta,\xi)=\frac{\psi(\eta,\xi)}{\sqrt{P(\xi)}}, \tag*{(52)}

which reduces (49) to

∂ψ∂ξ+Λ1Pψ∂ψ∂η+Λ2P2∂3ψ∂η3=0.(53)\frac{\partial\psi}{\partial\xi}+\frac{\Lambda_1}{\sqrt{P}}\psi\frac{\partial\psi}{\partial\eta}+\Lambda_2P^2\frac{\partial^3\psi}{\partial\eta^3}=0. \tag*{(53)}

Before turning to the solutions of (53), we examine its coefficients. The three coefficients of (49) admit simple mechanical interpretations, which also provide independent checks of the asymptotic analysis. Let

q=1−2cT2cL2=λλ+2μ=ν1−ν(54)q=1-\frac{2c_T^2}{c_L^2}=\frac{\lambda}{\lambda+2\mu}=\frac{\nu}{1-\nu} \tag*{(54)}

denote the ratio of transverse contraction to longitudinal extension in a layer with traction-free surfaces under plane strain; by (30), ∂v1/∂y=−q ∂u1/∂x0\partial v_1/\partial y=-q\,\partial u_1/\partial x_0, and (34) is equivalent to c2=cL2(1−q2)c^2=c_L^2(1-q^2).

Geometric term. Multiplying (49) by 2Pϕ2P\phi and integrating with respect to η\eta, for ϕ\phi vanishing sufficiently fast as ∣η∣→∞|\eta|\to\infty, we find

ddξ∫−∞∞Pϕ2 dη=ddξ∫−∞∞ψ2 dη=0.(55)\frac{d}{d\xi}\int_{-\infty}^{\infty}P\phi^2\,d\eta=\frac{d}{d\xi}\int_{-\infty}^{\infty}\psi^2\,d\eta=0. \tag*{(55)}

To leading order, the energy carried by the wave through a cross-section is proportional to the thickness times the square of the strain; (55) states that this quantity is conserved, and the term P′ϕ/(2P)P'\phi/(2P) is precisely what this conservation requires. A wave entering a thinner part of the layer is thereby amplified, in proportion to P−1/2P^{-1/2} as long as nonlinearity and dispersion are negligible.

Nonlinear coefficient. In terms of qq, (50) takes the form

Λ1=Γc2,Γ=η1(1−q3)−3η2q(1−q).(56)\Lambda_1=\frac{\Gamma}{c^2},\qquad\Gamma=\eta_1(1-q^3)-3\eta_2q(1-q). \tag*{(56)}

The quantity Γ\Gamma is the coefficient of the quadratic term in the uniaxial response of the layer: for a homogeneous deformation u1,X=eu_{1,X}=e with traction-free surfaces, the stress components of Appendix A give T11/ρ0=c2e+Γe2+O(e3)T_{11}/\rho_0=c^2e+\Gamma e^2+O(e^3). The nonlinear coefficient is thus independent of the local thickness and involves the third-order elastic constants only through η1\eta_1 and η2\eta_2.

Dispersive coefficient. Similarly, (51) may be written as

Λ2=−q26=−16(1−c2cL2)=−ν26(1−ν)2.(57)\Lambda_2=-\frac{q^2}{6}=-\frac{1}{6}\left(1-\frac{c^2}{c_L^2}\right)=-\frac{\nu^2}{6(1-\nu)^2}. \tag*{(57)}

It depends on the linear elastic constants only, is negative for every material with ν≠0\nu\ne0, and vanishes for ν=0\nu= 0, in which case dispersion does not appear at this order and the present scaling ceases to be appropriate. For P≡1P \equiv1 and infinitesimal amplitudes, (49) admits the plane-wave solutions ϕ∝exp⁡[i(Kη+Λ2K3ξ)]\phi\propto\exp[i(K\eta+\Lambda_2K^3\xi)], where KK is a dimensionless wavenumber. Returning to the dimensional variables by means of (12) and (15), we obtain the frequency ω=cK/L\omega=cK/L and the wavenumber k=K(1−εΛ2K2)/Lk=K(1-\varepsilon\Lambda_2K^2)/L. By (14), ε=h2/L2\varepsilon=h^2/L^2, so that the correction εK2\varepsilon K^2 equals (kh)2(kh)^2 to leading order, hh being the half-thickness of the layer. Elimination of KK then gives the dispersion relation

ω=ck[1−16q2(kh)2],\omega=ck\left[1-\frac{1}{6}q^2(kh)^2\right],

which coincides with the first two terms of the long-wave expansion of the fundamental symmetric branch of the Rayleigh–Lamb frequency equation [39]. In the irregular layer the dispersive term is multiplied by P2P^2, that is, by the square of the local thickness.

Soliton-type wave evolution in the variable-coefficient KdV equation

The reduced equation (53) is now used to study the evolution of localized long waves in the irregular layer. It is first brought to a normalised form in which the dispersive coefficient is unity and the whole influence of the material and of the geometry is carried by a single variable coefficient. The sign of this coefficient, which decides the polarity of the solitary waves, is discussed next, and its reduction to forms known to be integrable is then examined. The formulation of the numerical problem, the numerical method and its resolution tests, and the numerical results follow.

Normalised evolution equation

By (57), Λ2<0\Lambda_2 < 0 for every material with ν≠0\nu\ne0. Since P>0P > 0, the function

θ(ξ)=∣Λ2∣∫0ξP2(s) ds(58)\theta(\xi)=\lvert\Lambda_2\rvert\int_0^\xi P^2(s)\,\mathrm{d}s \tag*{(58)}

is strictly increasing and may be used as an evolution variable in place of ξ\xi. Together with θ\theta we introduce the coordinate

ζ=−η=x0−τ,(59)\zeta=-\eta=x_0-\tau, \tag*{(59)}

which measures the distance ahead of the frame moving with the velocity cc. Regarding ψ\psi as a function of (ζ,θ)(\zeta,\theta) and dividing (53) by ∣Λ2∣P2\lvert\Lambda_2\rvert P^2, we obtain

∂ψ∂θ+N(θ)ψ∂ψ∂ζ+∂3ψ∂ζ3=0,N(θ)=N0[P(ξ(θ))]−5/2,N0=Λ1Λ2.(60)\frac{\partial\psi}{\partial\theta}+\mathcal{N}(\theta)\psi\frac{\partial\psi}{\partial\zeta}+\frac{\partial^3\psi}{\partial\zeta^3}=0,\qquad\mathcal{N}(\theta)=\mathcal{N}_0\left[P(\xi(\theta))\right]^{-5/2},\qquad\mathcal{N}_0=\frac{\Lambda_1}{\Lambda_2}. \tag*{(60)}

The change of sign of both independent variables leaves the form of the equation unaltered and ensures that increasing θ\theta corresponds to increasing distance of propagation. In (60) the dispersive coefficient is unity; the geometry enters through the factor P−5/2P^{-5/2} of the nonlinear coefficient, and the material through the single number N0\mathcal{N}_0. This normalisation is preferred to the equivalent one in which the nonlinear coefficient is reduced to a constant, because Λ2\Lambda_2 has a fixed sign and does not vanish, whereas Λ1\Lambda_1 may have either sign and may vanish, as shown below.

Two consequences of (60) are noted. First, in the normalised variables infinitesimal disturbances satisfy the linearised KdV equation of the uniform layer whatever the profile PP; the variation of the thickness then acts only through the amplitude factor in (52) and through the stretching (58), and any further effect of the geometry on ψ\psi is a nonlinear one. Secondly, the substitution ψ→6ψ/N0\psi\to6\psi/\sqrt{\mathcal{N}_0} replaces N0\mathcal{N}_0 by 66, so that different materials are distinguished, in the normalised variables, only by the scale of the sign of the amplitude.

Since N\mathcal{N} does not depend on ζ\zeta, for solutions decaying sufficiently fast as ∣ζ∣→∞\lvert\zeta\rvert\to\infty, or periodic in ζ\zeta, equation (60) possesses the invariants

I1=∫ψ dζ,I2=∫ψ2 dζ,(61)I_1=\int\psi\,\mathrm{d}\zeta,\qquad I_2=\int\psi^2\,\mathrm{d}\zeta, \tag*{(61)}

the second of which is the energy integral (55).

