We consider stratified steady water waves in a two dimensional channel. Our subject is branches of Stokes waves, bifurcating from laminar flows. We assume that the mass flux and the Bernoulli constant are fixed and consider the period of the wave as a parameter, which can change its value along the branch. A new class of density and Bernoulli functions is presented, for which laminar flows generate global bifurcation branches. The laminar flows are not necessary unidirectional and we show that the bifurcation branch can bifurcate from the laminar flow with arbitrary large period.
Introduction
We consider stratified steady water waves in a two-dimensional channel. We use the classical formulation of the problem based on the Euler equations. The surface tension is neglected and the water motion can be rotational. Here we construct global branches of water waves. We assume that the flux and the Bernoulli constant are fixed and consider the period as a bifurcation parameter. We describe global branches of solutions bifurcating from a laminar flow. We present a class of vorticity and density functions, where you can pick up a laminar flow whose dispersion equation has an arbitrary, a-priori fixed frequency as a solution. This class contains laminar flows with counter-currents.
Statement of the main result
Our object of study is two-dimensional stratified steady water waves traveling with constant speed c under the influence of gravity. To eliminate the dependence on time we use a moving reference frame, where the fluid occupies a domain
D=Dξ={(x,y):−d<y<ξ(x),x∈R}
in the channel with the flat bottom B given by y=−d and with the free surface of the flow S=Sξ given by y=ξ(x). The density of the fluid ρ, defined in D, is assumed to be positive and not necessarily constant. To describe a water wave we use the velocity of the flow (u,v), the pressure P and the density ρ. If ρ is non-constant the fluid is referred to as stratified and it is stably stratified if ρ is non-decreasing with depth.
Corresponding relations describing the model can be found in the paper Walsh [15] (see also the survey [17], Sect.3). We recall these relations for readers convenience:
ux+vy=0in D(incompressibility),(1)
(u−c)ρx+vρy=0in D(the conservation of mass)(2)
and the conservation of momentum
(u−c)ux+vuy+ρPx=0in D,
(u−c)vx+vvy+ρPy=−gin D,(3)
where g is the gravitational constant. The boundary conditions are
v=(u−c)ξxandP=Patmon S,(4)
where Patm is the atmospheric pressure, and
v=0 on B.(5)
The pseudostream function ψ=ψ(x,y) is defined by
ψx(x,y)=−ρu(x,y),ψy(x,y)=ρ(u(x,y)−c).
Then equations (1.1) and (1.2) are satisfied if ρ is a function of ψ. In what follows we assume that ρ=ρ(−ψ) and ρ(p) is a positive function.
The relative pseudomass flux is defined by
p0=∫−dξ(x)ρ(u(x,y)−c)dY
and it does not depend on x. We will assume that p0<0.
The energy in the system
E:=21(ψx2+ψy2)+P+gρy(6)
is constant along the stream lines of ψ. This allows to define the Bernoulli function
β(ψ)=dψdE.(7)
It can be verified that
β(ψ)=Δψ−gyρp(−ψ)=0 in D.(8)
Boundary conditions for ψ are
ψ(x,ξ(x))=0 and ψ(x,−d)=−p0(9)
together with the Bernoulli boundary condition
21∣∇ψ∣2+gρ(0)(ξ(x)+d)=R for y=ξ(x),(10)
which is obtained from (1.6) by setting y=ξ(x) and using that ψ(x,ξ(x))=0. The constant R is called the Bernoulli constant.
We set
ω(y,ψ)=−gyρ′(−ψ)−β(ψ),(11)
where ρ′(−ψ)=−∂ψρ(−ψ). Then equation (1.8) takes the form
Δψ+ω(y,ψ)=0 in D.(12)
In the formulation (1.8)–(1.10), we assume that the constants p0 and R are given and the functions ψ and ξ are unknowns.
Our main assumption in our study concerns laminar solutions, i.e. solutions of (1.12) depending only on y. A laminar solution Ψ(y) satisfies
Ψ′′(y)+ω(y,Ψ(y))=0 on (−d,0)(13)
together with the boundary conditions
Ψ(0)=0 and Ψ(−d)=−p0,(14)
21∣Ψ′(0)∣2+gρ(0)d=R.(15)
Here Ψ and d>0 are considered as unknowns. We assume that ρ and β are C2,α and C1,α functions respectively, for a certain α∈(0,1), satisfying
has only positive eigenvalues μ for each d>0, where Ψ is a solution of the equation (1.13) with Ψ(0)=0. Here
ωp((y,Ψ)=−β′(Ψ)+yρ′′(−Ψ)(18)
i.e. it is the derivative of ω(y,p) with respect to p.