Sign of the nonlinear coefficient and polarity of solitary waves

The nonlinear coefficient of (60) keeps the sign of N0\mathcal{N}_0 along the whole layer: the irregularity changes the magnitude of N\mathcal{N}, but cannot reverse its sign as long as P>0P>0. From (56) and (57),

N0=−6Γq2c2=−3F(cL2−cT2)(cL2−2cT2)2,F=(3cL4−6cL2cT2+4cT4)η1−3cL2(cL2−2cT2)η2,(62)\mathcal{N}_0=-\frac{6\Gamma}{q^2c^2}=-\frac{3\mathcal{F}}{(c_L^2-c_T^2)(c_L^2-2c_T^2)^2},\qquad\mathcal{F}=(3c_L^4-6c_L^2c_T^2+4c_T^4)\eta_1-3c_L^2(c_L^2-2c_T^2)\eta_2, \tag*{(62)}

so that sgn⁡N0=−sgn⁡Γ=−sgn⁡F\operatorname{sgn}\mathcal{N}_0=-\operatorname{sgn}\Gamma=-\operatorname{sgn}\mathcal{F}. The numerator involves the third-order elastic constants, whereas the denominator involves the linear constants only. The coefficient vanishes when

η2=η2,cr=3cL4−6cL2cT2+4cT43cL2(cL2−2cT2)η1,(63)\eta_2=\eta_{2,\mathrm{cr}}=\frac{3c_L^4-6c_L^2c_T^2+4c_T^4}{3c_L^2(c_L^2-2c_T^2)}\eta_1, \tag*{(63)}

and, for cL2>2cT2c_L^2>2c_T^2, that is, for ν>0\nu>0, we have N0>0\mathcal{N}_0>0 if η2>η2,cr\eta_2>\eta_{2,\mathrm{cr}} and N0<0\mathcal{N}_0<0 if η2<η2,cr\eta_2<\eta_{2,\mathrm{cr}}. At η2=η2,cr\eta_2=\eta_{2,\mathrm{cr}} the quadratic nonlinearity disappears, (60) becomes linear, and the description of nonlinear waves requires a different scaling in which higher-order terms are retained; this case is not considered here.

For a layer of uniform thickness, N≡N0\mathcal{N}\equiv\mathcal{N}_0, and (60) has the solitary-wave solution [41]

ψ(ζ,θ)=3csN0sech⁡2[cs2(ζ−ζ0−csθ)],cs>0,(64)\psi(\zeta,\theta)=\frac{3c_s}{\mathcal{N}_0}\operatorname{sech}^2\left[\sqrt{\frac{c_s}{2}}(\zeta-\zeta_0-c_s\theta)\right],\qquad c_s>0, \tag*{(64)}

where ζ0\zeta_0 is the initial position of the crest and csc_s is the velocity in the (ζ,θ)(\zeta,\theta)-plane. The solitary wave advances towards positive ζ\zeta, that is, it travels faster than cc, and its amplitude has the sign of N0\mathcal{N}_0. Recalling that positive ϕ\phi means compression, we conclude that a material with Γ<0\Gamma<0, which by (56) is stiffer in compression than in extension, supports compressive solitary waves, whereas a material with Γ>0\Gamma>0 supports tensile ones. A pulse whose polarity is opposite to that of (64) does not evolve into solitary waves but disperses into an oscillatory wave train [41]. When PP varies, (64) is no longer an exact solution; it will serve as the initial condition and as the reference state in the numerical experiments.

Integrable cases

It is natural to ask whether, for some profiles of the free surfaces, (60) reduces to an equation known to be integrable. For equations of the form ψθ+f(θ)ψψζ+g(θ)ψζζζ=0\psi_\theta+f(\theta)\psi\psi_\zeta+g(\theta)\psi_{\zeta\zeta\zeta}=0, the Painlevé analysis leads to the integrability condition

g=f(α∫f dθ+β),g=f\left(\alpha\int f\,\mathrm{d}\theta+\beta\right),

with constants α\alpha and β\beta [37, 36, 38]. For g=1g=1 and f=Nf=\mathcal{N} this condition is satisfied when N\mathcal{N} is constant, which is the uniform layer, and N∝(θ+θ0)−1/2\mathcal{N}\propto(\theta+\theta_0)^{-1/2} with a constant θ0\theta_0. The second case is made explicit by the substitution

w=N06P−5/2ψ,(65)w=\frac{\mathcal{N}_0}{6}P^{-5/2}\psi, \tag*{(65)}

which transforms (60) into

∂w∂θ+52PdPdθw+6w∂w∂ζ+∂3w∂ζ3=0.(66)\frac{\partial w}{\partial\theta}+\frac{5}{2P}\frac{\mathrm{d}P}{\mathrm{d}\theta}w+6w\frac{\partial w}{\partial\zeta}+\frac{\partial^3w}{\partial\zeta^3}=0. \tag*{(66)}

This is the cylindrical KdV equation [36, 31, 42], with the linear term w/[2(θ+θ0)]w/[2(\theta+\theta_0)], if and only if P5∝θ+θ0P^5\propto\theta+\theta_0. Normalising P=1P=1 at ξ=0\xi=0 and returning to the physical variable by means of (58), we find

P(ξ)=(1+ξξ0)1/3,ξ0=5θ03∣Λ2∣.(67)P(\xi)=\left(1+\frac{\xi}{\xi_0}\right)^{1/3},\qquad\xi_0=\frac{5\theta_0}{3|\Lambda_2|}. \tag*{(67)}

Hence, for a layer of uniform thickness and for the monotone profile (67), (60) reduces to the KdV and to the cylindrical KdV equation, respectively, both of which are known to be integrable. In the second case the thickness increases (ξ0>0\xi_0 > 0) or decreases (ξ0<0\xi_0 < 0) as the cube root of the distance. For ∣ξ∣≪∣ξ0∣|\xi| \ll|\xi_0| this profile coincides, to first order, with a linear taper of small slope.

The profile (67) is, however, admissible in a bounded region only. For ξ0>0\xi_0 > 0 the thickness grows without bound, and the long-wave assumption underlying the present analysis, which requires εP2≪1\varepsilon P^2 \ll1, fails at distances ξ/ξ0=O(ε−3/2)\xi/\xi_0 = O(\varepsilon^{-3/2}); for ξ0<0\xi_0 < 0 the thickness vanishes at ξ=∣ξ0∣\xi= |\xi_0|. For this reason the cylindrical case is not examined further. On the other hand, bounded non-constant profiles, and in particular periodic or localized irregularities, do not satisfy the above condition, so that (60) does not reduce to either of these integrable forms. Its solutions are therefore obtained numerically in what follows.

Formulation of the numerical problem

For the numerical computations we write

x=ζ,t=θ,u(x,t)=ψ(ζ,θ),(68)x = \zeta,\qquad t = \theta,\qquad u(x,t) = \psi(\zeta,\theta), \tag*{(68)}

so that, throughout this section, xx and tt denote the normalised coordinate and the normalised propagation distance, and uu denotes the normalised strain amplitude ψ\psi; they are not related to the physical coordinate and time of § 3 or to the dimensionless displacement component uu of § 3.3. Equation (60) then reads

∂u∂t+N(t)u∂u∂x+∂3u∂x3=0,N(t)=N0[P(t)]−5/2,P(t)=P(ξ(t)).(69)\frac{\partial u}{\partial t}+\mathcal{N}(t)u\frac{\partial u}{\partial x}+\frac{\partial^3u}{\partial x^3}=0,\qquad\mathcal{N}(t)=\mathcal{N}_0\left[\mathcal{P}(t)\right]^{-5/2},\qquad\mathcal{P}(t)=P(\xi(t)). \tag*{(69)}

The geometry may be specified either by P(ξ)P(\xi), in which case tt follows from (58), or directly by the composed function P(t)\mathcal{P}(t), in which case the physical profile is recovered from

ξ(t)=1∣Λ2∣∫0tdt′P2(t′).(70)\xi(t)=\frac{1}{|\Lambda_2|}\int_0^t\frac{dt'}{\mathcal{P}^2(t')}. \tag*{(70)}

The second alternative is adopted here, with the periodic modulation

P(t)=1+Acos⁡(ωt),0≤A<1.(71)\mathcal{P}(t)=1+A\cos(\omega t),\qquad0\leq A<1. \tag*{(71)}

By (70), this corresponds to a surface profile that is periodic in ξ\xi, with the same extreme values 1±A1\pm A of the half-thickness and with the period

Ξ=2πω∣Λ2∣(1−A2)3/2.(72)\Xi=\frac{2\pi}{\omega|\Lambda_2|(1-A^2)^{3/2}}. \tag*{(72)}

The nonlinear coefficient accordingly oscillates between N0(1+A)−5/2\mathcal{N}_0(1+A)^{-5/2} and N0(1−A)−5/2\mathcal{N}_0(1-A)^{-5/2}.

The values of N0\mathcal{N}_0 used in the computations are collected in Table 1. For iron and copper, the second- and third-order elastic constants are taken directly from Truesdell and Noll [43 Sect. 66]. Because N0\mathcal{N}_0 depends on the material only through ratios of elastic constants, the mass density is not required. In addition to the two metals, model materials with cL2/cT2=3c_L^2/c_T^2=3 (ν=1/4\nu=1/4) are considered, for which N0=3(9η2−13η1)/(2cT2)\mathcal{N}_0=3(9\eta_2-13\eta_1)/(2c_T^2); the ratio η2/η1\eta_2/\eta_1 is chosen so as to give prescribed values of N0\mathcal{N}_0 of either sign.