We consider the following branch of the laminar solutions (Ψ(y;s),d(s)), which solve the problem
Ψ′′(y)+ω(y,Ψ(y))=0on (−∞,0],
Ψ(0)=0,Ψ′(0)=s,(19)
together with
Ψ(d(s))=−p0.(20)
Due to (1.16) the problem (1.19) is uniquely solvable and we denote its solution by Ψ(y;s). We take the maximal interval (−∞,s∗) such that equation (1.20) holds and Ψy(d(s);s)<0 for s<s∗. Clearly the pair (Ψ(y;s),d(s)) solves the problem (1.13)–(1.15) with
R=R(s):=21Ψy(0;s)2+gρ(0)d(s).(21)
The dispersion equation is expressed through the solvability of the spectral problem (see Sect. 3.3)
(∂y2−τ2)u+ωp(y,Ψ)u=0in (−d,0),
Ψyuy−κu=0for y=0,
u(−d)=0,(22)
where
κ=Ψy(0)Ψy(0)Ψyy(0)+gρ(0).(23)
We are looking for a nontrivial solution u and a positive number τ. So we may assume that u(0)=1. Having this in mind, we introduce the function γ(y;τ), which is the solution of
(∂y2−τ2)γ+ωp(y,Ψ)γ=0in (−d,0),
γ(0)=1,
γ(−d)=0.(24)
From the boundary condition for y=0 in (1.22), we obtain the following scalar equation for τ:
σ(τ;s):=(Ψyγy−κ)∣y=0=0,(25)
which is called the dispersion equation. The corresponding eigenfunction is γ(y;τ). The function σ is strongly increasing with respect to τ and tends to infinity as τ→∞. Therefore existence of a positive root to (1.25) is equivalent to σ(0;s)<0. In Sect. 2.3 we prove that
Proposition 1.1.The dispersion equation(25)has a positive solution for a certain s<0 in the cases
(i) The density ρ is stably stratified, i.e. ρ′(p)≤0;
(ii) s∗≥0ors∗is a small negative number.
(iii) gρ0+s∗Vy(0;s∗)>0.
In order to formulate the main theorem, let us introduce some functional spaces. For k=0,1,…, α∈(0,1) and −∞<a<b<∞, we denote by Ck,α(Da,b) and Ck,α([a,b]) the Hölder spaces of functions in Da,b={(x,y)∈D:a≤x≤b} and [a,b] respectively.
The space Cbk,α(D) consists of functions u defined on D such that
a∈Rsup∥u∥Ck,α(Da,a+1)<∞.
The subspace CΛk,α(D) in Cbk,α(D) consists of Λ-periodic functions and the subspace C0,e,Λk,α(D) consists of Λ-periodic, even functions vanishing for y=ξ(x).
Similarly, one can define the spaces Cbk,α(R), CΛk,α(R) and Ce,Λk,α(R).
In the just introduced functional spaces we shall use the norms
∥u∥CΛk,α(D)=∥u∥CΛk,α(D−Λ/2,Λ/2).
Similarly, we define the norms in other spaces of Λ-periodic functions.
We always assume that ψy=0 on the free surface Sξ.
Theorem 1.2.Let the laminar flow(Ψ,d)is chosen such that,s<0and the dispersion equation(25)has a solutionτ∗>0with the corresponding eigenfunctionγ(y;τ∗), and letRbe given by(21). Then there existε>0and a mapping
In what follows we shall use the notations τ∗ and Λ∗ introduced in the theorem and the corresponding laminar solution we denote by (Ψ∗(y),d∗). The parameter s, which is included in the definition of the laminar solution, is fixed and we will not indicate the dependence of the function on s, if it is not needed.
Remark 1.3. 1). If the functions ρ and β are real analytic then the functions Λ(t) and ξ[t] are also real analytic. This can be prove by complementing the Crandall–Rabinowitz theorem by results from Chapter I.16, [9].
2). Smoothness properties of the function ψ[t] will be clarified latter. It can be obtain also from the analysis of the boundary value problem (1.13)–(1.15) if we know smoothness properties of the functions Λ(t) and ξ[t].
In order to formulate a global version of the above theorem, we introduce the set
for δ>0, such that ψ and ξ are Λ–periodic and even functions satisfying
δ≤Λ≤δ−1,(35)
xmaxξ(x)≤R−δon Sξ,(36)
(x,y)max∣ψ(x,y)∣≤δ−1,(37)
xmax∣ξ′(x)∣≤δ−1,(38)
ψy≤−δon Sξ(39)
and
xmin(ξ(x)+d)≥δ.(40)
Let ε be another positive number. We will consider ψ satisfying ψ(x,ξ(x))=0 and ψy(x,ξ(x))<0. This motivates the introduction of the following set.
Definition of Uδ,ε. Assume that (ψ,ξ,Λ)∈Uδ. Then for each x∈R there exists y∗∈[−d,ξ(x)), such that either −ψy(x,y)>δ on (y∗(x),ξ(x)) and this interval is the largest, i.e. −ψy(x,y∗(x))=δ (in this case y∗(x)<−d) or y∗(x)=−d. We define
ε∗(ψ)=xminψ(x,y∗(x)).(41)
Let also x∗ be a point, where the minimum is attained. Observe that this is a positive function for all ψ∈Uδ. The function ψ belongs to Uδ,ε if and only if ε∗(ψ)>ε. If y∗(x)=−d for all x, then ψ(x,y∗(x))=−p0 in the case when (ψ,ξ,Λ) satisfies the problem (1.8)–(1.10).
The important property, which follows from the definition, is the following
∫y∗(x)ξ(x)ψy(x,s)ds=−ψ(x,y∗(x)),(42)
and therefore
−ψ(x∗,y∗(x∗))>δ(ξ(x∗)−y∗(x∗)).(43)
One can check that
Uδ1,ε1⊂Uδ2,ε2if δ2≤δ1 and ε2≤ε1
and that
Uδ=ε⋃Uδ,ε.(44)
Let also
Uε=δ>0⋃Uδ,ε.
U=δ⋃Uδ(45)
and
O={(ψ,ξ,Λ)∈U:(ψ,ξ,Λ) satisfy the problem xref‘(1.13)‘−−xref‘(1.15)‘}.(46)
Theorem 1.4. We assume that all assumptions in Theorem 1.2 are fulfilled and the functions ρ and β are real analytic. Then there exists the branch
where (Ψ[t],ξ[t]) is a Λ(t)–periodic solution to (13)–(15) coinciding with the branch (29) for small t. The vector function
R∋t↦(λ(t),ξ[t](λ−1(t)x^))∈R+×C0,e,Λ∗2,α(R)(48)
is real analytic with respect to t (up to a local reparametrization). Furthermore, one of the following alternatives occurs:
(a) there exists ε∗>0 such all element of the branch belongs to Uε∗ and that for every δ>0 there exists tδ such that all elements (ψ[t],ξ[t],Λ(t)) do not belong to Uδ,ε∗ for ∣t∣>tδ;
(b) There exist sequences {tj} approaching ∞ and {δj} approaching zero as j→∞, such that (ψ[tj],ξ[tj],Λ(j))∈Uδj and
We note that from (50) it follows that the product δj and ξ[tj](x∗j)−y∗j(x∗j) tends to zero when j→∞.
Laminar flows and the dispersion equation
The laminar flows
In this section we will study laminar solutions, i.e. solutions to the problem (13)–(15). The pair (Ψ,d) is considered as an unknown. We assume that p0<0 and the functions ρ and β satisfy (16). Under this condition the Cauchy problem (19) has a unique solution for each real s, which is denoted by Ψ(y;s).
Lemma 2.1. There exists d1>0 such that the problem (13), (14) has a unique solution for each d≤d1. This solution satisfies
Therefore F is a contraction operator for small d and ∣F(y,w)∣≤cd2. Applying Banach fixed point theorem, we obtain existence of a unique solution w for small d satisfying (51).
Remark 2.2. Let us construct solutions to the problem (19) for large ∣s∣, s<0. We are looking for the solution in the form
In the previous proposition Assumption (A) is not used but it is essential in proving of the next proposition.
Proposition 2.3.Solutions to the problem (1.19) satisfy the following monotonicity property
Ψ(y,s2)>Ψ(y,s1)for y<0 if s2<s1.(56)
Proof. Since s2<s1 the inequality (56) is true for small y, Assume that Ψ(y,s2)=Ψ(y,s1) for a certain y0, which is the largest negative y with this property. Let s3=(s2+s1)/2. Then there exists y1>y0 such that Ψ(y1,s3)=Ψ(y1,s1) or Ψ(y1,s3)=Ψ(y1,s2) and this is the largest such y. We put s~1=s2, s^1=s3 or s~1=s3, s^1=s1 depending on the above choice of s3. Then we have Ψ(y1,s~1)=Ψ(y1,s^1) and s~1<s^1.
Continuing this procedure we can construct a sequences {s^j}, {s~j} and {yj} with the following properties
We denote by s† the limit point of the sequences {s^j} and {s~j}, which is the same due to (2.7) and (2.8). Let also y† the limit of the sequence {yj}.
If y†=0 then due to uniqueness in Lemma 2.1 there are no two different solution coinciding for a small ∣y∣. If y†>0 then application of the assumption (A) to the function Ψ(⋅,s†) on the interval (y†,0) implies that the Dirichlet problem for the Fréchet derivative at the function Ψ(⋅,s†) has positive first eigenvalue.
Since both Ψ(y;s^j) and Ψ(y;s~j) approach Ψ(y;s†) when j→∞ on the interval [M,0] for a fixed M, M<yj, we have that
∣Aj(y)−ωp(y,Ψ(y;s†))∣≤εjon [M,0],
where εj→o as j→∞. This implies that Ψ(y;s^j) must coincide with Ψ(y;s~j) for large j. This contradiction proves Lemma.