For both metals N0>0\mathcal{N}_0>0, that is, Γ<0\Gamma<0: iron and copper support compressive solitary waves, and their normalised equations differ only in the scale of the amplitude.

Numerical method and resolution tests

In all computations the initial condition is the pulse

u(x,0)=cs2sech⁡2(cs2x),(73)u(x,0)=\frac{c_s}{2}\operatorname{sech}^2\left(\frac{\sqrt{c_s}}{2}x\right), \tag*{(73)}

which has the amplitude cs/2c_s/2 and the characteristic width 2/cs2/\sqrt{c_s}. By (64), it is an exact solitary wave of the uniform layer when N0=6\mathcal{N}_0=6; for other values of N0\mathcal{N}_0 it is not, and its evolution in the uniform layer (A=0A=0) then provides the baseline against which the effect of the irregularity is assessed.

Equation (69) is solved on a periodic interval −R≤x<R-R\le x<R by a Fourier pseudospectral method with NN grid points. The dispersive term is integrated exactly through its exponential propagator and the nonlinear term is advanced by the fourth-order exponential time-differencing Runge–Kutta scheme (ETDRK4) [44, 45]. The coefficient N(t)\mathcal{N}(t) is evaluated at the beginning of each step and kept fixed within the step. Unless stated otherwise,

R=400,N=1024,Δt=0.01,tmax⁡=100,ω=0.2,(74)R=400,\qquad N=1024,\qquad\Delta t=0.01,\qquad t_{\max}=100,\qquad\omega=0.2, \tag*{(74)}

so that Δx=0.78125\Delta x=0.78125 and the interval 0≤t≤1000\le t\le100 contains about three periods 2π/ω≃31.422\pi/\omega\simeq31.42 of the modulation (71).

The resolution was examined for the reference case defined below (N0=6\mathcal{N}_0=6, A=0.01A=0.01, cs=0.1c_s=0.1) by means of the relative differences

E2=∥utest−uref∥2∥uref∥2,E∞=∥utest−uref∥∞∥uref∥∞(75)E_2=\frac{\lVert u_{\mathrm{test}}-u_{\mathrm{ref}}\rVert_2}{\lVert u_{\mathrm{ref}}\rVert_2},\qquad E_\infty=\frac{\lVert u_{\mathrm{test}}-u_{\mathrm{ref}}\rVert_\infty}{\lVert u_{\mathrm{ref}}\rVert_\infty} \tag*{(75)}

between profiles at t=100t=100, evaluated at common grid points without any translation. In the grid test Δt=0.01\Delta t=0.01 is fixed and the solution with N=4096N=4096 is the reference; in the step test N=1024N=1024 is fixed and the solution with Δt=0.0025\Delta t=0.0025 is the reference. The results are given in Table 2. The spatial error is at the level of round-off already for N=1024N=1024. The differences in the step test decrease in proportion to Δt−Δtref\Delta t-\Delta t_{\mathrm{ref}}, as expected when the coefficient is frozen within a step; at the adopted step Δt=0.01\Delta t=0.01 they are E2=2.5×10−6E_2=2.5\times10^{-6} and E∞=1.5×10−6E_\infty=1.5\times10^{-6}, which is far below the differences between the cases compared in what follows. The temporal discretization is therefore the dominant, and still negligible, source of numerical error.

TestResolutionE2E_2E∞E_\infty
GridN=512N=5121.48×10−81.48\times10^{-8}1.02×10−81.02\times10^{-8}
GridN=1024N=10246.3×10−166.3\times10^{-16}4.2×10−164.2\times10^{-16}
GridN=2048N=20486.4×10−166.4\times10^{-16}5.6×10−165.6\times10^{-16}
StepΔt=0.04\Delta t=0.041.2515×10−51.2515\times10^{-5}7.3895×10−67.3895\times10^{-6}
StepΔt=0.02\Delta t=0.025.8413×10−65.8413\times10^{-6}3.4477×10−63.4477\times10^{-6}
StepΔt=0.01\Delta t=0.012.5036×10−62.5036\times10^{-6}1.4774×10−61.4774\times10^{-6}
StepΔt=0.005\Delta t=0.0058.3458×10−78.3458\times10^{-7}4.9245×10−74.9245\times10^{-7}

Table 2. Grid and step refinement for the reference case at t=100t=100.

As a further check, the invariant I2I_2 of (61) is monitored in every computation. Its conservation tests the consistency of the discretization but is not, by itself, a measure of the error of the profile, which is why it is used together with Table 2.

Numerical results

Reference computation. We begin with the model material for which N0=6\mathcal{N}_0=6 (η2/η1=17/9\eta_2/\eta_1=17/9), a weak irregularity with A=0.01A=0.01, and the pulse (73) with cs=0.1c_s=0.1, whose amplitude is 0.050.05 and whose width is 2/cs≃6.322/\sqrt{c_s}\simeq6.32. In this case

N(t)=6[1+0.01cos⁡(0.2t)]−5/2,5.853≤N(t)≤6.153,(76)\mathcal{N}(t)=6[1+0.01\cos(0.2t)]^{-5/2},\qquad5.853\le\mathcal{N}(t)\le6.153, \tag*{(76)}

so that the coefficient departs from the value for which the pulse is an exact solitary wave by abo 2.5%. The results are shown in Figure 1. The pulse advances towards positive xx and retains its shape; at t=100t = 100 its maximum is 0.049937, which is 0.13% below the initial value, and it is attained at the grid point x=10.156x = 10.156, close to the displacement cst=10c_s t = 10 of the solitary wave of the uniform layer. No secondary structure is visible on the scale of the figure. The deviation of I2I_2 from its initial value does not exceed 1.5×10−151.5 \times10^{-15} in absolute and 7.3×10−147.3 \times10^{-14} in relative terms, and is indistinguishable from zero in Figure 1(b). A weak periodic irregularity thus leaves the solitary wave practically unaffected over the distance considered, and this computation serves as the baseline for those that follow.

Figure 1. Reference computation for the model material with N0=6\mathcal{N}_0 = 6 (cL2/cT2=3c_L^2/c_T^2 = 3, η2/η1=17/9\eta_2/\eta_1 = 17/9) and the irregularity A=0.01A = 0.01, ω=0.2\omega= 0.2: (a) initial profile and profile at t=100t = 100 for cs=0.1c_s = 0.1; (b) deviation of the invariant I2I_2 from its initial value, on a fixed scale of ±10−7\pm10^{-7}; (c) nonlinear coefficient N(t)\mathcal{N}(t), the dotted line marking N=6\mathcal{N} = 6.

Iron. The effect of the amplitude of the irregularity is examined next for iron, for which N0=171.29\mathcal{N}_0 = 171.29 (Table 1). The amplitude takes the values A=0A = 0, 0.01, 0.05 and 0.10, the case A=0A = 0 being the layer of uniform thickness, and

Materialν\nuΛ1\Lambda_1Λ2\Lambda_2N0\mathcal{N}_0
Iron0.291−4.804-4.804−0.02805-0.02805171.29
Copper0.346−5.856-5.856−0.04666-0.04666125.49
Model, η2/η1=17/9\eta_2/\eta_1=17/91/41/4−1/9-1/9−1/54-1/546
Model, η2/η1=43/27\eta_2/\eta_1=43/271/41/4−1/27-1/27−1/54-1/542
Model, η2/η1=29/9\eta_2/\eta_1=29/91/41/4−4/9-4/9−1/54-1/5424
Model, η2/η1=1\eta_2/\eta_1=11/41/41/91/9−1/54-1/54−6-6

Table 1. Material parameters of the normalised equation (69). For the model materials cL2/cT2=3c_L^2/c_T^2=3 and η1/cT2=1\eta_1/c_T^2=1.

N(t)=171.29[1+Acos⁡(0.2t)]−5/2(77)\mathcal{N}(t) = 171.29 [1 + A \cos(0.2t)]^{-5/2} \tag*{(77)}

varies over the intervals [167.08, 175.64], [151.62, 194.72] and [134.97, 222.90] for A=0.01A = 0.01, 0.05 and 0.10, respectively. Since the geometric factor enters through an inverse power, the variation is not symmetric about N0\mathcal{N}_0,

N(t)N0=1−52Acos⁡(ωt)+358A2cos⁡2(ωt)+O(A3),(78)\frac{\mathcal{N}(t)}{\mathcal{N}_0} = 1 - \frac{5}{2} A \cos(\omega t) + \frac{35}{8} A^2 \cos^2(\omega t) + O(A^3), \tag*{(78)}

and its mean over a period exceeds N0\mathcal{N}_0 by a term of order A2A^2. The initial pulse is (73) with cs=0.01c_s = 0.01, that is, with the amplitude 0.005 and the width 20; the same pulse is used for the four values of AA. Because N0>0\mathcal{N}_0 > 0, this pulse has the polarity of the solitary waves of the material, but it is much larger than the solitary wave of the same width, whose amplitude is 3cs/N0=1.75×10−43c_s/\mathcal{N}_0 = 1.75 \times10^{-4}.

Figure 2 shows that in all four cases the pulse steepens on its leading side and disintegrates into a train of solitary waves ordered by amplitude, the largest one in front. For the uniform layer, the first three maxima at t=100t = 100 are 8.623 ×10−3\times10^{-3} at x=60.94x = 60.94, 6.187 ×10−3\times10^{-3} at x=42.97x = 42.97 and 3.801 ×10−3\times10^{-3} at x=26.56x = 26.56, while the rear part of the pulse has not yet separated. No negative values of any significance occur. This is the classical disintegration of an initial pulse into a train of solitary waves ordered by amplitude [41].

Figure 2. Effect of the amplitude of the irregularity on the evolution of a pulse in iron (N0=171.29\mathcal{N}_0=171.29), for A=0A=0, 0.010.01, 0.050.05 and 0.100.10 and ω=0.2\omega=0.2. Dashed curves: common initial profile with cs=0.01c_s=0.01; solid curves: solutions at t=100t=100. The axis limits are the same in all panels.

The irregularity does not alter this picture qualitatively, but it changes the individual waves measurably. As AA increases through 0.01, 0.05 and 0.10, the leading maximum decreases to 8.519 ×10−3\times10^{-3}, 8.360 ×10−3\times10^{-3} and 8.06 ×10−3\times10^{-3}, the last value being 6.5% below that of the uniform layer, whereas the second and third maxima increase, to 6.366 ×10−3\times10^{-3} and 4.038 ×10−3\times10^{-3} for A=0.10A = 0.10, that is, by 2.9% and 6.2%. The positions of the maxima change by at most one grid spacing. For A=0.10A = 0.10 additional undulations appear in the rear part of the pulse, and a shallow depression of depth 2.3×10−42.3 \times10^{-4} forms behind the leading wave. The irregularity thus redistributes the amplitudes within the train rather than amplifying or attenuating it as a whole. The overall difference is measured by

EA=∥uA(⋅,100)−u0(⋅,100)∥2∥u0(⋅,100)∥2,(79)E_A = \frac{\left\lVert u_A(\cdot,100)-u_0(\cdot,100)\right\rVert_2}{\left\lVert u_0(\cdot,100)\right\rVert_2}, \tag*{(79)}

where uAu_A denotes the solution for the amplitude AA; we find E0.01=2.1%E_{0.01}=2.1\%, E0.05=7.0%E_{0.05}=7.0\% and E0.10=16.8%E_{0.10}=16.8\%. These differences are several orders of magnitude larger than the numerical errors of Table 2. In these computations the relative deviation of I2I_2 remains below 7×10−107 \times10^{-10}, and the solution in the outer region ∣x∣≥360|x| \ge360 never exceeds 9×10−99 \times10^{-9} of its overall maximum, so that the periodic boundary conditions have no influence on the results. Doubling NN and halving Δt\Delta t for A=0.10A=0.10 change the final profile by 4×10−74 \times10^{-7} and 5×10−55 \times10^{-5} in the sense of E2E_2, respectively.

Copper. The same comparison is made for copper, for which N0=125.49\mathcal{N}_0=125.49, with the initial pulse, the frequency and the numerical parameters of the iron computations. The coefficient now varies over [122.41,128.68][122.41, 128.68], [111.08,142.66][111.08, 142.66] and [98.88,163.31][98.88, 163.31] for A=0.01A=0.01, 0.050.05 and 0.100.10. For a given AA the relative modulation N(t)/N0\mathcal{N}(t)/\mathcal{N}_0 is the same for the two metals; they differ in the strength of the nonlinearity relative to the dispersion, the value for copper being 73%73\% of that for iron. It should be kept in mind that equal values of tt do not correspond to equal distances in the two materials, because the relation (58) between tt and ξ\xi involves Λ2\Lambda_2.

The evolution, shown in Figure 3, is of the same type as in iron, but it proceeds more slowly. For the uniform layer the first three maxima are 8.245×10−38.245 \times10^{-3} at x=46.09x = 46.09, 5.004×10−35.004 \times10^{-3} at x=29.69x = 29.69 and 2.834×10−32.834 \times10^{-3} at x=14.06x = 14.06; the leading wave has travelled a shorter distance than in iron, and fewer waves have emerged. With increasing AA the leading maximum decreases to 8.196×10−38.196 \times10^{-3}, 8.044×10−38.044 \times10^{-3} and 7.937×10−37.937 \times10^{-3}, the reduction for A=0.10A = 0.10 being 3.7%, while the second and third maxima increase to 5.246×10−35.246 \times10^{-3} and 3.038×10−33.038 \times10^{-3}, that is, by 4.8% and 7.2%. The profile differences (79), referred to the uniform copper layer, are 1.3%, 4.3% and 9.3%, compared with 2.1%, 7.0% and 16.8% for iron. Under identical normalised conditions copper is therefore less sensitive to the irregularity than iron, in accordance with its smaller nonlinear coefficient. The relative deviation of I2I_2 is below 8×10−138 \times10^{-13} and the solution in ∣x∣≥360|x| \ge360 is below 1.1×10−111.1 \times10^{-11} of its maximum.

Figure 3. Effect of the amplitude of the irregularity on the evolution of a pulse in copper (N0=125.49\mathcal{N}_0 = 125.49), for A=0A = 0, 0.01, 0.05 and 0.10 and ω=0.2\omega= 0.2. Dashed curves: common initial profile with cs=0.01c_s = 0.01; solid curves: solutions at t=100t = 100. The axis limits are the same in all panels.

Velocity of the leading wave. To quantify the influence of the irregularity on the motion of the leading wave, the iron computations are repeated for A=0A = 0, 0.02, …\ldots, 0.10, and the position xp(t;A)x_p(t; A) of the largest maximum is determined. Since the maximum falls on the same or on a neighbouring grid point in all cases, its position is refined by a parabola through the largest grid value uju_j and its two neighbours,

xp≃xj+Δx2uj−1−uj+1uj−1−2uj+uj+1.(80)x_p \simeq x_j + \frac{\Delta x}{2}\frac{u_{j-1} - u_{j+1}}{u_{j-1} - 2u_j + u_{j+1}}. \tag*{(80)}

Table 3 gives xpx_{\mathrm{p}} at T=100T = 100, the mean velocity v‾p=xp(T;A)/T\overline{v}_{\mathrm{p}} = x_{\mathrm{p}}(T; A)/T and its change relative to the uniform layer, δv‾p=100[v‾p(A)/v‾p(0)−1]\delta\overline{v}_{\mathrm{p}} = 100[\overline{v}_{\mathrm{p}}(A)/\overline{v}_{\mathrm{p}}(0) - 1]. The mean velocity depends on AA weakly and not monotonically: it is smallest for A=0.04A = 0.04, where it is 0.37% below the value for the uniform layer, and exceeds that value by 0.19% for A=0.10A = 0.10. These figures refer to the normalised variables and to an interval that is not an integer multiple of the period of the modulation; they show that the irregularity affects the velocity of the leading wave much less than its amplitude.

AAxp(100;A)x_{\mathrm{p}}(100; A)v‾p(A)\overline{v}_{\mathrm{p}}(A)δv‾p\delta\overline{v}_{\mathrm{p}} (%)
0.0060.65630.606560.000
0.0260.49870.60499−0.260
0.0460.43350.60433−0.367
0.0660.45720.60457−0.328
0.0860.57320.60573−0.137
0.1060.77210.607720.191

Table 3. Position and mean velocity of the leading maximum in iron at T=100T = 100.

Model materials. Finally, in order to separate the role of the material nonlinearity from that of the geometry, we consider the four model materials of Table 1, for which N0=6\mathcal{N}_0 = 6, 2, 24 and −6, with the same irregularity, A=0.05A = 0.05 and ω=0.2\omega= 0.2, and the same initial pulse, cs=0.1c_{\mathrm{s}} = 0.1. Here N(t)=N0[1+0.05cos⁡(0.2 t)]−5/2\mathcal{N}(t) = \mathcal{N}_0[1 + 0.05\cos(0.2\,t)]^{-5/2} varies between 0.885 N00.885\,\mathcal{N}_0 and 1.137 N01.137\,\mathcal{N}_0. The computations use R=800R = 800, N=4096N = 4096 and Δt=0.005\Delta t = 0.005; the larger interval accommodates the oscillatory tails. The results are shown in Figure 4.