Consider the equation
Ψ(y;s)=−p0.(59)
The largest negative y satisfying the equation we denote by −d(s). According to Lemma 2.1 and Proposition 2.3 the equation is solvable for s in a neighborhood of −∞, and d(s)→0 and Ψ′(−d(s);s)=p0/d(s)+O(1) as s→−∞. The function s↦d(s) can be continued up to the first s∗ such that Ψ′(−d(s∗);s∗)=0 if s∗ is finite or s∗=+∞. Therefore, Ψ′(−d(s∗);s∗)<0 for s∈(−∞,s∗).
In the next proposition and in what follows we shall denote the derivative with respect to s by F˙(y;s)=∂sF(y;s).
Proposition 2.4. (i) Ψ˙(y;s)<0 for all s∈R and for all y<0;
(ii) If s∗ and d(s∗) are finite then d˙(s)→∞ as s→s∗.
Proof. (i) We have that Ψ˙(0;s)=0 and Ψys(0;s)=1 for all s<s∗. Assume that Ψs(y1;s)=0 for a certain y1<0. Then the function w(y)=Ψs(y;s) satisfies the problem (1.17) on the interval (y1,0) and by this assumption Ψs=0 on the interval (y1,0) and hence everywhere. This contradicts to Ψys(0;s)=1.
where H is a bounded function. We assume that the least eigenvalue of the Dirichlet problem for the operator −∂y2−H on the interval (y1,0) is positive.
Introduce functions h1(y) and h2(y) solving the problem with f=0 and with (α1,α2)=(0,1) and (α1,α2)=(1,0) respectively. According to the generalized maximum principle (see [4]),
Let a laminar flow (Ψ(y),d) solve the problem (1.13)–(1.15) on the interval (−d,0). We are looking for a solution to the problem (12), (9), (10) in the form
ψ(x,y)=Ψ(y)+εu(y)cos(τx)+O(ε2)
and
η(x)=εαcos(τx)+O(ε2),
where ε is a small number. Then α=−u(0)/Ψ′(0) and u together with τ solves the spectral problem (22). In order to obtain a scalar equation to express solvability of the spectral problem, we introduce the function γ=γ(y;τ) as the solution of the problem
Due to Assumption (A) this problem is uniquely solvable and by Sect. 2.2 γ>0 on (−d,0]. Let σ(τ;s)=Ψyγy−κ∣y=0, where κ is defined by (23). Then the spectral problem has a solution (τ,u) if and only if the dispersion equation (25) holds and the corresponding eigenfunction in (22) is u=γ(y;τ∗), where τ∗ denotes the solution of the dispersion equation.
The first simple property is that the function σ(τ)=σ(τ;s) is strongly increasing with respect with τ∈[0,∞). To prove this, we differentiate the problem (2.15) with respect to τ and obtain
(∂y2−τ2)h+ωp(y,Ψ)h=2γin (−d,0),
h(0)=0,
h(−d)=0,
where h(y)=h(y;τ)=τ−1γτ(y;τ). By Assumption (A) the operator −∂y2+τ2−ωp(y,Ψ) considered on (−d,0) with the Dirichlet boundary condition is positive definite for τ=0 and hence for all τ. Therefore we can use results of Sect. 2.2. The right hand side of the problem is positive for τ>0. Applying formula (2.13), we get h<0 for y∈(−d,0) and hence γyτ(0;τ)≥hy(0)τ>0. Therefore
στ(τ)=γyτ(0;τ)>0for τ>0andσ(τ)→∞as τ→∞.(66)
Due to this property in order to prove the existence of the solution to the dispersion equation (1.25) it is sufficient to investigate the sign of σ(0;s). In the next lemma we present a formula for σ(0;s).
Lemma 2.5. Let s=0 and s<s∗. Then (1.26) holds, where V is the solution of the problem (1.27) and
Vy(0)=∫−d0γ(z;0)ρp(−Ψ(z;s))dz≤0(67)
if the fluid is stably stratified, i.e. ρp≤0.
Proof. First let us show that
γ(y;0)=sd˙(s)−Ψ˙(y;s)+d˙(s)(Ψ′(y;s)−V(y)),(68)
where V is the solution of the problem (1.27).
Indeed the relations (1.27) guarantee that the function γ satisfies the first equation in (2.15). Using that −Ψ′(−d;s)d˙(s)+Ψ˙(−d;s)=0, one can check that γ satisfies the last relation in (2.15). The second relation in (2.15) is verified directly. This leads to the formula (2.18).
A straightforward application of this lemma is the following
Corollary 2.6. We have
σ(0;s)=sd˙(s)−s+O(1)<0for large negative s.
If the fluid is stably stratified then
σ(0;s)=sd˙(s)−d˙(s)(gρ0+sVy(0;s))+O(1)>0
when s∗≤0 and s→s∗. If s∗>0, then
σ(0;s)=−sgρ0+O(1)>0for small negative s.