Figure 4. Evolution of the same pulse (cs=0.1c_s = 0.1) in four model materials with cL2/cT2=3c_L^2/c_T^2 = 3 and η1/cT2=1\eta_1/c_T^2 = 1: (a) η2/η1=17/9\eta_2/\eta_1 = 17/9, (b) 43/2743/27, (c) 29/929/9 and (d) 11, corresponding to N0=6\mathcal{N}_0 = 6, 22, 2424 and −6-6. Dashed curves: initial profile; solid curves: solutions at t=100t = 100. In all cases A=0.05A = 0.05 and ω=0.2\omega= 0.2; the axis limits are the same in all panels.

For N0=6\mathcal{N}_0 = 6 (panel a) the pulse remains a solitary wave: its maximum at t=100t = 100 is 0.04992, which is 0.16% below the initial value, and only a tail of very small amplitude is formed. For N0=2\mathcal{N}_0 = 2 (panel b) the nonlinearity is too weak to balance the dispersion of the given pulse; the maximum falls to 0.04136, that is, by 17.3%, and an oscillatory tail with a first depression of −0.0092 develops behind it. For N0=24\mathcal{N}_0 = 24 (panel c) the pulse is larger than the solitary wave of its width and disintegrates: at t=100t = 100 two separated waves with the maxima 0.06734 at x=57.8x = 57.8 and 0.02470 at x=18.0x = 18.0 are present, the leading one being 34.7% higher than the initial pulse. For the uniform layer the inverse scattering theory gives in this case three solitary waves with the amplitudes 0.0703, 0.0235 and 0.0017, and one wave with the amplitude 0.0314 for N0=2\mathcal{N}_0 = 2. For N0=−6\mathcal{N}_0 = -6 (panel d) the polarity of the pulse is opposite to that of the solitary waves of the material; no solitary wave is formed, and the pulse disperses into an extended oscillatory wave train, its maximum decreasing to 0.02883 and the deepest depression reaching −0.02297. In these computations the relative deviation of I2I_2 does not exceed 2.6×10−122.6 \times10^{-12}, and repeating them on the interval with R=400R = 400 and N=2048N = 2048 changes the profiles in −200≤x≤100-200 \le x \le100 by less than 5.4×10−75.4 \times10^{-7}.

Taken together, the computations show that, for the amplitudes of the irregularity and the distance considered here, the character of the evolution of a given pulse (propagation as a solitary wave, disintegration into several solitary waves, or dispersion into a wave train) is decided by the material through the sign and the magnitude of N0\mathcal{N}_0, whereas the irregularity of the free surfaces modifies the amplitudes of the individual waves by a few per cent and their velocities by a fraction of a per cent. All amplitudes and positions quoted in this section refer to the normalised variables (68).

Conclusions

The propagation of long nonlinear extensional waves in a compressible hyperelastic layer with irregular, traction-free surfaces has been studied within the framework of finite elasticity. The free surfaces were taken to be symmetric with respect to the middle plane, and their irregularity was allowed to change the local thickness by a finite amount on a length scale long compared with the wavelength. In a distinguished limit in which weak nonlinearity, dispersion and the slow variation of the thickness appear at the same order, the method of multiple scales applied to the two-dimensional boundary-value problem led to a Korteweg–de Vries equation with variable coefficients for the leading-order longitudinal strain.

The structure of this equation is simple and has a direct mechanical interpretation. The velocity of the long wave is that of extensional waves in a plate and is not affected by the irregularity, whose influence is deferred to the order at which the amplitude evolves. There, the geometry enters in only two ways: through a linear term that expresses the conservation of the energy flux through a cross-section, and through the factor P2P^2, the square of the local thickness, multiplying the dispersive term. The nonlinear coefficient is independent of the geometry; it is the coefficient of the quadratic term in the uniaxial response of a layer with free surfaces and involves the third-order elastic constants only through two combinations. The dispersive coefficient, in contrast, depends on Poisson’s ratio alone and coincides with the long-wave limit of the fundamental symmetric branch of the Rayleigh–Lamb frequency equation; it arises from the inertia of the transverse contraction associated with the longitudinal strain. The linear and the nonlinear elastic properties of the material thus enter the long-wave model in clearly separated ways.

After the linear geometric term was removed and the dispersive coefficient was normalised, the effect of the material and of the geometry was found to be carried entirely by a single coefficient, N0P−5/2\mathcal{N}_0P^{-5/2}. Its sign is fixed by the material; for iron and copper it is such that compressive solitary waves of strain exist. For special choices of the surface irregularity, namely a uniform layer and a thickness varying as the cube root of the distance, the normalised equation reduces to forms known to be integrable, the Korteweg–de Vries and the cylindrical Korteweg–de Vries equations; since the latter profile is admissible in a bounded region only, bounded irregularities, and periodic ones in particular, require a numerical treatment. The computations carried out for periodically irregular layers of iron and copper and for model materials showed that the type of evolution of a given pulse, namely propagation as a solitary wave, disintegration into a train of solitary waves, or dispersion into an oscillatory wave train, is decided by the material, while the irregularity of the free surfaces redistributes the amplitudes of the individual waves and changes their velocities only slightly.

Several extensions of the present work suggest themselves. The response of a solitary wave to a localized change of thickness, such as a step or a bump, is of interest in its own right: an abrupt thinning is expected to produce secondary solitary waves, and an abrupt thickening a dispersive tail together with a shelf behind the transmitted wave, and these features can be described by combining the inverse scattering theory of the uniform layer with the adiabatic theory of slowly varying solitary waves [33, 34]. The formation and the dynamics of the trailing shelf generated by a slowly varying thickness deserve a separate analysis [32, 35]. For step-like initial data, and for pulses of the polarity opposite to that of the solitary waves, the evolution leads to dispersive shock waves, whose structure in a layer of variable thickness could be studied by means of the Whitham modulation theory with variable coefficients [32, 42]. Finally, the case in which the quadratic nonlinearity vanishes, the antisymmetric (flexural) mode, layers with non-symmetric free surfaces, and layers bonded to a substrate [29, 28] call for different scalings and lead to evolution equations of other types.

CRediT authorship contribution statement

Ali Demirci: Writing – review and editing, Writing-original draft, Visualization, Investigation, Software, Formal analysis, Funding acquisition. Haldun Taha Uzunoglu: Writing – review and editing, Writing-original draft, Visualization, Investigation, Software, Formal analysis, Methodology.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgements

A. D. was supported under Grant No. TGA-2025-46559 by Istanbul Technical University Office of Scientific Research Projects (ITU BAPSIS). This study constitutes part of the Ph.D. thesis of H. T. U. at Istanbul Technical University.

Data availability

No data was used for the research described in the article.

Stress components

For the plane motion (3), the strain energy function of § 3.2 and the relation (7) give, up to and including terms of the second degree in the displacement gradients,

T11ρ0=cL2u1,X+(cL2−2cT2)u2,Y+η1u1,X2+η2(u2,Y2+2u1,Xu2,Y)+η3u1,Yu2,X+η4K1,(81)\frac{T_{11}}{\rho_0}=c_L^2u_{1,X}+\left(c_L^2-2c_T^2\right)u_{2,Y}+\eta_1u_{1,X}^2+\eta_2\left(u_{2,Y}^2+2u_{1,X}u_{2,Y}\right)+\eta_3u_{1,Y}u_{2,X}+\eta_4\mathcal{K}_1, \tag*{(81)}
T21ρ0=cT2(u1,Y+u2,X)+(η3u2,X+2η4u1,Y)K2,(82)\frac{T_{21}}{\rho_0}=c_T^2\left(u_{1,Y}+u_{2,X}\right)+\left(\eta_3u_{2,X}+2\eta_4u_{1,Y}\right)\mathcal{K}_2, \tag*{(82)}
T12ρ0=cT2(u1,Y+u2,X)+(2η4u2,X+η3u1,Y)K2,(83)\frac{T_{12}}{\rho_0}=c_T^2\left(u_{1,Y}+u_{2,X}\right)+\left(2\eta_4u_{2,X}+\eta_3u_{1,Y}\right)\mathcal{K}_2, \tag*{(83)}
T22ρ0=cL2u2,Y+(cL2−2cT2)u1,X+η1u2,Y2+η2(u1,X2+2u1,Xu2,Y)+η3u1,Yu2,X+η4K1,(84)\frac{T_{22}}{\rho_0}=c_L^2u_{2,Y}+\left(c_L^2-2c_T^2\right)u_{1,X}+\eta_1u_{2,Y}^2+\eta_2\left(u_{1,X}^2+2u_{1,X}u_{2,Y}\right)+\eta_3u_{1,Y}u_{2,X}+\eta_4\mathcal{K}_1, \tag*{(84)}

where

K1=u1,Y2+u2,X2,K2=u1,X+u2,Y,(85)\mathcal{K}_1=u_{1,Y}^2+u_{2,X}^2,\qquad\mathcal{K}_2=u_{1,X}+u_{2,Y}, \tag*{(85)}

and cLc_L, cTc_T and ηi\eta_i are defined by (9)–(11). In components, the balance equations (4) are