Historical remarks
Laminar flows and dispersion equations for stratified steady Stokes waves were considered in many papers. In the paper [15], unidirectional laminar flows together with the dispersion equations were considered. The author obtained an estimate for the parameters of the problem which guarantees the existence of roots of the dispersion equation. Another type of laminar flows were suggested in the paper [16], where the vorticity and the density functions depend on the parameter d. Such laminar flows allows to include in the consideration the dispersion equation with small roots, which is important in study of solitary waves.
Here we present a condition on vorticity and the density functions, which guarantees the existence of laminar flows with dispersion equations having small roots. We note that the laminar flows are not necessary unidirectional in our case.
Bifurcation branches of water waves
In this section we use the study of the laminar solutions and the dispersion equation performed in the previous sections. The first step here is a reduction of the water wave problem with unknown boundary to a problem defined on a fixed domain.
Flattening the boundary
We assume that ψy=0 only on the boundary S and use the following flattening change of variables
x^=λx,y^=ξ(x)+dy+d(69)
where
λ=ΛΛ∗(70)
to reduce the problem to a fixed period Λ∗ and to a strip with fixed depth
Q={(x^,y^):x^∈R,0<y^<1}.
Since
∂x=λ(∂x^−η+dy^η′∂y^),∂y=η+d1∂y^,
where
η(x^)=ξ(λ−1x^)
and ′ means ∂x^, the equations (1.12) and (1.10) takes the form
F(ψ^,η;λ):=(λ2(∂x^−η+dy^η′∂y^)2+(η+d1∂y^)2)ψ^+ω^(y^,ψ^)=0in Q,
which is defined on Λ∗–periodic, even functions from C2,α(Q)×C2,α(R) satisfying ψ(x,0)=−p0, ψ(x,1)=0 and η+d>0. Here and in what follows we use the spaces Cbk,α(Q), CΛ∗k,α(Q), C0,e,Λ∗k,α(Q) which are defined similar to the spaces in D introduced before Theorem 1.2.
We remind that it is assumed that the density function ρ is of class C2,α for a certain α∈(0,1) and the Bernoulli function β is of class C1,α.
The introduced above operators are continuous in the following spaces
Theorem 3.1.We assume that the laminar flow (Ψ∗,d∗) is chosen such that the dispersion equation (1.25) has a solution τ∗>0 with the corresponding eigenfunction γ(y;τ∗). Then there exist ε>0 and a mapping
where (ψ[t],η[t],λ(t)) is an even, Λ∗–periodic solution to (3.3). The mapping is continuous together with the first derivatives with respect to t. Furthermore,
λ(0)=1,η[t]=−Ψ∗y(0)tcos(τ∗x)+o(∣t∣)and
ψ[t]=Ψ∗(y)+tγ(y;τ∗)cos(τ∗x)+o(∣t∣).(74)
Clearly, Theorem 1.2 follows from this assertion. Remark 1.3 is a consequence of the next
Remark 3.2. 1). If the functions ρ and β are real analytic then the functions λ(t), ψ[t] and η[t] are also real analytic. This can be prove by complementing the Crandall–Rabinowitz theorem by results from Chapter I.16, [9].
2). The smoothness property of the function ψ[t] mentioned in Remark 1.3 is equivalent to the smoothness property of the function ψ formulated in the above theorem.
Theorem 1.4 in the new variables has the following form.
Theorem 3.3.We assume that all assumptions from Theorem 3.1 are fulfilled. Moreover the functions ρ and β are real analytic. Then there exists the branch
where (ψ[t],η[t],λ(t)) is a even, Λ∗–periodic solution to (3.3) coinciding with the branch (3.5) for small t. The vector function (3.7) is real analytic with respect to t (up to a local reparameterization).
Furthermore, the second part of Theorem 1.4 concerning the alternatives (a) or (b) holds.
Laminar solutions
Laminar solutions of the problem (3.3) are solutions which depend only on y, when η=0 and ψ and d are considered as unknowns. Then they satisfy
d21∂y^2ψ^+ω^(y^,ψ^)=0in Q,(76)
2d21∣∂y^ψ^∣2+gρ(0)d−R=0for y^=d,
ψ^=0for y^=1,
ψ^=−p0for y^=0.
After the change of variable y=d(y^−1) this problem transforms to the problem (13)–(15).
If we are looking for solutions of the problem in the form
v(x,y)=γ(y;τ)cos(τx)andζ(x)=αcos(τx),(89)
then we obtain the following boundary condition for γ
Ψyγy−Ψyσ^=0for y=0.(90)
Here we have used also the second relation in (3.16). The relation (3.22) represents the dispersion equation for finding the frequency τ. The corresponding element of the kernel is given by (3.21), where α=−1/Ψy(0).
Small amplitude water waves. Proof of Theorem 3.1
Our proof is based on the classical Crandall–Rabinowitz Theorem. It deals with the following problem. Let X, Y be Banach spaces and I be a finite or infinite interval in R. Let also U be an open subset in X×I and
F:U→Y(91)
be a continuous function. We are interested in solutions of the equation
F(Φ,λ)=0,(Φ,λ)∈U(92)
such that the derivatives
∂ΦF,∂ΦΦF,∂λFand∂λΦF(93)
are also continuous.