T11,X+T21,Y=ρ0u¨1,T12,X+T22,Y=ρ0u¨2,(86)T_{11,X}+T_{21,Y}=\rho_0\ddot{u}_1,\qquad T_{12,X}+T_{22,Y}=\rho_0\ddot{u}_2, \tag*{(86)}

and the traction conditions (6) on the upper free surface Y=H(X)Y=H(X) are

−H′(X)T11+T21=0,−H′(X)T12+T22=0.(87)-H'(X)T_{11}+T_{21}=0,\qquad-H'(X)T_{12}+T_{22}=0. \tag*{(87)}

Dimensionless operators

Equations (81)–(87) are expressed in the variables (12) by means of (13) and (14). The first balance equation and the first traction condition are multiplied by εL/ρ0\varepsilon L/\rho_0 and ε/(δρ0)\varepsilon/(\delta\rho_0), and the second ones by ε2L/(δρ0)\varepsilon^2L/(\delta\rho_0) and ε/ρ0\varepsilon/\rho_0, respectively, so that the powers of ε\varepsilon in the operators below coincide with the orders of the perturbation problems of § 4. With

k1=(∂u∂y)2+ε2(∂v∂x)2,k2=∂u∂x+∂v∂y,(88)k_1=\left(\frac{\partial u}{\partial y}\right)^2+\varepsilon^2\left(\frac{\partial v}{\partial x}\right)^2,\qquad k_2=\frac{\partial u}{\partial x}+\frac{\partial v}{\partial y}, \tag*{(88)}

which are the dimensionless forms of εK1\varepsilon\mathcal{K}_1 and K2\mathcal{K}_2, and with subscripts xx, yy and tt denoting partial derivatives, the linear operators are

L1(u,v)=εcL2uττ−εcL2uxx−cT2uyy−ε(cL2−cT2)vxy(89)\mathcal{L}_1(u,v)=\varepsilon c_L^2u_{\tau\tau}-\varepsilon c_L^2u_{xx}-c_T^2u_{yy}-\varepsilon\left(c_L^2-c_T^2\right)v_{xy} \tag*{(89)}
L2(u,v)=ε2c2uττ−ε2cL2uxx−εcL2vyy−ε(cL2−cT2)uxy,(90)\mathcal{L}_2(u,v)=\varepsilon^2 c^2 u_{\tau\tau}-\varepsilon^2 c_L^2 u_{xx}-\varepsilon c_L^2 v_{yy}-\varepsilon\left(c_L^2-c_T^2\right)u_{xy}, \tag*{(90)}

and the nonlinear operators are

N1(u,v)=∂∂y[(2η4uy+εη3vx)k2]+ε∂∂x[η1ux2+η2vy2+2η2uxvy+η3uyvx]+η4∂k1∂x,(91)\mathcal{N}_1(u,v)=\frac{\partial}{\partial y}\left[(2\eta_4u_y+\varepsilon\eta_3v_x)k_2\right]+\varepsilon\frac{\partial}{\partial x}\left[\eta_1u_x^2+\eta_2v_y^2+2\eta_2u_xv_y+\eta_3u_yv_x\right]+\eta_4\frac{\partial k_1}{\partial x}, \tag*{(91)}
N2(u,v)=ε∂∂x[(2εη4vx+η3uy)k2]+ε∂∂y[η2ux2+η1vy2+2η2uxvy+η3uyvx]+η4∂k1∂y.(92)\mathcal{N}_2(u,v)=\varepsilon\frac{\partial}{\partial x}\left[(2\varepsilon\eta_4v_x+\eta_3u_y)k_2\right]+\varepsilon\frac{\partial}{\partial y}\left[\eta_2u_x^2+\eta_1v_y^2+2\eta_2u_xv_y+\eta_3u_yv_x\right]+\eta_4\frac{\partial k_1}{\partial y}. \tag*{(92)}

On the upper free surface y=Py=P, where ∂P/∂x=εP′(x1)\partial P/\partial x=\varepsilon P'(x_1), the traction operators are

B1+(u,v)=cT2(uy+εvx)+(2η4uy+εη3vx)k2−εP′{ε[cL2ux+(cL2−2cT2)vy]+η1ux2+η2vy2+2η2uxvy+η3uyvx}+η4k1,(93)\mathcal{B}_1^+(u,v)=c_T^2\left(u_y+\varepsilon v_x\right)+(2\eta_4u_y+\varepsilon\eta_3v_x)k_2 -\varepsilon P'\left\{\varepsilon\left[c_L^2u_x+\left(c_L^2-2c_T^2\right)v_y\right]+\eta_1u_x^2+\eta_2v_y^2+2\eta_2u_xv_y+\eta_3u_yv_x\right\}+\eta_4k_1, \tag*{(93)}
B2+(u,v)=ε{cL2vy+(cL2−2cT2)ux+η1vy2+η2ux2+2η2uxvy+η3uyvx}+η4k1−ε2P′[cT2(uy+εvx)+(η3vy+2εη4vx)k2].(94)\mathcal{B}_2^+(u,v)=\varepsilon\left\{c_L^2v_y+\left(c_L^2-2c_T^2\right)u_x+\eta_1v_y^2+\eta_2u_x^2+2\eta_2u_xv_y+\eta_3u_yv_x\right\}+\eta_4k_1 -\varepsilon^2P'\left[c_T^2\left(u_y+\varepsilon v_x\right)+(\eta_3v_y+2\varepsilon\eta_4v_x)k_2\right]. \tag*{(94)}

The operators Bi−\mathcal{B}_i^- on the lower free surface y=−Py=-P are obtained from (93) and (94) by replacing P′P' with −P′-P'. In all these expressions cLc_L, cTc_T, cc and ηi\eta_i retain their dimensions, every term having the dimension of a squared velocity.

Operators of the perturbation problems

With (22) and (23), the coefficients of ε2\varepsilon^2 and ε3\varepsilon^3 in (89)–(94) define the following operators. In the second-order problem,

L21=c2∂2u1∂τ2−cL2∂2u1∂x02−(cL2−cT2)∂2v1∂x0∂y,(95)\mathcal{L}^{21}=c^2\frac{\partial^2u_1}{\partial\tau^2}-c_L^2\frac{\partial^2u_1}{\partial x_0^2}-\left(c_L^2-c_T^2\right)\frac{\partial^2v_1}{\partial x_0\partial y}, \tag*{(95)}
L22=−cL2∂2v1∂y2−(cL2−cT2)∂2u1∂x0∂y,(96)\mathcal{L}^{22}=-c_L^2\frac{\partial^2v_1}{\partial y^2}-\left(c_L^2-c_T^2\right)\frac{\partial^2u_1}{\partial x_0\partial y}, \tag*{(96)}
N21=η4∂∂x0[(∂u1∂y)2]+∂∂y[2η4∂u1∂y(∂u1∂x0+∂v1∂y)],(97)\mathcal{N}^{21}=\eta_4\frac{\partial}{\partial x_0}\left[\left(\frac{\partial u_1}{\partial y}\right)^2\right]+\frac{\partial}{\partial y}\left[2\eta_4\frac{\partial u_1}{\partial y}\left(\frac{\partial u_1}{\partial x_0}+\frac{\partial v_1}{\partial y}\right)\right], \tag*{(97)}
N22=η4∂∂y[(∂u1∂y)2],(98)\mathcal{N}^{22}=\eta_4\frac{\partial}{\partial y}\left[\left(\frac{\partial u_1}{\partial y}\right)^2\right], \tag*{(98)}
B21=cT2∂v1∂x0+2η4∂u1∂y(∂u1∂x0+∂v1∂y),(99)\mathcal{B}^{21}=c_T^2\frac{\partial v_1}{\partial x_0}+2\eta_4\frac{\partial u_1}{\partial y}\left(\frac{\partial u_1}{\partial x_0}+\frac{\partial v_1}{\partial y}\right), \tag*{(99)}
B22=cL2∂v1∂y+(cL2−2cT2)∂u1∂x0+η4(∂u1∂y)2.(100)\mathcal{B}^{22}=c_L^2\frac{\partial v_1}{\partial y}+\left(c_L^2-2c_T^2\right)\frac{\partial u_1}{\partial x_0}+\eta_4\left(\frac{\partial u_1}{\partial y}\right)^2. \tag*{(100)}