Theorem 3.5. (Crandall–Rabinowitz, [3]) Let the derivatives of the operator F:
∂ΦF,∂ΦΦF,∂λFand∂λΦF(94)
are continuous. Suppose that
(i) F(0,λ)=0 holds for all (0,λ)∈U,
(ii) For some λ∗∈I such that (0,λ∗)∈U the operator ∂ΦF(0,λ∗) is a Fredholm operator with zero index and the null-space of FΦ(0,λ∗) is one-dimensional and generated by Φ0 and
FΦ,λ(0,λ∗)Φ0does not belong to the range of FΦ(0,λ∗).(95)
If (i) and (ii) hold, then a sufficiently small ε>0 exists and a C1 curve of solutions to (92):
{(Φ(t),λ(t)):∣t∣<ε}⊂X×I,
which bifurcates from (0,λ∗). Moreover, for pairs belonging to this curve the following properties hold:
Φ(t)=tΦ0+o(t)when 0<∣t∣<ε,
and λ(0)=λ∗,
{(λ,Φ)∈V:λ=λ∗ and F(λ,Φ)=0}={(λ(t),Φ(t)):0<∣t∣<ε},
where V⊂X×I is a certain neighbourhood of (0,λ∗).
For application of this theorem we choose d=d∗, λ∗=1, I=R+ and
Let U be the set of functions Φ=(w,ζ) in X, given by (1.45).
We introduce
F(Φ,λ)=(F(Ψ∗+w,ζ,λ),G(Ψ∗+w,ζ,λ)).
It is clear that the operator
F(Φ,λ):U→Y(96)
is continuous together with derivatives (93) and F(0,λ)=0.
It remains to verify the condition (ii) of the above theorem.
According to Sect. 3.3 the Fréchet derivative of the operator (3.4) at a laminar solution (Ψ∗,d∗) in the variables (x,y) is given by Lemma 3.13. Therefore the condition for existence of a positive τ is given by equation (3.22), which have a solution for a certain parameter s, when for example ρ′(−Ψ)≤0. This solution is unique and it was denoted by τ∗ in Theorem 1.2. The corresponding eigenfunction is u∗(y)=γ(y;τ∗)cos(τ∗x).
Simplicity of the eigenvalue follows from the simplicity of the eigenvalue of the problem (1.22) and monotonicity of the function σ(τ). The Fréchet derivative is a Fredholm operator since it is a self-adjoint operator in (x,y) variables.
In x^,y variables the Fréchet derivative has the form
and to prove (3.27), we must show that −2τ∗2∫−d∗0u∗u∗ydy=0, which is certainly true. Thus, all assumptions in the Crandall–Rabinowitz Theorem are verified and its application leads to the proof of Theorem 3.1.
Water waves of large amplitude
Here we use the same notations as in Theorem 3.5. Introduce
S={(Φ,λ)∈U:F(Φ,λ)=0}.(98)
Let
Bε={(Φ(t),λ(t)):t∈(−ε,ε)}(99)
be the bifurcating branch of small amplitude Stokes waves constructed in Theorem 3.5. We use the following version of the global bifurcation theorem taken from [6], which is a corrected and modified version of the bifurcation theorem from [5].
Theorem 3.6. Suppose that all conditions of Theorem 3.5 hold and for some sequence Kδ, δ∈(0,δ∗), of bounded closed subsets of U with U=⋃δ∈(0,δ∗)Kδ, the set S∩Kδ is compact for each δ∈(0,δ∗) and the Fréchet derivatives of the operator F is a Fredholm operator of index zero at all point from S. Then there exists a continuous curve B, which extends Bε as follows
(a) B={(Φ(t),λ(t)):t∈R}⊂U, where (Φ,λ):R→X×R+ is continuous;
(b) Bε⊂B⊂S;
(c) B has a real-analytic reparametrization locally around each of its points;
(d) One of the following alternatives occurs:
(α) for every δ∈(0,δ∗), there exists tδ>0 such that (Φ(t),λ(t)) does not belong to Kδ for all t∈R with ∣t∣>tδ;
(β) there exists T>0 such that (Φ(t+T),λ(t+T))=(Φ(t),λ(t)) for all t>0.
Moreover, such a curve of solutions to F(Φ,λ)=0 having the properties (a)-(d) is unique (up to reparametrization).
In order to apply this theorem, we choose
Kδ=Uδ,(100)
where Uδ is defined by (1.34). Then the formula
U=δ⋃Uδ(101)
holds because of (1.45) and monotonicity of Uδ with respect to δ. We note also that S coincides with O given by (1.46).
Let
M=(x,y)∈Dmaxψ(x,y),(102)
m=x∈Rmin(ξ(x)+d)(103)
and
L=xmax∣ξ′(x)∣.(104)
It is clear that M≥∣p0∣.
The following lemma will be important in proving of compactness property, which is one of the assumptions in the above theorem.
Lemma 3.7. There exists C∗=C∗(C1,L,M,m) depending on the constant C1 from the inequalities (16) and the constants L, M and m, such that if a Λ–periodic function (ψ,ξ) solves the problem (13)–(15), then the inequality
ψx2(x,y)+ψy2(x,y)≤C∗(105)
holds for all (x,y)∈D.
The proof of this assertion repeats the proof of Proposition 2 in [11]. The only difference is that we can not estimate m from below by a constant depending only on R, C1, L and M. We overcome this by including the dependence of C∗ on m in the final estimate.
Proof. Denote
Ωs={(x,y)∈Dξ:s<x<s+1}.
Repeating the proof of (4.12) in Proposition 2 in [11], we get
∫Ωs∣∇ψ∣2dxdy≤C(R,C1,M,L),
where C does not depend on s.
Differentiating the equation (12) with respect to x, we obtain
Δψx+ωp(y,ψ)ψx=0in D.
Applying Theorem 8.25 in [8], we arrive at the estimate (105) for the function ψx.
Differentiating the equation (12) with respect to y, we obtain
Δψy+ωp(y,ψ)ψy=gρ′(ψ)in D.
The value of ψ on the bottom y=−d is estimated already by a constant depending on C1, M, m and p0. The pointwise estimate of ψy on the bottom can be obtained by applying standard local estimates near the bottom for the Laplace operator. It is estimated by a constant depending on C1, M, m and p0. The value of ψy on the free surface y=ξ is estimated by using the boundary condition (10). Now application of Theorem 8.25 in [8] leads to the estimate (105) for the function ψy.
Proof of Theorem 3.6. In order to satisfy the assumptions of the theorem, we will check that
(i) the Frechet derivative at each point (w,η;λ) in O is a Fredholm operator of index zero;
(ii) Let
Kδ,ε=Uδ,ε,
where Uδ,ε is defined by (1). The set Kδ,ε∩O is compact in X;
(iii) the alternative (β) in (d) does not hold.
Let us prove (i). We shall use the following lemma, proved in [13], where we applied a slightly different change of variables. Let
x^=x,y^=ξ(x)+dy+d.
and let us introduce the transformation
v(x,y)=u(x^,y^)−ψy(x,y)ξ+d(y+d)ζ.(106)
Lemma 3.8. (i) Assume that the functions ψ and ξ satisfy (1.12) in the domain Dξ. If the function v is given by (3.38) then
(∂x2+∂y2)v+ω∗v=F(u,ζ),
where ω∗ is defined by (3.11).
(ii) Furthermore
ψxvx+ψyvy+σ^ζ=G(u,ζ)on Sξ,
where
σ^=ψxψxy+ψyψyy+gρ(0),v=−ψyζ.
By this lemma, the Fredholm property of the operator F follows from similar property for the operator
Av:=(∂x2+∂y2)v+gyρ′′(−ψ)v−β′(ψ)vin D,
Bv:=ψxvx+ψyvy−ψyσ^vfor y=ξ(x).(107)
defined on functions subject to
v(x,−d)=0.(108)
Since this operator is self-adjoint, we obtain the required property.
(ii) In order to prove the compactness property, we need some lemmas. Let ψ∈Uδ,ε. We apply the transformation
q=x,p=−ψ(x,y),(x,y)∈{(x,y):x∈R,y∗(x)≤y≤ξ(x)},
where y∗(x) was introduced in the definition 1.1. Then
In the next lemma we use the notation Bρ(q,p) for the disc with the center at (q,p) and with the radius ρ.
Lemma 3.9. Let h∈C2,α(Q) be a solution to (3.41) and δ and ε satisfy (1.1). Suppose that the corresponding (ψ,η) satisfies (1.13)–(1.15) and the constants M, m and L are defined by (3.34), (3.35) and (3.36). Let p∈[−ε/2,0] and ρ≤ε/2. Then there exists constants C∗ and α1∈(0,1) depending also on C1, L, M, m, δ and ρ such that
∥h∥C3,α1(Q∩Bρ/2((q,p)))≤C∗.
The proof of this assertion is the same as the proof of Proposition 3 in [11], where the results from [8], [14] and [1] are used. The important role plays by the inequality (3.37) in Lemma 3.7 and its analog in (q,p) variables:
hq2+hp2≤C∗δ−2,hp2≥C∗−1,(110)
where we have used also that ∣ψy∣≥δ. The main step here is the local estimates presented and proved in Sect. 2, Chapter 10, [14], and due to (3.42) the conditions required for the local estimate are satisfied.
Lemma 3.10. For each δ>0 and ε>0 the set
Kδ,ε∩O(111)
is compact in X.