In the third-order problem,

L31=c2∂2u2∂τ2−cL2(∂2u2∂x02+2∂2u1∂x0∂x1)−(cL2−cT2)(∂2v1∂x1∂y+∂2v2∂x0∂y),(101)\mathcal{L}^{31}=c^2\frac{\partial^2u_2}{\partial\tau^2}-c_L^2\left(\frac{\partial^2u_2}{\partial x_0^2}+2\frac{\partial^2u_1}{\partial x_0\partial x_1}\right)-\left(c_L^2-c_T^2\right)\left(\frac{\partial^2v_1}{\partial x_1\partial y}+\frac{\partial^2v_2}{\partial x_0\partial y}\right), \tag*{(101)}
L32=c2∂2v1∂τ2−cT2∂2v1∂x02−cL2∂2v1∂y2−(cL2−cT2)(∂2u1∂x1∂y+∂2u2∂x0∂y),(102)\mathcal{L}^{32}=c^2\frac{\partial^2v_1}{\partial\tau^2}-c_T^2\frac{\partial^2v_1}{\partial x_0^2}-c_L^2\frac{\partial^2v_1}{\partial y^2}-\left(c_L^2-c_T^2\right)\left(\frac{\partial^2u_1}{\partial x_1\partial y}+\frac{\partial^2u_2}{\partial x_0\partial y}\right), \tag*{(102)}
N31=∂∂x0[η1(∂u1∂x0)2+η2(∂v1∂y)2+2η2∂u1∂x0∂v1∂y+η3∂u1∂y∂v1∂x0+2η4∂u1∂y∂u2∂y]+η4∂∂x1[(∂u1∂y)2]+∂∂y[2η4∂u1∂y(∂u1∂x1+∂u2∂x0+∂v2∂y)+(η3∂v1∂x0+2η4∂u2∂y)(∂u1∂x0+∂v1∂y)].(103)\begin{aligned} \mathcal{N}^{31}={}&\frac{\partial}{\partial x_0}\left[\eta_1\left(\frac{\partial u_1}{\partial x_0}\right)^2+\eta_2\left(\frac{\partial v_1}{\partial y}\right)^2+2\eta_2\frac{\partial u_1}{\partial x_0}\frac{\partial v_1}{\partial y}+\eta_3\frac{\partial u_1}{\partial y}\frac{\partial v_1}{\partial x_0}+2\eta_4\frac{\partial u_1}{\partial y}\frac{\partial u_2}{\partial y}\right]\\ &+\eta_4\frac{\partial}{\partial x_1}\left[\left(\frac{\partial u_1}{\partial y}\right)^2\right]\\ &+\frac{\partial}{\partial y}\left[2\eta_4\frac{\partial u_1}{\partial y}\left(\frac{\partial u_1}{\partial x_1}+\frac{\partial u_2}{\partial x_0}+\frac{\partial v_2}{\partial y}\right)+\left(\eta_3\frac{\partial v_1}{\partial x_0}+2\eta_4\frac{\partial u_2}{\partial y}\right)\left(\frac{\partial u_1}{\partial x_0}+\frac{\partial v_1}{\partial y}\right)\right]. \tag*{(103)} \end{aligned}
N32=∂∂y[η2(∂u1∂x0)2+η1(∂v1∂y)2+2η2∂u1∂x0∂v1∂y+η3∂u1∂y∂v1∂x0+2η4∂u1∂y∂u2∂y]+∂∂x0[η3∂u1∂y(∂u1∂x0+∂v1∂y)](104)\mathcal{N}^{32} = \frac{\partial}{\partial y}\left[\eta_{2}\left(\frac{\partial u_{1}}{\partial x_{0}}\right)^{2}+\eta_{1}\left(\frac{\partial v_{1}}{\partial y}\right)^{2}+2\eta_{2}\frac{\partial u_{1}}{\partial x_{0}}\frac{\partial v_{1}}{\partial y}+\eta_{3}\frac{\partial u_{1}}{\partial y}\frac{\partial v_{1}}{\partial x_{0}}+2\eta_{4}\frac{\partial u_{1}}{\partial y}\frac{\partial u_{2}}{\partial y}\right]+\frac{\partial}{\partial x_{0}}\left[\eta_{3}\frac{\partial u_{1}}{\partial y}\left(\frac{\partial u_{1}}{\partial x_{0}}+\frac{\partial v_{1}}{\partial y}\right)\right] \tag*{(104)}
B31=cT2(∂v2∂x0+∂v1∂x1)+2η4∂u1∂y(∂u1∂x1+∂u2∂x0+∂v2∂y)+(η3∂v1∂x0+2η4∂u2∂y)(∂u1∂x0+∂v1∂y)−P′[cL2∂u1∂x0+(cL2−2cT2)∂v1∂y+η4(∂u1∂y)2](105)\mathcal{B}^{31}=c_{\mathrm{T}}^{2}\left(\frac{\partial v_{2}}{\partial x_{0}}+\frac{\partial v_{1}}{\partial x_{1}}\right)+2\eta_{4}\frac{\partial u_{1}}{\partial y}\left(\frac{\partial u_{1}}{\partial x_{1}}+\frac{\partial u_{2}}{\partial x_{0}}+\frac{\partial v_{2}}{\partial y}\right)+\left(\eta_{3}\frac{\partial v_{1}}{\partial x_{0}}+2\eta_{4}\frac{\partial u_{2}}{\partial y}\right)\left(\frac{\partial u_{1}}{\partial x_{0}}+\frac{\partial v_{1}}{\partial y}\right)-P'\left[c_{\mathrm{L}}^{2}\frac{\partial u_{1}}{\partial x_{0}}+\left(c_{\mathrm{L}}^{2}-2c_{\mathrm{T}}^{2}\right)\frac{\partial v_{1}}{\partial y}+\eta_{4}\left(\frac{\partial u_{1}}{\partial y}\right)^{2}\right] \tag*{(105)}
B32=cL2∂v2∂y+(cL2−2cT2)(∂u1∂x1+∂u2∂x0)+η1(∂v1∂y)2+η2(∂u1∂x0)2+2η2∂u1∂x0∂v1∂y+η3∂u1∂y∂v1∂x0+2η4∂u1∂y∂u1∂y−cT2P′∂u1∂y.(106)\mathcal{B}^{32}=c_{\mathrm{L}}^{2}\frac{\partial v_{2}}{\partial y}+\left(c_{\mathrm{L}}^{2}-2c_{\mathrm{T}}^{2}\right)\left(\frac{\partial u_{1}}{\partial x_{1}}+\frac{\partial u_{2}}{\partial x_{0}}\right)+\eta_{1}\left(\frac{\partial v_{1}}{\partial y}\right)^{2}+\eta_{2}\left(\frac{\partial u_{1}}{\partial x_{0}}\right)^{2}+2\eta_{2}\frac{\partial u_{1}}{\partial x_{0}}\frac{\partial v_{1}}{\partial y}+\eta_{3}\frac{\partial u_{1}}{\partial y}\frac{\partial v_{1}}{\partial x_{0}}+2\eta_{4}\frac{\partial u_{1}}{\partial y}\frac{\partial u_{1}}{\partial y}-c_{\mathrm{T}}^{2}P'\frac{\partial u_{1}}{\partial y}. \tag*{(106)}

All terms containing ∂u1/∂y\partial u_{1}/\partial y vanish identically once the first-order solution (25) is used; they are retained here so that the operators are valid for arbitrary arguments.