Proof. First we verify the following property. Let (ψ,ξ,Λ) satisfy (1.8)–(1.10). Then there exists constants C∗ and α1∈(0,1) depending on C1, L, M, m, δ and ε such that
∥ψ∥C3,α1(Dξ)≤C∗,∥ξ∥C3,α1(R)≤C∗.(112)
The proof of this inequality (proved in Proposition 3.2 [12]) follows from the lemma 3.9 and the local estimates from [1]. Now the required compactness property in the space X follows from (3.44) and compactness of inclusion of the Hölder spaces:
Kδ,ε∩O∈C3,α1(Dξ)×C3,α1(R),
which leads to compactness of (3.43). □
To derive assertions (a) and (b) from Theorem 1.4 and similar assertions from Theorem 3.3 we proceed as follows.
(a) Assume that there exists ε∗>0 such all element of the branch belongs to Uε∗. Then we can apply Theorem 3.6 for Kδ=Uδ,ε∗ and U=Uε∗, which leads to the proof of (a).
(b) Assume that (a) is not true. Then for every ε there exists δ and an element (ψ[t],ξ[t],Λ(t))∈Uδ on the branch such that
ε∗(ψ[t])<ε.(113)
One can choose εj satisfying εj→0 as j→∞ and (ψ[tj],ξ[tj],Λ(tj)) in such a way that tj→∞ as j→∞ and the inequality (3.45) still holds, when you include there the dependence on j. This proves the item (b) in Theorem 1.4.
(iii) Let (Ψ∗(y),d∗) be the laminar solution from Theorem 3.1 and let τ∗ be the positive root of the dispersion equation (1.25). Then the solution γ(y;τ∗) of the equation (1.24) is positive on the interval (−d,0]. By (1.33) the derivative ψx(x,y) is positive in (0,Λ∗)×[0,1). In the paper [10] the density is a constant, but the basis of considerations there is Lemma 2.1 and the maximum principles, which can be found in Theorem 2.1, [2], is still true. Therefore, all result there can be applied to this case also.
Acknowledgments
This work was supported by the Ministry of Science and Higher Education of the Russian Federation (agreement 075-15-2025-344 dated 29/04/2025 for Saint Petersburg Leonard Euler International Mathematical Institute at PDMI RAS).
References
References
[1]Agmon, S., Douglis, A. and Nirenberg, L., Estimates near the boundary for solutions of elliptic partial differential equations satisfying general boundary conditions. I. Commun. Pure Appl. Maths 12, 623–727, 1959.DOI
[2]DE Apushkinskaya, AI Nazarov, The normal derivative lemma and surrounding issues, Russian Mathematical Surveys, 2022.DOI
[3]Crandall, M.G., Rabinowitz, P.H.: Bifurcation from simple eigenvalues. J. Funct. Anal. 8, 321–340 (1971).DOI
[4]MH Protter, HF Weinberger, Maximum principles in differential equations. Prentice-Hall, 1967.
[5]B. Buffoni, J. Toland, Analytic theory of global bifurcation: an introduction, Princeton University Press, 2003.
[6]A Constantin, W Strauss, E Varvaruca, Global bifurcation of steady gravity water waves with critical layers, Acta Mathematica 217 (2), 195-262, 2016.arxiv.org/abs/1407.0092
[8]Gilbarg, D. and Trudinger, N. S. Elliptic Partial Differential Equations of Second Order, Springer, 2001.
[9]H. Kielhöfer, Bifurcation theory: An introduction with applications to PDEs, Springer New York Dordrecht Heidelberg London, 2011.
[10]V Kozlov, On loops in water wave branches and monotonicity of water waves, Journal of Mathematical Fluid Mechanics, 25 (1), 9, 2023.DOI
[11]V. Kozlov, N. Kuznetsov, E. Lokharu, On the Benjamin–Lighthill conjecture for water waves with vorticity, Journal of Fluid Mechanics 825, 961-1001, 2017.arxiv.org/abs/1511.02818
[12]V. Kozlov, E. Lokharu, Global bifurcation and highest waves on water of finite depth, ARMA, 2023. arXiv preprint arXiv:2010.14156, 2020.arxiv.org/abs/2010.14156
[13]V. Kozlov, Spectral properties of the Fréchet derivatives of stratified steady Stokes waves, arXiv:2512.03831 [math.AP], 2025.arxiv.org/abs/2512.03831
[14]Ladyzhenskaya, O. A. and Uraltseva, N. N. 1964 Linear and Quasilinear Elliptic Equations, Nauka (in Russian); English transl., Academic, 1968.
[15]S Walsh, Stratified steady periodic water waves, SIAM journal on mathematical analysis, 2009.arxiv.org/abs/0807.0474
[16]RM Chen, S Walsh, MH Wheeler, Existence and qualitative theory for stratified solitary water waves, Annales de l’Institut Henri Poincaré C 35 (2), 517-576, 2018.DOI
[17]S Haziot, V Hur, W Strauss, J Toland, E Wahlen, S Walsh, M Wheeler, Traveling water waves—the ebb and flow of two centuries, Quarterly of applied mathematics 80 (2), 317-401, 2022.arxiv.org/abs/2109.09208