References

  1. [1]A. M. Samsonov, Nonlinear strain waves in elastic waveguides, in: A. Jeffrey, J. Engelbrecht (Eds.), Nonlinear Waves in Solids, volume 341 of CISM Courses and Lectures, Springer-Verlag, Wien, 1994, pp. 349–382. doi:10.1007/978–3–7091–2444–4.DOI
  2. [2]G. A. Maugin, Physical and mathematical models of nonlinear waves in solids, in: A. Jeffrey, J. Engelbrecht (Eds.), Nonlinear Waves in Solids, volume 341 of CISM Courses and Lectures, Springer-Verlag, Wien, 1994, pp. 109–233. doi:10.1007/978–3–7091–2444–4.DOI
  3. [3]V. I. Erofeev, N. V. Klyueva, Solitons and nonlinear periodic strain waves in rods, plates, and shells (a review), Acoustical Physics 48 (2002) 643–655.DOI
  4. [4]A. V. Porubov, A. M. Samsonov, M. G. Velarde, A. V. Bukhanovsky, Strain solitary waves in an elastic rod embedded in another elastic external medium with sliding, Physical Review E 58 (1998) 3854–3864.DOI
  5. [5]A. V. Porubov, G. A. Maugin, V. V. Mareev, Localization of two-dimensional non-linear strain waves in a plate, International Journal of Non-Linear Mechanics 39 (2004) 1359–1370.DOI
  6. [6]A. V. Porubov, G. A. Maugin, Propagation of localized longitudinal strain waves in a plate in the presence of cubic nonlinearity, Physical Review E 74 (2006) 046617.pubmed.ncbi.nlm.nih.gov/17155202
  7. [7]A. V. Porubov, G. A. Maugin, Improved description of longitudinal strain solitary waves, Journal of Sound and Vibration 310 (2008) 694–701.DOI
  8. [8]A. M. Samsonov, G. V. Dreiden, A. V. Porubov, V. I. Semenova, Longitudinal-strain solitary solution focusing in a narrowing nonlinearly elastic rod, Physical Review B 57 (1998) 5778–5787.DOI
  9. [9]K. R. Khusnutdinova, M. R. Tranter, On radiating solitary waves in bi-layers with delamination and coupled Ostrovsky equations, Chaos 27 (2017) 013112.DOI
  10. [10]M. Teymür, Propagation of long extensional nonlinear waves in a hyper-elastic layer, in: E. İnan, A. Kırış (Eds.), Vibration Problems ICoVP 2005, Springer, Dordrecht, 2007, pp. 481–486.DOI
  11. [11]R. M. Chen, Some nonlinear dispersive waves arising in compressible hyperelastic plates, International Journal of Engineering Science 44 (2006) 1188–1204.DOI
  12. [12]M. Destrade, G. Saccomandi, Nonlinear transverse waves in deformed dispersive solids, Wave Motion 45 (2008) 325–336.DOI
  13. [13]M. Destrade, A. Goriely, G. Saccomandi, Scalar evolution equations for shear waves in incompressible solids: a simple derivation of the Z, ZK, KZK and KP equations, Proceedings of the Royal Society A 467 (2011) 1823–1834.DOI
  14. [14]H. A. Erbay, S. Erbay, A. Erkip, Derivation of the Camassa–Holm equations for elastic waves, Physics Letters A 379 (2015) 956–961.
  15. [15]H. A. Erbay, S. Erbay, A. Erkip, On the full dispersion Kadomtsev–Petviashvili equations for dispersive elastic waves, Wave Motion 114 (2022) 103015.DOI
  16. [16]H. A. Erbay, K. R. Rajagopal, G. Saccomandi, Y. Şengül, Dispersive transverse waves for a strain-limiting continuum model, Mathematics and Mechanics of Solids 29 (2024) 1216–1227.DOI
  17. [17]H. Berjamin, M. Destrade, G. Saccomandi, The KP equation of plane elastodynamics, SIAM Journal on Applied Mathematics 85 (2025) 1458–1474.
  18. [18]B. Wolf, Propagation of Love waves in layers with irregular boundaries, Pure and Applied Geophysics 78 (1970) 48–57.DOI
  19. [19]J. H. Woodhouse, Surface waves in a laterally varying layered structure, Geophysical Journal International 37 (1974) 461–490.DOI
  20. [20]X. Markenscoff, S. G. Lekoudis, Love waves in slowly varying layered media, Pure and Applied Geophysics 114 (1976) 805–810.DOI
  21. [21]S. S. Singh, Love wave at a layer medium bounded by irregular boundary surfaces, Journal of Vibration and Control 17 (2011) 789–795.DOI
  22. [22]P. Alam, S. Kundu, S. Gupta, Dispersion study of SH-wave propagation in an irregular magneto-elastic anisotropic crustal layer over an irregular heterogeneous half-space, Journal of King Saud University – Science 30 (2018) 301–310.
  23. [23]R. Kaur, S. K. Vishwakarma, T. R. Panigrahi, Influence of irregular geologies and inhomogeneity on SH-wave propagation, Acta Mechanica 231 (2020) 1821–1836.
  24. [24]R. Kaur, A. Arora, Love wave along the corrugated interface between an elastic half-space and the corrugated porous layer containing two fluids, Journal of Vibration and Control 32 (2026) 615–629.
  25. [25]Y. Song, X. Li, Z. Yang, E. Carrera, Y. Yang, A series solution for SH wave scattering by irregularly shaped surface topographies, Journal of Earthquake Engineering 27 (2023) 2145–2165.DOI
  26. [26]V. Tavaf, S. Banerjee, Generalized analytical dispersion equations for guided Rayleigh–Lamb (RL) waves and shear horizontal (SH) waves in corrugated waveguides, International Journal of Solids and Structures 202 (2020) 75–86.DOI
  27. [27]S. Ahmetolan, A. Peker-Dobie, E. Deliktas-Ozdemir, E. Caglayan, Propagation of Lamb waves in an elastic layer with irregular surfaces, Wave Motion 119 (2023) 103136.DOI
  28. [28]T. Sezer, S. Ahmetolan, A. Peker-Dobie, A. Demirci, Effects of surface roughness on generalised Rayleigh waves in elastic waveguides, Wave Motion 141 (2026) 103658.DOI
  29. [29]E. Deliktas-Ozdemir, M. Teymür, Nonlinear surface SH waves in a half-space covered by an irregular layer, Zeitschrift für angewandte Mathematik und Physik 73 (2022) 145.
  30. [30]Z. Hu, Z. An, Y. Kong, G. Lian, X. Wang, The nonlinear S0 Lamb mode in a plate with a linearly-varying thickness, Ultrasonics 94 (2019) 102–108.DOI
  31. [31]H.-H. Dai, Y. Huo, Solitary waves in an inhomogeneous rod composed of a general hyperelastic material, Wave Motion 35 (2002) 55–69.DOI
  32. [32]G. A. El, R. H. J. Grimshaw, Generation of undular bores in the shelves of slowly-varying solitary waves, Chaos 12 (2001) 1015–1026.pubmed.ncbi.nlm.nih.gov/12779625
  33. [33]R. H. J. Grimshaw, S. R. Pudjaprasetya, Generation of secondary solitary waves in the variable-coefficient Korteweg–de Vries equation, Studies in Applied Mathematics 112 (2004) 271–279.DOI
  34. [34]R. Grimshaw, C. Yuan, Crest depression and elevation tsunami waves in the framework of the Korteweg–de Vries equation, Natural Hazards 84 (2016) 493–511.DOI
  35. [35]C. Zhi, H. Wang, K. Chen, Y. You, Theoretical and experimental investigation on strongly nonlinear internal solitary waves moving over slope-shelf topography, Ocean Engineering 223 (2021) 108645.DOI
  36. [36]B. Tian, G.-M. Wei, C.-Y. Zhang, W.-R. Shan, Y.-T. Gao, Transformations for a generalized variable-coefficient Korteweg–de Vries model from blood vessels, Bose–Einstein condensates, rods and positons with symbolic computation, Physics Letters A 356 (2006) 8–16.DOI
  37. [37]C.-Y. Zhang, Y.-T. Gao, X.-H. Meng, J. Li, T. Xu, G.-M. Wei, H.-W. Zhu, Integrable properties of a variable-coefficient Korteweg–de Vries model from Bose–Einstein condensates and fluid dynamics, Journal of Physics A: Mathematical and General 39 (2006) 14353–14362.DOI
  38. [38]S.-F. Tian, H.-Q. Zhang, On the integrability of a generalized variable-coefficient forced Korteweg–de Vries equation in fluids, Studies in Applied Mathematics 132 (2014) 212–246.DOI
  39. [39]A. C. Eringen, E. S. Şuhubi, Elastodynamics, Volume II: Linear Theory, Academic Press, New York, 1975.
  40. [40]A. Jeffrey, Exact and asymptotic methods in nonlinear wave theory, in: A. Jeffrey, J. Engelbrecht (Eds.), Nonlinear Waves in Solids, volume 341 of CISM Courses and Lectures, Springer-Verlag, Wien, 1994, pp. 1–49. doi:10.1007/978–3–7091–2444–4.DOI
  41. [41]M. Remoissenet, Waves Called Solitons: Concepts and Experiments, Springer Science & Business Media, 2013.
  42. [42]M. J. Ablowitz, A. Demirci, Y.-P. Ma, Dispersive shock waves in the Kadomtsev–Petviashvili and two dimensional Benjamin–Ono equations, Physica D: Nonlinear Phenomena 333 (2016) 84–98.DOI
  43. [43]C. Truesdell, W. Noll, The Non-Linear Field Theories of Mechanics, second ed., Springer-Verlag, Berlin, Heidelberg, 1992. doi:10.1007/978–3–662–13183–1.
  44. [44]S. M. Cox, P. C. Matthews, Exponential time differencing for stiff systems, Journal of Computational Physics 176 (2002) 430–455.DOI
  45. [45]A.-K. Kassam, L. N. Trefethen, Fourth-order time-stepping for stiff PDEs, SIAM Journal on Scientific Computing 26 (2005) 1214–1233.DOI

Paper details

vlnc:W_cqzdzqdw
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents